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 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 110 // getrusage() is prepared to handle the associated failure.
 111 #ifndef RUSAGE_THREAD
 112   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 113 #endif
 114 
 115 #define MAX_PATH    (2 * K)
 116 
 117 #define MAX_SECS 100000000
 118 
 119 // for timer info max values which include all bits
 120 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 121 
 122 #define LARGEPAGES_BIT (1 << 6)
 123 ////////////////////////////////////////////////////////////////////////////////
 124 // global variables
 125 julong os::Linux::_physical_memory = 0;
 126 
 127 address   os::Linux::_initial_thread_stack_bottom = NULL;
 128 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 129 
 130 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 131 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 132 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 133 Mutex* os::Linux::_createThread_lock = NULL;
 134 pthread_t os::Linux::_main_thread;
 135 int os::Linux::_page_size = -1;
 136 const int os::Linux::_vm_default_page_size = (8 * K);
 137 bool os::Linux::_supports_fast_thread_cpu_time = false;
 138 const char * os::Linux::_glibc_version = NULL;
 139 const char * os::Linux::_libpthread_version = NULL;
 140 pthread_condattr_t os::Linux::_condattr[1];
 141 
 142 static jlong initial_time_count=0;
 143 
 144 static int clock_tics_per_sec = 100;
 145 
 146 // For diagnostics to print a message once. see run_periodic_checks
 147 static sigset_t check_signal_done;
 148 static bool check_signals = true;
 149 
 150 // Signal number used to suspend/resume a thread
 151 
 152 // do not use any signal number less than SIGSEGV, see 4355769
 153 static int SR_signum = SIGUSR2;
 154 sigset_t SR_sigset;
 155 
 156 // Declarations
 157 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 158 
 159 // utility functions
 160 
 161 static int SR_initialize();
 162 
 163 julong os::available_memory() {
 164   return Linux::available_memory();
 165 }
 166 
 167 julong os::Linux::available_memory() {
 168   // values in struct sysinfo are "unsigned long"
 169   struct sysinfo si;
 170   sysinfo(&si);
 171 
 172   return (julong)si.freeram * si.mem_unit;
 173 }
 174 
 175 julong os::physical_memory() {
 176   return Linux::physical_memory();
 177 }
 178 
 179 // Return true if user is running as root.
 180 
 181 bool os::have_special_privileges() {
 182   static bool init = false;
 183   static bool privileges = false;
 184   if (!init) {
 185     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 186     init = true;
 187   }
 188   return privileges;
 189 }
 190 
 191 
 192 #ifndef SYS_gettid
 193 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 194   #ifdef __ia64__
 195     #define SYS_gettid 1105
 196   #else
 197     #ifdef __i386__
 198       #define SYS_gettid 224
 199     #else
 200       #ifdef __amd64__
 201         #define SYS_gettid 186
 202       #else
 203         #ifdef __sparc__
 204           #define SYS_gettid 143
 205         #else
 206           #error define gettid for the arch
 207         #endif
 208       #endif
 209     #endif
 210   #endif
 211 #endif
 212 
 213 // Cpu architecture string
 214 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
 215 
 216 
 217 // pid_t gettid()
 218 //
 219 // Returns the kernel thread id of the currently running thread. Kernel
 220 // thread id is used to access /proc.
 221 pid_t os::Linux::gettid() {
 222   int rslt = syscall(SYS_gettid);
 223   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 224   return (pid_t)rslt;
 225 }
 226 
 227 // Most versions of linux have a bug where the number of processors are
 228 // determined by looking at the /proc file system.  In a chroot environment,
 229 // the system call returns 1.  This causes the VM to act as if it is
 230 // a single processor and elide locking (see is_MP() call).
 231 static bool unsafe_chroot_detected = false;
 232 static const char *unstable_chroot_error = "/proc file system not found.\n"
 233                      "Java may be unstable running multithreaded in a chroot "
 234                      "environment on Linux when /proc filesystem is not mounted.";
 235 
 236 void os::Linux::initialize_system_info() {
 237   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 238   if (processor_count() == 1) {
 239     pid_t pid = os::Linux::gettid();
 240     char fname[32];
 241     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 242     FILE *fp = fopen(fname, "r");
 243     if (fp == NULL) {
 244       unsafe_chroot_detected = true;
 245     } else {
 246       fclose(fp);
 247     }
 248   }
 249   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 250   assert(processor_count() > 0, "linux error");
 251 }
 252 
 253 void os::init_system_properties_values() {
 254   // The next steps are taken in the product version:
 255   //
 256   // Obtain the JAVA_HOME value from the location of libjvm.so.
 257   // This library should be located at:
 258   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 259   //
 260   // If "/jre/lib/" appears at the right place in the path, then we
 261   // assume libjvm.so is installed in a JDK and we use this path.
 262   //
 263   // Otherwise exit with message: "Could not create the Java virtual machine."
 264   //
 265   // The following extra steps are taken in the debugging version:
 266   //
 267   // If "/jre/lib/" does NOT appear at the right place in the path
 268   // instead of exit check for $JAVA_HOME environment variable.
 269   //
 270   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 271   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 272   // it looks like libjvm.so is installed there
 273   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 274   //
 275   // Otherwise exit.
 276   //
 277   // Important note: if the location of libjvm.so changes this
 278   // code needs to be changed accordingly.
 279 
 280   // See ld(1):
 281   //      The linker uses the following search paths to locate required
 282   //      shared libraries:
 283   //        1: ...
 284   //        ...
 285   //        7: The default directories, normally /lib and /usr/lib.
 286 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 287   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 288 #else
 289   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 290 #endif
 291 
 292 // Base path of extensions installed on the system.
 293 #define SYS_EXT_DIR     "/usr/java/packages"
 294 #define EXTENSIONS_DIR  "/lib/ext"
 295 
 296   // Buffer that fits several sprintfs.
 297   // Note that the space for the colon and the trailing null are provided
 298   // by the nulls included by the sizeof operator.
 299   const size_t bufsize =
 300     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 301          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 302   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 303 
 304   // sysclasspath, java_home, dll_dir
 305   {
 306     char *pslash;
 307     os::jvm_path(buf, bufsize);
 308 
 309     // Found the full path to libjvm.so.
 310     // Now cut the path to <java_home>/jre if we can.
 311     pslash = strrchr(buf, '/');
 312     if (pslash != NULL) {
 313       *pslash = '\0';            // Get rid of /libjvm.so.
 314     }
 315     pslash = strrchr(buf, '/');
 316     if (pslash != NULL) {
 317       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 318     }
 319     Arguments::set_dll_dir(buf);
 320 
 321     if (pslash != NULL) {
 322       pslash = strrchr(buf, '/');
 323       if (pslash != NULL) {
 324         *pslash = '\0';          // Get rid of /<arch>.
 325         pslash = strrchr(buf, '/');
 326         if (pslash != NULL) {
 327           *pslash = '\0';        // Get rid of /lib.
 328         }
 329       }
 330     }
 331     Arguments::set_java_home(buf);
 332     set_boot_path('/', ':');
 333   }
 334 
 335   // Where to look for native libraries.
 336   //
 337   // Note: Due to a legacy implementation, most of the library path
 338   // is set in the launcher. This was to accomodate linking restrictions
 339   // on legacy Linux implementations (which are no longer supported).
 340   // Eventually, all the library path setting will be done here.
 341   //
 342   // However, to prevent the proliferation of improperly built native
 343   // libraries, the new path component /usr/java/packages is added here.
 344   // Eventually, all the library path setting will be done here.
 345   {
 346     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 347     // should always exist (until the legacy problem cited above is
 348     // addressed).
 349     const char *v = ::getenv("LD_LIBRARY_PATH");
 350     const char *v_colon = ":";
 351     if (v == NULL) { v = ""; v_colon = ""; }
 352     // That's +1 for the colon and +1 for the trailing '\0'.
 353     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 354                                                      strlen(v) + 1 +
 355                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 356                                                      mtInternal);
 357     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 358     Arguments::set_library_path(ld_library_path);
 359     FREE_C_HEAP_ARRAY(char, ld_library_path);
 360   }
 361 
 362   // Extensions directories.
 363   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 364   Arguments::set_ext_dirs(buf);
 365 
 366   FREE_C_HEAP_ARRAY(char, buf);
 367 
 368 #undef DEFAULT_LIBPATH
 369 #undef SYS_EXT_DIR
 370 #undef EXTENSIONS_DIR
 371 }
 372 
 373 ////////////////////////////////////////////////////////////////////////////////
 374 // breakpoint support
 375 
 376 void os::breakpoint() {
 377   BREAKPOINT;
 378 }
 379 
 380 extern "C" void breakpoint() {
 381   // use debugger to set breakpoint here
 382 }
 383 
 384 ////////////////////////////////////////////////////////////////////////////////
 385 // signal support
 386 
 387 debug_only(static bool signal_sets_initialized = false);
 388 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 389 
 390 bool os::Linux::is_sig_ignored(int sig) {
 391   struct sigaction oact;
 392   sigaction(sig, (struct sigaction*)NULL, &oact);
 393   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 394                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 395   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 396     return true;
 397   } else {
 398     return false;
 399   }
 400 }
 401 
 402 void os::Linux::signal_sets_init() {
 403   // Should also have an assertion stating we are still single-threaded.
 404   assert(!signal_sets_initialized, "Already initialized");
 405   // Fill in signals that are necessarily unblocked for all threads in
 406   // the VM. Currently, we unblock the following signals:
 407   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 408   //                         by -Xrs (=ReduceSignalUsage));
 409   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 410   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 411   // the dispositions or masks wrt these signals.
 412   // Programs embedding the VM that want to use the above signals for their
 413   // own purposes must, at this time, use the "-Xrs" option to prevent
 414   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 415   // (See bug 4345157, and other related bugs).
 416   // In reality, though, unblocking these signals is really a nop, since
 417   // these signals are not blocked by default.
 418   sigemptyset(&unblocked_sigs);
 419   sigemptyset(&allowdebug_blocked_sigs);
 420   sigaddset(&unblocked_sigs, SIGILL);
 421   sigaddset(&unblocked_sigs, SIGSEGV);
 422   sigaddset(&unblocked_sigs, SIGBUS);
 423   sigaddset(&unblocked_sigs, SIGFPE);
 424 #if defined(PPC64)
 425   sigaddset(&unblocked_sigs, SIGTRAP);
 426 #endif
 427   sigaddset(&unblocked_sigs, SR_signum);
 428 
 429   if (!ReduceSignalUsage) {
 430     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 431       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 432       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 433     }
 434     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 435       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 436       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 437     }
 438     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 439       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 440       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 441     }
 442   }
 443   // Fill in signals that are blocked by all but the VM thread.
 444   sigemptyset(&vm_sigs);
 445   if (!ReduceSignalUsage) {
 446     sigaddset(&vm_sigs, BREAK_SIGNAL);
 447   }
 448   debug_only(signal_sets_initialized = true);
 449 
 450 }
 451 
 452 // These are signals that are unblocked while a thread is running Java.
 453 // (For some reason, they get blocked by default.)
 454 sigset_t* os::Linux::unblocked_signals() {
 455   assert(signal_sets_initialized, "Not initialized");
 456   return &unblocked_sigs;
 457 }
 458 
 459 // These are the signals that are blocked while a (non-VM) thread is
 460 // running Java. Only the VM thread handles these signals.
 461 sigset_t* os::Linux::vm_signals() {
 462   assert(signal_sets_initialized, "Not initialized");
 463   return &vm_sigs;
 464 }
 465 
 466 // These are signals that are blocked during cond_wait to allow debugger in
 467 sigset_t* os::Linux::allowdebug_blocked_signals() {
 468   assert(signal_sets_initialized, "Not initialized");
 469   return &allowdebug_blocked_sigs;
 470 }
 471 
 472 void os::Linux::hotspot_sigmask(Thread* thread) {
 473 
 474   //Save caller's signal mask before setting VM signal mask
 475   sigset_t caller_sigmask;
 476   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 477 
 478   OSThread* osthread = thread->osthread();
 479   osthread->set_caller_sigmask(caller_sigmask);
 480 
 481   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 482 
 483   if (!ReduceSignalUsage) {
 484     if (thread->is_VM_thread()) {
 485       // Only the VM thread handles BREAK_SIGNAL ...
 486       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 487     } else {
 488       // ... all other threads block BREAK_SIGNAL
 489       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 490     }
 491   }
 492 }
 493 
 494 //////////////////////////////////////////////////////////////////////////////
 495 // detecting pthread library
 496 
 497 void os::Linux::libpthread_init() {
 498   // Save glibc and pthread version strings.
 499 #if !defined(_CS_GNU_LIBC_VERSION) || \
 500     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 501   #error "glibc too old (< 2.3.2)"
 502 #endif
 503 
 504   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 505   assert(n > 0, "cannot retrieve glibc version");
 506   char *str = (char *)malloc(n, mtInternal);
 507   confstr(_CS_GNU_LIBC_VERSION, str, n);
 508   os::Linux::set_glibc_version(str);
 509 
 510   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 511   assert(n > 0, "cannot retrieve pthread version");
 512   str = (char *)malloc(n, mtInternal);
 513   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 514   os::Linux::set_libpthread_version(str);
 515 }
 516 
 517 /////////////////////////////////////////////////////////////////////////////
 518 // thread stack expansion
 519 
 520 // os::Linux::manually_expand_stack() takes care of expanding the thread
 521 // stack. Note that this is normally not needed: pthread stacks allocate
 522 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 523 // committed. Therefore it is not necessary to expand the stack manually.
 524 //
 525 // Manually expanding the stack was historically needed on LinuxThreads
 526 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 527 // it is kept to deal with very rare corner cases:
 528 //
 529 // For one, user may run the VM on an own implementation of threads
 530 // whose stacks are - like the old LinuxThreads - implemented using
 531 // mmap(MAP_GROWSDOWN).
 532 //
 533 // Also, this coding may be needed if the VM is running on the primordial
 534 // thread. Normally we avoid running on the primordial thread; however,
 535 // user may still invoke the VM on the primordial thread.
 536 //
 537 // The following historical comment describes the details about running
 538 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 539 
 540 
 541 // Force Linux kernel to expand current thread stack. If "bottom" is close
 542 // to the stack guard, caller should block all signals.
 543 //
 544 // MAP_GROWSDOWN:
 545 //   A special mmap() flag that is used to implement thread stacks. It tells
 546 //   kernel that the memory region should extend downwards when needed. This
 547 //   allows early versions of LinuxThreads to only mmap the first few pages
 548 //   when creating a new thread. Linux kernel will automatically expand thread
 549 //   stack as needed (on page faults).
 550 //
 551 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 552 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 553 //   region, it's hard to tell if the fault is due to a legitimate stack
 554 //   access or because of reading/writing non-exist memory (e.g. buffer
 555 //   overrun). As a rule, if the fault happens below current stack pointer,
 556 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 557 //   application (see Linux kernel fault.c).
 558 //
 559 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 560 //   stack overflow detection.
 561 //
 562 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 563 //   not use MAP_GROWSDOWN.
 564 //
 565 // To get around the problem and allow stack banging on Linux, we need to
 566 // manually expand thread stack after receiving the SIGSEGV.
 567 //
 568 // There are two ways to expand thread stack to address "bottom", we used
 569 // both of them in JVM before 1.5:
 570 //   1. adjust stack pointer first so that it is below "bottom", and then
 571 //      touch "bottom"
 572 //   2. mmap() the page in question
 573 //
 574 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 575 // if current sp is already near the lower end of page 101, and we need to
 576 // call mmap() to map page 100, it is possible that part of the mmap() frame
 577 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 578 // That will destroy the mmap() frame and cause VM to crash.
 579 //
 580 // The following code works by adjusting sp first, then accessing the "bottom"
 581 // page to force a page fault. Linux kernel will then automatically expand the
 582 // stack mapping.
 583 //
 584 // _expand_stack_to() assumes its frame size is less than page size, which
 585 // should always be true if the function is not inlined.
 586 
 587 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 588   #define NOINLINE
 589 #else
 590   #define NOINLINE __attribute__ ((noinline))
 591 #endif
 592 
 593 static void _expand_stack_to(address bottom) NOINLINE;
 594 
 595 static void _expand_stack_to(address bottom) {
 596   address sp;
 597   size_t size;
 598   volatile char *p;
 599 
 600   // Adjust bottom to point to the largest address within the same page, it
 601   // gives us a one-page buffer if alloca() allocates slightly more memory.
 602   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 603   bottom += os::Linux::page_size() - 1;
 604 
 605   // sp might be slightly above current stack pointer; if that's the case, we
 606   // will alloca() a little more space than necessary, which is OK. Don't use
 607   // os::current_stack_pointer(), as its result can be slightly below current
 608   // stack pointer, causing us to not alloca enough to reach "bottom".
 609   sp = (address)&sp;
 610 
 611   if (sp > bottom) {
 612     size = sp - bottom;
 613     p = (volatile char *)alloca(size);
 614     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 615     p[0] = '\0';
 616   }
 617 }
 618 
 619 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 620   assert(t!=NULL, "just checking");
 621   assert(t->osthread()->expanding_stack(), "expand should be set");
 622   assert(t->stack_base() != NULL, "stack_base was not initialized");
 623 
 624   if (addr <  t->stack_base() && addr >= t->stack_yellow_zone_base()) {
 625     sigset_t mask_all, old_sigset;
 626     sigfillset(&mask_all);
 627     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 628     _expand_stack_to(addr);
 629     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 630     return true;
 631   }
 632   return false;
 633 }
 634 
 635 //////////////////////////////////////////////////////////////////////////////
 636 // create new thread
 637 
 638 // Thread start routine for all newly created threads
 639 static void *java_start(Thread *thread) {
 640   // Try to randomize the cache line index of hot stack frames.
 641   // This helps when threads of the same stack traces evict each other's
 642   // cache lines. The threads can be either from the same JVM instance, or
 643   // from different JVM instances. The benefit is especially true for
 644   // processors with hyperthreading technology.
 645   static int counter = 0;
 646   int pid = os::current_process_id();
 647   alloca(((pid ^ counter++) & 7) * 128);
 648 
 649   ThreadLocalStorage::set_thread(thread);
 650 
 651   OSThread* osthread = thread->osthread();
 652   Monitor* sync = osthread->startThread_lock();
 653 
 654   osthread->set_thread_id(os::current_thread_id());
 655 
 656   if (UseNUMA) {
 657     int lgrp_id = os::numa_get_group_id();
 658     if (lgrp_id != -1) {
 659       thread->set_lgrp_id(lgrp_id);
 660     }
 661   }
 662   // initialize signal mask for this thread
 663   os::Linux::hotspot_sigmask(thread);
 664 
 665   // initialize floating point control register
 666   os::Linux::init_thread_fpu_state();
 667 
 668   // handshaking with parent thread
 669   {
 670     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 671 
 672     // notify parent thread
 673     osthread->set_state(INITIALIZED);
 674     sync->notify_all();
 675 
 676     // wait until os::start_thread()
 677     while (osthread->get_state() == INITIALIZED) {
 678       sync->wait(Mutex::_no_safepoint_check_flag);
 679     }
 680   }
 681 
 682   // call one more level start routine
 683   thread->run();
 684 
 685   return 0;
 686 }
 687 
 688 bool os::create_thread(Thread* thread, ThreadType thr_type,
 689                        size_t stack_size) {
 690   assert(thread->osthread() == NULL, "caller responsible");
 691 
 692   // Allocate the OSThread object
 693   OSThread* osthread = new OSThread(NULL, NULL);
 694   if (osthread == NULL) {
 695     return false;
 696   }
 697 
 698   // set the correct thread state
 699   osthread->set_thread_type(thr_type);
 700 
 701   // Initial state is ALLOCATED but not INITIALIZED
 702   osthread->set_state(ALLOCATED);
 703 
 704   thread->set_osthread(osthread);
 705 
 706   // init thread attributes
 707   pthread_attr_t attr;
 708   pthread_attr_init(&attr);
 709   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 710 
 711   // stack size
 712   // calculate stack size if it's not specified by caller
 713   if (stack_size == 0) {
 714     stack_size = os::Linux::default_stack_size(thr_type);
 715 
 716     switch (thr_type) {
 717     case os::java_thread:
 718       // Java threads use ThreadStackSize which default value can be
 719       // changed with the flag -Xss
 720       assert(JavaThread::stack_size_at_create() > 0, "this should be set");
 721       stack_size = JavaThread::stack_size_at_create();
 722       break;
 723     case os::compiler_thread:
 724       if (CompilerThreadStackSize > 0) {
 725         stack_size = (size_t)(CompilerThreadStackSize * K);
 726         break;
 727       } // else fall through:
 728         // use VMThreadStackSize if CompilerThreadStackSize is not defined
 729     case os::vm_thread:
 730     case os::pgc_thread:
 731     case os::cgc_thread:
 732     case os::watcher_thread:
 733       if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 734       break;
 735     }
 736   }
 737 
 738   stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 739   pthread_attr_setstacksize(&attr, stack_size);
 740 
 741   // glibc guard page
 742   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 743 
 744   ThreadState state;
 745 
 746   {
 747     pthread_t tid;
 748     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 749 
 750     pthread_attr_destroy(&attr);
 751 
 752     if (ret != 0) {
 753       if (PrintMiscellaneous && (Verbose || WizardMode)) {
 754         perror("pthread_create()");
 755       }
 756       // Need to clean up stuff we've allocated so far
 757       thread->set_osthread(NULL);
 758       delete osthread;
 759       return false;
 760     }
 761 
 762     // Store pthread info into the OSThread
 763     osthread->set_pthread_id(tid);
 764 
 765     // Wait until child thread is either initialized or aborted
 766     {
 767       Monitor* sync_with_child = osthread->startThread_lock();
 768       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 769       while ((state = osthread->get_state()) == ALLOCATED) {
 770         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 771       }
 772     }
 773   }
 774 
 775   // Aborted due to thread limit being reached
 776   if (state == ZOMBIE) {
 777     thread->set_osthread(NULL);
 778     delete osthread;
 779     return false;
 780   }
 781 
 782   // The thread is returned suspended (in state INITIALIZED),
 783   // and is started higher up in the call chain
 784   assert(state == INITIALIZED, "race condition");
 785   return true;
 786 }
 787 
 788 /////////////////////////////////////////////////////////////////////////////
 789 // attach existing thread
 790 
 791 // bootstrap the main thread
 792 bool os::create_main_thread(JavaThread* thread) {
 793   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 794   return create_attached_thread(thread);
 795 }
 796 
 797 bool os::create_attached_thread(JavaThread* thread) {
 798 #ifdef ASSERT
 799   thread->verify_not_published();
 800 #endif
 801 
 802   // Allocate the OSThread object
 803   OSThread* osthread = new OSThread(NULL, NULL);
 804 
 805   if (osthread == NULL) {
 806     return false;
 807   }
 808 
 809   // Store pthread info into the OSThread
 810   osthread->set_thread_id(os::Linux::gettid());
 811   osthread->set_pthread_id(::pthread_self());
 812 
 813   // initialize floating point control register
 814   os::Linux::init_thread_fpu_state();
 815 
 816   // Initial thread state is RUNNABLE
 817   osthread->set_state(RUNNABLE);
 818 
 819   thread->set_osthread(osthread);
 820 
 821   if (UseNUMA) {
 822     int lgrp_id = os::numa_get_group_id();
 823     if (lgrp_id != -1) {
 824       thread->set_lgrp_id(lgrp_id);
 825     }
 826   }
 827 
 828   if (os::Linux::is_initial_thread()) {
 829     // If current thread is initial thread, its stack is mapped on demand,
 830     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 831     // the entire stack region to avoid SEGV in stack banging.
 832     // It is also useful to get around the heap-stack-gap problem on SuSE
 833     // kernel (see 4821821 for details). We first expand stack to the top
 834     // of yellow zone, then enable stack yellow zone (order is significant,
 835     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 836     // is no gap between the last two virtual memory regions.
 837 
 838     JavaThread *jt = (JavaThread *)thread;
 839     address addr = jt->stack_yellow_zone_base();
 840     assert(addr != NULL, "initialization problem?");
 841     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 842 
 843     osthread->set_expanding_stack();
 844     os::Linux::manually_expand_stack(jt, addr);
 845     osthread->clear_expanding_stack();
 846   }
 847 
 848   // initialize signal mask for this thread
 849   // and save the caller's signal mask
 850   os::Linux::hotspot_sigmask(thread);
 851 
 852   return true;
 853 }
 854 
 855 void os::pd_start_thread(Thread* thread) {
 856   OSThread * osthread = thread->osthread();
 857   assert(osthread->get_state() != INITIALIZED, "just checking");
 858   Monitor* sync_with_child = osthread->startThread_lock();
 859   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 860   sync_with_child->notify();
 861 }
 862 
 863 // Free Linux resources related to the OSThread
 864 void os::free_thread(OSThread* osthread) {
 865   assert(osthread != NULL, "osthread not set");
 866 
 867   if (Thread::current()->osthread() == osthread) {
 868     // Restore caller's signal mask
 869     sigset_t sigmask = osthread->caller_sigmask();
 870     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 871   }
 872 
 873   delete osthread;
 874 }
 875 
 876 //////////////////////////////////////////////////////////////////////////////
 877 // thread local storage
 878 
 879 // Restore the thread pointer if the destructor is called. This is in case
 880 // someone from JNI code sets up a destructor with pthread_key_create to run
 881 // detachCurrentThread on thread death. Unless we restore the thread pointer we
 882 // will hang or crash. When detachCurrentThread is called the key will be set
 883 // to null and we will not be called again. If detachCurrentThread is never
 884 // called we could loop forever depending on the pthread implementation.
 885 static void restore_thread_pointer(void* p) {
 886   Thread* thread = (Thread*) p;
 887   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
 888 }
 889 
 890 int os::allocate_thread_local_storage() {
 891   pthread_key_t key;
 892   int rslt = pthread_key_create(&key, restore_thread_pointer);
 893   assert(rslt == 0, "cannot allocate thread local storage");
 894   return (int)key;
 895 }
 896 
 897 // Note: This is currently not used by VM, as we don't destroy TLS key
 898 // on VM exit.
 899 void os::free_thread_local_storage(int index) {
 900   int rslt = pthread_key_delete((pthread_key_t)index);
 901   assert(rslt == 0, "invalid index");
 902 }
 903 
 904 void os::thread_local_storage_at_put(int index, void* value) {
 905   int rslt = pthread_setspecific((pthread_key_t)index, value);
 906   assert(rslt == 0, "pthread_setspecific failed");
 907 }
 908 
 909 extern "C" Thread* get_thread() {
 910   return ThreadLocalStorage::thread();
 911 }
 912 
 913 //////////////////////////////////////////////////////////////////////////////
 914 // initial thread
 915 
 916 // Check if current thread is the initial thread, similar to Solaris thr_main.
 917 bool os::Linux::is_initial_thread(void) {
 918   char dummy;
 919   // If called before init complete, thread stack bottom will be null.
 920   // Can be called if fatal error occurs before initialization.
 921   if (initial_thread_stack_bottom() == NULL) return false;
 922   assert(initial_thread_stack_bottom() != NULL &&
 923          initial_thread_stack_size()   != 0,
 924          "os::init did not locate initial thread's stack region");
 925   if ((address)&dummy >= initial_thread_stack_bottom() &&
 926       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 927     return true;
 928   } else {
 929     return false;
 930   }
 931 }
 932 
 933 // Find the virtual memory area that contains addr
 934 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 935   FILE *fp = fopen("/proc/self/maps", "r");
 936   if (fp) {
 937     address low, high;
 938     while (!feof(fp)) {
 939       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 940         if (low <= addr && addr < high) {
 941           if (vma_low)  *vma_low  = low;
 942           if (vma_high) *vma_high = high;
 943           fclose(fp);
 944           return true;
 945         }
 946       }
 947       for (;;) {
 948         int ch = fgetc(fp);
 949         if (ch == EOF || ch == (int)'\n') break;
 950       }
 951     }
 952     fclose(fp);
 953   }
 954   return false;
 955 }
 956 
 957 // Locate initial thread stack. This special handling of initial thread stack
 958 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 959 // bogus value for initial thread.
 960 void os::Linux::capture_initial_stack(size_t max_size) {
 961   // stack size is the easy part, get it from RLIMIT_STACK
 962   size_t stack_size;
 963   struct rlimit rlim;
 964   getrlimit(RLIMIT_STACK, &rlim);
 965   stack_size = rlim.rlim_cur;
 966 
 967   // 6308388: a bug in ld.so will relocate its own .data section to the
 968   //   lower end of primordial stack; reduce ulimit -s value a little bit
 969   //   so we won't install guard page on ld.so's data section.
 970   stack_size -= 2 * page_size();
 971 
 972   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
 973   //   7.1, in both cases we will get 2G in return value.
 974   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
 975   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
 976   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
 977   //   in case other parts in glibc still assumes 2M max stack size.
 978   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
 979   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
 980   if (stack_size > 2 * K * K IA64_ONLY(*2)) {
 981     stack_size = 2 * K * K IA64_ONLY(*2);
 982   }
 983   // Try to figure out where the stack base (top) is. This is harder.
 984   //
 985   // When an application is started, glibc saves the initial stack pointer in
 986   // a global variable "__libc_stack_end", which is then used by system
 987   // libraries. __libc_stack_end should be pretty close to stack top. The
 988   // variable is available since the very early days. However, because it is
 989   // a private interface, it could disappear in the future.
 990   //
 991   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 992   // to __libc_stack_end, it is very close to stack top, but isn't the real
 993   // stack top. Note that /proc may not exist if VM is running as a chroot
 994   // program, so reading /proc/<pid>/stat could fail. Also the contents of
 995   // /proc/<pid>/stat could change in the future (though unlikely).
 996   //
 997   // We try __libc_stack_end first. If that doesn't work, look for
 998   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
 999   // as a hint, which should work well in most cases.
1000 
1001   uintptr_t stack_start;
1002 
1003   // try __libc_stack_end first
1004   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1005   if (p && *p) {
1006     stack_start = *p;
1007   } else {
1008     // see if we can get the start_stack field from /proc/self/stat
1009     FILE *fp;
1010     int pid;
1011     char state;
1012     int ppid;
1013     int pgrp;
1014     int session;
1015     int nr;
1016     int tpgrp;
1017     unsigned long flags;
1018     unsigned long minflt;
1019     unsigned long cminflt;
1020     unsigned long majflt;
1021     unsigned long cmajflt;
1022     unsigned long utime;
1023     unsigned long stime;
1024     long cutime;
1025     long cstime;
1026     long prio;
1027     long nice;
1028     long junk;
1029     long it_real;
1030     uintptr_t start;
1031     uintptr_t vsize;
1032     intptr_t rss;
1033     uintptr_t rsslim;
1034     uintptr_t scodes;
1035     uintptr_t ecode;
1036     int i;
1037 
1038     // Figure what the primordial thread stack base is. Code is inspired
1039     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1040     // followed by command name surrounded by parentheses, state, etc.
1041     char stat[2048];
1042     int statlen;
1043 
1044     fp = fopen("/proc/self/stat", "r");
1045     if (fp) {
1046       statlen = fread(stat, 1, 2047, fp);
1047       stat[statlen] = '\0';
1048       fclose(fp);
1049 
1050       // Skip pid and the command string. Note that we could be dealing with
1051       // weird command names, e.g. user could decide to rename java launcher
1052       // to "java 1.4.2 :)", then the stat file would look like
1053       //                1234 (java 1.4.2 :)) R ... ...
1054       // We don't really need to know the command string, just find the last
1055       // occurrence of ")" and then start parsing from there. See bug 4726580.
1056       char * s = strrchr(stat, ')');
1057 
1058       i = 0;
1059       if (s) {
1060         // Skip blank chars
1061         do { s++; } while (s && isspace(*s));
1062 
1063 #define _UFM UINTX_FORMAT
1064 #define _DFM INTX_FORMAT
1065 
1066         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1067         //              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
1068         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,
1069                    &state,          // 3  %c
1070                    &ppid,           // 4  %d
1071                    &pgrp,           // 5  %d
1072                    &session,        // 6  %d
1073                    &nr,             // 7  %d
1074                    &tpgrp,          // 8  %d
1075                    &flags,          // 9  %lu
1076                    &minflt,         // 10 %lu
1077                    &cminflt,        // 11 %lu
1078                    &majflt,         // 12 %lu
1079                    &cmajflt,        // 13 %lu
1080                    &utime,          // 14 %lu
1081                    &stime,          // 15 %lu
1082                    &cutime,         // 16 %ld
1083                    &cstime,         // 17 %ld
1084                    &prio,           // 18 %ld
1085                    &nice,           // 19 %ld
1086                    &junk,           // 20 %ld
1087                    &it_real,        // 21 %ld
1088                    &start,          // 22 UINTX_FORMAT
1089                    &vsize,          // 23 UINTX_FORMAT
1090                    &rss,            // 24 INTX_FORMAT
1091                    &rsslim,         // 25 UINTX_FORMAT
1092                    &scodes,         // 26 UINTX_FORMAT
1093                    &ecode,          // 27 UINTX_FORMAT
1094                    &stack_start);   // 28 UINTX_FORMAT
1095       }
1096 
1097 #undef _UFM
1098 #undef _DFM
1099 
1100       if (i != 28 - 2) {
1101         assert(false, "Bad conversion from /proc/self/stat");
1102         // product mode - assume we are the initial thread, good luck in the
1103         // embedded case.
1104         warning("Can't detect initial thread stack location - bad conversion");
1105         stack_start = (uintptr_t) &rlim;
1106       }
1107     } else {
1108       // For some reason we can't open /proc/self/stat (for example, running on
1109       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1110       // most cases, so don't abort:
1111       warning("Can't detect initial thread stack location - no /proc/self/stat");
1112       stack_start = (uintptr_t) &rlim;
1113     }
1114   }
1115 
1116   // Now we have a pointer (stack_start) very close to the stack top, the
1117   // next thing to do is to figure out the exact location of stack top. We
1118   // can find out the virtual memory area that contains stack_start by
1119   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1120   // and its upper limit is the real stack top. (again, this would fail if
1121   // running inside chroot, because /proc may not exist.)
1122 
1123   uintptr_t stack_top;
1124   address low, high;
1125   if (find_vma((address)stack_start, &low, &high)) {
1126     // success, "high" is the true stack top. (ignore "low", because initial
1127     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1128     stack_top = (uintptr_t)high;
1129   } else {
1130     // failed, likely because /proc/self/maps does not exist
1131     warning("Can't detect initial thread stack location - find_vma failed");
1132     // best effort: stack_start is normally within a few pages below the real
1133     // stack top, use it as stack top, and reduce stack size so we won't put
1134     // guard page outside stack.
1135     stack_top = stack_start;
1136     stack_size -= 16 * page_size();
1137   }
1138 
1139   // stack_top could be partially down the page so align it
1140   stack_top = align_size_up(stack_top, page_size());
1141 
1142   if (max_size && stack_size > max_size) {
1143     _initial_thread_stack_size = max_size;
1144   } else {
1145     _initial_thread_stack_size = stack_size;
1146   }
1147 
1148   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1149   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1150 }
1151 
1152 ////////////////////////////////////////////////////////////////////////////////
1153 // time support
1154 
1155 // Time since start-up in seconds to a fine granularity.
1156 // Used by VMSelfDestructTimer and the MemProfiler.
1157 double os::elapsedTime() {
1158 
1159   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1160 }
1161 
1162 jlong os::elapsed_counter() {
1163   return javaTimeNanos() - initial_time_count;
1164 }
1165 
1166 jlong os::elapsed_frequency() {
1167   return NANOSECS_PER_SEC; // nanosecond resolution
1168 }
1169 
1170 bool os::supports_vtime() { return true; }
1171 bool os::enable_vtime()   { return false; }
1172 bool os::vtime_enabled()  { return false; }
1173 
1174 double os::elapsedVTime() {
1175   struct rusage usage;
1176   int retval = getrusage(RUSAGE_THREAD, &usage);
1177   if (retval == 0) {
1178     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);
1179   } else {
1180     // better than nothing, but not much
1181     return elapsedTime();
1182   }
1183 }
1184 
1185 jlong os::javaTimeMillis() {
1186   timeval time;
1187   int status = gettimeofday(&time, NULL);
1188   assert(status != -1, "linux error");
1189   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1190 }
1191 
1192 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1193   timeval time;
1194   int status = gettimeofday(&time, NULL);
1195   assert(status != -1, "linux error");
1196   seconds = jlong(time.tv_sec);
1197   nanos = jlong(time.tv_usec) * 1000;
1198 }
1199 
1200 
1201 #ifndef CLOCK_MONOTONIC
1202   #define CLOCK_MONOTONIC (1)
1203 #endif
1204 
1205 void os::Linux::clock_init() {
1206   // we do dlopen's in this particular order due to bug in linux
1207   // dynamical loader (see 6348968) leading to crash on exit
1208   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1209   if (handle == NULL) {
1210     handle = dlopen("librt.so", RTLD_LAZY);
1211   }
1212 
1213   if (handle) {
1214     int (*clock_getres_func)(clockid_t, struct timespec*) =
1215            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1216     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1217            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1218     if (clock_getres_func && clock_gettime_func) {
1219       // See if monotonic clock is supported by the kernel. Note that some
1220       // early implementations simply return kernel jiffies (updated every
1221       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1222       // for nano time (though the monotonic property is still nice to have).
1223       // It's fixed in newer kernels, however clock_getres() still returns
1224       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1225       // resolution for now. Hopefully as people move to new kernels, this
1226       // won't be a problem.
1227       struct timespec res;
1228       struct timespec tp;
1229       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1230           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1231         // yes, monotonic clock is supported
1232         _clock_gettime = clock_gettime_func;
1233         return;
1234       } else {
1235         // close librt if there is no monotonic clock
1236         dlclose(handle);
1237       }
1238     }
1239   }
1240   warning("No monotonic clock was available - timed services may " \
1241           "be adversely affected if the time-of-day clock changes");
1242 }
1243 
1244 #ifndef SYS_clock_getres
1245   #if defined(IA32) || defined(AMD64)
1246     #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1247     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1248   #else
1249     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1250     #define sys_clock_getres(x,y)  -1
1251   #endif
1252 #else
1253   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1254 #endif
1255 
1256 void os::Linux::fast_thread_clock_init() {
1257   if (!UseLinuxPosixThreadCPUClocks) {
1258     return;
1259   }
1260   clockid_t clockid;
1261   struct timespec tp;
1262   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1263       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1264 
1265   // Switch to using fast clocks for thread cpu time if
1266   // the sys_clock_getres() returns 0 error code.
1267   // Note, that some kernels may support the current thread
1268   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1269   // returned by the pthread_getcpuclockid().
1270   // If the fast Posix clocks are supported then the sys_clock_getres()
1271   // must return at least tp.tv_sec == 0 which means a resolution
1272   // better than 1 sec. This is extra check for reliability.
1273 
1274   if (pthread_getcpuclockid_func &&
1275       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1276       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1277     _supports_fast_thread_cpu_time = true;
1278     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1279   }
1280 }
1281 
1282 jlong os::javaTimeNanos() {
1283   if (os::supports_monotonic_clock()) {
1284     struct timespec tp;
1285     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1286     assert(status == 0, "gettime error");
1287     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1288     return result;
1289   } else {
1290     timeval time;
1291     int status = gettimeofday(&time, NULL);
1292     assert(status != -1, "linux error");
1293     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1294     return 1000 * usecs;
1295   }
1296 }
1297 
1298 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1299   if (os::supports_monotonic_clock()) {
1300     info_ptr->max_value = ALL_64_BITS;
1301 
1302     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1303     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1304     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1305   } else {
1306     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1307     info_ptr->max_value = ALL_64_BITS;
1308 
1309     // gettimeofday is a real time clock so it skips
1310     info_ptr->may_skip_backward = true;
1311     info_ptr->may_skip_forward = true;
1312   }
1313 
1314   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1315 }
1316 
1317 // Return the real, user, and system times in seconds from an
1318 // arbitrary fixed point in the past.
1319 bool os::getTimesSecs(double* process_real_time,
1320                       double* process_user_time,
1321                       double* process_system_time) {
1322   struct tms ticks;
1323   clock_t real_ticks = times(&ticks);
1324 
1325   if (real_ticks == (clock_t) (-1)) {
1326     return false;
1327   } else {
1328     double ticks_per_second = (double) clock_tics_per_sec;
1329     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1330     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1331     *process_real_time = ((double) real_ticks) / ticks_per_second;
1332 
1333     return true;
1334   }
1335 }
1336 
1337 
1338 char * os::local_time_string(char *buf, size_t buflen) {
1339   struct tm t;
1340   time_t long_time;
1341   time(&long_time);
1342   localtime_r(&long_time, &t);
1343   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1344                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1345                t.tm_hour, t.tm_min, t.tm_sec);
1346   return buf;
1347 }
1348 
1349 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1350   return localtime_r(clock, res);
1351 }
1352 
1353 ////////////////////////////////////////////////////////////////////////////////
1354 // runtime exit support
1355 
1356 // Note: os::shutdown() might be called very early during initialization, or
1357 // called from signal handler. Before adding something to os::shutdown(), make
1358 // sure it is async-safe and can handle partially initialized VM.
1359 void os::shutdown() {
1360 
1361   // allow PerfMemory to attempt cleanup of any persistent resources
1362   perfMemory_exit();
1363 
1364   // needs to remove object in file system
1365   AttachListener::abort();
1366 
1367   // flush buffered output, finish log files
1368   ostream_abort();
1369 
1370   // Check for abort hook
1371   abort_hook_t abort_hook = Arguments::abort_hook();
1372   if (abort_hook != NULL) {
1373     abort_hook();
1374   }
1375 
1376 }
1377 
1378 // Note: os::abort() might be called very early during initialization, or
1379 // called from signal handler. Before adding something to os::abort(), make
1380 // sure it is async-safe and can handle partially initialized VM.
1381 void os::abort(bool dump_core, void* siginfo, void* context) {
1382   os::shutdown();
1383   if (dump_core) {
1384 #ifndef PRODUCT
1385     fdStream out(defaultStream::output_fd());
1386     out.print_raw("Current thread is ");
1387     char buf[16];
1388     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1389     out.print_raw_cr(buf);
1390     out.print_raw_cr("Dumping core ...");
1391 #endif
1392     ::abort(); // dump core
1393   }
1394 
1395   ::exit(1);
1396 }
1397 
1398 // Die immediately, no exit hook, no abort hook, no cleanup.
1399 void os::die() {
1400   ::abort();
1401 }
1402 
1403 
1404 // This method is a copy of JDK's sysGetLastErrorString
1405 // from src/solaris/hpi/src/system_md.c
1406 
1407 size_t os::lasterror(char *buf, size_t len) {
1408   if (errno == 0)  return 0;
1409 
1410   const char *s = ::strerror(errno);
1411   size_t n = ::strlen(s);
1412   if (n >= len) {
1413     n = len - 1;
1414   }
1415   ::strncpy(buf, s, n);
1416   buf[n] = '\0';
1417   return n;
1418 }
1419 
1420 // thread_id is kernel thread id (similar to Solaris LWP id)
1421 intx os::current_thread_id() { return os::Linux::gettid(); }
1422 int os::current_process_id() {
1423   return ::getpid();
1424 }
1425 
1426 // DLL functions
1427 
1428 const char* os::dll_file_extension() { return ".so"; }
1429 
1430 // This must be hard coded because it's the system's temporary
1431 // directory not the java application's temp directory, ala java.io.tmpdir.
1432 const char* os::get_temp_directory() { return "/tmp"; }
1433 
1434 static bool file_exists(const char* filename) {
1435   struct stat statbuf;
1436   if (filename == NULL || strlen(filename) == 0) {
1437     return false;
1438   }
1439   return os::stat(filename, &statbuf) == 0;
1440 }
1441 
1442 bool os::dll_build_name(char* buffer, size_t buflen,
1443                         const char* pname, const char* fname) {
1444   bool retval = false;
1445   // Copied from libhpi
1446   const size_t pnamelen = pname ? strlen(pname) : 0;
1447 
1448   // Return error on buffer overflow.
1449   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1450     return retval;
1451   }
1452 
1453   if (pnamelen == 0) {
1454     snprintf(buffer, buflen, "lib%s.so", fname);
1455     retval = true;
1456   } else if (strchr(pname, *os::path_separator()) != NULL) {
1457     int n;
1458     char** pelements = split_path(pname, &n);
1459     if (pelements == NULL) {
1460       return false;
1461     }
1462     for (int i = 0; i < n; i++) {
1463       // Really shouldn't be NULL, but check can't hurt
1464       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1465         continue; // skip the empty path values
1466       }
1467       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1468       if (file_exists(buffer)) {
1469         retval = true;
1470         break;
1471       }
1472     }
1473     // release the storage
1474     for (int i = 0; i < n; i++) {
1475       if (pelements[i] != NULL) {
1476         FREE_C_HEAP_ARRAY(char, pelements[i]);
1477       }
1478     }
1479     if (pelements != NULL) {
1480       FREE_C_HEAP_ARRAY(char*, pelements);
1481     }
1482   } else {
1483     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1484     retval = true;
1485   }
1486   return retval;
1487 }
1488 
1489 // check if addr is inside libjvm.so
1490 bool os::address_is_in_vm(address addr) {
1491   static address libjvm_base_addr;
1492   Dl_info dlinfo;
1493 
1494   if (libjvm_base_addr == NULL) {
1495     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1496       libjvm_base_addr = (address)dlinfo.dli_fbase;
1497     }
1498     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1499   }
1500 
1501   if (dladdr((void *)addr, &dlinfo) != 0) {
1502     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1503   }
1504 
1505   return false;
1506 }
1507 
1508 bool os::dll_address_to_function_name(address addr, char *buf,
1509                                       int buflen, int *offset,
1510                                       bool demangle) {
1511   // buf is not optional, but offset is optional
1512   assert(buf != NULL, "sanity check");
1513 
1514   Dl_info dlinfo;
1515 
1516   if (dladdr((void*)addr, &dlinfo) != 0) {
1517     // see if we have a matching symbol
1518     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1519       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1520         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1521       }
1522       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1523       return true;
1524     }
1525     // no matching symbol so try for just file info
1526     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1527       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1528                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1529         return true;
1530       }
1531     }
1532   }
1533 
1534   buf[0] = '\0';
1535   if (offset != NULL) *offset = -1;
1536   return false;
1537 }
1538 
1539 struct _address_to_library_name {
1540   address addr;          // input : memory address
1541   size_t  buflen;        //         size of fname
1542   char*   fname;         // output: library name
1543   address base;          //         library base addr
1544 };
1545 
1546 static int address_to_library_name_callback(struct dl_phdr_info *info,
1547                                             size_t size, void *data) {
1548   int i;
1549   bool found = false;
1550   address libbase = NULL;
1551   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1552 
1553   // iterate through all loadable segments
1554   for (i = 0; i < info->dlpi_phnum; i++) {
1555     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1556     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1557       // base address of a library is the lowest address of its loaded
1558       // segments.
1559       if (libbase == NULL || libbase > segbase) {
1560         libbase = segbase;
1561       }
1562       // see if 'addr' is within current segment
1563       if (segbase <= d->addr &&
1564           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1565         found = true;
1566       }
1567     }
1568   }
1569 
1570   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1571   // so dll_address_to_library_name() can fall through to use dladdr() which
1572   // can figure out executable name from argv[0].
1573   if (found && info->dlpi_name && info->dlpi_name[0]) {
1574     d->base = libbase;
1575     if (d->fname) {
1576       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1577     }
1578     return 1;
1579   }
1580   return 0;
1581 }
1582 
1583 bool os::dll_address_to_library_name(address addr, char* buf,
1584                                      int buflen, int* offset) {
1585   // buf is not optional, but offset is optional
1586   assert(buf != NULL, "sanity check");
1587 
1588   Dl_info dlinfo;
1589   struct _address_to_library_name data;
1590 
1591   // There is a bug in old glibc dladdr() implementation that it could resolve
1592   // to wrong library name if the .so file has a base address != NULL. Here
1593   // we iterate through the program headers of all loaded libraries to find
1594   // out which library 'addr' really belongs to. This workaround can be
1595   // removed once the minimum requirement for glibc is moved to 2.3.x.
1596   data.addr = addr;
1597   data.fname = buf;
1598   data.buflen = buflen;
1599   data.base = NULL;
1600   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1601 
1602   if (rslt) {
1603     // buf already contains library name
1604     if (offset) *offset = addr - data.base;
1605     return true;
1606   }
1607   if (dladdr((void*)addr, &dlinfo) != 0) {
1608     if (dlinfo.dli_fname != NULL) {
1609       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1610     }
1611     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1612       *offset = addr - (address)dlinfo.dli_fbase;
1613     }
1614     return true;
1615   }
1616 
1617   buf[0] = '\0';
1618   if (offset) *offset = -1;
1619   return false;
1620 }
1621 
1622 // Loads .dll/.so and
1623 // in case of error it checks if .dll/.so was built for the
1624 // same architecture as Hotspot is running on
1625 
1626 
1627 // Remember the stack's state. The Linux dynamic linker will change
1628 // the stack to 'executable' at most once, so we must safepoint only once.
1629 bool os::Linux::_stack_is_executable = false;
1630 
1631 // VM operation that loads a library.  This is necessary if stack protection
1632 // of the Java stacks can be lost during loading the library.  If we
1633 // do not stop the Java threads, they can stack overflow before the stacks
1634 // are protected again.
1635 class VM_LinuxDllLoad: public VM_Operation {
1636  private:
1637   const char *_filename;
1638   char *_ebuf;
1639   int _ebuflen;
1640   void *_lib;
1641  public:
1642   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1643     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1644   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1645   void doit() {
1646     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1647     os::Linux::_stack_is_executable = true;
1648   }
1649   void* loaded_library() { return _lib; }
1650 };
1651 
1652 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1653   void * result = NULL;
1654   bool load_attempted = false;
1655 
1656   // Check whether the library to load might change execution rights
1657   // of the stack. If they are changed, the protection of the stack
1658   // guard pages will be lost. We need a safepoint to fix this.
1659   //
1660   // See Linux man page execstack(8) for more info.
1661   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1662     ElfFile ef(filename);
1663     if (!ef.specifies_noexecstack()) {
1664       if (!is_init_completed()) {
1665         os::Linux::_stack_is_executable = true;
1666         // This is OK - No Java threads have been created yet, and hence no
1667         // stack guard pages to fix.
1668         //
1669         // This should happen only when you are building JDK7 using a very
1670         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1671         //
1672         // Dynamic loader will make all stacks executable after
1673         // this function returns, and will not do that again.
1674         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1675       } else {
1676         warning("You have loaded library %s which might have disabled stack guard. "
1677                 "The VM will try to fix the stack guard now.\n"
1678                 "It's highly recommended that you fix the library with "
1679                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1680                 filename);
1681 
1682         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1683         JavaThread *jt = JavaThread::current();
1684         if (jt->thread_state() != _thread_in_native) {
1685           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1686           // that requires ExecStack. Cannot enter safe point. Let's give up.
1687           warning("Unable to fix stack guard. Giving up.");
1688         } else {
1689           if (!LoadExecStackDllInVMThread) {
1690             // This is for the case where the DLL has an static
1691             // constructor function that executes JNI code. We cannot
1692             // load such DLLs in the VMThread.
1693             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1694           }
1695 
1696           ThreadInVMfromNative tiv(jt);
1697           debug_only(VMNativeEntryWrapper vew;)
1698 
1699           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1700           VMThread::execute(&op);
1701           if (LoadExecStackDllInVMThread) {
1702             result = op.loaded_library();
1703           }
1704           load_attempted = true;
1705         }
1706       }
1707     }
1708   }
1709 
1710   if (!load_attempted) {
1711     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1712   }
1713 
1714   if (result != NULL) {
1715     // Successful loading
1716     return result;
1717   }
1718 
1719   Elf32_Ehdr elf_head;
1720   int diag_msg_max_length=ebuflen-strlen(ebuf);
1721   char* diag_msg_buf=ebuf+strlen(ebuf);
1722 
1723   if (diag_msg_max_length==0) {
1724     // No more space in ebuf for additional diagnostics message
1725     return NULL;
1726   }
1727 
1728 
1729   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1730 
1731   if (file_descriptor < 0) {
1732     // Can't open library, report dlerror() message
1733     return NULL;
1734   }
1735 
1736   bool failed_to_read_elf_head=
1737     (sizeof(elf_head)!=
1738      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1739 
1740   ::close(file_descriptor);
1741   if (failed_to_read_elf_head) {
1742     // file i/o error - report dlerror() msg
1743     return NULL;
1744   }
1745 
1746   typedef struct {
1747     Elf32_Half  code;         // Actual value as defined in elf.h
1748     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1749     char        elf_class;    // 32 or 64 bit
1750     char        endianess;    // MSB or LSB
1751     char*       name;         // String representation
1752   } arch_t;
1753 
1754 #ifndef EM_486
1755   #define EM_486          6               /* Intel 80486 */
1756 #endif
1757 #ifndef EM_AARCH64
1758   #define EM_AARCH64    183               /* ARM AARCH64 */
1759 #endif
1760 
1761   static const arch_t arch_array[]={
1762     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1763     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1764     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1765     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1766     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1767     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1768     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1769     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1770 #if defined(VM_LITTLE_ENDIAN)
1771     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1772 #else
1773     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1774 #endif
1775     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1776     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1777     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1778     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1779     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1780     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1781     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1782     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1783   };
1784 
1785 #if  (defined IA32)
1786   static  Elf32_Half running_arch_code=EM_386;
1787 #elif   (defined AMD64)
1788   static  Elf32_Half running_arch_code=EM_X86_64;
1789 #elif  (defined IA64)
1790   static  Elf32_Half running_arch_code=EM_IA_64;
1791 #elif  (defined __sparc) && (defined _LP64)
1792   static  Elf32_Half running_arch_code=EM_SPARCV9;
1793 #elif  (defined __sparc) && (!defined _LP64)
1794   static  Elf32_Half running_arch_code=EM_SPARC;
1795 #elif  (defined __powerpc64__)
1796   static  Elf32_Half running_arch_code=EM_PPC64;
1797 #elif  (defined __powerpc__)
1798   static  Elf32_Half running_arch_code=EM_PPC;
1799 #elif  (defined ARM)
1800   static  Elf32_Half running_arch_code=EM_ARM;
1801 #elif  (defined S390)
1802   static  Elf32_Half running_arch_code=EM_S390;
1803 #elif  (defined ALPHA)
1804   static  Elf32_Half running_arch_code=EM_ALPHA;
1805 #elif  (defined MIPSEL)
1806   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1807 #elif  (defined PARISC)
1808   static  Elf32_Half running_arch_code=EM_PARISC;
1809 #elif  (defined MIPS)
1810   static  Elf32_Half running_arch_code=EM_MIPS;
1811 #elif  (defined M68K)
1812   static  Elf32_Half running_arch_code=EM_68K;
1813 #elif  (defined AARCH64)
1814   static  Elf32_Half running_arch_code=EM_AARCH64;
1815 #else
1816     #error Method os::dll_load requires that one of following is defined:\
1817          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1818 #endif
1819 
1820   // Identify compatability class for VM's architecture and library's architecture
1821   // Obtain string descriptions for architectures
1822 
1823   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1824   int running_arch_index=-1;
1825 
1826   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1827     if (running_arch_code == arch_array[i].code) {
1828       running_arch_index    = i;
1829     }
1830     if (lib_arch.code == arch_array[i].code) {
1831       lib_arch.compat_class = arch_array[i].compat_class;
1832       lib_arch.name         = arch_array[i].name;
1833     }
1834   }
1835 
1836   assert(running_arch_index != -1,
1837          "Didn't find running architecture code (running_arch_code) in arch_array");
1838   if (running_arch_index == -1) {
1839     // Even though running architecture detection failed
1840     // we may still continue with reporting dlerror() message
1841     return NULL;
1842   }
1843 
1844   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1845     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1846     return NULL;
1847   }
1848 
1849 #ifndef S390
1850   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1851     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1852     return NULL;
1853   }
1854 #endif // !S390
1855 
1856   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1857     if (lib_arch.name!=NULL) {
1858       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1859                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1860                  lib_arch.name, arch_array[running_arch_index].name);
1861     } else {
1862       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1863                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1864                  lib_arch.code,
1865                  arch_array[running_arch_index].name);
1866     }
1867   }
1868 
1869   return NULL;
1870 }
1871 
1872 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1873                                 int ebuflen) {
1874   void * result = ::dlopen(filename, RTLD_LAZY);
1875   if (result == NULL) {
1876     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1877     ebuf[ebuflen-1] = '\0';
1878   }
1879   return result;
1880 }
1881 
1882 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1883                                        int ebuflen) {
1884   void * result = NULL;
1885   if (LoadExecStackDllInVMThread) {
1886     result = dlopen_helper(filename, ebuf, ebuflen);
1887   }
1888 
1889   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1890   // library that requires an executable stack, or which does not have this
1891   // stack attribute set, dlopen changes the stack attribute to executable. The
1892   // read protection of the guard pages gets lost.
1893   //
1894   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1895   // may have been queued at the same time.
1896 
1897   if (!_stack_is_executable) {
1898     JavaThread *jt = Threads::first();
1899 
1900     while (jt) {
1901       if (!jt->stack_guard_zone_unused() &&        // Stack not yet fully initialized
1902           jt->stack_guards_enabled()) {       // No pending stack overflow exceptions
1903         if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
1904                               jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
1905           warning("Attempt to reguard stack yellow zone failed.");
1906         }
1907       }
1908       jt = jt->next();
1909     }
1910   }
1911 
1912   return result;
1913 }
1914 
1915 void* os::dll_lookup(void* handle, const char* name) {
1916   void* res = dlsym(handle, name);
1917   return res;
1918 }
1919 
1920 void* os::get_default_process_handle() {
1921   return (void*)::dlopen(NULL, RTLD_LAZY);
1922 }
1923 
1924 static bool _print_ascii_file(const char* filename, outputStream* st) {
1925   int fd = ::open(filename, O_RDONLY);
1926   if (fd == -1) {
1927     return false;
1928   }
1929 
1930   char buf[32];
1931   int bytes;
1932   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1933     st->print_raw(buf, bytes);
1934   }
1935 
1936   ::close(fd);
1937 
1938   return true;
1939 }
1940 
1941 void os::print_dll_info(outputStream *st) {
1942   st->print_cr("Dynamic libraries:");
1943 
1944   char fname[32];
1945   pid_t pid = os::Linux::gettid();
1946 
1947   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1948 
1949   if (!_print_ascii_file(fname, st)) {
1950     st->print("Can not get library information for pid = %d\n", pid);
1951   }
1952 }
1953 
1954 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1955   FILE *procmapsFile = NULL;
1956 
1957   // Open the procfs maps file for the current process
1958   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1959     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1960     char line[PATH_MAX + 100];
1961 
1962     // Read line by line from 'file'
1963     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1964       u8 base, top, offset, inode;
1965       char permissions[5];
1966       char device[6];
1967       char name[PATH_MAX + 1];
1968 
1969       // Parse fields from line
1970       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1971              &base, &top, permissions, &offset, device, &inode, name);
1972 
1973       // Filter by device id '00:00' so that we only get file system mapped files.
1974       if (strcmp(device, "00:00") != 0) {
1975 
1976         // Call callback with the fields of interest
1977         if(callback(name, (address)base, (address)top, param)) {
1978           // Oops abort, callback aborted
1979           fclose(procmapsFile);
1980           return 1;
1981         }
1982       }
1983     }
1984     fclose(procmapsFile);
1985   }
1986   return 0;
1987 }
1988 
1989 void os::print_os_info_brief(outputStream* st) {
1990   os::Linux::print_distro_info(st);
1991 
1992   os::Posix::print_uname_info(st);
1993 
1994   os::Linux::print_libversion_info(st);
1995 
1996 }
1997 
1998 void os::print_os_info(outputStream* st) {
1999   st->print("OS:");
2000 
2001   os::Linux::print_distro_info(st);
2002 
2003   os::Posix::print_uname_info(st);
2004 
2005   // Print warning if unsafe chroot environment detected
2006   if (unsafe_chroot_detected) {
2007     st->print("WARNING!! ");
2008     st->print_cr("%s", unstable_chroot_error);
2009   }
2010 
2011   os::Linux::print_libversion_info(st);
2012 
2013   os::Posix::print_rlimit_info(st);
2014 
2015   os::Posix::print_load_average(st);
2016 
2017   os::Linux::print_full_memory_info(st);
2018 }
2019 
2020 // Try to identify popular distros.
2021 // Most Linux distributions have a /etc/XXX-release file, which contains
2022 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2023 // file that also contains the OS version string. Some have more than one
2024 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2025 // /etc/redhat-release.), so the order is important.
2026 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2027 // their own specific XXX-release file as well as a redhat-release file.
2028 // Because of this the XXX-release file needs to be searched for before the
2029 // redhat-release file.
2030 // Since Red Hat has a lsb-release file that is not very descriptive the
2031 // search for redhat-release needs to be before lsb-release.
2032 // Since the lsb-release file is the new standard it needs to be searched
2033 // before the older style release files.
2034 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2035 // next to last resort.  The os-release file is a new standard that contains
2036 // distribution information and the system-release file seems to be an old
2037 // standard that has been replaced by the lsb-release and os-release files.
2038 // Searching for the debian_version file is the last resort.  It contains
2039 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2040 // "Debian " is printed before the contents of the debian_version file.
2041 
2042 const char* distro_files[] = {
2043   "/etc/oracle-release",
2044   "/etc/mandriva-release",
2045   "/etc/mandrake-release",
2046   "/etc/sun-release",
2047   "/etc/redhat-release",
2048   "/etc/lsb-release",
2049   "/etc/SuSE-release",
2050   "/etc/turbolinux-release",
2051   "/etc/gentoo-release",
2052   "/etc/ltib-release",
2053   "/etc/angstrom-version",
2054   "/etc/system-release",
2055   "/etc/os-release",
2056   NULL };
2057 
2058 void os::Linux::print_distro_info(outputStream* st) {
2059   for (int i = 0;; i++) {
2060     const char* file = distro_files[i];
2061     if (file == NULL) {
2062       break;  // done
2063     }
2064     // If file prints, we found it.
2065     if (_print_ascii_file(file, st)) {
2066       return;
2067     }
2068   }
2069 
2070   if (file_exists("/etc/debian_version")) {
2071     st->print("Debian ");
2072     _print_ascii_file("/etc/debian_version", st);
2073   } else {
2074     st->print("Linux");
2075   }
2076   st->cr();
2077 }
2078 
2079 static void parse_os_info(char* distro, size_t length, const char* file) {
2080   FILE* fp = fopen(file, "r");
2081   if (fp != NULL) {
2082     char buf[256];
2083     // get last line of the file.
2084     while (fgets(buf, sizeof(buf), fp)) { }
2085     // Edit out extra stuff in expected ubuntu format
2086     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL) {
2087       char* ptr = strstr(buf, "\"");  // the name is in quotes
2088       if (ptr != NULL) {
2089         ptr++; // go beyond first quote
2090         char* nl = strchr(ptr, '\"');
2091         if (nl != NULL) *nl = '\0';
2092         strncpy(distro, ptr, length);
2093       } else {
2094         ptr = strstr(buf, "=");
2095         ptr++; // go beyond equals then
2096         char* nl = strchr(ptr, '\n');
2097         if (nl != NULL) *nl = '\0';
2098         strncpy(distro, ptr, length);
2099       }
2100     } else {
2101       // if not in expected Ubuntu format, print out whole line minus \n
2102       char* nl = strchr(buf, '\n');
2103       if (nl != NULL) *nl = '\0';
2104       strncpy(distro, buf, length);
2105     }
2106     // close distro file
2107     fclose(fp);
2108   }
2109 }
2110 
2111 void os::get_summary_os_info(char* buf, size_t buflen) {
2112   for (int i = 0;; i++) {
2113     const char* file = distro_files[i];
2114     if (file == NULL) {
2115       break; // ran out of distro_files
2116     }
2117     if (file_exists(file)) {
2118       parse_os_info(buf, buflen, file);
2119       return;
2120     }
2121   }
2122   // special case for debian
2123   if (file_exists("/etc/debian_version")) {
2124     strncpy(buf, "Debian ", buflen);
2125     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2126   } else {
2127     strncpy(buf, "Linux", buflen);
2128   }
2129 }
2130 
2131 void os::Linux::print_libversion_info(outputStream* st) {
2132   // libc, pthread
2133   st->print("libc:");
2134   st->print("%s ", os::Linux::glibc_version());
2135   st->print("%s ", os::Linux::libpthread_version());
2136   st->cr();
2137 }
2138 
2139 void os::Linux::print_full_memory_info(outputStream* st) {
2140   st->print("\n/proc/meminfo:\n");
2141   _print_ascii_file("/proc/meminfo", st);
2142   st->cr();
2143 }
2144 
2145 void os::print_memory_info(outputStream* st) {
2146 
2147   st->print("Memory:");
2148   st->print(" %dk page", os::vm_page_size()>>10);
2149 
2150   // values in struct sysinfo are "unsigned long"
2151   struct sysinfo si;
2152   sysinfo(&si);
2153 
2154   st->print(", physical " UINT64_FORMAT "k",
2155             os::physical_memory() >> 10);
2156   st->print("(" UINT64_FORMAT "k free)",
2157             os::available_memory() >> 10);
2158   st->print(", swap " UINT64_FORMAT "k",
2159             ((jlong)si.totalswap * si.mem_unit) >> 10);
2160   st->print("(" UINT64_FORMAT "k free)",
2161             ((jlong)si.freeswap * si.mem_unit) >> 10);
2162   st->cr();
2163 }
2164 
2165 // Print the first "model name" line and the first "flags" line
2166 // that we find and nothing more. We assume "model name" comes
2167 // before "flags" so if we find a second "model name", then the
2168 // "flags" field is considered missing.
2169 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2170 #if defined(IA32) || defined(AMD64)
2171   // Other platforms have less repetitive cpuinfo files
2172   FILE *fp = fopen("/proc/cpuinfo", "r");
2173   if (fp) {
2174     while (!feof(fp)) {
2175       if (fgets(buf, buflen, fp)) {
2176         // Assume model name comes before flags
2177         bool model_name_printed = false;
2178         if (strstr(buf, "model name") != NULL) {
2179           if (!model_name_printed) {
2180             st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n");
2181             st->print_raw(buf);
2182             model_name_printed = true;
2183           } else {
2184             // model name printed but not flags?  Odd, just return
2185             fclose(fp);
2186             return true;
2187           }
2188         }
2189         // print the flags line too
2190         if (strstr(buf, "flags") != NULL) {
2191           st->print_raw(buf);
2192           fclose(fp);
2193           return true;
2194         }
2195       }
2196     }
2197     fclose(fp);
2198   }
2199 #endif // x86 platforms
2200   return false;
2201 }
2202 
2203 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2204   // Only print the model name if the platform provides this as a summary
2205   if (!print_model_name_and_flags(st, buf, buflen)) {
2206     st->print("\n/proc/cpuinfo:\n");
2207     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2208       st->print_cr("  <Not Available>");
2209     }
2210   }
2211 }
2212 
2213 #if defined(AMD64) || defined(IA32) || defined(X32)
2214 const char* search_string = "model name";
2215 #elif defined(SPARC)
2216 const char* search_string = "cpu";
2217 #else
2218 const char* search_string = "Processor";
2219 #endif
2220 
2221 // Parses the cpuinfo file for string representing the model name.
2222 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2223   FILE* fp = fopen("/proc/cpuinfo", "r");
2224   if (fp != NULL) {
2225     while (!feof(fp)) {
2226       char buf[256];
2227       if (fgets(buf, sizeof(buf), fp)) {
2228         char* start = strstr(buf, search_string);
2229         if (start != NULL) {
2230           char *ptr = start + strlen(search_string);
2231           char *end = buf + strlen(buf);
2232           while (ptr != end) {
2233              // skip whitespace and colon for the rest of the name.
2234              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2235                break;
2236              }
2237              ptr++;
2238           }
2239           if (ptr != end) {
2240             // reasonable string, get rid of newline and keep the rest
2241             char* nl = strchr(buf, '\n');
2242             if (nl != NULL) *nl = '\0';
2243             strncpy(cpuinfo, ptr, length);
2244             fclose(fp);
2245             return;
2246           }
2247         }
2248       }
2249     }
2250     fclose(fp);
2251   }
2252   // cpuinfo not found or parsing failed, just print generic string.  The entire
2253   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2254 #if defined(AMD64)
2255   strncpy(cpuinfo, "x86_64", length);
2256 #elif defined(IA32)
2257   strncpy(cpuinfo, "x86_32", length);
2258 #elif defined(IA64)
2259   strncpy(cpuinfo, "IA64", length);
2260 #elif defined(SPARC)
2261   strncpy(cpuinfo, "sparcv9", length);
2262 #elif defined(AARCH64)
2263   strncpy(cpuinfo, "AArch64", length);
2264 #elif defined(ARM)
2265   strncpy(cpuinfo, "ARM", length);
2266 #elif defined(PPC)
2267   strncpy(cpuinfo, "PPC64", length);
2268 #elif defined(ZERO_LIBARCH)
2269   strncpy(cpuinfo, ZERO_LIBARCH, length);
2270 #else
2271   strncpy(cpuinfo, "unknown", length);
2272 #endif
2273 }
2274 
2275 void os::print_siginfo(outputStream* st, void* siginfo) {
2276   const siginfo_t* si = (const siginfo_t*)siginfo;
2277 
2278   os::Posix::print_siginfo_brief(st, si);
2279 #if INCLUDE_CDS
2280   if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2281       UseSharedSpaces) {
2282     FileMapInfo* mapinfo = FileMapInfo::current_info();
2283     if (mapinfo->is_in_shared_space(si->si_addr)) {
2284       st->print("\n\nError accessing class data sharing archive."   \
2285                 " Mapped file inaccessible during execution, "      \
2286                 " possible disk/network problem.");
2287     }
2288   }
2289 #endif
2290   st->cr();
2291 }
2292 
2293 
2294 static void print_signal_handler(outputStream* st, int sig,
2295                                  char* buf, size_t buflen);
2296 
2297 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2298   st->print_cr("Signal Handlers:");
2299   print_signal_handler(st, SIGSEGV, buf, buflen);
2300   print_signal_handler(st, SIGBUS , buf, buflen);
2301   print_signal_handler(st, SIGFPE , buf, buflen);
2302   print_signal_handler(st, SIGPIPE, buf, buflen);
2303   print_signal_handler(st, SIGXFSZ, buf, buflen);
2304   print_signal_handler(st, SIGILL , buf, buflen);
2305   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2306   print_signal_handler(st, SR_signum, buf, buflen);
2307   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2308   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2309   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2310   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2311 #if defined(PPC64)
2312   print_signal_handler(st, SIGTRAP, buf, buflen);
2313 #endif
2314 }
2315 
2316 static char saved_jvm_path[MAXPATHLEN] = {0};
2317 
2318 // Find the full path to the current module, libjvm.so
2319 void os::jvm_path(char *buf, jint buflen) {
2320   // Error checking.
2321   if (buflen < MAXPATHLEN) {
2322     assert(false, "must use a large-enough buffer");
2323     buf[0] = '\0';
2324     return;
2325   }
2326   // Lazy resolve the path to current module.
2327   if (saved_jvm_path[0] != 0) {
2328     strcpy(buf, saved_jvm_path);
2329     return;
2330   }
2331 
2332   char dli_fname[MAXPATHLEN];
2333   bool ret = dll_address_to_library_name(
2334                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2335                                          dli_fname, sizeof(dli_fname), NULL);
2336   assert(ret, "cannot locate libjvm");
2337   char *rp = NULL;
2338   if (ret && dli_fname[0] != '\0') {
2339     rp = realpath(dli_fname, buf);
2340   }
2341   if (rp == NULL) {
2342     return;
2343   }
2344 
2345   if (Arguments::sun_java_launcher_is_altjvm()) {
2346     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2347     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2348     // If "/jre/lib/" appears at the right place in the string, then
2349     // assume we are installed in a JDK and we're done. Otherwise, check
2350     // for a JAVA_HOME environment variable and fix up the path so it
2351     // looks like libjvm.so is installed there (append a fake suffix
2352     // hotspot/libjvm.so).
2353     const char *p = buf + strlen(buf) - 1;
2354     for (int count = 0; p > buf && count < 5; ++count) {
2355       for (--p; p > buf && *p != '/'; --p)
2356         /* empty */ ;
2357     }
2358 
2359     if (strncmp(p, "/jre/lib/", 9) != 0) {
2360       // Look for JAVA_HOME in the environment.
2361       char* java_home_var = ::getenv("JAVA_HOME");
2362       if (java_home_var != NULL && java_home_var[0] != 0) {
2363         char* jrelib_p;
2364         int len;
2365 
2366         // Check the current module name "libjvm.so".
2367         p = strrchr(buf, '/');
2368         if (p == NULL) {
2369           return;
2370         }
2371         assert(strstr(p, "/libjvm") == p, "invalid library name");
2372 
2373         rp = realpath(java_home_var, buf);
2374         if (rp == NULL) {
2375           return;
2376         }
2377 
2378         // determine if this is a legacy image or modules image
2379         // modules image doesn't have "jre" subdirectory
2380         len = strlen(buf);
2381         assert(len < buflen, "Ran out of buffer room");
2382         jrelib_p = buf + len;
2383         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2384         if (0 != access(buf, F_OK)) {
2385           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2386         }
2387 
2388         if (0 == access(buf, F_OK)) {
2389           // Use current module name "libjvm.so"
2390           len = strlen(buf);
2391           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2392         } else {
2393           // Go back to path of .so
2394           rp = realpath(dli_fname, buf);
2395           if (rp == NULL) {
2396             return;
2397           }
2398         }
2399       }
2400     }
2401   }
2402 
2403   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2404   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2405 }
2406 
2407 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2408   // no prefix required, not even "_"
2409 }
2410 
2411 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2412   // no suffix required
2413 }
2414 
2415 ////////////////////////////////////////////////////////////////////////////////
2416 // sun.misc.Signal support
2417 
2418 static volatile jint sigint_count = 0;
2419 
2420 static void UserHandler(int sig, void *siginfo, void *context) {
2421   // 4511530 - sem_post is serialized and handled by the manager thread. When
2422   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2423   // don't want to flood the manager thread with sem_post requests.
2424   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2425     return;
2426   }
2427 
2428   // Ctrl-C is pressed during error reporting, likely because the error
2429   // handler fails to abort. Let VM die immediately.
2430   if (sig == SIGINT && is_error_reported()) {
2431     os::die();
2432   }
2433 
2434   os::signal_notify(sig);
2435 }
2436 
2437 void* os::user_handler() {
2438   return CAST_FROM_FN_PTR(void*, UserHandler);
2439 }
2440 
2441 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2442   struct timespec ts;
2443   // Semaphore's are always associated with CLOCK_REALTIME
2444   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2445   // see unpackTime for discussion on overflow checking
2446   if (sec >= MAX_SECS) {
2447     ts.tv_sec += MAX_SECS;
2448     ts.tv_nsec = 0;
2449   } else {
2450     ts.tv_sec += sec;
2451     ts.tv_nsec += nsec;
2452     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2453       ts.tv_nsec -= NANOSECS_PER_SEC;
2454       ++ts.tv_sec; // note: this must be <= max_secs
2455     }
2456   }
2457 
2458   return ts;
2459 }
2460 
2461 extern "C" {
2462   typedef void (*sa_handler_t)(int);
2463   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2464 }
2465 
2466 void* os::signal(int signal_number, void* handler) {
2467   struct sigaction sigAct, oldSigAct;
2468 
2469   sigfillset(&(sigAct.sa_mask));
2470   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2471   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2472 
2473   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2474     // -1 means registration failed
2475     return (void *)-1;
2476   }
2477 
2478   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2479 }
2480 
2481 void os::signal_raise(int signal_number) {
2482   ::raise(signal_number);
2483 }
2484 
2485 // The following code is moved from os.cpp for making this
2486 // code platform specific, which it is by its very nature.
2487 
2488 // Will be modified when max signal is changed to be dynamic
2489 int os::sigexitnum_pd() {
2490   return NSIG;
2491 }
2492 
2493 // a counter for each possible signal value
2494 static volatile jint pending_signals[NSIG+1] = { 0 };
2495 
2496 // Linux(POSIX) specific hand shaking semaphore.
2497 static sem_t sig_sem;
2498 static PosixSemaphore sr_semaphore;
2499 
2500 void os::signal_init_pd() {
2501   // Initialize signal structures
2502   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2503 
2504   // Initialize signal semaphore
2505   ::sem_init(&sig_sem, 0, 0);
2506 }
2507 
2508 void os::signal_notify(int sig) {
2509   Atomic::inc(&pending_signals[sig]);
2510   ::sem_post(&sig_sem);
2511 }
2512 
2513 static int check_pending_signals(bool wait) {
2514   Atomic::store(0, &sigint_count);
2515   for (;;) {
2516     for (int i = 0; i < NSIG + 1; i++) {
2517       jint n = pending_signals[i];
2518       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2519         return i;
2520       }
2521     }
2522     if (!wait) {
2523       return -1;
2524     }
2525     JavaThread *thread = JavaThread::current();
2526     ThreadBlockInVM tbivm(thread);
2527 
2528     bool threadIsSuspended;
2529     do {
2530       thread->set_suspend_equivalent();
2531       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2532       ::sem_wait(&sig_sem);
2533 
2534       // were we externally suspended while we were waiting?
2535       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2536       if (threadIsSuspended) {
2537         // The semaphore has been incremented, but while we were waiting
2538         // another thread suspended us. We don't want to continue running
2539         // while suspended because that would surprise the thread that
2540         // suspended us.
2541         ::sem_post(&sig_sem);
2542 
2543         thread->java_suspend_self();
2544       }
2545     } while (threadIsSuspended);
2546   }
2547 }
2548 
2549 int os::signal_lookup() {
2550   return check_pending_signals(false);
2551 }
2552 
2553 int os::signal_wait() {
2554   return check_pending_signals(true);
2555 }
2556 
2557 ////////////////////////////////////////////////////////////////////////////////
2558 // Virtual Memory
2559 
2560 int os::vm_page_size() {
2561   // Seems redundant as all get out
2562   assert(os::Linux::page_size() != -1, "must call os::init");
2563   return os::Linux::page_size();
2564 }
2565 
2566 // Solaris allocates memory by pages.
2567 int os::vm_allocation_granularity() {
2568   assert(os::Linux::page_size() != -1, "must call os::init");
2569   return os::Linux::page_size();
2570 }
2571 
2572 // Rationale behind this function:
2573 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2574 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2575 //  samples for JITted code. Here we create private executable mapping over the code cache
2576 //  and then we can use standard (well, almost, as mapping can change) way to provide
2577 //  info for the reporting script by storing timestamp and location of symbol
2578 void linux_wrap_code(char* base, size_t size) {
2579   static volatile jint cnt = 0;
2580 
2581   if (!UseOprofile) {
2582     return;
2583   }
2584 
2585   char buf[PATH_MAX+1];
2586   int num = Atomic::add(1, &cnt);
2587 
2588   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2589            os::get_temp_directory(), os::current_process_id(), num);
2590   unlink(buf);
2591 
2592   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2593 
2594   if (fd != -1) {
2595     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2596     if (rv != (off_t)-1) {
2597       if (::write(fd, "", 1) == 1) {
2598         mmap(base, size,
2599              PROT_READ|PROT_WRITE|PROT_EXEC,
2600              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2601       }
2602     }
2603     ::close(fd);
2604     unlink(buf);
2605   }
2606 }
2607 
2608 static bool recoverable_mmap_error(int err) {
2609   // See if the error is one we can let the caller handle. This
2610   // list of errno values comes from JBS-6843484. I can't find a
2611   // Linux man page that documents this specific set of errno
2612   // values so while this list currently matches Solaris, it may
2613   // change as we gain experience with this failure mode.
2614   switch (err) {
2615   case EBADF:
2616   case EINVAL:
2617   case ENOTSUP:
2618     // let the caller deal with these errors
2619     return true;
2620 
2621   default:
2622     // Any remaining errors on this OS can cause our reserved mapping
2623     // to be lost. That can cause confusion where different data
2624     // structures think they have the same memory mapped. The worst
2625     // scenario is if both the VM and a library think they have the
2626     // same memory mapped.
2627     return false;
2628   }
2629 }
2630 
2631 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2632                                     int err) {
2633   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2634           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2635           strerror(err), err);
2636 }
2637 
2638 static void warn_fail_commit_memory(char* addr, size_t size,
2639                                     size_t alignment_hint, bool exec,
2640                                     int err) {
2641   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2642           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2643           alignment_hint, exec, strerror(err), err);
2644 }
2645 
2646 // NOTE: Linux kernel does not really reserve the pages for us.
2647 //       All it does is to check if there are enough free pages
2648 //       left at the time of mmap(). This could be a potential
2649 //       problem.
2650 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2651   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2652   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2653                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2654   if (res != (uintptr_t) MAP_FAILED) {
2655     if (UseNUMAInterleaving) {
2656       numa_make_global(addr, size);
2657     }
2658     return 0;
2659   }
2660 
2661   int err = errno;  // save errno from mmap() call above
2662 
2663   if (!recoverable_mmap_error(err)) {
2664     warn_fail_commit_memory(addr, size, exec, err);
2665     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2666   }
2667 
2668   return err;
2669 }
2670 
2671 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2672   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2673 }
2674 
2675 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2676                                   const char* mesg) {
2677   assert(mesg != NULL, "mesg must be specified");
2678   int err = os::Linux::commit_memory_impl(addr, size, exec);
2679   if (err != 0) {
2680     // the caller wants all commit errors to exit with the specified mesg:
2681     warn_fail_commit_memory(addr, size, exec, err);
2682     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2683   }
2684 }
2685 
2686 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2687 #ifndef MAP_HUGETLB
2688   #define MAP_HUGETLB 0x40000
2689 #endif
2690 
2691 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2692 #ifndef MADV_HUGEPAGE
2693   #define MADV_HUGEPAGE 14
2694 #endif
2695 
2696 int os::Linux::commit_memory_impl(char* addr, size_t size,
2697                                   size_t alignment_hint, bool exec) {
2698   int err = os::Linux::commit_memory_impl(addr, size, exec);
2699   if (err == 0) {
2700     realign_memory(addr, size, alignment_hint);
2701   }
2702   return err;
2703 }
2704 
2705 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2706                           bool exec) {
2707   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2708 }
2709 
2710 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2711                                   size_t alignment_hint, bool exec,
2712                                   const char* mesg) {
2713   assert(mesg != NULL, "mesg must be specified");
2714   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2715   if (err != 0) {
2716     // the caller wants all commit errors to exit with the specified mesg:
2717     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2718     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2719   }
2720 }
2721 
2722 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2723   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2724     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2725     // be supported or the memory may already be backed by huge pages.
2726     ::madvise(addr, bytes, MADV_HUGEPAGE);
2727   }
2728 }
2729 
2730 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2731   // This method works by doing an mmap over an existing mmaping and effectively discarding
2732   // the existing pages. However it won't work for SHM-based large pages that cannot be
2733   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2734   // small pages on top of the SHM segment. This method always works for small pages, so we
2735   // allow that in any case.
2736   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2737     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2738   }
2739 }
2740 
2741 void os::numa_make_global(char *addr, size_t bytes) {
2742   Linux::numa_interleave_memory(addr, bytes);
2743 }
2744 
2745 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2746 // bind policy to MPOL_PREFERRED for the current thread.
2747 #define USE_MPOL_PREFERRED 0
2748 
2749 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2750   // To make NUMA and large pages more robust when both enabled, we need to ease
2751   // the requirements on where the memory should be allocated. MPOL_BIND is the
2752   // default policy and it will force memory to be allocated on the specified
2753   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2754   // the specified node, but will not force it. Using this policy will prevent
2755   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2756   // free large pages.
2757   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2758   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2759 }
2760 
2761 bool os::numa_topology_changed() { return false; }
2762 
2763 size_t os::numa_get_groups_num() {
2764   int max_node = Linux::numa_max_node();
2765   return max_node > 0 ? max_node + 1 : 1;
2766 }
2767 
2768 int os::numa_get_group_id() {
2769   int cpu_id = Linux::sched_getcpu();
2770   if (cpu_id != -1) {
2771     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2772     if (lgrp_id != -1) {
2773       return lgrp_id;
2774     }
2775   }
2776   return 0;
2777 }
2778 
2779 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2780   for (size_t i = 0; i < size; i++) {
2781     ids[i] = i;
2782   }
2783   return size;
2784 }
2785 
2786 bool os::get_page_info(char *start, page_info* info) {
2787   return false;
2788 }
2789 
2790 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2791                      page_info* page_found) {
2792   return end;
2793 }
2794 
2795 
2796 int os::Linux::sched_getcpu_syscall(void) {
2797   unsigned int cpu;
2798   int retval = -1;
2799 
2800 #if defined(IA32)
2801   #ifndef SYS_getcpu
2802     #define SYS_getcpu 318
2803   #endif
2804   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2805 #elif defined(AMD64)
2806 // Unfortunately we have to bring all these macros here from vsyscall.h
2807 // to be able to compile on old linuxes.
2808   #define __NR_vgetcpu 2
2809   #define VSYSCALL_START (-10UL << 20)
2810   #define VSYSCALL_SIZE 1024
2811   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2812   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2813   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2814   retval = vgetcpu(&cpu, NULL, NULL);
2815 #endif
2816 
2817   return (retval == -1) ? retval : cpu;
2818 }
2819 
2820 // Something to do with the numa-aware allocator needs these symbols
2821 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2822 extern "C" JNIEXPORT void numa_error(char *where) { }
2823 extern "C" JNIEXPORT int fork1() { return fork(); }
2824 
2825 
2826 // If we are running with libnuma version > 2, then we should
2827 // be trying to use symbols with versions 1.1
2828 // If we are running with earlier version, which did not have symbol versions,
2829 // we should use the base version.
2830 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2831   void *f = dlvsym(handle, name, "libnuma_1.1");
2832   if (f == NULL) {
2833     f = dlsym(handle, name);
2834   }
2835   return f;
2836 }
2837 
2838 bool os::Linux::libnuma_init() {
2839   // sched_getcpu() should be in libc.
2840   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2841                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2842 
2843   // If it's not, try a direct syscall.
2844   if (sched_getcpu() == -1) {
2845     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2846                                     (void*)&sched_getcpu_syscall));
2847   }
2848 
2849   if (sched_getcpu() != -1) { // Does it work?
2850     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2851     if (handle != NULL) {
2852       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2853                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2854       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2855                                        libnuma_dlsym(handle, "numa_max_node")));
2856       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2857                                         libnuma_dlsym(handle, "numa_available")));
2858       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2859                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2860       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2861                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2862       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2863                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2864 
2865 
2866       if (numa_available() != -1) {
2867         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2868         // Create a cpu -> node mapping
2869         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2870         rebuild_cpu_to_node_map();
2871         return true;
2872       }
2873     }
2874   }
2875   return false;
2876 }
2877 
2878 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2879 // The table is later used in get_node_by_cpu().
2880 void os::Linux::rebuild_cpu_to_node_map() {
2881   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2882                               // in libnuma (possible values are starting from 16,
2883                               // and continuing up with every other power of 2, but less
2884                               // than the maximum number of CPUs supported by kernel), and
2885                               // is a subject to change (in libnuma version 2 the requirements
2886                               // are more reasonable) we'll just hardcode the number they use
2887                               // in the library.
2888   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2889 
2890   size_t cpu_num = os::active_processor_count();
2891   size_t cpu_map_size = NCPUS / BitsPerCLong;
2892   size_t cpu_map_valid_size =
2893     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2894 
2895   cpu_to_node()->clear();
2896   cpu_to_node()->at_grow(cpu_num - 1);
2897   size_t node_num = numa_get_groups_num();
2898 
2899   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2900   for (size_t i = 0; i < node_num; i++) {
2901     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2902       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2903         if (cpu_map[j] != 0) {
2904           for (size_t k = 0; k < BitsPerCLong; k++) {
2905             if (cpu_map[j] & (1UL << k)) {
2906               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2907             }
2908           }
2909         }
2910       }
2911     }
2912   }
2913   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2914 }
2915 
2916 int os::Linux::get_node_by_cpu(int cpu_id) {
2917   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2918     return cpu_to_node()->at(cpu_id);
2919   }
2920   return -1;
2921 }
2922 
2923 GrowableArray<int>* os::Linux::_cpu_to_node;
2924 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2925 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2926 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2927 os::Linux::numa_available_func_t os::Linux::_numa_available;
2928 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2929 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2930 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2931 unsigned long* os::Linux::_numa_all_nodes;
2932 
2933 bool os::pd_uncommit_memory(char* addr, size_t size) {
2934   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2935                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2936   return res  != (uintptr_t) MAP_FAILED;
2937 }
2938 
2939 static address get_stack_commited_bottom(address bottom, size_t size) {
2940   address nbot = bottom;
2941   address ntop = bottom + size;
2942 
2943   size_t page_sz = os::vm_page_size();
2944   unsigned pages = size / page_sz;
2945 
2946   unsigned char vec[1];
2947   unsigned imin = 1, imax = pages + 1, imid;
2948   int mincore_return_value = 0;
2949 
2950   assert(imin <= imax, "Unexpected page size");
2951 
2952   while (imin < imax) {
2953     imid = (imax + imin) / 2;
2954     nbot = ntop - (imid * page_sz);
2955 
2956     // Use a trick with mincore to check whether the page is mapped or not.
2957     // mincore sets vec to 1 if page resides in memory and to 0 if page
2958     // is swapped output but if page we are asking for is unmapped
2959     // it returns -1,ENOMEM
2960     mincore_return_value = mincore(nbot, page_sz, vec);
2961 
2962     if (mincore_return_value == -1) {
2963       // Page is not mapped go up
2964       // to find first mapped page
2965       if (errno != EAGAIN) {
2966         assert(errno == ENOMEM, "Unexpected mincore errno");
2967         imax = imid;
2968       }
2969     } else {
2970       // Page is mapped go down
2971       // to find first not mapped page
2972       imin = imid + 1;
2973     }
2974   }
2975 
2976   nbot = nbot + page_sz;
2977 
2978   // Adjust stack bottom one page up if last checked page is not mapped
2979   if (mincore_return_value == -1) {
2980     nbot = nbot + page_sz;
2981   }
2982 
2983   return nbot;
2984 }
2985 
2986 
2987 // Linux uses a growable mapping for the stack, and if the mapping for
2988 // the stack guard pages is not removed when we detach a thread the
2989 // stack cannot grow beyond the pages where the stack guard was
2990 // mapped.  If at some point later in the process the stack expands to
2991 // that point, the Linux kernel cannot expand the stack any further
2992 // because the guard pages are in the way, and a segfault occurs.
2993 //
2994 // However, it's essential not to split the stack region by unmapping
2995 // a region (leaving a hole) that's already part of the stack mapping,
2996 // so if the stack mapping has already grown beyond the guard pages at
2997 // the time we create them, we have to truncate the stack mapping.
2998 // So, we need to know the extent of the stack mapping when
2999 // create_stack_guard_pages() is called.
3000 
3001 // We only need this for stacks that are growable: at the time of
3002 // writing thread stacks don't use growable mappings (i.e. those
3003 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3004 // only applies to the main thread.
3005 
3006 // If the (growable) stack mapping already extends beyond the point
3007 // where we're going to put our guard pages, truncate the mapping at
3008 // that point by munmap()ping it.  This ensures that when we later
3009 // munmap() the guard pages we don't leave a hole in the stack
3010 // mapping. This only affects the main/initial thread
3011 
3012 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3013   if (os::Linux::is_initial_thread()) {
3014     // As we manually grow stack up to bottom inside create_attached_thread(),
3015     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3016     // we don't need to do anything special.
3017     // Check it first, before calling heavy function.
3018     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3019     unsigned char vec[1];
3020 
3021     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3022       // Fallback to slow path on all errors, including EAGAIN
3023       stack_extent = (uintptr_t) get_stack_commited_bottom(
3024                                                            os::Linux::initial_thread_stack_bottom(),
3025                                                            (size_t)addr - stack_extent);
3026     }
3027 
3028     if (stack_extent < (uintptr_t)addr) {
3029       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3030     }
3031   }
3032 
3033   return os::commit_memory(addr, size, !ExecMem);
3034 }
3035 
3036 // If this is a growable mapping, remove the guard pages entirely by
3037 // munmap()ping them.  If not, just call uncommit_memory(). This only
3038 // affects the main/initial thread, but guard against future OS changes
3039 // It's safe to always unmap guard pages for initial thread because we
3040 // always place it right after end of the mapped region
3041 
3042 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3043   uintptr_t stack_extent, stack_base;
3044 
3045   if (os::Linux::is_initial_thread()) {
3046     return ::munmap(addr, size) == 0;
3047   }
3048 
3049   return os::uncommit_memory(addr, size);
3050 }
3051 
3052 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3053 // at 'requested_addr'. If there are existing memory mappings at the same
3054 // location, however, they will be overwritten. If 'fixed' is false,
3055 // 'requested_addr' is only treated as a hint, the return value may or
3056 // may not start from the requested address. Unlike Linux mmap(), this
3057 // function returns NULL to indicate failure.
3058 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3059   char * addr;
3060   int flags;
3061 
3062   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3063   if (fixed) {
3064     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3065     flags |= MAP_FIXED;
3066   }
3067 
3068   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3069   // touch an uncommitted page. Otherwise, the read/write might
3070   // succeed if we have enough swap space to back the physical page.
3071   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3072                        flags, -1, 0);
3073 
3074   return addr == MAP_FAILED ? NULL : addr;
3075 }
3076 
3077 static int anon_munmap(char * addr, size_t size) {
3078   return ::munmap(addr, size) == 0;
3079 }
3080 
3081 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3082                             size_t alignment_hint) {
3083   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3084 }
3085 
3086 bool os::pd_release_memory(char* addr, size_t size) {
3087   return anon_munmap(addr, size);
3088 }
3089 
3090 static bool linux_mprotect(char* addr, size_t size, int prot) {
3091   // Linux wants the mprotect address argument to be page aligned.
3092   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3093 
3094   // According to SUSv3, mprotect() should only be used with mappings
3095   // established by mmap(), and mmap() always maps whole pages. Unaligned
3096   // 'addr' likely indicates problem in the VM (e.g. trying to change
3097   // protection of malloc'ed or statically allocated memory). Check the
3098   // caller if you hit this assert.
3099   assert(addr == bottom, "sanity check");
3100 
3101   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3102   return ::mprotect(bottom, size, prot) == 0;
3103 }
3104 
3105 // Set protections specified
3106 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3107                         bool is_committed) {
3108   unsigned int p = 0;
3109   switch (prot) {
3110   case MEM_PROT_NONE: p = PROT_NONE; break;
3111   case MEM_PROT_READ: p = PROT_READ; break;
3112   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3113   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3114   default:
3115     ShouldNotReachHere();
3116   }
3117   // is_committed is unused.
3118   return linux_mprotect(addr, bytes, p);
3119 }
3120 
3121 bool os::guard_memory(char* addr, size_t size) {
3122   return linux_mprotect(addr, size, PROT_NONE);
3123 }
3124 
3125 bool os::unguard_memory(char* addr, size_t size) {
3126   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3127 }
3128 
3129 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3130                                                     size_t page_size) {
3131   bool result = false;
3132   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3133                  MAP_ANONYMOUS|MAP_PRIVATE,
3134                  -1, 0);
3135   if (p != MAP_FAILED) {
3136     void *aligned_p = align_ptr_up(p, page_size);
3137 
3138     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3139 
3140     munmap(p, page_size * 2);
3141   }
3142 
3143   if (warn && !result) {
3144     warning("TransparentHugePages is not supported by the operating system.");
3145   }
3146 
3147   return result;
3148 }
3149 
3150 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3151   bool result = false;
3152   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3153                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3154                  -1, 0);
3155 
3156   if (p != MAP_FAILED) {
3157     // We don't know if this really is a huge page or not.
3158     FILE *fp = fopen("/proc/self/maps", "r");
3159     if (fp) {
3160       while (!feof(fp)) {
3161         char chars[257];
3162         long x = 0;
3163         if (fgets(chars, sizeof(chars), fp)) {
3164           if (sscanf(chars, "%lx-%*x", &x) == 1
3165               && x == (long)p) {
3166             if (strstr (chars, "hugepage")) {
3167               result = true;
3168               break;
3169             }
3170           }
3171         }
3172       }
3173       fclose(fp);
3174     }
3175     munmap(p, page_size);
3176   }
3177 
3178   if (warn && !result) {
3179     warning("HugeTLBFS is not supported by the operating system.");
3180   }
3181 
3182   return result;
3183 }
3184 
3185 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3186 //
3187 // From the coredump_filter documentation:
3188 //
3189 // - (bit 0) anonymous private memory
3190 // - (bit 1) anonymous shared memory
3191 // - (bit 2) file-backed private memory
3192 // - (bit 3) file-backed shared memory
3193 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3194 //           effective only if the bit 2 is cleared)
3195 // - (bit 5) hugetlb private memory
3196 // - (bit 6) hugetlb shared memory
3197 //
3198 static void set_coredump_filter(void) {
3199   FILE *f;
3200   long cdm;
3201 
3202   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3203     return;
3204   }
3205 
3206   if (fscanf(f, "%lx", &cdm) != 1) {
3207     fclose(f);
3208     return;
3209   }
3210 
3211   rewind(f);
3212 
3213   if ((cdm & LARGEPAGES_BIT) == 0) {
3214     cdm |= LARGEPAGES_BIT;
3215     fprintf(f, "%#lx", cdm);
3216   }
3217 
3218   fclose(f);
3219 }
3220 
3221 // Large page support
3222 
3223 static size_t _large_page_size = 0;
3224 
3225 size_t os::Linux::find_large_page_size() {
3226   size_t large_page_size = 0;
3227 
3228   // large_page_size on Linux is used to round up heap size. x86 uses either
3229   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3230   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3231   // page as large as 256M.
3232   //
3233   // Here we try to figure out page size by parsing /proc/meminfo and looking
3234   // for a line with the following format:
3235   //    Hugepagesize:     2048 kB
3236   //
3237   // If we can't determine the value (e.g. /proc is not mounted, or the text
3238   // format has been changed), we'll use the largest page size supported by
3239   // the processor.
3240 
3241 #ifndef ZERO
3242   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3243                      ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3244 #endif // ZERO
3245 
3246   FILE *fp = fopen("/proc/meminfo", "r");
3247   if (fp) {
3248     while (!feof(fp)) {
3249       int x = 0;
3250       char buf[16];
3251       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3252         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3253           large_page_size = x * K;
3254           break;
3255         }
3256       } else {
3257         // skip to next line
3258         for (;;) {
3259           int ch = fgetc(fp);
3260           if (ch == EOF || ch == (int)'\n') break;
3261         }
3262       }
3263     }
3264     fclose(fp);
3265   }
3266 
3267   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3268     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3269             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3270             proper_unit_for_byte_size(large_page_size));
3271   }
3272 
3273   return large_page_size;
3274 }
3275 
3276 size_t os::Linux::setup_large_page_size() {
3277   _large_page_size = Linux::find_large_page_size();
3278   const size_t default_page_size = (size_t)Linux::page_size();
3279   if (_large_page_size > default_page_size) {
3280     _page_sizes[0] = _large_page_size;
3281     _page_sizes[1] = default_page_size;
3282     _page_sizes[2] = 0;
3283   }
3284 
3285   return _large_page_size;
3286 }
3287 
3288 bool os::Linux::setup_large_page_type(size_t page_size) {
3289   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3290       FLAG_IS_DEFAULT(UseSHM) &&
3291       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3292 
3293     // The type of large pages has not been specified by the user.
3294 
3295     // Try UseHugeTLBFS and then UseSHM.
3296     UseHugeTLBFS = UseSHM = true;
3297 
3298     // Don't try UseTransparentHugePages since there are known
3299     // performance issues with it turned on. This might change in the future.
3300     UseTransparentHugePages = false;
3301   }
3302 
3303   if (UseTransparentHugePages) {
3304     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3305     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3306       UseHugeTLBFS = false;
3307       UseSHM = false;
3308       return true;
3309     }
3310     UseTransparentHugePages = false;
3311   }
3312 
3313   if (UseHugeTLBFS) {
3314     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3315     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3316       UseSHM = false;
3317       return true;
3318     }
3319     UseHugeTLBFS = false;
3320   }
3321 
3322   return UseSHM;
3323 }
3324 
3325 void os::large_page_init() {
3326   if (!UseLargePages &&
3327       !UseTransparentHugePages &&
3328       !UseHugeTLBFS &&
3329       !UseSHM) {
3330     // Not using large pages.
3331     return;
3332   }
3333 
3334   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3335     // The user explicitly turned off large pages.
3336     // Ignore the rest of the large pages flags.
3337     UseTransparentHugePages = false;
3338     UseHugeTLBFS = false;
3339     UseSHM = false;
3340     return;
3341   }
3342 
3343   size_t large_page_size = Linux::setup_large_page_size();
3344   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3345 
3346   set_coredump_filter();
3347 }
3348 
3349 #ifndef SHM_HUGETLB
3350   #define SHM_HUGETLB 04000
3351 #endif
3352 
3353 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3354                                             char* req_addr, bool exec) {
3355   // "exec" is passed in but not used.  Creating the shared image for
3356   // the code cache doesn't have an SHM_X executable permission to check.
3357   assert(UseLargePages && UseSHM, "only for SHM large pages");
3358   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3359 
3360   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3361     return NULL; // Fallback to small pages.
3362   }
3363 
3364   key_t key = IPC_PRIVATE;
3365   char *addr;
3366 
3367   bool warn_on_failure = UseLargePages &&
3368                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3369                          !FLAG_IS_DEFAULT(UseSHM) ||
3370                          !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3371   char msg[128];
3372 
3373   // Create a large shared memory region to attach to based on size.
3374   // Currently, size is the total size of the heap
3375   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3376   if (shmid == -1) {
3377     // Possible reasons for shmget failure:
3378     // 1. shmmax is too small for Java heap.
3379     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3380     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3381     // 2. not enough large page memory.
3382     //    > check available large pages: cat /proc/meminfo
3383     //    > increase amount of large pages:
3384     //          echo new_value > /proc/sys/vm/nr_hugepages
3385     //      Note 1: different Linux may use different name for this property,
3386     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3387     //      Note 2: it's possible there's enough physical memory available but
3388     //            they are so fragmented after a long run that they can't
3389     //            coalesce into large pages. Try to reserve large pages when
3390     //            the system is still "fresh".
3391     if (warn_on_failure) {
3392       jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3393       warning("%s", msg);
3394     }
3395     return NULL;
3396   }
3397 
3398   // attach to the region
3399   addr = (char*)shmat(shmid, req_addr, 0);
3400   int err = errno;
3401 
3402   // Remove shmid. If shmat() is successful, the actual shared memory segment
3403   // will be deleted when it's detached by shmdt() or when the process
3404   // terminates. If shmat() is not successful this will remove the shared
3405   // segment immediately.
3406   shmctl(shmid, IPC_RMID, NULL);
3407 
3408   if ((intptr_t)addr == -1) {
3409     if (warn_on_failure) {
3410       jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3411       warning("%s", msg);
3412     }
3413     return NULL;
3414   }
3415 
3416   return addr;
3417 }
3418 
3419 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3420                                         int error) {
3421   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3422 
3423   bool warn_on_failure = UseLargePages &&
3424       (!FLAG_IS_DEFAULT(UseLargePages) ||
3425        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3426        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3427 
3428   if (warn_on_failure) {
3429     char msg[128];
3430     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3431                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3432     warning("%s", msg);
3433   }
3434 }
3435 
3436 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3437                                                         char* req_addr,
3438                                                         bool exec) {
3439   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3440   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3441   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3442 
3443   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3444   char* addr = (char*)::mmap(req_addr, bytes, prot,
3445                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3446                              -1, 0);
3447 
3448   if (addr == MAP_FAILED) {
3449     warn_on_large_pages_failure(req_addr, bytes, errno);
3450     return NULL;
3451   }
3452 
3453   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3454 
3455   return addr;
3456 }
3457 
3458 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3459 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3460 //   (req_addr != NULL) or with a given alignment.
3461 //  - bytes shall be a multiple of alignment.
3462 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3463 //  - alignment sets the alignment at which memory shall be allocated.
3464 //     It must be a multiple of allocation granularity.
3465 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3466 //  req_addr or NULL.
3467 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3468 
3469   size_t extra_size = bytes;
3470   if (req_addr == NULL && alignment > 0) {
3471     extra_size += alignment;
3472   }
3473 
3474   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3475     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3476     -1, 0);
3477   if (start == MAP_FAILED) {
3478     start = NULL;
3479   } else {
3480     if (req_addr != NULL) {
3481       if (start != req_addr) {
3482         ::munmap(start, extra_size);
3483         start = NULL;
3484       }
3485     } else {
3486       char* const start_aligned = (char*) align_ptr_up(start, alignment);
3487       char* const end_aligned = start_aligned + bytes;
3488       char* const end = start + extra_size;
3489       if (start_aligned > start) {
3490         ::munmap(start, start_aligned - start);
3491       }
3492       if (end_aligned < end) {
3493         ::munmap(end_aligned, end - end_aligned);
3494       }
3495       start = start_aligned;
3496     }
3497   }
3498   return start;
3499 
3500 }
3501 
3502 // Reserve memory using mmap(MAP_HUGETLB).
3503 //  - bytes shall be a multiple of alignment.
3504 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3505 //  - alignment sets the alignment at which memory shall be allocated.
3506 //     It must be a multiple of allocation granularity.
3507 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3508 //  req_addr or NULL.
3509 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3510                                                          size_t alignment,
3511                                                          char* req_addr,
3512                                                          bool exec) {
3513   size_t large_page_size = os::large_page_size();
3514   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3515 
3516   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3517   assert(is_size_aligned(bytes, alignment), "Must be");
3518 
3519   // First reserve - but not commit - the address range in small pages.
3520   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3521 
3522   if (start == NULL) {
3523     return NULL;
3524   }
3525 
3526   assert(is_ptr_aligned(start, alignment), "Must be");
3527 
3528   char* end = start + bytes;
3529 
3530   // Find the regions of the allocated chunk that can be promoted to large pages.
3531   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3532   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3533 
3534   size_t lp_bytes = lp_end - lp_start;
3535 
3536   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3537 
3538   if (lp_bytes == 0) {
3539     // The mapped region doesn't even span the start and the end of a large page.
3540     // Fall back to allocate a non-special area.
3541     ::munmap(start, end - start);
3542     return NULL;
3543   }
3544 
3545   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3546 
3547   void* result;
3548 
3549   // Commit small-paged leading area.
3550   if (start != lp_start) {
3551     result = ::mmap(start, lp_start - start, prot,
3552                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3553                     -1, 0);
3554     if (result == MAP_FAILED) {
3555       ::munmap(lp_start, end - lp_start);
3556       return NULL;
3557     }
3558   }
3559 
3560   // Commit large-paged area.
3561   result = ::mmap(lp_start, lp_bytes, prot,
3562                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3563                   -1, 0);
3564   if (result == MAP_FAILED) {
3565     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3566     // If the mmap above fails, the large pages region will be unmapped and we
3567     // have regions before and after with small pages. Release these regions.
3568     //
3569     // |  mapped  |  unmapped  |  mapped  |
3570     // ^          ^            ^          ^
3571     // start      lp_start     lp_end     end
3572     //
3573     ::munmap(start, lp_start - start);
3574     ::munmap(lp_end, end - lp_end);
3575     return NULL;
3576   }
3577 
3578   // Commit small-paged trailing area.
3579   if (lp_end != end) {
3580     result = ::mmap(lp_end, end - lp_end, prot,
3581                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3582                     -1, 0);
3583     if (result == MAP_FAILED) {
3584       ::munmap(start, lp_end - start);
3585       return NULL;
3586     }
3587   }
3588 
3589   return start;
3590 }
3591 
3592 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3593                                                    size_t alignment,
3594                                                    char* req_addr,
3595                                                    bool exec) {
3596   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3597   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3598   assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3599   assert(is_power_of_2(os::large_page_size()), "Must be");
3600   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3601 
3602   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3603     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3604   } else {
3605     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3606   }
3607 }
3608 
3609 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3610                                  char* req_addr, bool exec) {
3611   assert(UseLargePages, "only for large pages");
3612 
3613   char* addr;
3614   if (UseSHM) {
3615     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3616   } else {
3617     assert(UseHugeTLBFS, "must be");
3618     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3619   }
3620 
3621   if (addr != NULL) {
3622     if (UseNUMAInterleaving) {
3623       numa_make_global(addr, bytes);
3624     }
3625 
3626     // The memory is committed
3627     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3628   }
3629 
3630   return addr;
3631 }
3632 
3633 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3634   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3635   return shmdt(base) == 0;
3636 }
3637 
3638 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3639   return pd_release_memory(base, bytes);
3640 }
3641 
3642 bool os::release_memory_special(char* base, size_t bytes) {
3643   bool res;
3644   if (MemTracker::tracking_level() > NMT_minimal) {
3645     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3646     res = os::Linux::release_memory_special_impl(base, bytes);
3647     if (res) {
3648       tkr.record((address)base, bytes);
3649     }
3650 
3651   } else {
3652     res = os::Linux::release_memory_special_impl(base, bytes);
3653   }
3654   return res;
3655 }
3656 
3657 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3658   assert(UseLargePages, "only for large pages");
3659   bool res;
3660 
3661   if (UseSHM) {
3662     res = os::Linux::release_memory_special_shm(base, bytes);
3663   } else {
3664     assert(UseHugeTLBFS, "must be");
3665     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3666   }
3667   return res;
3668 }
3669 
3670 size_t os::large_page_size() {
3671   return _large_page_size;
3672 }
3673 
3674 // With SysV SHM the entire memory region must be allocated as shared
3675 // memory.
3676 // HugeTLBFS allows application to commit large page memory on demand.
3677 // However, when committing memory with HugeTLBFS fails, the region
3678 // that was supposed to be committed will lose the old reservation
3679 // and allow other threads to steal that memory region. Because of this
3680 // behavior we can't commit HugeTLBFS memory.
3681 bool os::can_commit_large_page_memory() {
3682   return UseTransparentHugePages;
3683 }
3684 
3685 bool os::can_execute_large_page_memory() {
3686   return UseTransparentHugePages || UseHugeTLBFS;
3687 }
3688 
3689 // Reserve memory at an arbitrary address, only if that area is
3690 // available (and not reserved for something else).
3691 
3692 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3693   const int max_tries = 10;
3694   char* base[max_tries];
3695   size_t size[max_tries];
3696   const size_t gap = 0x000000;
3697 
3698   // Assert only that the size is a multiple of the page size, since
3699   // that's all that mmap requires, and since that's all we really know
3700   // about at this low abstraction level.  If we need higher alignment,
3701   // we can either pass an alignment to this method or verify alignment
3702   // in one of the methods further up the call chain.  See bug 5044738.
3703   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3704 
3705   // Repeatedly allocate blocks until the block is allocated at the
3706   // right spot.
3707 
3708   // Linux mmap allows caller to pass an address as hint; give it a try first,
3709   // if kernel honors the hint then we can return immediately.
3710   char * addr = anon_mmap(requested_addr, bytes, false);
3711   if (addr == requested_addr) {
3712     return requested_addr;
3713   }
3714 
3715   if (addr != NULL) {
3716     // mmap() is successful but it fails to reserve at the requested address
3717     anon_munmap(addr, bytes);
3718   }
3719 
3720   int i;
3721   for (i = 0; i < max_tries; ++i) {
3722     base[i] = reserve_memory(bytes);
3723 
3724     if (base[i] != NULL) {
3725       // Is this the block we wanted?
3726       if (base[i] == requested_addr) {
3727         size[i] = bytes;
3728         break;
3729       }
3730 
3731       // Does this overlap the block we wanted? Give back the overlapped
3732       // parts and try again.
3733 
3734       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3735       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3736         unmap_memory(base[i], top_overlap);
3737         base[i] += top_overlap;
3738         size[i] = bytes - top_overlap;
3739       } else {
3740         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3741         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3742           unmap_memory(requested_addr, bottom_overlap);
3743           size[i] = bytes - bottom_overlap;
3744         } else {
3745           size[i] = bytes;
3746         }
3747       }
3748     }
3749   }
3750 
3751   // Give back the unused reserved pieces.
3752 
3753   for (int j = 0; j < i; ++j) {
3754     if (base[j] != NULL) {
3755       unmap_memory(base[j], size[j]);
3756     }
3757   }
3758 
3759   if (i < max_tries) {
3760     return requested_addr;
3761   } else {
3762     return NULL;
3763   }
3764 }
3765 
3766 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3767   return ::read(fd, buf, nBytes);
3768 }
3769 
3770 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3771   return ::pread(fd, buf, nBytes, offset);
3772 }
3773 
3774 // Short sleep, direct OS call.
3775 //
3776 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3777 // sched_yield(2) will actually give up the CPU:
3778 //
3779 //   * Alone on this pariticular CPU, keeps running.
3780 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3781 //     (pre 2.6.39).
3782 //
3783 // So calling this with 0 is an alternative.
3784 //
3785 void os::naked_short_sleep(jlong ms) {
3786   struct timespec req;
3787 
3788   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3789   req.tv_sec = 0;
3790   if (ms > 0) {
3791     req.tv_nsec = (ms % 1000) * 1000000;
3792   } else {
3793     req.tv_nsec = 1;
3794   }
3795 
3796   nanosleep(&req, NULL);
3797 
3798   return;
3799 }
3800 
3801 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3802 void os::infinite_sleep() {
3803   while (true) {    // sleep forever ...
3804     ::sleep(100);   // ... 100 seconds at a time
3805   }
3806 }
3807 
3808 // Used to convert frequent JVM_Yield() to nops
3809 bool os::dont_yield() {
3810   return DontYieldALot;
3811 }
3812 
3813 void os::naked_yield() {
3814   sched_yield();
3815 }
3816 
3817 ////////////////////////////////////////////////////////////////////////////////
3818 // thread priority support
3819 
3820 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3821 // only supports dynamic priority, static priority must be zero. For real-time
3822 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3823 // However, for large multi-threaded applications, SCHED_RR is not only slower
3824 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3825 // of 5 runs - Sep 2005).
3826 //
3827 // The following code actually changes the niceness of kernel-thread/LWP. It
3828 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3829 // not the entire user process, and user level threads are 1:1 mapped to kernel
3830 // threads. It has always been the case, but could change in the future. For
3831 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3832 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3833 
3834 int os::java_to_os_priority[CriticalPriority + 1] = {
3835   19,              // 0 Entry should never be used
3836 
3837    4,              // 1 MinPriority
3838    3,              // 2
3839    2,              // 3
3840 
3841    1,              // 4
3842    0,              // 5 NormPriority
3843   -1,              // 6
3844 
3845   -2,              // 7
3846   -3,              // 8
3847   -4,              // 9 NearMaxPriority
3848 
3849   -5,              // 10 MaxPriority
3850 
3851   -5               // 11 CriticalPriority
3852 };
3853 
3854 static int prio_init() {
3855   if (ThreadPriorityPolicy == 1) {
3856     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3857     // if effective uid is not root. Perhaps, a more elegant way of doing
3858     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3859     if (geteuid() != 0) {
3860       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3861         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3862       }
3863       ThreadPriorityPolicy = 0;
3864     }
3865   }
3866   if (UseCriticalJavaThreadPriority) {
3867     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3868   }
3869   return 0;
3870 }
3871 
3872 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3873   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3874 
3875   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3876   return (ret == 0) ? OS_OK : OS_ERR;
3877 }
3878 
3879 OSReturn os::get_native_priority(const Thread* const thread,
3880                                  int *priority_ptr) {
3881   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3882     *priority_ptr = java_to_os_priority[NormPriority];
3883     return OS_OK;
3884   }
3885 
3886   errno = 0;
3887   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3888   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3889 }
3890 
3891 // Hint to the underlying OS that a task switch would not be good.
3892 // Void return because it's a hint and can fail.
3893 void os::hint_no_preempt() {}
3894 
3895 ////////////////////////////////////////////////////////////////////////////////
3896 // suspend/resume support
3897 
3898 //  the low-level signal-based suspend/resume support is a remnant from the
3899 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3900 //  within hotspot. Now there is a single use-case for this:
3901 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3902 //      that runs in the watcher thread.
3903 //  The remaining code is greatly simplified from the more general suspension
3904 //  code that used to be used.
3905 //
3906 //  The protocol is quite simple:
3907 //  - suspend:
3908 //      - sends a signal to the target thread
3909 //      - polls the suspend state of the osthread using a yield loop
3910 //      - target thread signal handler (SR_handler) sets suspend state
3911 //        and blocks in sigsuspend until continued
3912 //  - resume:
3913 //      - sets target osthread state to continue
3914 //      - sends signal to end the sigsuspend loop in the SR_handler
3915 //
3916 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3917 
3918 static void resume_clear_context(OSThread *osthread) {
3919   osthread->set_ucontext(NULL);
3920   osthread->set_siginfo(NULL);
3921 }
3922 
3923 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3924                                  ucontext_t* context) {
3925   osthread->set_ucontext(context);
3926   osthread->set_siginfo(siginfo);
3927 }
3928 
3929 // Handler function invoked when a thread's execution is suspended or
3930 // resumed. We have to be careful that only async-safe functions are
3931 // called here (Note: most pthread functions are not async safe and
3932 // should be avoided.)
3933 //
3934 // Note: sigwait() is a more natural fit than sigsuspend() from an
3935 // interface point of view, but sigwait() prevents the signal hander
3936 // from being run. libpthread would get very confused by not having
3937 // its signal handlers run and prevents sigwait()'s use with the
3938 // mutex granting granting signal.
3939 //
3940 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3941 //
3942 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3943   // Save and restore errno to avoid confusing native code with EINTR
3944   // after sigsuspend.
3945   int old_errno = errno;
3946 
3947   Thread* thread = Thread::current();
3948   OSThread* osthread = thread->osthread();
3949   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3950 
3951   os::SuspendResume::State current = osthread->sr.state();
3952   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3953     suspend_save_context(osthread, siginfo, context);
3954 
3955     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3956     os::SuspendResume::State state = osthread->sr.suspended();
3957     if (state == os::SuspendResume::SR_SUSPENDED) {
3958       sigset_t suspend_set;  // signals for sigsuspend()
3959 
3960       // get current set of blocked signals and unblock resume signal
3961       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3962       sigdelset(&suspend_set, SR_signum);
3963 
3964       sr_semaphore.signal();
3965       // wait here until we are resumed
3966       while (1) {
3967         sigsuspend(&suspend_set);
3968 
3969         os::SuspendResume::State result = osthread->sr.running();
3970         if (result == os::SuspendResume::SR_RUNNING) {
3971           sr_semaphore.signal();
3972           break;
3973         }
3974       }
3975 
3976     } else if (state == os::SuspendResume::SR_RUNNING) {
3977       // request was cancelled, continue
3978     } else {
3979       ShouldNotReachHere();
3980     }
3981 
3982     resume_clear_context(osthread);
3983   } else if (current == os::SuspendResume::SR_RUNNING) {
3984     // request was cancelled, continue
3985   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3986     // ignore
3987   } else {
3988     // ignore
3989   }
3990 
3991   errno = old_errno;
3992 }
3993 
3994 
3995 static int SR_initialize() {
3996   struct sigaction act;
3997   char *s;
3998   // Get signal number to use for suspend/resume
3999   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4000     int sig = ::strtol(s, 0, 10);
4001     if (sig > 0 || sig < _NSIG) {
4002       SR_signum = sig;
4003     }
4004   }
4005 
4006   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4007          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4008 
4009   sigemptyset(&SR_sigset);
4010   sigaddset(&SR_sigset, SR_signum);
4011 
4012   // Set up signal handler for suspend/resume
4013   act.sa_flags = SA_RESTART|SA_SIGINFO;
4014   act.sa_handler = (void (*)(int)) SR_handler;
4015 
4016   // SR_signum is blocked by default.
4017   // 4528190 - We also need to block pthread restart signal (32 on all
4018   // supported Linux platforms). Note that LinuxThreads need to block
4019   // this signal for all threads to work properly. So we don't have
4020   // to use hard-coded signal number when setting up the mask.
4021   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4022 
4023   if (sigaction(SR_signum, &act, 0) == -1) {
4024     return -1;
4025   }
4026 
4027   // Save signal flag
4028   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4029   return 0;
4030 }
4031 
4032 static int sr_notify(OSThread* osthread) {
4033   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4034   assert_status(status == 0, status, "pthread_kill");
4035   return status;
4036 }
4037 
4038 // "Randomly" selected value for how long we want to spin
4039 // before bailing out on suspending a thread, also how often
4040 // we send a signal to a thread we want to resume
4041 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4042 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4043 
4044 // returns true on success and false on error - really an error is fatal
4045 // but this seems the normal response to library errors
4046 static bool do_suspend(OSThread* osthread) {
4047   assert(osthread->sr.is_running(), "thread should be running");
4048   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4049 
4050   // mark as suspended and send signal
4051   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4052     // failed to switch, state wasn't running?
4053     ShouldNotReachHere();
4054     return false;
4055   }
4056 
4057   if (sr_notify(osthread) != 0) {
4058     ShouldNotReachHere();
4059   }
4060 
4061   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4062   while (true) {
4063     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4064       break;
4065     } else {
4066       // timeout
4067       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4068       if (cancelled == os::SuspendResume::SR_RUNNING) {
4069         return false;
4070       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4071         // make sure that we consume the signal on the semaphore as well
4072         sr_semaphore.wait();
4073         break;
4074       } else {
4075         ShouldNotReachHere();
4076         return false;
4077       }
4078     }
4079   }
4080 
4081   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4082   return true;
4083 }
4084 
4085 static void do_resume(OSThread* osthread) {
4086   assert(osthread->sr.is_suspended(), "thread should be suspended");
4087   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4088 
4089   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4090     // failed to switch to WAKEUP_REQUEST
4091     ShouldNotReachHere();
4092     return;
4093   }
4094 
4095   while (true) {
4096     if (sr_notify(osthread) == 0) {
4097       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4098         if (osthread->sr.is_running()) {
4099           return;
4100         }
4101       }
4102     } else {
4103       ShouldNotReachHere();
4104     }
4105   }
4106 
4107   guarantee(osthread->sr.is_running(), "Must be running!");
4108 }
4109 
4110 ///////////////////////////////////////////////////////////////////////////////////
4111 // signal handling (except suspend/resume)
4112 
4113 // This routine may be used by user applications as a "hook" to catch signals.
4114 // The user-defined signal handler must pass unrecognized signals to this
4115 // routine, and if it returns true (non-zero), then the signal handler must
4116 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4117 // routine will never retun false (zero), but instead will execute a VM panic
4118 // routine kill the process.
4119 //
4120 // If this routine returns false, it is OK to call it again.  This allows
4121 // the user-defined signal handler to perform checks either before or after
4122 // the VM performs its own checks.  Naturally, the user code would be making
4123 // a serious error if it tried to handle an exception (such as a null check
4124 // or breakpoint) that the VM was generating for its own correct operation.
4125 //
4126 // This routine may recognize any of the following kinds of signals:
4127 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4128 // It should be consulted by handlers for any of those signals.
4129 //
4130 // The caller of this routine must pass in the three arguments supplied
4131 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4132 // field of the structure passed to sigaction().  This routine assumes that
4133 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4134 //
4135 // Note that the VM will print warnings if it detects conflicting signal
4136 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4137 //
4138 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4139                                                  siginfo_t* siginfo,
4140                                                  void* ucontext,
4141                                                  int abort_if_unrecognized);
4142 
4143 void signalHandler(int sig, siginfo_t* info, void* uc) {
4144   assert(info != NULL && uc != NULL, "it must be old kernel");
4145   int orig_errno = errno;  // Preserve errno value over signal handler.
4146   JVM_handle_linux_signal(sig, info, uc, true);
4147   errno = orig_errno;
4148 }
4149 
4150 
4151 // This boolean allows users to forward their own non-matching signals
4152 // to JVM_handle_linux_signal, harmlessly.
4153 bool os::Linux::signal_handlers_are_installed = false;
4154 
4155 // For signal-chaining
4156 struct sigaction os::Linux::sigact[MAXSIGNUM];
4157 unsigned int os::Linux::sigs = 0;
4158 bool os::Linux::libjsig_is_loaded = false;
4159 typedef struct sigaction *(*get_signal_t)(int);
4160 get_signal_t os::Linux::get_signal_action = NULL;
4161 
4162 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4163   struct sigaction *actp = NULL;
4164 
4165   if (libjsig_is_loaded) {
4166     // Retrieve the old signal handler from libjsig
4167     actp = (*get_signal_action)(sig);
4168   }
4169   if (actp == NULL) {
4170     // Retrieve the preinstalled signal handler from jvm
4171     actp = get_preinstalled_handler(sig);
4172   }
4173 
4174   return actp;
4175 }
4176 
4177 static bool call_chained_handler(struct sigaction *actp, int sig,
4178                                  siginfo_t *siginfo, void *context) {
4179   // Call the old signal handler
4180   if (actp->sa_handler == SIG_DFL) {
4181     // It's more reasonable to let jvm treat it as an unexpected exception
4182     // instead of taking the default action.
4183     return false;
4184   } else if (actp->sa_handler != SIG_IGN) {
4185     if ((actp->sa_flags & SA_NODEFER) == 0) {
4186       // automaticlly block the signal
4187       sigaddset(&(actp->sa_mask), sig);
4188     }
4189 
4190     sa_handler_t hand;
4191     sa_sigaction_t sa;
4192     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4193     // retrieve the chained handler
4194     if (siginfo_flag_set) {
4195       sa = actp->sa_sigaction;
4196     } else {
4197       hand = actp->sa_handler;
4198     }
4199 
4200     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4201       actp->sa_handler = SIG_DFL;
4202     }
4203 
4204     // try to honor the signal mask
4205     sigset_t oset;
4206     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4207 
4208     // call into the chained handler
4209     if (siginfo_flag_set) {
4210       (*sa)(sig, siginfo, context);
4211     } else {
4212       (*hand)(sig);
4213     }
4214 
4215     // restore the signal mask
4216     pthread_sigmask(SIG_SETMASK, &oset, 0);
4217   }
4218   // Tell jvm's signal handler the signal is taken care of.
4219   return true;
4220 }
4221 
4222 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4223   bool chained = false;
4224   // signal-chaining
4225   if (UseSignalChaining) {
4226     struct sigaction *actp = get_chained_signal_action(sig);
4227     if (actp != NULL) {
4228       chained = call_chained_handler(actp, sig, siginfo, context);
4229     }
4230   }
4231   return chained;
4232 }
4233 
4234 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4235   if ((((unsigned int)1 << sig) & sigs) != 0) {
4236     return &sigact[sig];
4237   }
4238   return NULL;
4239 }
4240 
4241 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4242   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4243   sigact[sig] = oldAct;
4244   sigs |= (unsigned int)1 << sig;
4245 }
4246 
4247 // for diagnostic
4248 int os::Linux::sigflags[MAXSIGNUM];
4249 
4250 int os::Linux::get_our_sigflags(int sig) {
4251   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4252   return sigflags[sig];
4253 }
4254 
4255 void os::Linux::set_our_sigflags(int sig, int flags) {
4256   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4257   sigflags[sig] = flags;
4258 }
4259 
4260 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4261   // Check for overwrite.
4262   struct sigaction oldAct;
4263   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4264 
4265   void* oldhand = oldAct.sa_sigaction
4266                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4267                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4268   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4269       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4270       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4271     if (AllowUserSignalHandlers || !set_installed) {
4272       // Do not overwrite; user takes responsibility to forward to us.
4273       return;
4274     } else if (UseSignalChaining) {
4275       // save the old handler in jvm
4276       save_preinstalled_handler(sig, oldAct);
4277       // libjsig also interposes the sigaction() call below and saves the
4278       // old sigaction on it own.
4279     } else {
4280       fatal("Encountered unexpected pre-existing sigaction handler "
4281             "%#lx for signal %d.", (long)oldhand, sig);
4282     }
4283   }
4284 
4285   struct sigaction sigAct;
4286   sigfillset(&(sigAct.sa_mask));
4287   sigAct.sa_handler = SIG_DFL;
4288   if (!set_installed) {
4289     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4290   } else {
4291     sigAct.sa_sigaction = signalHandler;
4292     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4293   }
4294   // Save flags, which are set by ours
4295   assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4296   sigflags[sig] = sigAct.sa_flags;
4297 
4298   int ret = sigaction(sig, &sigAct, &oldAct);
4299   assert(ret == 0, "check");
4300 
4301   void* oldhand2  = oldAct.sa_sigaction
4302                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4303                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4304   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4305 }
4306 
4307 // install signal handlers for signals that HotSpot needs to
4308 // handle in order to support Java-level exception handling.
4309 
4310 void os::Linux::install_signal_handlers() {
4311   if (!signal_handlers_are_installed) {
4312     signal_handlers_are_installed = true;
4313 
4314     // signal-chaining
4315     typedef void (*signal_setting_t)();
4316     signal_setting_t begin_signal_setting = NULL;
4317     signal_setting_t end_signal_setting = NULL;
4318     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4319                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4320     if (begin_signal_setting != NULL) {
4321       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4322                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4323       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4324                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4325       libjsig_is_loaded = true;
4326       assert(UseSignalChaining, "should enable signal-chaining");
4327     }
4328     if (libjsig_is_loaded) {
4329       // Tell libjsig jvm is setting signal handlers
4330       (*begin_signal_setting)();
4331     }
4332 
4333     set_signal_handler(SIGSEGV, true);
4334     set_signal_handler(SIGPIPE, true);
4335     set_signal_handler(SIGBUS, true);
4336     set_signal_handler(SIGILL, true);
4337     set_signal_handler(SIGFPE, true);
4338 #if defined(PPC64)
4339     set_signal_handler(SIGTRAP, true);
4340 #endif
4341     set_signal_handler(SIGXFSZ, true);
4342 
4343     if (libjsig_is_loaded) {
4344       // Tell libjsig jvm finishes setting signal handlers
4345       (*end_signal_setting)();
4346     }
4347 
4348     // We don't activate signal checker if libjsig is in place, we trust ourselves
4349     // and if UserSignalHandler is installed all bets are off.
4350     // Log that signal checking is off only if -verbose:jni is specified.
4351     if (CheckJNICalls) {
4352       if (libjsig_is_loaded) {
4353         if (PrintJNIResolving) {
4354           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4355         }
4356         check_signals = false;
4357       }
4358       if (AllowUserSignalHandlers) {
4359         if (PrintJNIResolving) {
4360           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4361         }
4362         check_signals = false;
4363       }
4364     }
4365   }
4366 }
4367 
4368 // This is the fastest way to get thread cpu time on Linux.
4369 // Returns cpu time (user+sys) for any thread, not only for current.
4370 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4371 // It might work on 2.6.10+ with a special kernel/glibc patch.
4372 // For reference, please, see IEEE Std 1003.1-2004:
4373 //   http://www.unix.org/single_unix_specification
4374 
4375 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4376   struct timespec tp;
4377   int rc = os::Linux::clock_gettime(clockid, &tp);
4378   assert(rc == 0, "clock_gettime is expected to return 0 code");
4379 
4380   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4381 }
4382 
4383 /////
4384 // glibc on Linux platform uses non-documented flag
4385 // to indicate, that some special sort of signal
4386 // trampoline is used.
4387 // We will never set this flag, and we should
4388 // ignore this flag in our diagnostic
4389 #ifdef SIGNIFICANT_SIGNAL_MASK
4390   #undef SIGNIFICANT_SIGNAL_MASK
4391 #endif
4392 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4393 
4394 static const char* get_signal_handler_name(address handler,
4395                                            char* buf, int buflen) {
4396   int offset;
4397   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4398   if (found) {
4399     // skip directory names
4400     const char *p1, *p2;
4401     p1 = buf;
4402     size_t len = strlen(os::file_separator());
4403     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4404     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4405   } else {
4406     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4407   }
4408   return buf;
4409 }
4410 
4411 static void print_signal_handler(outputStream* st, int sig,
4412                                  char* buf, size_t buflen) {
4413   struct sigaction sa;
4414 
4415   sigaction(sig, NULL, &sa);
4416 
4417   // See comment for SIGNIFICANT_SIGNAL_MASK define
4418   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4419 
4420   st->print("%s: ", os::exception_name(sig, buf, buflen));
4421 
4422   address handler = (sa.sa_flags & SA_SIGINFO)
4423     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4424     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4425 
4426   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4427     st->print("SIG_DFL");
4428   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4429     st->print("SIG_IGN");
4430   } else {
4431     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4432   }
4433 
4434   st->print(", sa_mask[0]=");
4435   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4436 
4437   address rh = VMError::get_resetted_sighandler(sig);
4438   // May be, handler was resetted by VMError?
4439   if (rh != NULL) {
4440     handler = rh;
4441     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4442   }
4443 
4444   st->print(", sa_flags=");
4445   os::Posix::print_sa_flags(st, sa.sa_flags);
4446 
4447   // Check: is it our handler?
4448   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4449       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4450     // It is our signal handler
4451     // check for flags, reset system-used one!
4452     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4453       st->print(
4454                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4455                 os::Linux::get_our_sigflags(sig));
4456     }
4457   }
4458   st->cr();
4459 }
4460 
4461 
4462 #define DO_SIGNAL_CHECK(sig)                      \
4463   do {                                            \
4464     if (!sigismember(&check_signal_done, sig)) {  \
4465       os::Linux::check_signal_handler(sig);       \
4466     }                                             \
4467   } while (0)
4468 
4469 // This method is a periodic task to check for misbehaving JNI applications
4470 // under CheckJNI, we can add any periodic checks here
4471 
4472 void os::run_periodic_checks() {
4473   if (check_signals == false) return;
4474 
4475   // SEGV and BUS if overridden could potentially prevent
4476   // generation of hs*.log in the event of a crash, debugging
4477   // such a case can be very challenging, so we absolutely
4478   // check the following for a good measure:
4479   DO_SIGNAL_CHECK(SIGSEGV);
4480   DO_SIGNAL_CHECK(SIGILL);
4481   DO_SIGNAL_CHECK(SIGFPE);
4482   DO_SIGNAL_CHECK(SIGBUS);
4483   DO_SIGNAL_CHECK(SIGPIPE);
4484   DO_SIGNAL_CHECK(SIGXFSZ);
4485 #if defined(PPC64)
4486   DO_SIGNAL_CHECK(SIGTRAP);
4487 #endif
4488 
4489   // ReduceSignalUsage allows the user to override these handlers
4490   // see comments at the very top and jvm_solaris.h
4491   if (!ReduceSignalUsage) {
4492     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4493     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4494     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4495     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4496   }
4497 
4498   DO_SIGNAL_CHECK(SR_signum);
4499   DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4500 }
4501 
4502 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4503 
4504 static os_sigaction_t os_sigaction = NULL;
4505 
4506 void os::Linux::check_signal_handler(int sig) {
4507   char buf[O_BUFLEN];
4508   address jvmHandler = NULL;
4509 
4510 
4511   struct sigaction act;
4512   if (os_sigaction == NULL) {
4513     // only trust the default sigaction, in case it has been interposed
4514     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4515     if (os_sigaction == NULL) return;
4516   }
4517 
4518   os_sigaction(sig, (struct sigaction*)NULL, &act);
4519 
4520 
4521   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4522 
4523   address thisHandler = (act.sa_flags & SA_SIGINFO)
4524     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4525     : CAST_FROM_FN_PTR(address, act.sa_handler);
4526 
4527 
4528   switch (sig) {
4529   case SIGSEGV:
4530   case SIGBUS:
4531   case SIGFPE:
4532   case SIGPIPE:
4533   case SIGILL:
4534   case SIGXFSZ:
4535     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4536     break;
4537 
4538   case SHUTDOWN1_SIGNAL:
4539   case SHUTDOWN2_SIGNAL:
4540   case SHUTDOWN3_SIGNAL:
4541   case BREAK_SIGNAL:
4542     jvmHandler = (address)user_handler();
4543     break;
4544 
4545   case INTERRUPT_SIGNAL:
4546     jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4547     break;
4548 
4549   default:
4550     if (sig == SR_signum) {
4551       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4552     } else {
4553       return;
4554     }
4555     break;
4556   }
4557 
4558   if (thisHandler != jvmHandler) {
4559     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4560     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4561     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4562     // No need to check this sig any longer
4563     sigaddset(&check_signal_done, sig);
4564     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4565     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4566       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4567                     exception_name(sig, buf, O_BUFLEN));
4568     }
4569   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4570     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4571     tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4572     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4573     // No need to check this sig any longer
4574     sigaddset(&check_signal_done, sig);
4575   }
4576 
4577   // Dump all the signal
4578   if (sigismember(&check_signal_done, sig)) {
4579     print_signal_handlers(tty, buf, O_BUFLEN);
4580   }
4581 }
4582 
4583 extern void report_error(char* file_name, int line_no, char* title,
4584                          char* format, ...);
4585 
4586 extern bool signal_name(int signo, char* buf, size_t len);
4587 
4588 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4589   if (0 < exception_code && exception_code <= SIGRTMAX) {
4590     // signal
4591     if (!signal_name(exception_code, buf, size)) {
4592       jio_snprintf(buf, size, "SIG%d", exception_code);
4593     }
4594     return buf;
4595   } else {
4596     return NULL;
4597   }
4598 }
4599 
4600 // this is called _before_ the most of global arguments have been parsed
4601 void os::init(void) {
4602   char dummy;   // used to get a guess on initial stack address
4603 //  first_hrtime = gethrtime();
4604 
4605   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4606 
4607   init_random(1234567);
4608 
4609   ThreadCritical::initialize();
4610 
4611   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4612   if (Linux::page_size() == -1) {
4613     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4614           strerror(errno));
4615   }
4616   init_page_sizes((size_t) Linux::page_size());
4617 
4618   Linux::initialize_system_info();
4619 
4620   // main_thread points to the aboriginal thread
4621   Linux::_main_thread = pthread_self();
4622 
4623   Linux::clock_init();
4624   initial_time_count = javaTimeNanos();
4625 
4626   // pthread_condattr initialization for monotonic clock
4627   int status;
4628   pthread_condattr_t* _condattr = os::Linux::condAttr();
4629   if ((status = pthread_condattr_init(_condattr)) != 0) {
4630     fatal("pthread_condattr_init: %s", strerror(status));
4631   }
4632   // Only set the clock if CLOCK_MONOTONIC is available
4633   if (os::supports_monotonic_clock()) {
4634     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4635       if (status == EINVAL) {
4636         warning("Unable to use monotonic clock with relative timed-waits" \
4637                 " - changes to the time-of-day clock may have adverse affects");
4638       } else {
4639         fatal("pthread_condattr_setclock: %s", strerror(status));
4640       }
4641     }
4642   }
4643   // else it defaults to CLOCK_REALTIME
4644 
4645   // If the pagesize of the VM is greater than 8K determine the appropriate
4646   // number of initial guard pages.  The user can change this with the
4647   // command line arguments, if needed.
4648   if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4649     StackYellowPages = 1;
4650     StackRedPages = 1;
4651 #if defined(IA32) || defined(IA64)
4652     StackReservedPages = 1;
4653 #else
4654     StackReservedPages = 0;
4655 #endif
4656     StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4657   }
4658 
4659   // retrieve entry point for pthread_setname_np
4660   Linux::_pthread_setname_np =
4661     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4662 
4663 }
4664 
4665 // To install functions for atexit system call
4666 extern "C" {
4667   static void perfMemory_exit_helper() {
4668     perfMemory_exit();
4669   }
4670 }
4671 
4672 // this is called _after_ the global arguments have been parsed
4673 jint os::init_2(void) {
4674   Linux::fast_thread_clock_init();
4675 
4676   // Allocate a single page and mark it as readable for safepoint polling
4677   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4678   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4679 
4680   os::set_polling_page(polling_page);
4681 
4682 #ifndef PRODUCT
4683   if (Verbose && PrintMiscellaneous) {
4684     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4685                (intptr_t)polling_page);
4686   }
4687 #endif
4688 
4689   if (!UseMembar) {
4690     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4691     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4692     os::set_memory_serialize_page(mem_serialize_page);
4693 
4694 #ifndef PRODUCT
4695     if (Verbose && PrintMiscellaneous) {
4696       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n",
4697                  (intptr_t)mem_serialize_page);
4698     }
4699 #endif
4700   }
4701 
4702   // initialize suspend/resume support - must do this before signal_sets_init()
4703   if (SR_initialize() != 0) {
4704     perror("SR_initialize failed");
4705     return JNI_ERR;
4706   }
4707 
4708   Linux::signal_sets_init();
4709   Linux::install_signal_handlers();
4710 
4711   // Check minimum allowable stack size for thread creation and to initialize
4712   // the java system classes, including StackOverflowError - depends on page
4713   // size.  Add a page for compiler2 recursion in main thread.
4714   // Add in 2*BytesPerWord times page size to account for VM stack during
4715   // class initialization depending on 32 or 64 bit VM.
4716   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4717                                       (size_t)(StackReservedPages+StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4718                                       (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4719 
4720   size_t threadStackSizeInBytes = ThreadStackSize * K;
4721   if (threadStackSizeInBytes != 0 &&
4722       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4723     tty->print_cr("\nThe stack size specified is too small, "
4724                   "Specify at least " SIZE_FORMAT "k",
4725                   os::Linux::min_stack_allowed/ K);
4726     return JNI_ERR;
4727   }
4728 
4729   // Make the stack size a multiple of the page size so that
4730   // the yellow/red zones can be guarded.
4731   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4732                                                 vm_page_size()));
4733 
4734   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4735 
4736 #if defined(IA32)
4737   workaround_expand_exec_shield_cs_limit();
4738 #endif
4739 
4740   Linux::libpthread_init();
4741   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4742     tty->print_cr("[HotSpot is running with %s, %s]\n",
4743                   Linux::glibc_version(), Linux::libpthread_version());
4744   }
4745 
4746   if (UseNUMA) {
4747     if (!Linux::libnuma_init()) {
4748       UseNUMA = false;
4749     } else {
4750       if ((Linux::numa_max_node() < 1)) {
4751         // There's only one node(they start from 0), disable NUMA.
4752         UseNUMA = false;
4753       }
4754     }
4755     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4756     // we can make the adaptive lgrp chunk resizing work. If the user specified
4757     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4758     // disable adaptive resizing.
4759     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4760       if (FLAG_IS_DEFAULT(UseNUMA)) {
4761         UseNUMA = false;
4762       } else {
4763         if (FLAG_IS_DEFAULT(UseLargePages) &&
4764             FLAG_IS_DEFAULT(UseSHM) &&
4765             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4766           UseLargePages = false;
4767         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4768           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4769           UseAdaptiveSizePolicy = false;
4770           UseAdaptiveNUMAChunkSizing = false;
4771         }
4772       }
4773     }
4774     if (!UseNUMA && ForceNUMA) {
4775       UseNUMA = true;
4776     }
4777   }
4778 
4779   if (MaxFDLimit) {
4780     // set the number of file descriptors to max. print out error
4781     // if getrlimit/setrlimit fails but continue regardless.
4782     struct rlimit nbr_files;
4783     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4784     if (status != 0) {
4785       if (PrintMiscellaneous && (Verbose || WizardMode)) {
4786         perror("os::init_2 getrlimit failed");
4787       }
4788     } else {
4789       nbr_files.rlim_cur = nbr_files.rlim_max;
4790       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4791       if (status != 0) {
4792         if (PrintMiscellaneous && (Verbose || WizardMode)) {
4793           perror("os::init_2 setrlimit failed");
4794         }
4795       }
4796     }
4797   }
4798 
4799   // Initialize lock used to serialize thread creation (see os::create_thread)
4800   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4801 
4802   // at-exit methods are called in the reverse order of their registration.
4803   // atexit functions are called on return from main or as a result of a
4804   // call to exit(3C). There can be only 32 of these functions registered
4805   // and atexit() does not set errno.
4806 
4807   if (PerfAllowAtExitRegistration) {
4808     // only register atexit functions if PerfAllowAtExitRegistration is set.
4809     // atexit functions can be delayed until process exit time, which
4810     // can be problematic for embedded VM situations. Embedded VMs should
4811     // call DestroyJavaVM() to assure that VM resources are released.
4812 
4813     // note: perfMemory_exit_helper atexit function may be removed in
4814     // the future if the appropriate cleanup code can be added to the
4815     // VM_Exit VMOperation's doit method.
4816     if (atexit(perfMemory_exit_helper) != 0) {
4817       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4818     }
4819   }
4820 
4821   // initialize thread priority policy
4822   prio_init();
4823 
4824   return JNI_OK;
4825 }
4826 
4827 // Mark the polling page as unreadable
4828 void os::make_polling_page_unreadable(void) {
4829   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4830     fatal("Could not disable polling page");
4831   }
4832 }
4833 
4834 // Mark the polling page as readable
4835 void os::make_polling_page_readable(void) {
4836   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4837     fatal("Could not enable polling page");
4838   }
4839 }
4840 
4841 int os::active_processor_count() {
4842   // Linux doesn't yet have a (official) notion of processor sets,
4843   // so just return the number of online processors.
4844   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4845   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4846   return online_cpus;
4847 }
4848 
4849 void os::set_native_thread_name(const char *name) {
4850   if (Linux::_pthread_setname_np) {
4851     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4852     snprintf(buf, sizeof(buf), "%s", name);
4853     buf[sizeof(buf) - 1] = '\0';
4854     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4855     // ERANGE should not happen; all other errors should just be ignored.
4856     assert(rc != ERANGE, "pthread_setname_np failed");
4857   }
4858 }
4859 
4860 bool os::distribute_processes(uint length, uint* distribution) {
4861   // Not yet implemented.
4862   return false;
4863 }
4864 
4865 bool os::bind_to_processor(uint processor_id) {
4866   // Not yet implemented.
4867   return false;
4868 }
4869 
4870 ///
4871 
4872 void os::SuspendedThreadTask::internal_do_task() {
4873   if (do_suspend(_thread->osthread())) {
4874     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4875     do_task(context);
4876     do_resume(_thread->osthread());
4877   }
4878 }
4879 
4880 class PcFetcher : public os::SuspendedThreadTask {
4881  public:
4882   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4883   ExtendedPC result();
4884  protected:
4885   void do_task(const os::SuspendedThreadTaskContext& context);
4886  private:
4887   ExtendedPC _epc;
4888 };
4889 
4890 ExtendedPC PcFetcher::result() {
4891   guarantee(is_done(), "task is not done yet.");
4892   return _epc;
4893 }
4894 
4895 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4896   Thread* thread = context.thread();
4897   OSThread* osthread = thread->osthread();
4898   if (osthread->ucontext() != NULL) {
4899     _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4900   } else {
4901     // NULL context is unexpected, double-check this is the VMThread
4902     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4903   }
4904 }
4905 
4906 // Suspends the target using the signal mechanism and then grabs the PC before
4907 // resuming the target. Used by the flat-profiler only
4908 ExtendedPC os::get_thread_pc(Thread* thread) {
4909   // Make sure that it is called by the watcher for the VMThread
4910   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4911   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4912 
4913   PcFetcher fetcher(thread);
4914   fetcher.run();
4915   return fetcher.result();
4916 }
4917 
4918 ////////////////////////////////////////////////////////////////////////////////
4919 // debug support
4920 
4921 bool os::find(address addr, outputStream* st) {
4922   Dl_info dlinfo;
4923   memset(&dlinfo, 0, sizeof(dlinfo));
4924   if (dladdr(addr, &dlinfo) != 0) {
4925     st->print(PTR_FORMAT ": ", p2i(addr));
4926     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4927       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
4928                 p2i(addr) - p2i(dlinfo.dli_saddr));
4929     } else if (dlinfo.dli_fbase != NULL) {
4930       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
4931     } else {
4932       st->print("<absolute address>");
4933     }
4934     if (dlinfo.dli_fname != NULL) {
4935       st->print(" in %s", dlinfo.dli_fname);
4936     }
4937     if (dlinfo.dli_fbase != NULL) {
4938       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
4939     }
4940     st->cr();
4941 
4942     if (Verbose) {
4943       // decode some bytes around the PC
4944       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4945       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4946       address       lowest = (address) dlinfo.dli_sname;
4947       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4948       if (begin < lowest)  begin = lowest;
4949       Dl_info dlinfo2;
4950       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4951           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4952         end = (address) dlinfo2.dli_saddr;
4953       }
4954       Disassembler::decode(begin, end, st);
4955     }
4956     return true;
4957   }
4958   return false;
4959 }
4960 
4961 ////////////////////////////////////////////////////////////////////////////////
4962 // misc
4963 
4964 // This does not do anything on Linux. This is basically a hook for being
4965 // able to use structured exception handling (thread-local exception filters)
4966 // on, e.g., Win32.
4967 void
4968 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4969                          JavaCallArguments* args, Thread* thread) {
4970   f(value, method, args, thread);
4971 }
4972 
4973 void os::print_statistics() {
4974 }
4975 
4976 int os::message_box(const char* title, const char* message) {
4977   int i;
4978   fdStream err(defaultStream::error_fd());
4979   for (i = 0; i < 78; i++) err.print_raw("=");
4980   err.cr();
4981   err.print_raw_cr(title);
4982   for (i = 0; i < 78; i++) err.print_raw("-");
4983   err.cr();
4984   err.print_raw_cr(message);
4985   for (i = 0; i < 78; i++) err.print_raw("=");
4986   err.cr();
4987 
4988   char buf[16];
4989   // Prevent process from exiting upon "read error" without consuming all CPU
4990   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4991 
4992   return buf[0] == 'y' || buf[0] == 'Y';
4993 }
4994 
4995 int os::stat(const char *path, struct stat *sbuf) {
4996   char pathbuf[MAX_PATH];
4997   if (strlen(path) > MAX_PATH - 1) {
4998     errno = ENAMETOOLONG;
4999     return -1;
5000   }
5001   os::native_path(strcpy(pathbuf, path));
5002   return ::stat(pathbuf, sbuf);
5003 }
5004 
5005 bool os::check_heap(bool force) {
5006   return true;
5007 }
5008 
5009 // Is a (classpath) directory empty?
5010 bool os::dir_is_empty(const char* path) {
5011   DIR *dir = NULL;
5012   struct dirent *ptr;
5013 
5014   dir = opendir(path);
5015   if (dir == NULL) return true;
5016 
5017   // Scan the directory
5018   bool result = true;
5019   char buf[sizeof(struct dirent) + MAX_PATH];
5020   while (result && (ptr = ::readdir(dir)) != NULL) {
5021     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5022       result = false;
5023     }
5024   }
5025   closedir(dir);
5026   return result;
5027 }
5028 
5029 // This code originates from JDK's sysOpen and open64_w
5030 // from src/solaris/hpi/src/system_md.c
5031 
5032 int os::open(const char *path, int oflag, int mode) {
5033   if (strlen(path) > MAX_PATH - 1) {
5034     errno = ENAMETOOLONG;
5035     return -1;
5036   }
5037 
5038   // All file descriptors that are opened in the Java process and not
5039   // specifically destined for a subprocess should have the close-on-exec
5040   // flag set.  If we don't set it, then careless 3rd party native code
5041   // might fork and exec without closing all appropriate file descriptors
5042   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5043   // turn might:
5044   //
5045   // - cause end-of-file to fail to be detected on some file
5046   //   descriptors, resulting in mysterious hangs, or
5047   //
5048   // - might cause an fopen in the subprocess to fail on a system
5049   //   suffering from bug 1085341.
5050   //
5051   // (Yes, the default setting of the close-on-exec flag is a Unix
5052   // design flaw)
5053   //
5054   // See:
5055   // 1085341: 32-bit stdio routines should support file descriptors >255
5056   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5057   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5058   //
5059   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5060   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5061   // because it saves a system call and removes a small window where the flag
5062   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5063   // and we fall back to using FD_CLOEXEC (see below).
5064 #ifdef O_CLOEXEC
5065   oflag |= O_CLOEXEC;
5066 #endif
5067 
5068   int fd = ::open64(path, oflag, mode);
5069   if (fd == -1) return -1;
5070 
5071   //If the open succeeded, the file might still be a directory
5072   {
5073     struct stat64 buf64;
5074     int ret = ::fstat64(fd, &buf64);
5075     int st_mode = buf64.st_mode;
5076 
5077     if (ret != -1) {
5078       if ((st_mode & S_IFMT) == S_IFDIR) {
5079         errno = EISDIR;
5080         ::close(fd);
5081         return -1;
5082       }
5083     } else {
5084       ::close(fd);
5085       return -1;
5086     }
5087   }
5088 
5089 #ifdef FD_CLOEXEC
5090   // Validate that the use of the O_CLOEXEC flag on open above worked.
5091   // With recent kernels, we will perform this check exactly once.
5092   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5093   if (!O_CLOEXEC_is_known_to_work) {
5094     int flags = ::fcntl(fd, F_GETFD);
5095     if (flags != -1) {
5096       if ((flags & FD_CLOEXEC) != 0)
5097         O_CLOEXEC_is_known_to_work = 1;
5098       else
5099         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5100     }
5101   }
5102 #endif
5103 
5104   return fd;
5105 }
5106 
5107 
5108 // create binary file, rewriting existing file if required
5109 int os::create_binary_file(const char* path, bool rewrite_existing) {
5110   int oflags = O_WRONLY | O_CREAT;
5111   if (!rewrite_existing) {
5112     oflags |= O_EXCL;
5113   }
5114   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5115 }
5116 
5117 // return current position of file pointer
5118 jlong os::current_file_offset(int fd) {
5119   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5120 }
5121 
5122 // move file pointer to the specified offset
5123 jlong os::seek_to_file_offset(int fd, jlong offset) {
5124   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5125 }
5126 
5127 // This code originates from JDK's sysAvailable
5128 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5129 
5130 int os::available(int fd, jlong *bytes) {
5131   jlong cur, end;
5132   int mode;
5133   struct stat64 buf64;
5134 
5135   if (::fstat64(fd, &buf64) >= 0) {
5136     mode = buf64.st_mode;
5137     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5138       // XXX: is the following call interruptible? If so, this might
5139       // need to go through the INTERRUPT_IO() wrapper as for other
5140       // blocking, interruptible calls in this file.
5141       int n;
5142       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5143         *bytes = n;
5144         return 1;
5145       }
5146     }
5147   }
5148   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5149     return 0;
5150   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5151     return 0;
5152   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5153     return 0;
5154   }
5155   *bytes = end - cur;
5156   return 1;
5157 }
5158 
5159 // Map a block of memory.
5160 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5161                         char *addr, size_t bytes, bool read_only,
5162                         bool allow_exec) {
5163   int prot;
5164   int flags = MAP_PRIVATE;
5165 
5166   if (read_only) {
5167     prot = PROT_READ;
5168   } else {
5169     prot = PROT_READ | PROT_WRITE;
5170   }
5171 
5172   if (allow_exec) {
5173     prot |= PROT_EXEC;
5174   }
5175 
5176   if (addr != NULL) {
5177     flags |= MAP_FIXED;
5178   }
5179 
5180   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5181                                      fd, file_offset);
5182   if (mapped_address == MAP_FAILED) {
5183     return NULL;
5184   }
5185   return mapped_address;
5186 }
5187 
5188 
5189 // Remap a block of memory.
5190 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5191                           char *addr, size_t bytes, bool read_only,
5192                           bool allow_exec) {
5193   // same as map_memory() on this OS
5194   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5195                         allow_exec);
5196 }
5197 
5198 
5199 // Unmap a block of memory.
5200 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5201   return munmap(addr, bytes) == 0;
5202 }
5203 
5204 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5205 
5206 static clockid_t thread_cpu_clockid(Thread* thread) {
5207   pthread_t tid = thread->osthread()->pthread_id();
5208   clockid_t clockid;
5209 
5210   // Get thread clockid
5211   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5212   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5213   return clockid;
5214 }
5215 
5216 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5217 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5218 // of a thread.
5219 //
5220 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5221 // the fast estimate available on the platform.
5222 
5223 jlong os::current_thread_cpu_time() {
5224   if (os::Linux::supports_fast_thread_cpu_time()) {
5225     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5226   } else {
5227     // return user + sys since the cost is the same
5228     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5229   }
5230 }
5231 
5232 jlong os::thread_cpu_time(Thread* thread) {
5233   // consistent with what current_thread_cpu_time() returns
5234   if (os::Linux::supports_fast_thread_cpu_time()) {
5235     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5236   } else {
5237     return slow_thread_cpu_time(thread, true /* user + sys */);
5238   }
5239 }
5240 
5241 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5242   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5243     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5244   } else {
5245     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5246   }
5247 }
5248 
5249 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5250   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5251     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5252   } else {
5253     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5254   }
5255 }
5256 
5257 //  -1 on error.
5258 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5259   pid_t  tid = thread->osthread()->thread_id();
5260   char *s;
5261   char stat[2048];
5262   int statlen;
5263   char proc_name[64];
5264   int count;
5265   long sys_time, user_time;
5266   char cdummy;
5267   int idummy;
5268   long ldummy;
5269   FILE *fp;
5270 
5271   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5272   fp = fopen(proc_name, "r");
5273   if (fp == NULL) return -1;
5274   statlen = fread(stat, 1, 2047, fp);
5275   stat[statlen] = '\0';
5276   fclose(fp);
5277 
5278   // Skip pid and the command string. Note that we could be dealing with
5279   // weird command names, e.g. user could decide to rename java launcher
5280   // to "java 1.4.2 :)", then the stat file would look like
5281   //                1234 (java 1.4.2 :)) R ... ...
5282   // We don't really need to know the command string, just find the last
5283   // occurrence of ")" and then start parsing from there. See bug 4726580.
5284   s = strrchr(stat, ')');
5285   if (s == NULL) return -1;
5286 
5287   // Skip blank chars
5288   do { s++; } while (s && isspace(*s));
5289 
5290   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5291                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5292                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5293                  &user_time, &sys_time);
5294   if (count != 13) return -1;
5295   if (user_sys_cpu_time) {
5296     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5297   } else {
5298     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5299   }
5300 }
5301 
5302 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5303   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5304   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5305   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5306   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5307 }
5308 
5309 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5310   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5311   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5312   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5313   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5314 }
5315 
5316 bool os::is_thread_cpu_time_supported() {
5317   return true;
5318 }
5319 
5320 // System loadavg support.  Returns -1 if load average cannot be obtained.
5321 // Linux doesn't yet have a (official) notion of processor sets,
5322 // so just return the system wide load average.
5323 int os::loadavg(double loadavg[], int nelem) {
5324   return ::getloadavg(loadavg, nelem);
5325 }
5326 
5327 void os::pause() {
5328   char filename[MAX_PATH];
5329   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5330     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5331   } else {
5332     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5333   }
5334 
5335   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5336   if (fd != -1) {
5337     struct stat buf;
5338     ::close(fd);
5339     while (::stat(filename, &buf) == 0) {
5340       (void)::poll(NULL, 0, 100);
5341     }
5342   } else {
5343     jio_fprintf(stderr,
5344                 "Could not open pause file '%s', continuing immediately.\n", filename);
5345   }
5346 }
5347 
5348 
5349 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5350 // comment paragraphs are worth repeating here:
5351 //
5352 // Assumption:
5353 //    Only one parker can exist on an event, which is why we allocate
5354 //    them per-thread. Multiple unparkers can coexist.
5355 //
5356 // _Event serves as a restricted-range semaphore.
5357 //   -1 : thread is blocked, i.e. there is a waiter
5358 //    0 : neutral: thread is running or ready,
5359 //        could have been signaled after a wait started
5360 //    1 : signaled - thread is running or ready
5361 //
5362 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5363 // hang indefinitely.  For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5364 // For specifics regarding the bug see GLIBC BUGID 261237 :
5365 //    http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5366 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5367 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5368 // is used.  (The simple C test-case provided in the GLIBC bug report manifests the
5369 // hang).  The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5370 // and monitorenter when we're using 1-0 locking.  All those operations may result in
5371 // calls to pthread_cond_timedwait().  Using LD_ASSUME_KERNEL to use an older version
5372 // of libpthread avoids the problem, but isn't practical.
5373 //
5374 // Possible remedies:
5375 //
5376 // 1.   Establish a minimum relative wait time.  50 to 100 msecs seems to work.
5377 //      This is palliative and probabilistic, however.  If the thread is preempted
5378 //      between the call to compute_abstime() and pthread_cond_timedwait(), more
5379 //      than the minimum period may have passed, and the abstime may be stale (in the
5380 //      past) resultin in a hang.   Using this technique reduces the odds of a hang
5381 //      but the JVM is still vulnerable, particularly on heavily loaded systems.
5382 //
5383 // 2.   Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5384 //      of the usual flag-condvar-mutex idiom.  The write side of the pipe is set
5385 //      NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5386 //      reduces to poll()+read().  This works well, but consumes 2 FDs per extant
5387 //      thread.
5388 //
5389 // 3.   Embargo pthread_cond_timedwait() and implement a native "chron" thread
5390 //      that manages timeouts.  We'd emulate pthread_cond_timedwait() by enqueuing
5391 //      a timeout request to the chron thread and then blocking via pthread_cond_wait().
5392 //      This also works well.  In fact it avoids kernel-level scalability impediments
5393 //      on certain platforms that don't handle lots of active pthread_cond_timedwait()
5394 //      timers in a graceful fashion.
5395 //
5396 // 4.   When the abstime value is in the past it appears that control returns
5397 //      correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5398 //      Subsequent timedwait/wait calls may hang indefinitely.  Given that, we
5399 //      can avoid the problem by reinitializing the condvar -- by cond_destroy()
5400 //      followed by cond_init() -- after all calls to pthread_cond_timedwait().
5401 //      It may be possible to avoid reinitialization by checking the return
5402 //      value from pthread_cond_timedwait().  In addition to reinitializing the
5403 //      condvar we must establish the invariant that cond_signal() is only called
5404 //      within critical sections protected by the adjunct mutex.  This prevents
5405 //      cond_signal() from "seeing" a condvar that's in the midst of being
5406 //      reinitialized or that is corrupt.  Sadly, this invariant obviates the
5407 //      desirable signal-after-unlock optimization that avoids futile context switching.
5408 //
5409 //      I'm also concerned that some versions of NTPL might allocate an auxilliary
5410 //      structure when a condvar is used or initialized.  cond_destroy()  would
5411 //      release the helper structure.  Our reinitialize-after-timedwait fix
5412 //      put excessive stress on malloc/free and locks protecting the c-heap.
5413 //
5414 // We currently use (4).  See the WorkAroundNTPLTimedWaitHang flag.
5415 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5416 // and only enabling the work-around for vulnerable environments.
5417 
5418 // utility to compute the abstime argument to timedwait:
5419 // millis is the relative timeout time
5420 // abstime will be the absolute timeout time
5421 // TODO: replace compute_abstime() with unpackTime()
5422 
5423 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5424   if (millis < 0)  millis = 0;
5425 
5426   jlong seconds = millis / 1000;
5427   millis %= 1000;
5428   if (seconds > 50000000) { // see man cond_timedwait(3T)
5429     seconds = 50000000;
5430   }
5431 
5432   if (os::supports_monotonic_clock()) {
5433     struct timespec now;
5434     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5435     assert_status(status == 0, status, "clock_gettime");
5436     abstime->tv_sec = now.tv_sec  + seconds;
5437     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5438     if (nanos >= NANOSECS_PER_SEC) {
5439       abstime->tv_sec += 1;
5440       nanos -= NANOSECS_PER_SEC;
5441     }
5442     abstime->tv_nsec = nanos;
5443   } else {
5444     struct timeval now;
5445     int status = gettimeofday(&now, NULL);
5446     assert(status == 0, "gettimeofday");
5447     abstime->tv_sec = now.tv_sec  + seconds;
5448     long usec = now.tv_usec + millis * 1000;
5449     if (usec >= 1000000) {
5450       abstime->tv_sec += 1;
5451       usec -= 1000000;
5452     }
5453     abstime->tv_nsec = usec * 1000;
5454   }
5455   return abstime;
5456 }
5457 
5458 void os::PlatformEvent::park() {       // AKA "down()"
5459   // Transitions for _Event:
5460   //   -1 => -1 : illegal
5461   //    1 =>  0 : pass - return immediately
5462   //    0 => -1 : block; then set _Event to 0 before returning
5463 
5464   // Invariant: Only the thread associated with the Event/PlatformEvent
5465   // may call park().
5466   // TODO: assert that _Assoc != NULL or _Assoc == Self
5467   assert(_nParked == 0, "invariant");
5468 
5469   int v;
5470   for (;;) {
5471     v = _Event;
5472     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5473   }
5474   guarantee(v >= 0, "invariant");
5475   if (v == 0) {
5476     // Do this the hard way by blocking ...
5477     int status = pthread_mutex_lock(_mutex);
5478     assert_status(status == 0, status, "mutex_lock");
5479     guarantee(_nParked == 0, "invariant");
5480     ++_nParked;
5481     while (_Event < 0) {
5482       status = pthread_cond_wait(_cond, _mutex);
5483       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5484       // Treat this the same as if the wait was interrupted
5485       if (status == ETIME) { status = EINTR; }
5486       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5487     }
5488     --_nParked;
5489 
5490     _Event = 0;
5491     status = pthread_mutex_unlock(_mutex);
5492     assert_status(status == 0, status, "mutex_unlock");
5493     // Paranoia to ensure our locked and lock-free paths interact
5494     // correctly with each other.
5495     OrderAccess::fence();
5496   }
5497   guarantee(_Event >= 0, "invariant");
5498 }
5499 
5500 int os::PlatformEvent::park(jlong millis) {
5501   // Transitions for _Event:
5502   //   -1 => -1 : illegal
5503   //    1 =>  0 : pass - return immediately
5504   //    0 => -1 : block; then set _Event to 0 before returning
5505 
5506   guarantee(_nParked == 0, "invariant");
5507 
5508   int v;
5509   for (;;) {
5510     v = _Event;
5511     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5512   }
5513   guarantee(v >= 0, "invariant");
5514   if (v != 0) return OS_OK;
5515 
5516   // We do this the hard way, by blocking the thread.
5517   // Consider enforcing a minimum timeout value.
5518   struct timespec abst;
5519   compute_abstime(&abst, millis);
5520 
5521   int ret = OS_TIMEOUT;
5522   int status = pthread_mutex_lock(_mutex);
5523   assert_status(status == 0, status, "mutex_lock");
5524   guarantee(_nParked == 0, "invariant");
5525   ++_nParked;
5526 
5527   // Object.wait(timo) will return because of
5528   // (a) notification
5529   // (b) timeout
5530   // (c) thread.interrupt
5531   //
5532   // Thread.interrupt and object.notify{All} both call Event::set.
5533   // That is, we treat thread.interrupt as a special case of notification.
5534   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5535   // We assume all ETIME returns are valid.
5536   //
5537   // TODO: properly differentiate simultaneous notify+interrupt.
5538   // In that case, we should propagate the notify to another waiter.
5539 
5540   while (_Event < 0) {
5541     status = pthread_cond_timedwait(_cond, _mutex, &abst);
5542     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5543       pthread_cond_destroy(_cond);
5544       pthread_cond_init(_cond, os::Linux::condAttr());
5545     }
5546     assert_status(status == 0 || status == EINTR ||
5547                   status == ETIME || status == ETIMEDOUT,
5548                   status, "cond_timedwait");
5549     if (!FilterSpuriousWakeups) break;                 // previous semantics
5550     if (status == ETIME || status == ETIMEDOUT) break;
5551     // We consume and ignore EINTR and spurious wakeups.
5552   }
5553   --_nParked;
5554   if (_Event >= 0) {
5555     ret = OS_OK;
5556   }
5557   _Event = 0;
5558   status = pthread_mutex_unlock(_mutex);
5559   assert_status(status == 0, status, "mutex_unlock");
5560   assert(_nParked == 0, "invariant");
5561   // Paranoia to ensure our locked and lock-free paths interact
5562   // correctly with each other.
5563   OrderAccess::fence();
5564   return ret;
5565 }
5566 
5567 void os::PlatformEvent::unpark() {
5568   // Transitions for _Event:
5569   //    0 => 1 : just return
5570   //    1 => 1 : just return
5571   //   -1 => either 0 or 1; must signal target thread
5572   //         That is, we can safely transition _Event from -1 to either
5573   //         0 or 1.
5574   // See also: "Semaphores in Plan 9" by Mullender & Cox
5575   //
5576   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5577   // that it will take two back-to-back park() calls for the owning
5578   // thread to block. This has the benefit of forcing a spurious return
5579   // from the first park() call after an unpark() call which will help
5580   // shake out uses of park() and unpark() without condition variables.
5581 
5582   if (Atomic::xchg(1, &_Event) >= 0) return;
5583 
5584   // Wait for the thread associated with the event to vacate
5585   int status = pthread_mutex_lock(_mutex);
5586   assert_status(status == 0, status, "mutex_lock");
5587   int AnyWaiters = _nParked;
5588   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5589   if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5590     AnyWaiters = 0;
5591     pthread_cond_signal(_cond);
5592   }
5593   status = pthread_mutex_unlock(_mutex);
5594   assert_status(status == 0, status, "mutex_unlock");
5595   if (AnyWaiters != 0) {
5596     // Note that we signal() *after* dropping the lock for "immortal" Events.
5597     // This is safe and avoids a common class of  futile wakeups.  In rare
5598     // circumstances this can cause a thread to return prematurely from
5599     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5600     // will simply re-test the condition and re-park itself.
5601     // This provides particular benefit if the underlying platform does not
5602     // provide wait morphing.
5603     status = pthread_cond_signal(_cond);
5604     assert_status(status == 0, status, "cond_signal");
5605   }
5606 }
5607 
5608 
5609 // JSR166
5610 // -------------------------------------------------------
5611 
5612 // The solaris and linux implementations of park/unpark are fairly
5613 // conservative for now, but can be improved. They currently use a
5614 // mutex/condvar pair, plus a a count.
5615 // Park decrements count if > 0, else does a condvar wait.  Unpark
5616 // sets count to 1 and signals condvar.  Only one thread ever waits
5617 // on the condvar. Contention seen when trying to park implies that someone
5618 // is unparking you, so don't wait. And spurious returns are fine, so there
5619 // is no need to track notifications.
5620 
5621 // This code is common to linux and solaris and will be moved to a
5622 // common place in dolphin.
5623 //
5624 // The passed in time value is either a relative time in nanoseconds
5625 // or an absolute time in milliseconds. Either way it has to be unpacked
5626 // into suitable seconds and nanoseconds components and stored in the
5627 // given timespec structure.
5628 // Given time is a 64-bit value and the time_t used in the timespec is only
5629 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5630 // overflow if times way in the future are given. Further on Solaris versions
5631 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5632 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5633 // As it will be 28 years before "now + 100000000" will overflow we can
5634 // ignore overflow and just impose a hard-limit on seconds using the value
5635 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5636 // years from "now".
5637 
5638 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5639   assert(time > 0, "convertTime");
5640   time_t max_secs = 0;
5641 
5642   if (!os::supports_monotonic_clock() || isAbsolute) {
5643     struct timeval now;
5644     int status = gettimeofday(&now, NULL);
5645     assert(status == 0, "gettimeofday");
5646 
5647     max_secs = now.tv_sec + MAX_SECS;
5648 
5649     if (isAbsolute) {
5650       jlong secs = time / 1000;
5651       if (secs > max_secs) {
5652         absTime->tv_sec = max_secs;
5653       } else {
5654         absTime->tv_sec = secs;
5655       }
5656       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5657     } else {
5658       jlong secs = time / NANOSECS_PER_SEC;
5659       if (secs >= MAX_SECS) {
5660         absTime->tv_sec = max_secs;
5661         absTime->tv_nsec = 0;
5662       } else {
5663         absTime->tv_sec = now.tv_sec + secs;
5664         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5665         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5666           absTime->tv_nsec -= NANOSECS_PER_SEC;
5667           ++absTime->tv_sec; // note: this must be <= max_secs
5668         }
5669       }
5670     }
5671   } else {
5672     // must be relative using monotonic clock
5673     struct timespec now;
5674     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5675     assert_status(status == 0, status, "clock_gettime");
5676     max_secs = now.tv_sec + MAX_SECS;
5677     jlong secs = time / NANOSECS_PER_SEC;
5678     if (secs >= MAX_SECS) {
5679       absTime->tv_sec = max_secs;
5680       absTime->tv_nsec = 0;
5681     } else {
5682       absTime->tv_sec = now.tv_sec + secs;
5683       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5684       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5685         absTime->tv_nsec -= NANOSECS_PER_SEC;
5686         ++absTime->tv_sec; // note: this must be <= max_secs
5687       }
5688     }
5689   }
5690   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5691   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5692   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5693   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5694 }
5695 
5696 void Parker::park(bool isAbsolute, jlong time) {
5697   // Ideally we'd do something useful while spinning, such
5698   // as calling unpackTime().
5699 
5700   // Optional fast-path check:
5701   // Return immediately if a permit is available.
5702   // We depend on Atomic::xchg() having full barrier semantics
5703   // since we are doing a lock-free update to _counter.
5704   if (Atomic::xchg(0, &_counter) > 0) return;
5705 
5706   Thread* thread = Thread::current();
5707   assert(thread->is_Java_thread(), "Must be JavaThread");
5708   JavaThread *jt = (JavaThread *)thread;
5709 
5710   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5711   // Check interrupt before trying to wait
5712   if (Thread::is_interrupted(thread, false)) {
5713     return;
5714   }
5715 
5716   // Next, demultiplex/decode time arguments
5717   timespec absTime;
5718   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5719     return;
5720   }
5721   if (time > 0) {
5722     unpackTime(&absTime, isAbsolute, time);
5723   }
5724 
5725 
5726   // Enter safepoint region
5727   // Beware of deadlocks such as 6317397.
5728   // The per-thread Parker:: mutex is a classic leaf-lock.
5729   // In particular a thread must never block on the Threads_lock while
5730   // holding the Parker:: mutex.  If safepoints are pending both the
5731   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5732   ThreadBlockInVM tbivm(jt);
5733 
5734   // Don't wait if cannot get lock since interference arises from
5735   // unblocking.  Also. check interrupt before trying wait
5736   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5737     return;
5738   }
5739 
5740   int status;
5741   if (_counter > 0)  { // no wait needed
5742     _counter = 0;
5743     status = pthread_mutex_unlock(_mutex);
5744     assert(status == 0, "invariant");
5745     // Paranoia to ensure our locked and lock-free paths interact
5746     // correctly with each other and Java-level accesses.
5747     OrderAccess::fence();
5748     return;
5749   }
5750 
5751 #ifdef ASSERT
5752   // Don't catch signals while blocked; let the running threads have the signals.
5753   // (This allows a debugger to break into the running thread.)
5754   sigset_t oldsigs;
5755   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5756   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5757 #endif
5758 
5759   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5760   jt->set_suspend_equivalent();
5761   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5762 
5763   assert(_cur_index == -1, "invariant");
5764   if (time == 0) {
5765     _cur_index = REL_INDEX; // arbitrary choice when not timed
5766     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5767   } else {
5768     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5769     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5770     if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5771       pthread_cond_destroy(&_cond[_cur_index]);
5772       pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5773     }
5774   }
5775   _cur_index = -1;
5776   assert_status(status == 0 || status == EINTR ||
5777                 status == ETIME || status == ETIMEDOUT,
5778                 status, "cond_timedwait");
5779 
5780 #ifdef ASSERT
5781   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5782 #endif
5783 
5784   _counter = 0;
5785   status = pthread_mutex_unlock(_mutex);
5786   assert_status(status == 0, status, "invariant");
5787   // Paranoia to ensure our locked and lock-free paths interact
5788   // correctly with each other and Java-level accesses.
5789   OrderAccess::fence();
5790 
5791   // If externally suspended while waiting, re-suspend
5792   if (jt->handle_special_suspend_equivalent_condition()) {
5793     jt->java_suspend_self();
5794   }
5795 }
5796 
5797 void Parker::unpark() {
5798   int status = pthread_mutex_lock(_mutex);
5799   assert(status == 0, "invariant");
5800   const int s = _counter;
5801   _counter = 1;
5802   if (s < 1) {
5803     // thread might be parked
5804     if (_cur_index != -1) {
5805       // thread is definitely parked
5806       if (WorkAroundNPTLTimedWaitHang) {
5807         status = pthread_cond_signal(&_cond[_cur_index]);
5808         assert(status == 0, "invariant");
5809         status = pthread_mutex_unlock(_mutex);
5810         assert(status == 0, "invariant");
5811       } else {
5812         // must capture correct index before unlocking
5813         int index = _cur_index;
5814         status = pthread_mutex_unlock(_mutex);
5815         assert(status == 0, "invariant");
5816         status = pthread_cond_signal(&_cond[index]);
5817         assert(status == 0, "invariant");
5818       }
5819     } else {
5820       pthread_mutex_unlock(_mutex);
5821       assert(status == 0, "invariant");
5822     }
5823   } else {
5824     pthread_mutex_unlock(_mutex);
5825     assert(status == 0, "invariant");
5826   }
5827 }
5828 
5829 
5830 extern char** environ;
5831 
5832 // Run the specified command in a separate process. Return its exit value,
5833 // or -1 on failure (e.g. can't fork a new process).
5834 // Unlike system(), this function can be called from signal handler. It
5835 // doesn't block SIGINT et al.
5836 int os::fork_and_exec(char* cmd) {
5837   const char * argv[4] = {"sh", "-c", cmd, NULL};
5838 
5839   pid_t pid = fork();
5840 
5841   if (pid < 0) {
5842     // fork failed
5843     return -1;
5844 
5845   } else if (pid == 0) {
5846     // child process
5847 
5848     execve("/bin/sh", (char* const*)argv, environ);
5849 
5850     // execve failed
5851     _exit(-1);
5852 
5853   } else  {
5854     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5855     // care about the actual exit code, for now.
5856 
5857     int status;
5858 
5859     // Wait for the child process to exit.  This returns immediately if
5860     // the child has already exited. */
5861     while (waitpid(pid, &status, 0) < 0) {
5862       switch (errno) {
5863       case ECHILD: return 0;
5864       case EINTR: break;
5865       default: return -1;
5866       }
5867     }
5868 
5869     if (WIFEXITED(status)) {
5870       // The child exited normally; get its exit code.
5871       return WEXITSTATUS(status);
5872     } else if (WIFSIGNALED(status)) {
5873       // The child exited because of a signal
5874       // The best value to return is 0x80 + signal number,
5875       // because that is what all Unix shells do, and because
5876       // it allows callers to distinguish between process exit and
5877       // process death by signal.
5878       return 0x80 + WTERMSIG(status);
5879     } else {
5880       // Unknown exit code; pass it through
5881       return status;
5882     }
5883   }
5884 }
5885 
5886 // is_headless_jre()
5887 //
5888 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5889 // in order to report if we are running in a headless jre
5890 //
5891 // Since JDK8 xawt/libmawt.so was moved into the same directory
5892 // as libawt.so, and renamed libawt_xawt.so
5893 //
5894 bool os::is_headless_jre() {
5895   struct stat statbuf;
5896   char buf[MAXPATHLEN];
5897   char libmawtpath[MAXPATHLEN];
5898   const char *xawtstr  = "/xawt/libmawt.so";
5899   const char *new_xawtstr = "/libawt_xawt.so";
5900   char *p;
5901 
5902   // Get path to libjvm.so
5903   os::jvm_path(buf, sizeof(buf));
5904 
5905   // Get rid of libjvm.so
5906   p = strrchr(buf, '/');
5907   if (p == NULL) {
5908     return false;
5909   } else {
5910     *p = '\0';
5911   }
5912 
5913   // Get rid of client or server
5914   p = strrchr(buf, '/');
5915   if (p == NULL) {
5916     return false;
5917   } else {
5918     *p = '\0';
5919   }
5920 
5921   // check xawt/libmawt.so
5922   strcpy(libmawtpath, buf);
5923   strcat(libmawtpath, xawtstr);
5924   if (::stat(libmawtpath, &statbuf) == 0) return false;
5925 
5926   // check libawt_xawt.so
5927   strcpy(libmawtpath, buf);
5928   strcat(libmawtpath, new_xawtstr);
5929   if (::stat(libmawtpath, &statbuf) == 0) return false;
5930 
5931   return true;
5932 }
5933 
5934 // Get the default path to the core file
5935 // Returns the length of the string
5936 int os::get_core_path(char* buffer, size_t bufferSize) {
5937   /*
5938    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5939    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5940    */
5941   const int core_pattern_len = 129;
5942   char core_pattern[core_pattern_len] = {0};
5943 
5944   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5945   if (core_pattern_file != -1) {
5946     ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5947     ::close(core_pattern_file);
5948 
5949     if (ret > 0) {
5950       char *last_char = core_pattern + strlen(core_pattern) - 1;
5951 
5952       if (*last_char == '\n') {
5953         *last_char = '\0';
5954       }
5955     }
5956   }
5957 
5958   if (strlen(core_pattern) == 0) {
5959     return -1;
5960   }
5961 
5962   char *pid_pos = strstr(core_pattern, "%p");
5963   int written;
5964 
5965   if (core_pattern[0] == '/') {
5966     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5967   } else {
5968     char cwd[PATH_MAX];
5969 
5970     const char* p = get_current_directory(cwd, PATH_MAX);
5971     if (p == NULL) {
5972       return -1;
5973     }
5974 
5975     if (core_pattern[0] == '|') {
5976       written = jio_snprintf(buffer, bufferSize,
5977                         "\"%s\" (or dumping to %s/core.%d)",
5978                                      &core_pattern[1], p, current_process_id());
5979     } else {
5980       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5981     }
5982   }
5983 
5984   if (written < 0) {
5985     return -1;
5986   }
5987 
5988   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5989     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5990 
5991     if (core_uses_pid_file != -1) {
5992       char core_uses_pid = 0;
5993       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5994       ::close(core_uses_pid_file);
5995 
5996       if (core_uses_pid == '1') {
5997         jio_snprintf(buffer + written, bufferSize - written,
5998                                           ".%d", current_process_id());
5999       }
6000     }
6001   }
6002 
6003   return strlen(buffer);
6004 }
6005 
6006 /////////////// Unit tests ///////////////
6007 
6008 #ifndef PRODUCT
6009 
6010 #define test_log(...)              \
6011   do {                             \
6012     if (VerboseInternalVMTests) {  \
6013       tty->print_cr(__VA_ARGS__);  \
6014       tty->flush();                \
6015     }                              \
6016   } while (false)
6017 
6018 class TestReserveMemorySpecial : AllStatic {
6019  public:
6020   static void small_page_write(void* addr, size_t size) {
6021     size_t page_size = os::vm_page_size();
6022 
6023     char* end = (char*)addr + size;
6024     for (char* p = (char*)addr; p < end; p += page_size) {
6025       *p = 1;
6026     }
6027   }
6028 
6029   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6030     if (!UseHugeTLBFS) {
6031       return;
6032     }
6033 
6034     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6035 
6036     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6037 
6038     if (addr != NULL) {
6039       small_page_write(addr, size);
6040 
6041       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6042     }
6043   }
6044 
6045   static void test_reserve_memory_special_huge_tlbfs_only() {
6046     if (!UseHugeTLBFS) {
6047       return;
6048     }
6049 
6050     size_t lp = os::large_page_size();
6051 
6052     for (size_t size = lp; size <= lp * 10; size += lp) {
6053       test_reserve_memory_special_huge_tlbfs_only(size);
6054     }
6055   }
6056 
6057   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6058     size_t lp = os::large_page_size();
6059     size_t ag = os::vm_allocation_granularity();
6060 
6061     // sizes to test
6062     const size_t sizes[] = {
6063       lp, lp + ag, lp + lp / 2, lp * 2,
6064       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6065       lp * 10, lp * 10 + lp / 2
6066     };
6067     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6068 
6069     // For each size/alignment combination, we test three scenarios:
6070     // 1) with req_addr == NULL
6071     // 2) with a non-null req_addr at which we expect to successfully allocate
6072     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6073     //    expect the allocation to either fail or to ignore req_addr
6074 
6075     // Pre-allocate two areas; they shall be as large as the largest allocation
6076     //  and aligned to the largest alignment we will be testing.
6077     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6078     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6079       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6080       -1, 0);
6081     assert(mapping1 != MAP_FAILED, "should work");
6082 
6083     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6084       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6085       -1, 0);
6086     assert(mapping2 != MAP_FAILED, "should work");
6087 
6088     // Unmap the first mapping, but leave the second mapping intact: the first
6089     // mapping will serve as a value for a "good" req_addr (case 2). The second
6090     // mapping, still intact, as "bad" req_addr (case 3).
6091     ::munmap(mapping1, mapping_size);
6092 
6093     // Case 1
6094     test_log("%s, req_addr NULL:", __FUNCTION__);
6095     test_log("size            align           result");
6096 
6097     for (int i = 0; i < num_sizes; i++) {
6098       const size_t size = sizes[i];
6099       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6100         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6101         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6102                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6103         if (p != NULL) {
6104           assert(is_ptr_aligned(p, alignment), "must be");
6105           small_page_write(p, size);
6106           os::Linux::release_memory_special_huge_tlbfs(p, size);
6107         }
6108       }
6109     }
6110 
6111     // Case 2
6112     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6113     test_log("size            align           req_addr         result");
6114 
6115     for (int i = 0; i < num_sizes; i++) {
6116       const size_t size = sizes[i];
6117       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6118         char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6119         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6120         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6121                  size, alignment, p2i(req_addr), p2i(p),
6122                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6123         if (p != NULL) {
6124           assert(p == req_addr, "must be");
6125           small_page_write(p, size);
6126           os::Linux::release_memory_special_huge_tlbfs(p, size);
6127         }
6128       }
6129     }
6130 
6131     // Case 3
6132     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6133     test_log("size            align           req_addr         result");
6134 
6135     for (int i = 0; i < num_sizes; i++) {
6136       const size_t size = sizes[i];
6137       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6138         char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6139         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6140         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6141                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6142         // as the area around req_addr contains already existing mappings, the API should always
6143         // return NULL (as per contract, it cannot return another address)
6144         assert(p == NULL, "must be");
6145       }
6146     }
6147 
6148     ::munmap(mapping2, mapping_size);
6149 
6150   }
6151 
6152   static void test_reserve_memory_special_huge_tlbfs() {
6153     if (!UseHugeTLBFS) {
6154       return;
6155     }
6156 
6157     test_reserve_memory_special_huge_tlbfs_only();
6158     test_reserve_memory_special_huge_tlbfs_mixed();
6159   }
6160 
6161   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6162     if (!UseSHM) {
6163       return;
6164     }
6165 
6166     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6167 
6168     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6169 
6170     if (addr != NULL) {
6171       assert(is_ptr_aligned(addr, alignment), "Check");
6172       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6173 
6174       small_page_write(addr, size);
6175 
6176       os::Linux::release_memory_special_shm(addr, size);
6177     }
6178   }
6179 
6180   static void test_reserve_memory_special_shm() {
6181     size_t lp = os::large_page_size();
6182     size_t ag = os::vm_allocation_granularity();
6183 
6184     for (size_t size = ag; size < lp * 3; size += ag) {
6185       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6186         test_reserve_memory_special_shm(size, alignment);
6187       }
6188     }
6189   }
6190 
6191   static void test() {
6192     test_reserve_memory_special_huge_tlbfs();
6193     test_reserve_memory_special_shm();
6194   }
6195 };
6196 
6197 void TestReserveMemorySpecial_test() {
6198   TestReserveMemorySpecial::test();
6199 }
6200 
6201 #endif