1 /*
   2  * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_linux.h"
  35 #include "logging/log.hpp"
  36 #include "memory/allocation.inline.hpp"
  37 #include "memory/filemap.hpp"
  38 #include "mutex_linux.inline.hpp"
  39 #include "oops/oop.inline.hpp"
  40 #include "os_linux.inline.hpp"
  41 #include "os_share_linux.hpp"
  42 #include "prims/jniFastGetField.hpp"
  43 #include "prims/jvm.h"
  44 #include "prims/jvm_misc.hpp"
  45 #include "runtime/arguments.hpp"
  46 #include "runtime/atomic.inline.hpp"
  47 #include "runtime/extendedPC.hpp"
  48 #include "runtime/globals.hpp"
  49 #include "runtime/interfaceSupport.hpp"
  50 #include "runtime/init.hpp"
  51 #include "runtime/java.hpp"
  52 #include "runtime/javaCalls.hpp"
  53 #include "runtime/mutexLocker.hpp"
  54 #include "runtime/objectMonitor.hpp"
  55 #include "runtime/orderAccess.inline.hpp"
  56 #include "runtime/osThread.hpp"
  57 #include "runtime/perfMemory.hpp"
  58 #include "runtime/sharedRuntime.hpp"
  59 #include "runtime/statSampler.hpp"
  60 #include "runtime/stubRoutines.hpp"
  61 #include "runtime/thread.inline.hpp"
  62 #include "runtime/threadCritical.hpp"
  63 #include "runtime/timer.hpp"
  64 #include "semaphore_posix.hpp"
  65 #include "services/attachListener.hpp"
  66 #include "services/memTracker.hpp"
  67 #include "services/runtimeService.hpp"
  68 #include "utilities/decoder.hpp"
  69 #include "utilities/defaultStream.hpp"
  70 #include "utilities/events.hpp"
  71 #include "utilities/elfFile.hpp"
  72 #include "utilities/growableArray.hpp"
  73 #include "utilities/macros.hpp"
  74 #include "utilities/vmError.hpp"
  75 
  76 // put OS-includes here
  77 # include <sys/types.h>
  78 # include <sys/mman.h>
  79 # include <sys/stat.h>
  80 # include <sys/select.h>
  81 # include <pthread.h>
  82 # include <signal.h>
  83 # include <errno.h>
  84 # include <dlfcn.h>
  85 # include <stdio.h>
  86 # include <unistd.h>
  87 # include <sys/resource.h>
  88 # include <pthread.h>
  89 # include <sys/stat.h>
  90 # include <sys/time.h>
  91 # include <sys/times.h>
  92 # include <sys/utsname.h>
  93 # include <sys/socket.h>
  94 # include <sys/wait.h>
  95 # include <pwd.h>
  96 # include <poll.h>
  97 # include <semaphore.h>
  98 # include <fcntl.h>
  99 # include <string.h>
 100 # include <syscall.h>
 101 # include <sys/sysinfo.h>
 102 # include <gnu/libc-version.h>
 103 # include <sys/ipc.h>
 104 # include <sys/shm.h>
 105 # include <link.h>
 106 # include <stdint.h>
 107 # include <inttypes.h>
 108 # include <sys/ioctl.h>
 109 
 110 #ifndef _GNU_SOURCE
 111   #define _GNU_SOURCE
 112   #include <sched.h>
 113   #undef _GNU_SOURCE
 114 #else
 115   #include <sched.h>
 116 #endif
 117 
 118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 119 // getrusage() is prepared to handle the associated failure.
 120 #ifndef RUSAGE_THREAD
 121   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 122 #endif
 123 
 124 #define MAX_PATH    (2 * K)
 125 
 126 #define MAX_SECS 100000000
 127 
 128 // for timer info max values which include all bits
 129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 130 
 131 #define LARGEPAGES_BIT (1 << 6)
 132 ////////////////////////////////////////////////////////////////////////////////
 133 // global variables
 134 julong os::Linux::_physical_memory = 0;
 135 
 136 address   os::Linux::_initial_thread_stack_bottom = NULL;
 137 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 138 
 139 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 141 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 142 Mutex* os::Linux::_createThread_lock = NULL;
 143 pthread_t os::Linux::_main_thread;
 144 int os::Linux::_page_size = -1;
 145 const int os::Linux::_vm_default_page_size = (8 * K);
 146 bool os::Linux::_supports_fast_thread_cpu_time = false;
 147 uint32_t os::Linux::_os_version = 0;
 148 const char * os::Linux::_glibc_version = NULL;
 149 const char * os::Linux::_libpthread_version = NULL;
 150 pthread_condattr_t os::Linux::_condattr[1];
 151 
 152 static jlong initial_time_count=0;
 153 
 154 static int clock_tics_per_sec = 100;
 155 
 156 // For diagnostics to print a message once. see run_periodic_checks
 157 static sigset_t check_signal_done;
 158 static bool check_signals = true;
 159 
 160 // Signal number used to suspend/resume a thread
 161 
 162 // do not use any signal number less than SIGSEGV, see 4355769
 163 static int SR_signum = SIGUSR2;
 164 sigset_t SR_sigset;
 165 
 166 // Declarations
 167 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 168 
 169 // utility functions
 170 
 171 static int SR_initialize();
 172 
 173 julong os::available_memory() {
 174   return Linux::available_memory();
 175 }
 176 
 177 julong os::Linux::available_memory() {
 178   // values in struct sysinfo are "unsigned long"
 179   struct sysinfo si;
 180   sysinfo(&si);
 181 
 182   return (julong)si.freeram * si.mem_unit;
 183 }
 184 
 185 julong os::physical_memory() {
 186   return Linux::physical_memory();
 187 }
 188 
 189 // Return true if user is running as root.
 190 
 191 bool os::have_special_privileges() {
 192   static bool init = false;
 193   static bool privileges = false;
 194   if (!init) {
 195     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 196     init = true;
 197   }
 198   return privileges;
 199 }
 200 
 201 
 202 #ifndef SYS_gettid
 203 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 204   #ifdef __ia64__
 205     #define SYS_gettid 1105
 206   #else
 207     #ifdef __i386__
 208       #define SYS_gettid 224
 209     #else
 210       #ifdef __amd64__
 211         #define SYS_gettid 186
 212       #else
 213         #ifdef __sparc__
 214           #define SYS_gettid 143
 215         #else
 216           #error define gettid for the arch
 217         #endif
 218       #endif
 219     #endif
 220   #endif
 221 #endif
 222 
 223 // Cpu architecture string
 224 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
 225 
 226 
 227 // pid_t gettid()
 228 //
 229 // Returns the kernel thread id of the currently running thread. Kernel
 230 // thread id is used to access /proc.
 231 pid_t os::Linux::gettid() {
 232   int rslt = syscall(SYS_gettid);
 233   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 234   return (pid_t)rslt;
 235 }
 236 
 237 // Most versions of linux have a bug where the number of processors are
 238 // determined by looking at the /proc file system.  In a chroot environment,
 239 // the system call returns 1.  This causes the VM to act as if it is
 240 // a single processor and elide locking (see is_MP() call).
 241 static bool unsafe_chroot_detected = false;
 242 static const char *unstable_chroot_error = "/proc file system not found.\n"
 243                      "Java may be unstable running multithreaded in a chroot "
 244                      "environment on Linux when /proc filesystem is not mounted.";
 245 
 246 void os::Linux::initialize_system_info() {
 247   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 248   if (processor_count() == 1) {
 249     pid_t pid = os::Linux::gettid();
 250     char fname[32];
 251     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 252     FILE *fp = fopen(fname, "r");
 253     if (fp == NULL) {
 254       unsafe_chroot_detected = true;
 255     } else {
 256       fclose(fp);
 257     }
 258   }
 259   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 260   assert(processor_count() > 0, "linux error");
 261 }
 262 
 263 void os::init_system_properties_values() {
 264   // The next steps are taken in the product version:
 265   //
 266   // Obtain the JAVA_HOME value from the location of libjvm.so.
 267   // This library should be located at:
 268   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 269   //
 270   // If "/jre/lib/" appears at the right place in the path, then we
 271   // assume libjvm.so is installed in a JDK and we use this path.
 272   //
 273   // Otherwise exit with message: "Could not create the Java virtual machine."
 274   //
 275   // The following extra steps are taken in the debugging version:
 276   //
 277   // If "/jre/lib/" does NOT appear at the right place in the path
 278   // instead of exit check for $JAVA_HOME environment variable.
 279   //
 280   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 281   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 282   // it looks like libjvm.so is installed there
 283   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 284   //
 285   // Otherwise exit.
 286   //
 287   // Important note: if the location of libjvm.so changes this
 288   // code needs to be changed accordingly.
 289 
 290   // See ld(1):
 291   //      The linker uses the following search paths to locate required
 292   //      shared libraries:
 293   //        1: ...
 294   //        ...
 295   //        7: The default directories, normally /lib and /usr/lib.
 296 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 297   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 298 #else
 299   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 300 #endif
 301 
 302 // Base path of extensions installed on the system.
 303 #define SYS_EXT_DIR     "/usr/java/packages"
 304 #define EXTENSIONS_DIR  "/lib/ext"
 305 
 306   // Buffer that fits several sprintfs.
 307   // Note that the space for the colon and the trailing null are provided
 308   // by the nulls included by the sizeof operator.
 309   const size_t bufsize =
 310     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 311          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 312   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 313 
 314   // sysclasspath, java_home, dll_dir
 315   {
 316     char *pslash;
 317     os::jvm_path(buf, bufsize);
 318 
 319     // Found the full path to libjvm.so.
 320     // Now cut the path to <java_home>/jre if we can.
 321     pslash = strrchr(buf, '/');
 322     if (pslash != NULL) {
 323       *pslash = '\0';            // Get rid of /libjvm.so.
 324     }
 325     pslash = strrchr(buf, '/');
 326     if (pslash != NULL) {
 327       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 328     }
 329     Arguments::set_dll_dir(buf);
 330 
 331     if (pslash != NULL) {
 332       pslash = strrchr(buf, '/');
 333       if (pslash != NULL) {
 334         *pslash = '\0';          // Get rid of /<arch>.
 335         pslash = strrchr(buf, '/');
 336         if (pslash != NULL) {
 337           *pslash = '\0';        // Get rid of /lib.
 338         }
 339       }
 340     }
 341     Arguments::set_java_home(buf);
 342     set_boot_path('/', ':');
 343   }
 344 
 345   // Where to look for native libraries.
 346   //
 347   // Note: Due to a legacy implementation, most of the library path
 348   // is set in the launcher. This was to accomodate linking restrictions
 349   // on legacy Linux implementations (which are no longer supported).
 350   // Eventually, all the library path setting will be done here.
 351   //
 352   // However, to prevent the proliferation of improperly built native
 353   // libraries, the new path component /usr/java/packages is added here.
 354   // Eventually, all the library path setting will be done here.
 355   {
 356     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 357     // should always exist (until the legacy problem cited above is
 358     // addressed).
 359     const char *v = ::getenv("LD_LIBRARY_PATH");
 360     const char *v_colon = ":";
 361     if (v == NULL) { v = ""; v_colon = ""; }
 362     // That's +1 for the colon and +1 for the trailing '\0'.
 363     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 364                                                      strlen(v) + 1 +
 365                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 366                                                      mtInternal);
 367     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 368     Arguments::set_library_path(ld_library_path);
 369     FREE_C_HEAP_ARRAY(char, ld_library_path);
 370   }
 371 
 372   // Extensions directories.
 373   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 374   Arguments::set_ext_dirs(buf);
 375 
 376   FREE_C_HEAP_ARRAY(char, buf);
 377 
 378 #undef DEFAULT_LIBPATH
 379 #undef SYS_EXT_DIR
 380 #undef EXTENSIONS_DIR
 381 }
 382 
 383 ////////////////////////////////////////////////////////////////////////////////
 384 // breakpoint support
 385 
 386 void os::breakpoint() {
 387   BREAKPOINT;
 388 }
 389 
 390 extern "C" void breakpoint() {
 391   // use debugger to set breakpoint here
 392 }
 393 
 394 ////////////////////////////////////////////////////////////////////////////////
 395 // signal support
 396 
 397 debug_only(static bool signal_sets_initialized = false);
 398 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 399 
 400 bool os::Linux::is_sig_ignored(int sig) {
 401   struct sigaction oact;
 402   sigaction(sig, (struct sigaction*)NULL, &oact);
 403   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 404                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 405   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 406     return true;
 407   } else {
 408     return false;
 409   }
 410 }
 411 
 412 void os::Linux::signal_sets_init() {
 413   // Should also have an assertion stating we are still single-threaded.
 414   assert(!signal_sets_initialized, "Already initialized");
 415   // Fill in signals that are necessarily unblocked for all threads in
 416   // the VM. Currently, we unblock the following signals:
 417   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 418   //                         by -Xrs (=ReduceSignalUsage));
 419   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 420   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 421   // the dispositions or masks wrt these signals.
 422   // Programs embedding the VM that want to use the above signals for their
 423   // own purposes must, at this time, use the "-Xrs" option to prevent
 424   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 425   // (See bug 4345157, and other related bugs).
 426   // In reality, though, unblocking these signals is really a nop, since
 427   // these signals are not blocked by default.
 428   sigemptyset(&unblocked_sigs);
 429   sigemptyset(&allowdebug_blocked_sigs);
 430   sigaddset(&unblocked_sigs, SIGILL);
 431   sigaddset(&unblocked_sigs, SIGSEGV);
 432   sigaddset(&unblocked_sigs, SIGBUS);
 433   sigaddset(&unblocked_sigs, SIGFPE);
 434 #if defined(PPC64)
 435   sigaddset(&unblocked_sigs, SIGTRAP);
 436 #endif
 437   sigaddset(&unblocked_sigs, SR_signum);
 438 
 439   if (!ReduceSignalUsage) {
 440     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 441       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 442       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 443     }
 444     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 445       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 446       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 447     }
 448     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 449       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 450       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 451     }
 452   }
 453   // Fill in signals that are blocked by all but the VM thread.
 454   sigemptyset(&vm_sigs);
 455   if (!ReduceSignalUsage) {
 456     sigaddset(&vm_sigs, BREAK_SIGNAL);
 457   }
 458   debug_only(signal_sets_initialized = true);
 459 
 460 }
 461 
 462 // These are signals that are unblocked while a thread is running Java.
 463 // (For some reason, they get blocked by default.)
 464 sigset_t* os::Linux::unblocked_signals() {
 465   assert(signal_sets_initialized, "Not initialized");
 466   return &unblocked_sigs;
 467 }
 468 
 469 // These are the signals that are blocked while a (non-VM) thread is
 470 // running Java. Only the VM thread handles these signals.
 471 sigset_t* os::Linux::vm_signals() {
 472   assert(signal_sets_initialized, "Not initialized");
 473   return &vm_sigs;
 474 }
 475 
 476 // These are signals that are blocked during cond_wait to allow debugger in
 477 sigset_t* os::Linux::allowdebug_blocked_signals() {
 478   assert(signal_sets_initialized, "Not initialized");
 479   return &allowdebug_blocked_sigs;
 480 }
 481 
 482 void os::Linux::hotspot_sigmask(Thread* thread) {
 483 
 484   //Save caller's signal mask before setting VM signal mask
 485   sigset_t caller_sigmask;
 486   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 487 
 488   OSThread* osthread = thread->osthread();
 489   osthread->set_caller_sigmask(caller_sigmask);
 490 
 491   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 492 
 493   if (!ReduceSignalUsage) {
 494     if (thread->is_VM_thread()) {
 495       // Only the VM thread handles BREAK_SIGNAL ...
 496       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 497     } else {
 498       // ... all other threads block BREAK_SIGNAL
 499       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 500     }
 501   }
 502 }
 503 
 504 //////////////////////////////////////////////////////////////////////////////
 505 // detecting pthread library
 506 
 507 void os::Linux::libpthread_init() {
 508   // Save glibc and pthread version strings.
 509 #if !defined(_CS_GNU_LIBC_VERSION) || \
 510     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 511   #error "glibc too old (< 2.3.2)"
 512 #endif
 513 
 514   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 515   assert(n > 0, "cannot retrieve glibc version");
 516   char *str = (char *)malloc(n, mtInternal);
 517   confstr(_CS_GNU_LIBC_VERSION, str, n);
 518   os::Linux::set_glibc_version(str);
 519 
 520   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 521   assert(n > 0, "cannot retrieve pthread version");
 522   str = (char *)malloc(n, mtInternal);
 523   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 524   os::Linux::set_libpthread_version(str);
 525 }
 526 
 527 /////////////////////////////////////////////////////////////////////////////
 528 // thread stack expansion
 529 
 530 // os::Linux::manually_expand_stack() takes care of expanding the thread
 531 // stack. Note that this is normally not needed: pthread stacks allocate
 532 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 533 // committed. Therefore it is not necessary to expand the stack manually.
 534 //
 535 // Manually expanding the stack was historically needed on LinuxThreads
 536 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 537 // it is kept to deal with very rare corner cases:
 538 //
 539 // For one, user may run the VM on an own implementation of threads
 540 // whose stacks are - like the old LinuxThreads - implemented using
 541 // mmap(MAP_GROWSDOWN).
 542 //
 543 // Also, this coding may be needed if the VM is running on the primordial
 544 // thread. Normally we avoid running on the primordial thread; however,
 545 // user may still invoke the VM on the primordial thread.
 546 //
 547 // The following historical comment describes the details about running
 548 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 549 
 550 
 551 // Force Linux kernel to expand current thread stack. If "bottom" is close
 552 // to the stack guard, caller should block all signals.
 553 //
 554 // MAP_GROWSDOWN:
 555 //   A special mmap() flag that is used to implement thread stacks. It tells
 556 //   kernel that the memory region should extend downwards when needed. This
 557 //   allows early versions of LinuxThreads to only mmap the first few pages
 558 //   when creating a new thread. Linux kernel will automatically expand thread
 559 //   stack as needed (on page faults).
 560 //
 561 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 562 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 563 //   region, it's hard to tell if the fault is due to a legitimate stack
 564 //   access or because of reading/writing non-exist memory (e.g. buffer
 565 //   overrun). As a rule, if the fault happens below current stack pointer,
 566 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 567 //   application (see Linux kernel fault.c).
 568 //
 569 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 570 //   stack overflow detection.
 571 //
 572 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 573 //   not use MAP_GROWSDOWN.
 574 //
 575 // To get around the problem and allow stack banging on Linux, we need to
 576 // manually expand thread stack after receiving the SIGSEGV.
 577 //
 578 // There are two ways to expand thread stack to address "bottom", we used
 579 // both of them in JVM before 1.5:
 580 //   1. adjust stack pointer first so that it is below "bottom", and then
 581 //      touch "bottom"
 582 //   2. mmap() the page in question
 583 //
 584 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 585 // if current sp is already near the lower end of page 101, and we need to
 586 // call mmap() to map page 100, it is possible that part of the mmap() frame
 587 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 588 // That will destroy the mmap() frame and cause VM to crash.
 589 //
 590 // The following code works by adjusting sp first, then accessing the "bottom"
 591 // page to force a page fault. Linux kernel will then automatically expand the
 592 // stack mapping.
 593 //
 594 // _expand_stack_to() assumes its frame size is less than page size, which
 595 // should always be true if the function is not inlined.
 596 
 597 #if __GNUC__ < 3    // gcc 2.x does not support noinline attribute
 598   #define NOINLINE
 599 #else
 600   #define NOINLINE __attribute__ ((noinline))
 601 #endif
 602 
 603 static void _expand_stack_to(address bottom) NOINLINE;
 604 
 605 static void _expand_stack_to(address bottom) {
 606   address sp;
 607   size_t size;
 608   volatile char *p;
 609 
 610   // Adjust bottom to point to the largest address within the same page, it
 611   // gives us a one-page buffer if alloca() allocates slightly more memory.
 612   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 613   bottom += os::Linux::page_size() - 1;
 614 
 615   // sp might be slightly above current stack pointer; if that's the case, we
 616   // will alloca() a little more space than necessary, which is OK. Don't use
 617   // os::current_stack_pointer(), as its result can be slightly below current
 618   // stack pointer, causing us to not alloca enough to reach "bottom".
 619   sp = (address)&sp;
 620 
 621   if (sp > bottom) {
 622     size = sp - bottom;
 623     p = (volatile char *)alloca(size);
 624     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 625     p[0] = '\0';
 626   }
 627 }
 628 
 629 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 630   assert(t!=NULL, "just checking");
 631   assert(t->osthread()->expanding_stack(), "expand should be set");
 632   assert(t->stack_base() != NULL, "stack_base was not initialized");
 633 
 634   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 635     sigset_t mask_all, old_sigset;
 636     sigfillset(&mask_all);
 637     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 638     _expand_stack_to(addr);
 639     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 640     return true;
 641   }
 642   return false;
 643 }
 644 
 645 //////////////////////////////////////////////////////////////////////////////
 646 // create new thread
 647 
 648 // Thread start routine for all newly created threads
 649 static void *java_start(Thread *thread) {
 650   // Try to randomize the cache line index of hot stack frames.
 651   // This helps when threads of the same stack traces evict each other's
 652   // cache lines. The threads can be either from the same JVM instance, or
 653   // from different JVM instances. The benefit is especially true for
 654   // processors with hyperthreading technology.
 655   static int counter = 0;
 656   int pid = os::current_process_id();
 657   alloca(((pid ^ counter++) & 7) * 128);
 658 
 659   thread->initialize_thread_current();
 660 
 661   OSThread* osthread = thread->osthread();
 662   Monitor* sync = osthread->startThread_lock();
 663 
 664   osthread->set_thread_id(os::current_thread_id());
 665 
 666   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 667     os::current_thread_id(), (uintx) pthread_self());
 668 
 669   if (UseNUMA) {
 670     int lgrp_id = os::numa_get_group_id();
 671     if (lgrp_id != -1) {
 672       thread->set_lgrp_id(lgrp_id);
 673     }
 674   }
 675   // initialize signal mask for this thread
 676   os::Linux::hotspot_sigmask(thread);
 677 
 678   // initialize floating point control register
 679   os::Linux::init_thread_fpu_state();
 680 
 681   // handshaking with parent thread
 682   {
 683     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 684 
 685     // notify parent thread
 686     osthread->set_state(INITIALIZED);
 687     sync->notify_all();
 688 
 689     // wait until os::start_thread()
 690     while (osthread->get_state() == INITIALIZED) {
 691       sync->wait(Mutex::_no_safepoint_check_flag);
 692     }
 693   }
 694 
 695   // call one more level start routine
 696   thread->run();
 697 
 698   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 699     os::current_thread_id(), (uintx) pthread_self());
 700 
 701   return 0;
 702 }
 703 
 704 bool os::create_thread(Thread* thread, ThreadType thr_type,
 705                        size_t stack_size) {
 706   assert(thread->osthread() == NULL, "caller responsible");
 707 
 708   // Allocate the OSThread object
 709   OSThread* osthread = new OSThread(NULL, NULL);
 710   if (osthread == NULL) {
 711     return false;
 712   }
 713 
 714   // set the correct thread state
 715   osthread->set_thread_type(thr_type);
 716 
 717   // Initial state is ALLOCATED but not INITIALIZED
 718   osthread->set_state(ALLOCATED);
 719 
 720   thread->set_osthread(osthread);
 721 
 722   // init thread attributes
 723   pthread_attr_t attr;
 724   pthread_attr_init(&attr);
 725   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 726 
 727   // stack size
 728   // calculate stack size if it's not specified by caller
 729   if (stack_size == 0) {
 730     stack_size = os::Linux::default_stack_size(thr_type);
 731 
 732     switch (thr_type) {
 733     case os::java_thread:
 734       // Java threads use ThreadStackSize which default value can be
 735       // changed with the flag -Xss
 736       assert(JavaThread::stack_size_at_create() > 0, "this should be set");
 737       stack_size = JavaThread::stack_size_at_create();
 738       break;
 739     case os::compiler_thread:
 740       if (CompilerThreadStackSize > 0) {
 741         stack_size = (size_t)(CompilerThreadStackSize * K);
 742         break;
 743       } // else fall through:
 744         // use VMThreadStackSize if CompilerThreadStackSize is not defined
 745     case os::vm_thread:
 746     case os::pgc_thread:
 747     case os::cgc_thread:
 748     case os::watcher_thread:
 749       if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 750       break;
 751     }
 752   }
 753 
 754   stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 755   pthread_attr_setstacksize(&attr, stack_size);
 756 
 757   // glibc guard page
 758   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 759 
 760   ThreadState state;
 761 
 762   {
 763     pthread_t tid;
 764     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
 765 
 766     char buf[64];
 767     if (ret == 0) {
 768       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 769         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 770     } else {
 771       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 772         strerror(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 773     }
 774 
 775     pthread_attr_destroy(&attr);
 776 
 777     if (ret != 0) {
 778       // Need to clean up stuff we've allocated so far
 779       thread->set_osthread(NULL);
 780       delete osthread;
 781       return false;
 782     }
 783 
 784     // Store pthread info into the OSThread
 785     osthread->set_pthread_id(tid);
 786 
 787     // Wait until child thread is either initialized or aborted
 788     {
 789       Monitor* sync_with_child = osthread->startThread_lock();
 790       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 791       while ((state = osthread->get_state()) == ALLOCATED) {
 792         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 793       }
 794     }
 795   }
 796 
 797   // Aborted due to thread limit being reached
 798   if (state == ZOMBIE) {
 799     thread->set_osthread(NULL);
 800     delete osthread;
 801     return false;
 802   }
 803 
 804   // The thread is returned suspended (in state INITIALIZED),
 805   // and is started higher up in the call chain
 806   assert(state == INITIALIZED, "race condition");
 807   return true;
 808 }
 809 
 810 /////////////////////////////////////////////////////////////////////////////
 811 // attach existing thread
 812 
 813 // bootstrap the main thread
 814 bool os::create_main_thread(JavaThread* thread) {
 815   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 816   return create_attached_thread(thread);
 817 }
 818 
 819 bool os::create_attached_thread(JavaThread* thread) {
 820 #ifdef ASSERT
 821   thread->verify_not_published();
 822 #endif
 823 
 824   // Allocate the OSThread object
 825   OSThread* osthread = new OSThread(NULL, NULL);
 826 
 827   if (osthread == NULL) {
 828     return false;
 829   }
 830 
 831   // Store pthread info into the OSThread
 832   osthread->set_thread_id(os::Linux::gettid());
 833   osthread->set_pthread_id(::pthread_self());
 834 
 835   // initialize floating point control register
 836   os::Linux::init_thread_fpu_state();
 837 
 838   // Initial thread state is RUNNABLE
 839   osthread->set_state(RUNNABLE);
 840 
 841   thread->set_osthread(osthread);
 842 
 843   if (UseNUMA) {
 844     int lgrp_id = os::numa_get_group_id();
 845     if (lgrp_id != -1) {
 846       thread->set_lgrp_id(lgrp_id);
 847     }
 848   }
 849 
 850   if (os::Linux::is_initial_thread()) {
 851     // If current thread is initial thread, its stack is mapped on demand,
 852     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 853     // the entire stack region to avoid SEGV in stack banging.
 854     // It is also useful to get around the heap-stack-gap problem on SuSE
 855     // kernel (see 4821821 for details). We first expand stack to the top
 856     // of yellow zone, then enable stack yellow zone (order is significant,
 857     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 858     // is no gap between the last two virtual memory regions.
 859 
 860     JavaThread *jt = (JavaThread *)thread;
 861     address addr = jt->stack_reserved_zone_base();
 862     assert(addr != NULL, "initialization problem?");
 863     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 864 
 865     osthread->set_expanding_stack();
 866     os::Linux::manually_expand_stack(jt, addr);
 867     osthread->clear_expanding_stack();
 868   }
 869 
 870   // initialize signal mask for this thread
 871   // and save the caller's signal mask
 872   os::Linux::hotspot_sigmask(thread);
 873 
 874   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 875     os::current_thread_id(), (uintx) pthread_self());
 876 
 877   return true;
 878 }
 879 
 880 void os::pd_start_thread(Thread* thread) {
 881   OSThread * osthread = thread->osthread();
 882   assert(osthread->get_state() != INITIALIZED, "just checking");
 883   Monitor* sync_with_child = osthread->startThread_lock();
 884   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 885   sync_with_child->notify();
 886 }
 887 
 888 // Free Linux resources related to the OSThread
 889 void os::free_thread(OSThread* osthread) {
 890   assert(osthread != NULL, "osthread not set");
 891 
 892   if (Thread::current()->osthread() == osthread) {
 893     // Restore caller's signal mask
 894     sigset_t sigmask = osthread->caller_sigmask();
 895     pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 896   }
 897 
 898   delete osthread;
 899 }
 900 
 901 //////////////////////////////////////////////////////////////////////////////
 902 // initial thread
 903 
 904 // Check if current thread is the initial thread, similar to Solaris thr_main.
 905 bool os::Linux::is_initial_thread(void) {
 906   char dummy;
 907   // If called before init complete, thread stack bottom will be null.
 908   // Can be called if fatal error occurs before initialization.
 909   if (initial_thread_stack_bottom() == NULL) return false;
 910   assert(initial_thread_stack_bottom() != NULL &&
 911          initial_thread_stack_size()   != 0,
 912          "os::init did not locate initial thread's stack region");
 913   if ((address)&dummy >= initial_thread_stack_bottom() &&
 914       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 915     return true;
 916   } else {
 917     return false;
 918   }
 919 }
 920 
 921 // Find the virtual memory area that contains addr
 922 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 923   FILE *fp = fopen("/proc/self/maps", "r");
 924   if (fp) {
 925     address low, high;
 926     while (!feof(fp)) {
 927       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 928         if (low <= addr && addr < high) {
 929           if (vma_low)  *vma_low  = low;
 930           if (vma_high) *vma_high = high;
 931           fclose(fp);
 932           return true;
 933         }
 934       }
 935       for (;;) {
 936         int ch = fgetc(fp);
 937         if (ch == EOF || ch == (int)'\n') break;
 938       }
 939     }
 940     fclose(fp);
 941   }
 942   return false;
 943 }
 944 
 945 // Locate initial thread stack. This special handling of initial thread stack
 946 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 947 // bogus value for initial thread.
 948 void os::Linux::capture_initial_stack(size_t max_size) {
 949   // stack size is the easy part, get it from RLIMIT_STACK
 950   size_t stack_size;
 951   struct rlimit rlim;
 952   getrlimit(RLIMIT_STACK, &rlim);
 953   stack_size = rlim.rlim_cur;
 954 
 955   // 6308388: a bug in ld.so will relocate its own .data section to the
 956   //   lower end of primordial stack; reduce ulimit -s value a little bit
 957   //   so we won't install guard page on ld.so's data section.
 958   stack_size -= 2 * page_size();
 959 
 960   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
 961   //   7.1, in both cases we will get 2G in return value.
 962   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
 963   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
 964   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
 965   //   in case other parts in glibc still assumes 2M max stack size.
 966   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
 967   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
 968   if (stack_size > 2 * K * K IA64_ONLY(*2)) {
 969     stack_size = 2 * K * K IA64_ONLY(*2);
 970   }
 971   // Try to figure out where the stack base (top) is. This is harder.
 972   //
 973   // When an application is started, glibc saves the initial stack pointer in
 974   // a global variable "__libc_stack_end", which is then used by system
 975   // libraries. __libc_stack_end should be pretty close to stack top. The
 976   // variable is available since the very early days. However, because it is
 977   // a private interface, it could disappear in the future.
 978   //
 979   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 980   // to __libc_stack_end, it is very close to stack top, but isn't the real
 981   // stack top. Note that /proc may not exist if VM is running as a chroot
 982   // program, so reading /proc/<pid>/stat could fail. Also the contents of
 983   // /proc/<pid>/stat could change in the future (though unlikely).
 984   //
 985   // We try __libc_stack_end first. If that doesn't work, look for
 986   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
 987   // as a hint, which should work well in most cases.
 988 
 989   uintptr_t stack_start;
 990 
 991   // try __libc_stack_end first
 992   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
 993   if (p && *p) {
 994     stack_start = *p;
 995   } else {
 996     // see if we can get the start_stack field from /proc/self/stat
 997     FILE *fp;
 998     int pid;
 999     char state;
1000     int ppid;
1001     int pgrp;
1002     int session;
1003     int nr;
1004     int tpgrp;
1005     unsigned long flags;
1006     unsigned long minflt;
1007     unsigned long cminflt;
1008     unsigned long majflt;
1009     unsigned long cmajflt;
1010     unsigned long utime;
1011     unsigned long stime;
1012     long cutime;
1013     long cstime;
1014     long prio;
1015     long nice;
1016     long junk;
1017     long it_real;
1018     uintptr_t start;
1019     uintptr_t vsize;
1020     intptr_t rss;
1021     uintptr_t rsslim;
1022     uintptr_t scodes;
1023     uintptr_t ecode;
1024     int i;
1025 
1026     // Figure what the primordial thread stack base is. Code is inspired
1027     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1028     // followed by command name surrounded by parentheses, state, etc.
1029     char stat[2048];
1030     int statlen;
1031 
1032     fp = fopen("/proc/self/stat", "r");
1033     if (fp) {
1034       statlen = fread(stat, 1, 2047, fp);
1035       stat[statlen] = '\0';
1036       fclose(fp);
1037 
1038       // Skip pid and the command string. Note that we could be dealing with
1039       // weird command names, e.g. user could decide to rename java launcher
1040       // to "java 1.4.2 :)", then the stat file would look like
1041       //                1234 (java 1.4.2 :)) R ... ...
1042       // We don't really need to know the command string, just find the last
1043       // occurrence of ")" and then start parsing from there. See bug 4726580.
1044       char * s = strrchr(stat, ')');
1045 
1046       i = 0;
1047       if (s) {
1048         // Skip blank chars
1049         do { s++; } while (s && isspace(*s));
1050 
1051 #define _UFM UINTX_FORMAT
1052 #define _DFM INTX_FORMAT
1053 
1054         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1055         //              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
1056         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,
1057                    &state,          // 3  %c
1058                    &ppid,           // 4  %d
1059                    &pgrp,           // 5  %d
1060                    &session,        // 6  %d
1061                    &nr,             // 7  %d
1062                    &tpgrp,          // 8  %d
1063                    &flags,          // 9  %lu
1064                    &minflt,         // 10 %lu
1065                    &cminflt,        // 11 %lu
1066                    &majflt,         // 12 %lu
1067                    &cmajflt,        // 13 %lu
1068                    &utime,          // 14 %lu
1069                    &stime,          // 15 %lu
1070                    &cutime,         // 16 %ld
1071                    &cstime,         // 17 %ld
1072                    &prio,           // 18 %ld
1073                    &nice,           // 19 %ld
1074                    &junk,           // 20 %ld
1075                    &it_real,        // 21 %ld
1076                    &start,          // 22 UINTX_FORMAT
1077                    &vsize,          // 23 UINTX_FORMAT
1078                    &rss,            // 24 INTX_FORMAT
1079                    &rsslim,         // 25 UINTX_FORMAT
1080                    &scodes,         // 26 UINTX_FORMAT
1081                    &ecode,          // 27 UINTX_FORMAT
1082                    &stack_start);   // 28 UINTX_FORMAT
1083       }
1084 
1085 #undef _UFM
1086 #undef _DFM
1087 
1088       if (i != 28 - 2) {
1089         assert(false, "Bad conversion from /proc/self/stat");
1090         // product mode - assume we are the initial thread, good luck in the
1091         // embedded case.
1092         warning("Can't detect initial thread stack location - bad conversion");
1093         stack_start = (uintptr_t) &rlim;
1094       }
1095     } else {
1096       // For some reason we can't open /proc/self/stat (for example, running on
1097       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1098       // most cases, so don't abort:
1099       warning("Can't detect initial thread stack location - no /proc/self/stat");
1100       stack_start = (uintptr_t) &rlim;
1101     }
1102   }
1103 
1104   // Now we have a pointer (stack_start) very close to the stack top, the
1105   // next thing to do is to figure out the exact location of stack top. We
1106   // can find out the virtual memory area that contains stack_start by
1107   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1108   // and its upper limit is the real stack top. (again, this would fail if
1109   // running inside chroot, because /proc may not exist.)
1110 
1111   uintptr_t stack_top;
1112   address low, high;
1113   if (find_vma((address)stack_start, &low, &high)) {
1114     // success, "high" is the true stack top. (ignore "low", because initial
1115     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1116     stack_top = (uintptr_t)high;
1117   } else {
1118     // failed, likely because /proc/self/maps does not exist
1119     warning("Can't detect initial thread stack location - find_vma failed");
1120     // best effort: stack_start is normally within a few pages below the real
1121     // stack top, use it as stack top, and reduce stack size so we won't put
1122     // guard page outside stack.
1123     stack_top = stack_start;
1124     stack_size -= 16 * page_size();
1125   }
1126 
1127   // stack_top could be partially down the page so align it
1128   stack_top = align_size_up(stack_top, page_size());
1129 
1130   if (max_size && stack_size > max_size) {
1131     _initial_thread_stack_size = max_size;
1132   } else {
1133     _initial_thread_stack_size = stack_size;
1134   }
1135 
1136   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1137   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1138 }
1139 
1140 ////////////////////////////////////////////////////////////////////////////////
1141 // time support
1142 
1143 // Time since start-up in seconds to a fine granularity.
1144 // Used by VMSelfDestructTimer and the MemProfiler.
1145 double os::elapsedTime() {
1146 
1147   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1148 }
1149 
1150 jlong os::elapsed_counter() {
1151   return javaTimeNanos() - initial_time_count;
1152 }
1153 
1154 jlong os::elapsed_frequency() {
1155   return NANOSECS_PER_SEC; // nanosecond resolution
1156 }
1157 
1158 bool os::supports_vtime() { return true; }
1159 bool os::enable_vtime()   { return false; }
1160 bool os::vtime_enabled()  { return false; }
1161 
1162 double os::elapsedVTime() {
1163   struct rusage usage;
1164   int retval = getrusage(RUSAGE_THREAD, &usage);
1165   if (retval == 0) {
1166     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);
1167   } else {
1168     // better than nothing, but not much
1169     return elapsedTime();
1170   }
1171 }
1172 
1173 jlong os::javaTimeMillis() {
1174   timeval time;
1175   int status = gettimeofday(&time, NULL);
1176   assert(status != -1, "linux error");
1177   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1178 }
1179 
1180 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1181   timeval time;
1182   int status = gettimeofday(&time, NULL);
1183   assert(status != -1, "linux error");
1184   seconds = jlong(time.tv_sec);
1185   nanos = jlong(time.tv_usec) * 1000;
1186 }
1187 
1188 
1189 #ifndef CLOCK_MONOTONIC
1190   #define CLOCK_MONOTONIC (1)
1191 #endif
1192 
1193 void os::Linux::clock_init() {
1194   // we do dlopen's in this particular order due to bug in linux
1195   // dynamical loader (see 6348968) leading to crash on exit
1196   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1197   if (handle == NULL) {
1198     handle = dlopen("librt.so", RTLD_LAZY);
1199   }
1200 
1201   if (handle) {
1202     int (*clock_getres_func)(clockid_t, struct timespec*) =
1203            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1204     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1205            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1206     if (clock_getres_func && clock_gettime_func) {
1207       // See if monotonic clock is supported by the kernel. Note that some
1208       // early implementations simply return kernel jiffies (updated every
1209       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1210       // for nano time (though the monotonic property is still nice to have).
1211       // It's fixed in newer kernels, however clock_getres() still returns
1212       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1213       // resolution for now. Hopefully as people move to new kernels, this
1214       // won't be a problem.
1215       struct timespec res;
1216       struct timespec tp;
1217       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1218           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1219         // yes, monotonic clock is supported
1220         _clock_gettime = clock_gettime_func;
1221         return;
1222       } else {
1223         // close librt if there is no monotonic clock
1224         dlclose(handle);
1225       }
1226     }
1227   }
1228   warning("No monotonic clock was available - timed services may " \
1229           "be adversely affected if the time-of-day clock changes");
1230 }
1231 
1232 #ifndef SYS_clock_getres
1233   #if defined(IA32) || defined(AMD64)
1234     #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1235     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1236   #else
1237     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1238     #define sys_clock_getres(x,y)  -1
1239   #endif
1240 #else
1241   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1242 #endif
1243 
1244 void os::Linux::fast_thread_clock_init() {
1245   if (!UseLinuxPosixThreadCPUClocks) {
1246     return;
1247   }
1248   clockid_t clockid;
1249   struct timespec tp;
1250   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1251       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1252 
1253   // Switch to using fast clocks for thread cpu time if
1254   // the sys_clock_getres() returns 0 error code.
1255   // Note, that some kernels may support the current thread
1256   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1257   // returned by the pthread_getcpuclockid().
1258   // If the fast Posix clocks are supported then the sys_clock_getres()
1259   // must return at least tp.tv_sec == 0 which means a resolution
1260   // better than 1 sec. This is extra check for reliability.
1261 
1262   if (pthread_getcpuclockid_func &&
1263       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1264       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1265     _supports_fast_thread_cpu_time = true;
1266     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1267   }
1268 }
1269 
1270 jlong os::javaTimeNanos() {
1271   if (os::supports_monotonic_clock()) {
1272     struct timespec tp;
1273     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1274     assert(status == 0, "gettime error");
1275     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1276     return result;
1277   } else {
1278     timeval time;
1279     int status = gettimeofday(&time, NULL);
1280     assert(status != -1, "linux error");
1281     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1282     return 1000 * usecs;
1283   }
1284 }
1285 
1286 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1287   if (os::supports_monotonic_clock()) {
1288     info_ptr->max_value = ALL_64_BITS;
1289 
1290     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1291     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1292     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1293   } else {
1294     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1295     info_ptr->max_value = ALL_64_BITS;
1296 
1297     // gettimeofday is a real time clock so it skips
1298     info_ptr->may_skip_backward = true;
1299     info_ptr->may_skip_forward = true;
1300   }
1301 
1302   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1303 }
1304 
1305 // Return the real, user, and system times in seconds from an
1306 // arbitrary fixed point in the past.
1307 bool os::getTimesSecs(double* process_real_time,
1308                       double* process_user_time,
1309                       double* process_system_time) {
1310   struct tms ticks;
1311   clock_t real_ticks = times(&ticks);
1312 
1313   if (real_ticks == (clock_t) (-1)) {
1314     return false;
1315   } else {
1316     double ticks_per_second = (double) clock_tics_per_sec;
1317     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1318     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1319     *process_real_time = ((double) real_ticks) / ticks_per_second;
1320 
1321     return true;
1322   }
1323 }
1324 
1325 
1326 char * os::local_time_string(char *buf, size_t buflen) {
1327   struct tm t;
1328   time_t long_time;
1329   time(&long_time);
1330   localtime_r(&long_time, &t);
1331   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1332                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1333                t.tm_hour, t.tm_min, t.tm_sec);
1334   return buf;
1335 }
1336 
1337 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1338   return localtime_r(clock, res);
1339 }
1340 
1341 ////////////////////////////////////////////////////////////////////////////////
1342 // runtime exit support
1343 
1344 // Note: os::shutdown() might be called very early during initialization, or
1345 // called from signal handler. Before adding something to os::shutdown(), make
1346 // sure it is async-safe and can handle partially initialized VM.
1347 void os::shutdown() {
1348 
1349   // allow PerfMemory to attempt cleanup of any persistent resources
1350   perfMemory_exit();
1351 
1352   // needs to remove object in file system
1353   AttachListener::abort();
1354 
1355   // flush buffered output, finish log files
1356   ostream_abort();
1357 
1358   // Check for abort hook
1359   abort_hook_t abort_hook = Arguments::abort_hook();
1360   if (abort_hook != NULL) {
1361     abort_hook();
1362   }
1363 
1364 }
1365 
1366 // Note: os::abort() might be called very early during initialization, or
1367 // called from signal handler. Before adding something to os::abort(), make
1368 // sure it is async-safe and can handle partially initialized VM.
1369 void os::abort(bool dump_core, void* siginfo, const void* context) {
1370   os::shutdown();
1371   if (dump_core) {
1372 #ifndef PRODUCT
1373     fdStream out(defaultStream::output_fd());
1374     out.print_raw("Current thread is ");
1375     char buf[16];
1376     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1377     out.print_raw_cr(buf);
1378     out.print_raw_cr("Dumping core ...");
1379 #endif
1380     ::abort(); // dump core
1381   }
1382 
1383   ::exit(1);
1384 }
1385 
1386 // Die immediately, no exit hook, no abort hook, no cleanup.
1387 void os::die() {
1388   ::abort();
1389 }
1390 
1391 
1392 // This method is a copy of JDK's sysGetLastErrorString
1393 // from src/solaris/hpi/src/system_md.c
1394 
1395 size_t os::lasterror(char *buf, size_t len) {
1396   if (errno == 0)  return 0;
1397 
1398   const char *s = ::strerror(errno);
1399   size_t n = ::strlen(s);
1400   if (n >= len) {
1401     n = len - 1;
1402   }
1403   ::strncpy(buf, s, n);
1404   buf[n] = '\0';
1405   return n;
1406 }
1407 
1408 // thread_id is kernel thread id (similar to Solaris LWP id)
1409 intx os::current_thread_id() { return os::Linux::gettid(); }
1410 int os::current_process_id() {
1411   return ::getpid();
1412 }
1413 
1414 // DLL functions
1415 
1416 const char* os::dll_file_extension() { return ".so"; }
1417 
1418 // This must be hard coded because it's the system's temporary
1419 // directory not the java application's temp directory, ala java.io.tmpdir.
1420 const char* os::get_temp_directory() { return "/tmp"; }
1421 
1422 static bool file_exists(const char* filename) {
1423   struct stat statbuf;
1424   if (filename == NULL || strlen(filename) == 0) {
1425     return false;
1426   }
1427   return os::stat(filename, &statbuf) == 0;
1428 }
1429 
1430 bool os::dll_build_name(char* buffer, size_t buflen,
1431                         const char* pname, const char* fname) {
1432   bool retval = false;
1433   // Copied from libhpi
1434   const size_t pnamelen = pname ? strlen(pname) : 0;
1435 
1436   // Return error on buffer overflow.
1437   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1438     return retval;
1439   }
1440 
1441   if (pnamelen == 0) {
1442     snprintf(buffer, buflen, "lib%s.so", fname);
1443     retval = true;
1444   } else if (strchr(pname, *os::path_separator()) != NULL) {
1445     int n;
1446     char** pelements = split_path(pname, &n);
1447     if (pelements == NULL) {
1448       return false;
1449     }
1450     for (int i = 0; i < n; i++) {
1451       // Really shouldn't be NULL, but check can't hurt
1452       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1453         continue; // skip the empty path values
1454       }
1455       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1456       if (file_exists(buffer)) {
1457         retval = true;
1458         break;
1459       }
1460     }
1461     // release the storage
1462     for (int i = 0; i < n; i++) {
1463       if (pelements[i] != NULL) {
1464         FREE_C_HEAP_ARRAY(char, pelements[i]);
1465       }
1466     }
1467     if (pelements != NULL) {
1468       FREE_C_HEAP_ARRAY(char*, pelements);
1469     }
1470   } else {
1471     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1472     retval = true;
1473   }
1474   return retval;
1475 }
1476 
1477 // check if addr is inside libjvm.so
1478 bool os::address_is_in_vm(address addr) {
1479   static address libjvm_base_addr;
1480   Dl_info dlinfo;
1481 
1482   if (libjvm_base_addr == NULL) {
1483     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1484       libjvm_base_addr = (address)dlinfo.dli_fbase;
1485     }
1486     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1487   }
1488 
1489   if (dladdr((void *)addr, &dlinfo) != 0) {
1490     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1491   }
1492 
1493   return false;
1494 }
1495 
1496 bool os::dll_address_to_function_name(address addr, char *buf,
1497                                       int buflen, int *offset,
1498                                       bool demangle) {
1499   // buf is not optional, but offset is optional
1500   assert(buf != NULL, "sanity check");
1501 
1502   Dl_info dlinfo;
1503 
1504   if (dladdr((void*)addr, &dlinfo) != 0) {
1505     // see if we have a matching symbol
1506     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1507       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1508         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1509       }
1510       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1511       return true;
1512     }
1513     // no matching symbol so try for just file info
1514     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1515       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1516                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1517         return true;
1518       }
1519     }
1520   }
1521 
1522   buf[0] = '\0';
1523   if (offset != NULL) *offset = -1;
1524   return false;
1525 }
1526 
1527 struct _address_to_library_name {
1528   address addr;          // input : memory address
1529   size_t  buflen;        //         size of fname
1530   char*   fname;         // output: library name
1531   address base;          //         library base addr
1532 };
1533 
1534 static int address_to_library_name_callback(struct dl_phdr_info *info,
1535                                             size_t size, void *data) {
1536   int i;
1537   bool found = false;
1538   address libbase = NULL;
1539   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1540 
1541   // iterate through all loadable segments
1542   for (i = 0; i < info->dlpi_phnum; i++) {
1543     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1544     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1545       // base address of a library is the lowest address of its loaded
1546       // segments.
1547       if (libbase == NULL || libbase > segbase) {
1548         libbase = segbase;
1549       }
1550       // see if 'addr' is within current segment
1551       if (segbase <= d->addr &&
1552           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1553         found = true;
1554       }
1555     }
1556   }
1557 
1558   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1559   // so dll_address_to_library_name() can fall through to use dladdr() which
1560   // can figure out executable name from argv[0].
1561   if (found && info->dlpi_name && info->dlpi_name[0]) {
1562     d->base = libbase;
1563     if (d->fname) {
1564       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1565     }
1566     return 1;
1567   }
1568   return 0;
1569 }
1570 
1571 bool os::dll_address_to_library_name(address addr, char* buf,
1572                                      int buflen, int* offset) {
1573   // buf is not optional, but offset is optional
1574   assert(buf != NULL, "sanity check");
1575 
1576   Dl_info dlinfo;
1577   struct _address_to_library_name data;
1578 
1579   // There is a bug in old glibc dladdr() implementation that it could resolve
1580   // to wrong library name if the .so file has a base address != NULL. Here
1581   // we iterate through the program headers of all loaded libraries to find
1582   // out which library 'addr' really belongs to. This workaround can be
1583   // removed once the minimum requirement for glibc is moved to 2.3.x.
1584   data.addr = addr;
1585   data.fname = buf;
1586   data.buflen = buflen;
1587   data.base = NULL;
1588   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1589 
1590   if (rslt) {
1591     // buf already contains library name
1592     if (offset) *offset = addr - data.base;
1593     return true;
1594   }
1595   if (dladdr((void*)addr, &dlinfo) != 0) {
1596     if (dlinfo.dli_fname != NULL) {
1597       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1598     }
1599     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1600       *offset = addr - (address)dlinfo.dli_fbase;
1601     }
1602     return true;
1603   }
1604 
1605   buf[0] = '\0';
1606   if (offset) *offset = -1;
1607   return false;
1608 }
1609 
1610 // Loads .dll/.so and
1611 // in case of error it checks if .dll/.so was built for the
1612 // same architecture as Hotspot is running on
1613 
1614 
1615 // Remember the stack's state. The Linux dynamic linker will change
1616 // the stack to 'executable' at most once, so we must safepoint only once.
1617 bool os::Linux::_stack_is_executable = false;
1618 
1619 // VM operation that loads a library.  This is necessary if stack protection
1620 // of the Java stacks can be lost during loading the library.  If we
1621 // do not stop the Java threads, they can stack overflow before the stacks
1622 // are protected again.
1623 class VM_LinuxDllLoad: public VM_Operation {
1624  private:
1625   const char *_filename;
1626   char *_ebuf;
1627   int _ebuflen;
1628   void *_lib;
1629  public:
1630   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1631     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1632   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1633   void doit() {
1634     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1635     os::Linux::_stack_is_executable = true;
1636   }
1637   void* loaded_library() { return _lib; }
1638 };
1639 
1640 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1641   void * result = NULL;
1642   bool load_attempted = false;
1643 
1644   // Check whether the library to load might change execution rights
1645   // of the stack. If they are changed, the protection of the stack
1646   // guard pages will be lost. We need a safepoint to fix this.
1647   //
1648   // See Linux man page execstack(8) for more info.
1649   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1650     ElfFile ef(filename);
1651     if (!ef.specifies_noexecstack()) {
1652       if (!is_init_completed()) {
1653         os::Linux::_stack_is_executable = true;
1654         // This is OK - No Java threads have been created yet, and hence no
1655         // stack guard pages to fix.
1656         //
1657         // This should happen only when you are building JDK7 using a very
1658         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1659         //
1660         // Dynamic loader will make all stacks executable after
1661         // this function returns, and will not do that again.
1662         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1663       } else {
1664         warning("You have loaded library %s which might have disabled stack guard. "
1665                 "The VM will try to fix the stack guard now.\n"
1666                 "It's highly recommended that you fix the library with "
1667                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1668                 filename);
1669 
1670         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1671         JavaThread *jt = JavaThread::current();
1672         if (jt->thread_state() != _thread_in_native) {
1673           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1674           // that requires ExecStack. Cannot enter safe point. Let's give up.
1675           warning("Unable to fix stack guard. Giving up.");
1676         } else {
1677           if (!LoadExecStackDllInVMThread) {
1678             // This is for the case where the DLL has an static
1679             // constructor function that executes JNI code. We cannot
1680             // load such DLLs in the VMThread.
1681             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1682           }
1683 
1684           ThreadInVMfromNative tiv(jt);
1685           debug_only(VMNativeEntryWrapper vew;)
1686 
1687           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1688           VMThread::execute(&op);
1689           if (LoadExecStackDllInVMThread) {
1690             result = op.loaded_library();
1691           }
1692           load_attempted = true;
1693         }
1694       }
1695     }
1696   }
1697 
1698   if (!load_attempted) {
1699     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1700   }
1701 
1702   if (result != NULL) {
1703     // Successful loading
1704     return result;
1705   }
1706 
1707   Elf32_Ehdr elf_head;
1708   int diag_msg_max_length=ebuflen-strlen(ebuf);
1709   char* diag_msg_buf=ebuf+strlen(ebuf);
1710 
1711   if (diag_msg_max_length==0) {
1712     // No more space in ebuf for additional diagnostics message
1713     return NULL;
1714   }
1715 
1716 
1717   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1718 
1719   if (file_descriptor < 0) {
1720     // Can't open library, report dlerror() message
1721     return NULL;
1722   }
1723 
1724   bool failed_to_read_elf_head=
1725     (sizeof(elf_head)!=
1726      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1727 
1728   ::close(file_descriptor);
1729   if (failed_to_read_elf_head) {
1730     // file i/o error - report dlerror() msg
1731     return NULL;
1732   }
1733 
1734   typedef struct {
1735     Elf32_Half  code;         // Actual value as defined in elf.h
1736     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1737     char        elf_class;    // 32 or 64 bit
1738     char        endianess;    // MSB or LSB
1739     char*       name;         // String representation
1740   } arch_t;
1741 
1742 #ifndef EM_486
1743   #define EM_486          6               /* Intel 80486 */
1744 #endif
1745 #ifndef EM_AARCH64
1746   #define EM_AARCH64    183               /* ARM AARCH64 */
1747 #endif
1748 
1749   static const arch_t arch_array[]={
1750     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1751     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1752     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1753     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1754     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1755     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1756     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1757     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1758 #if defined(VM_LITTLE_ENDIAN)
1759     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1760 #else
1761     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1762 #endif
1763     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1764     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1765     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1766     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1767     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1768     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1769     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1770     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1771   };
1772 
1773 #if  (defined IA32)
1774   static  Elf32_Half running_arch_code=EM_386;
1775 #elif   (defined AMD64)
1776   static  Elf32_Half running_arch_code=EM_X86_64;
1777 #elif  (defined IA64)
1778   static  Elf32_Half running_arch_code=EM_IA_64;
1779 #elif  (defined __sparc) && (defined _LP64)
1780   static  Elf32_Half running_arch_code=EM_SPARCV9;
1781 #elif  (defined __sparc) && (!defined _LP64)
1782   static  Elf32_Half running_arch_code=EM_SPARC;
1783 #elif  (defined __powerpc64__)
1784   static  Elf32_Half running_arch_code=EM_PPC64;
1785 #elif  (defined __powerpc__)
1786   static  Elf32_Half running_arch_code=EM_PPC;
1787 #elif  (defined ARM)
1788   static  Elf32_Half running_arch_code=EM_ARM;
1789 #elif  (defined S390)
1790   static  Elf32_Half running_arch_code=EM_S390;
1791 #elif  (defined ALPHA)
1792   static  Elf32_Half running_arch_code=EM_ALPHA;
1793 #elif  (defined MIPSEL)
1794   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1795 #elif  (defined PARISC)
1796   static  Elf32_Half running_arch_code=EM_PARISC;
1797 #elif  (defined MIPS)
1798   static  Elf32_Half running_arch_code=EM_MIPS;
1799 #elif  (defined M68K)
1800   static  Elf32_Half running_arch_code=EM_68K;
1801 #elif  (defined AARCH64)
1802   static  Elf32_Half running_arch_code=EM_AARCH64;
1803 #else
1804     #error Method os::dll_load requires that one of following is defined:\
1805          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1806 #endif
1807 
1808   // Identify compatability class for VM's architecture and library's architecture
1809   // Obtain string descriptions for architectures
1810 
1811   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1812   int running_arch_index=-1;
1813 
1814   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1815     if (running_arch_code == arch_array[i].code) {
1816       running_arch_index    = i;
1817     }
1818     if (lib_arch.code == arch_array[i].code) {
1819       lib_arch.compat_class = arch_array[i].compat_class;
1820       lib_arch.name         = arch_array[i].name;
1821     }
1822   }
1823 
1824   assert(running_arch_index != -1,
1825          "Didn't find running architecture code (running_arch_code) in arch_array");
1826   if (running_arch_index == -1) {
1827     // Even though running architecture detection failed
1828     // we may still continue with reporting dlerror() message
1829     return NULL;
1830   }
1831 
1832   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1833     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1834     return NULL;
1835   }
1836 
1837 #ifndef S390
1838   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1839     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1840     return NULL;
1841   }
1842 #endif // !S390
1843 
1844   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1845     if (lib_arch.name!=NULL) {
1846       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1847                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1848                  lib_arch.name, arch_array[running_arch_index].name);
1849     } else {
1850       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1851                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1852                  lib_arch.code,
1853                  arch_array[running_arch_index].name);
1854     }
1855   }
1856 
1857   return NULL;
1858 }
1859 
1860 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1861                                 int ebuflen) {
1862   void * result = ::dlopen(filename, RTLD_LAZY);
1863   if (result == NULL) {
1864     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1865     ebuf[ebuflen-1] = '\0';
1866   }
1867   return result;
1868 }
1869 
1870 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1871                                        int ebuflen) {
1872   void * result = NULL;
1873   if (LoadExecStackDllInVMThread) {
1874     result = dlopen_helper(filename, ebuf, ebuflen);
1875   }
1876 
1877   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1878   // library that requires an executable stack, or which does not have this
1879   // stack attribute set, dlopen changes the stack attribute to executable. The
1880   // read protection of the guard pages gets lost.
1881   //
1882   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1883   // may have been queued at the same time.
1884 
1885   if (!_stack_is_executable) {
1886     JavaThread *jt = Threads::first();
1887 
1888     while (jt) {
1889       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1890           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1891         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1892           warning("Attempt to reguard stack yellow zone failed.");
1893         }
1894       }
1895       jt = jt->next();
1896     }
1897   }
1898 
1899   return result;
1900 }
1901 
1902 void* os::dll_lookup(void* handle, const char* name) {
1903   void* res = dlsym(handle, name);
1904   return res;
1905 }
1906 
1907 void* os::get_default_process_handle() {
1908   return (void*)::dlopen(NULL, RTLD_LAZY);
1909 }
1910 
1911 static bool _print_ascii_file(const char* filename, outputStream* st) {
1912   int fd = ::open(filename, O_RDONLY);
1913   if (fd == -1) {
1914     return false;
1915   }
1916 
1917   char buf[32];
1918   int bytes;
1919   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1920     st->print_raw(buf, bytes);
1921   }
1922 
1923   ::close(fd);
1924 
1925   return true;
1926 }
1927 
1928 void os::print_dll_info(outputStream *st) {
1929   st->print_cr("Dynamic libraries:");
1930 
1931   char fname[32];
1932   pid_t pid = os::Linux::gettid();
1933 
1934   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1935 
1936   if (!_print_ascii_file(fname, st)) {
1937     st->print("Can not get library information for pid = %d\n", pid);
1938   }
1939 }
1940 
1941 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1942   FILE *procmapsFile = NULL;
1943 
1944   // Open the procfs maps file for the current process
1945   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1946     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1947     char line[PATH_MAX + 100];
1948 
1949     // Read line by line from 'file'
1950     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1951       u8 base, top, offset, inode;
1952       char permissions[5];
1953       char device[6];
1954       char name[PATH_MAX + 1];
1955 
1956       // Parse fields from line
1957       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1958              &base, &top, permissions, &offset, device, &inode, name);
1959 
1960       // Filter by device id '00:00' so that we only get file system mapped files.
1961       if (strcmp(device, "00:00") != 0) {
1962 
1963         // Call callback with the fields of interest
1964         if(callback(name, (address)base, (address)top, param)) {
1965           // Oops abort, callback aborted
1966           fclose(procmapsFile);
1967           return 1;
1968         }
1969       }
1970     }
1971     fclose(procmapsFile);
1972   }
1973   return 0;
1974 }
1975 
1976 void os::print_os_info_brief(outputStream* st) {
1977   os::Linux::print_distro_info(st);
1978 
1979   os::Posix::print_uname_info(st);
1980 
1981   os::Linux::print_libversion_info(st);
1982 
1983 }
1984 
1985 void os::print_os_info(outputStream* st) {
1986   st->print("OS:");
1987 
1988   os::Linux::print_distro_info(st);
1989 
1990   os::Posix::print_uname_info(st);
1991 
1992   // Print warning if unsafe chroot environment detected
1993   if (unsafe_chroot_detected) {
1994     st->print("WARNING!! ");
1995     st->print_cr("%s", unstable_chroot_error);
1996   }
1997 
1998   os::Linux::print_libversion_info(st);
1999 
2000   os::Posix::print_rlimit_info(st);
2001 
2002   os::Posix::print_load_average(st);
2003 
2004   os::Linux::print_full_memory_info(st);
2005 }
2006 
2007 // Try to identify popular distros.
2008 // Most Linux distributions have a /etc/XXX-release file, which contains
2009 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2010 // file that also contains the OS version string. Some have more than one
2011 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2012 // /etc/redhat-release.), so the order is important.
2013 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2014 // their own specific XXX-release file as well as a redhat-release file.
2015 // Because of this the XXX-release file needs to be searched for before the
2016 // redhat-release file.
2017 // Since Red Hat has a lsb-release file that is not very descriptive the
2018 // search for redhat-release needs to be before lsb-release.
2019 // Since the lsb-release file is the new standard it needs to be searched
2020 // before the older style release files.
2021 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2022 // next to last resort.  The os-release file is a new standard that contains
2023 // distribution information and the system-release file seems to be an old
2024 // standard that has been replaced by the lsb-release and os-release files.
2025 // Searching for the debian_version file is the last resort.  It contains
2026 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2027 // "Debian " is printed before the contents of the debian_version file.
2028 
2029 const char* distro_files[] = {
2030   "/etc/oracle-release",
2031   "/etc/mandriva-release",
2032   "/etc/mandrake-release",
2033   "/etc/sun-release",
2034   "/etc/redhat-release",
2035   "/etc/lsb-release",
2036   "/etc/SuSE-release",
2037   "/etc/turbolinux-release",
2038   "/etc/gentoo-release",
2039   "/etc/ltib-release",
2040   "/etc/angstrom-version",
2041   "/etc/system-release",
2042   "/etc/os-release",
2043   NULL };
2044 
2045 void os::Linux::print_distro_info(outputStream* st) {
2046   for (int i = 0;; i++) {
2047     const char* file = distro_files[i];
2048     if (file == NULL) {
2049       break;  // done
2050     }
2051     // If file prints, we found it.
2052     if (_print_ascii_file(file, st)) {
2053       return;
2054     }
2055   }
2056 
2057   if (file_exists("/etc/debian_version")) {
2058     st->print("Debian ");
2059     _print_ascii_file("/etc/debian_version", st);
2060   } else {
2061     st->print("Linux");
2062   }
2063   st->cr();
2064 }
2065 
2066 static void parse_os_info(char* distro, size_t length, const char* file) {
2067   FILE* fp = fopen(file, "r");
2068   if (fp != NULL) {
2069     char buf[256];
2070     // get last line of the file.
2071     while (fgets(buf, sizeof(buf), fp)) { }
2072     // Edit out extra stuff in expected ubuntu format
2073     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL) {
2074       char* ptr = strstr(buf, "\"");  // the name is in quotes
2075       if (ptr != NULL) {
2076         ptr++; // go beyond first quote
2077         char* nl = strchr(ptr, '\"');
2078         if (nl != NULL) *nl = '\0';
2079         strncpy(distro, ptr, length);
2080       } else {
2081         ptr = strstr(buf, "=");
2082         ptr++; // go beyond equals then
2083         char* nl = strchr(ptr, '\n');
2084         if (nl != NULL) *nl = '\0';
2085         strncpy(distro, ptr, length);
2086       }
2087     } else {
2088       // if not in expected Ubuntu format, print out whole line minus \n
2089       char* nl = strchr(buf, '\n');
2090       if (nl != NULL) *nl = '\0';
2091       strncpy(distro, buf, length);
2092     }
2093     // close distro file
2094     fclose(fp);
2095   }
2096 }
2097 
2098 void os::get_summary_os_info(char* buf, size_t buflen) {
2099   for (int i = 0;; i++) {
2100     const char* file = distro_files[i];
2101     if (file == NULL) {
2102       break; // ran out of distro_files
2103     }
2104     if (file_exists(file)) {
2105       parse_os_info(buf, buflen, file);
2106       return;
2107     }
2108   }
2109   // special case for debian
2110   if (file_exists("/etc/debian_version")) {
2111     strncpy(buf, "Debian ", buflen);
2112     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2113   } else {
2114     strncpy(buf, "Linux", buflen);
2115   }
2116 }
2117 
2118 void os::Linux::print_libversion_info(outputStream* st) {
2119   // libc, pthread
2120   st->print("libc:");
2121   st->print("%s ", os::Linux::glibc_version());
2122   st->print("%s ", os::Linux::libpthread_version());
2123   st->cr();
2124 }
2125 
2126 void os::Linux::print_full_memory_info(outputStream* st) {
2127   st->print("\n/proc/meminfo:\n");
2128   _print_ascii_file("/proc/meminfo", st);
2129   st->cr();
2130 }
2131 
2132 void os::print_memory_info(outputStream* st) {
2133 
2134   st->print("Memory:");
2135   st->print(" %dk page", os::vm_page_size()>>10);
2136 
2137   // values in struct sysinfo are "unsigned long"
2138   struct sysinfo si;
2139   sysinfo(&si);
2140 
2141   st->print(", physical " UINT64_FORMAT "k",
2142             os::physical_memory() >> 10);
2143   st->print("(" UINT64_FORMAT "k free)",
2144             os::available_memory() >> 10);
2145   st->print(", swap " UINT64_FORMAT "k",
2146             ((jlong)si.totalswap * si.mem_unit) >> 10);
2147   st->print("(" UINT64_FORMAT "k free)",
2148             ((jlong)si.freeswap * si.mem_unit) >> 10);
2149   st->cr();
2150 }
2151 
2152 // Print the first "model name" line and the first "flags" line
2153 // that we find and nothing more. We assume "model name" comes
2154 // before "flags" so if we find a second "model name", then the
2155 // "flags" field is considered missing.
2156 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2157 #if defined(IA32) || defined(AMD64)
2158   // Other platforms have less repetitive cpuinfo files
2159   FILE *fp = fopen("/proc/cpuinfo", "r");
2160   if (fp) {
2161     while (!feof(fp)) {
2162       if (fgets(buf, buflen, fp)) {
2163         // Assume model name comes before flags
2164         bool model_name_printed = false;
2165         if (strstr(buf, "model name") != NULL) {
2166           if (!model_name_printed) {
2167             st->print_raw("\nCPU Model and flags from /proc/cpuinfo:\n");
2168             st->print_raw(buf);
2169             model_name_printed = true;
2170           } else {
2171             // model name printed but not flags?  Odd, just return
2172             fclose(fp);
2173             return true;
2174           }
2175         }
2176         // print the flags line too
2177         if (strstr(buf, "flags") != NULL) {
2178           st->print_raw(buf);
2179           fclose(fp);
2180           return true;
2181         }
2182       }
2183     }
2184     fclose(fp);
2185   }
2186 #endif // x86 platforms
2187   return false;
2188 }
2189 
2190 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2191   // Only print the model name if the platform provides this as a summary
2192   if (!print_model_name_and_flags(st, buf, buflen)) {
2193     st->print("\n/proc/cpuinfo:\n");
2194     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2195       st->print_cr("  <Not Available>");
2196     }
2197   }
2198 }
2199 
2200 #if defined(AMD64) || defined(IA32) || defined(X32)
2201 const char* search_string = "model name";
2202 #elif defined(SPARC)
2203 const char* search_string = "cpu";
2204 #elif defined(PPC64)
2205 const char* search_string = "cpu";
2206 #else
2207 const char* search_string = "Processor";
2208 #endif
2209 
2210 // Parses the cpuinfo file for string representing the model name.
2211 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2212   FILE* fp = fopen("/proc/cpuinfo", "r");
2213   if (fp != NULL) {
2214     while (!feof(fp)) {
2215       char buf[256];
2216       if (fgets(buf, sizeof(buf), fp)) {
2217         char* start = strstr(buf, search_string);
2218         if (start != NULL) {
2219           char *ptr = start + strlen(search_string);
2220           char *end = buf + strlen(buf);
2221           while (ptr != end) {
2222              // skip whitespace and colon for the rest of the name.
2223              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2224                break;
2225              }
2226              ptr++;
2227           }
2228           if (ptr != end) {
2229             // reasonable string, get rid of newline and keep the rest
2230             char* nl = strchr(buf, '\n');
2231             if (nl != NULL) *nl = '\0';
2232             strncpy(cpuinfo, ptr, length);
2233             fclose(fp);
2234             return;
2235           }
2236         }
2237       }
2238     }
2239     fclose(fp);
2240   }
2241   // cpuinfo not found or parsing failed, just print generic string.  The entire
2242   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2243 #if defined(AMD64)
2244   strncpy(cpuinfo, "x86_64", length);
2245 #elif defined(IA32)
2246   strncpy(cpuinfo, "x86_32", length);
2247 #elif defined(IA64)
2248   strncpy(cpuinfo, "IA64", length);
2249 #elif defined(SPARC)
2250   strncpy(cpuinfo, "sparcv9", length);
2251 #elif defined(AARCH64)
2252   strncpy(cpuinfo, "AArch64", length);
2253 #elif defined(ARM)
2254   strncpy(cpuinfo, "ARM", length);
2255 #elif defined(PPC)
2256   strncpy(cpuinfo, "PPC64", length);
2257 #elif defined(ZERO_LIBARCH)
2258   strncpy(cpuinfo, ZERO_LIBARCH, length);
2259 #else
2260   strncpy(cpuinfo, "unknown", length);
2261 #endif
2262 }
2263 
2264 static void print_signal_handler(outputStream* st, int sig,
2265                                  char* buf, size_t buflen);
2266 
2267 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2268   st->print_cr("Signal Handlers:");
2269   print_signal_handler(st, SIGSEGV, buf, buflen);
2270   print_signal_handler(st, SIGBUS , buf, buflen);
2271   print_signal_handler(st, SIGFPE , buf, buflen);
2272   print_signal_handler(st, SIGPIPE, buf, buflen);
2273   print_signal_handler(st, SIGXFSZ, buf, buflen);
2274   print_signal_handler(st, SIGILL , buf, buflen);
2275   print_signal_handler(st, SR_signum, buf, buflen);
2276   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2277   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2278   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2279   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2280 #if defined(PPC64)
2281   print_signal_handler(st, SIGTRAP, buf, buflen);
2282 #endif
2283 }
2284 
2285 static char saved_jvm_path[MAXPATHLEN] = {0};
2286 
2287 // Find the full path to the current module, libjvm.so
2288 void os::jvm_path(char *buf, jint buflen) {
2289   // Error checking.
2290   if (buflen < MAXPATHLEN) {
2291     assert(false, "must use a large-enough buffer");
2292     buf[0] = '\0';
2293     return;
2294   }
2295   // Lazy resolve the path to current module.
2296   if (saved_jvm_path[0] != 0) {
2297     strcpy(buf, saved_jvm_path);
2298     return;
2299   }
2300 
2301   char dli_fname[MAXPATHLEN];
2302   bool ret = dll_address_to_library_name(
2303                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2304                                          dli_fname, sizeof(dli_fname), NULL);
2305   assert(ret, "cannot locate libjvm");
2306   char *rp = NULL;
2307   if (ret && dli_fname[0] != '\0') {
2308     rp = realpath(dli_fname, buf);
2309   }
2310   if (rp == NULL) {
2311     return;
2312   }
2313 
2314   if (Arguments::sun_java_launcher_is_altjvm()) {
2315     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2316     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2317     // If "/jre/lib/" appears at the right place in the string, then
2318     // assume we are installed in a JDK and we're done. Otherwise, check
2319     // for a JAVA_HOME environment variable and fix up the path so it
2320     // looks like libjvm.so is installed there (append a fake suffix
2321     // hotspot/libjvm.so).
2322     const char *p = buf + strlen(buf) - 1;
2323     for (int count = 0; p > buf && count < 5; ++count) {
2324       for (--p; p > buf && *p != '/'; --p)
2325         /* empty */ ;
2326     }
2327 
2328     if (strncmp(p, "/jre/lib/", 9) != 0) {
2329       // Look for JAVA_HOME in the environment.
2330       char* java_home_var = ::getenv("JAVA_HOME");
2331       if (java_home_var != NULL && java_home_var[0] != 0) {
2332         char* jrelib_p;
2333         int len;
2334 
2335         // Check the current module name "libjvm.so".
2336         p = strrchr(buf, '/');
2337         if (p == NULL) {
2338           return;
2339         }
2340         assert(strstr(p, "/libjvm") == p, "invalid library name");
2341 
2342         rp = realpath(java_home_var, buf);
2343         if (rp == NULL) {
2344           return;
2345         }
2346 
2347         // determine if this is a legacy image or modules image
2348         // modules image doesn't have "jre" subdirectory
2349         len = strlen(buf);
2350         assert(len < buflen, "Ran out of buffer room");
2351         jrelib_p = buf + len;
2352         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2353         if (0 != access(buf, F_OK)) {
2354           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2355         }
2356 
2357         if (0 == access(buf, F_OK)) {
2358           // Use current module name "libjvm.so"
2359           len = strlen(buf);
2360           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2361         } else {
2362           // Go back to path of .so
2363           rp = realpath(dli_fname, buf);
2364           if (rp == NULL) {
2365             return;
2366           }
2367         }
2368       }
2369     }
2370   }
2371 
2372   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2373   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2374 }
2375 
2376 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2377   // no prefix required, not even "_"
2378 }
2379 
2380 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2381   // no suffix required
2382 }
2383 
2384 ////////////////////////////////////////////////////////////////////////////////
2385 // sun.misc.Signal support
2386 
2387 static volatile jint sigint_count = 0;
2388 
2389 static void UserHandler(int sig, void *siginfo, void *context) {
2390   // 4511530 - sem_post is serialized and handled by the manager thread. When
2391   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2392   // don't want to flood the manager thread with sem_post requests.
2393   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2394     return;
2395   }
2396 
2397   // Ctrl-C is pressed during error reporting, likely because the error
2398   // handler fails to abort. Let VM die immediately.
2399   if (sig == SIGINT && is_error_reported()) {
2400     os::die();
2401   }
2402 
2403   os::signal_notify(sig);
2404 }
2405 
2406 void* os::user_handler() {
2407   return CAST_FROM_FN_PTR(void*, UserHandler);
2408 }
2409 
2410 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2411   struct timespec ts;
2412   // Semaphore's are always associated with CLOCK_REALTIME
2413   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2414   // see unpackTime for discussion on overflow checking
2415   if (sec >= MAX_SECS) {
2416     ts.tv_sec += MAX_SECS;
2417     ts.tv_nsec = 0;
2418   } else {
2419     ts.tv_sec += sec;
2420     ts.tv_nsec += nsec;
2421     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2422       ts.tv_nsec -= NANOSECS_PER_SEC;
2423       ++ts.tv_sec; // note: this must be <= max_secs
2424     }
2425   }
2426 
2427   return ts;
2428 }
2429 
2430 extern "C" {
2431   typedef void (*sa_handler_t)(int);
2432   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2433 }
2434 
2435 void* os::signal(int signal_number, void* handler) {
2436   struct sigaction sigAct, oldSigAct;
2437 
2438   sigfillset(&(sigAct.sa_mask));
2439   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2440   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2441 
2442   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2443     // -1 means registration failed
2444     return (void *)-1;
2445   }
2446 
2447   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2448 }
2449 
2450 void os::signal_raise(int signal_number) {
2451   ::raise(signal_number);
2452 }
2453 
2454 // The following code is moved from os.cpp for making this
2455 // code platform specific, which it is by its very nature.
2456 
2457 // Will be modified when max signal is changed to be dynamic
2458 int os::sigexitnum_pd() {
2459   return NSIG;
2460 }
2461 
2462 // a counter for each possible signal value
2463 static volatile jint pending_signals[NSIG+1] = { 0 };
2464 
2465 // Linux(POSIX) specific hand shaking semaphore.
2466 static sem_t sig_sem;
2467 static PosixSemaphore sr_semaphore;
2468 
2469 void os::signal_init_pd() {
2470   // Initialize signal structures
2471   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2472 
2473   // Initialize signal semaphore
2474   ::sem_init(&sig_sem, 0, 0);
2475 }
2476 
2477 void os::signal_notify(int sig) {
2478   Atomic::inc(&pending_signals[sig]);
2479   ::sem_post(&sig_sem);
2480 }
2481 
2482 static int check_pending_signals(bool wait) {
2483   Atomic::store(0, &sigint_count);
2484   for (;;) {
2485     for (int i = 0; i < NSIG + 1; i++) {
2486       jint n = pending_signals[i];
2487       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2488         return i;
2489       }
2490     }
2491     if (!wait) {
2492       return -1;
2493     }
2494     JavaThread *thread = JavaThread::current();
2495     ThreadBlockInVM tbivm(thread);
2496 
2497     bool threadIsSuspended;
2498     do {
2499       thread->set_suspend_equivalent();
2500       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2501       ::sem_wait(&sig_sem);
2502 
2503       // were we externally suspended while we were waiting?
2504       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2505       if (threadIsSuspended) {
2506         // The semaphore has been incremented, but while we were waiting
2507         // another thread suspended us. We don't want to continue running
2508         // while suspended because that would surprise the thread that
2509         // suspended us.
2510         ::sem_post(&sig_sem);
2511 
2512         thread->java_suspend_self();
2513       }
2514     } while (threadIsSuspended);
2515   }
2516 }
2517 
2518 int os::signal_lookup() {
2519   return check_pending_signals(false);
2520 }
2521 
2522 int os::signal_wait() {
2523   return check_pending_signals(true);
2524 }
2525 
2526 ////////////////////////////////////////////////////////////////////////////////
2527 // Virtual Memory
2528 
2529 int os::vm_page_size() {
2530   // Seems redundant as all get out
2531   assert(os::Linux::page_size() != -1, "must call os::init");
2532   return os::Linux::page_size();
2533 }
2534 
2535 // Solaris allocates memory by pages.
2536 int os::vm_allocation_granularity() {
2537   assert(os::Linux::page_size() != -1, "must call os::init");
2538   return os::Linux::page_size();
2539 }
2540 
2541 // Rationale behind this function:
2542 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2543 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2544 //  samples for JITted code. Here we create private executable mapping over the code cache
2545 //  and then we can use standard (well, almost, as mapping can change) way to provide
2546 //  info for the reporting script by storing timestamp and location of symbol
2547 void linux_wrap_code(char* base, size_t size) {
2548   static volatile jint cnt = 0;
2549 
2550   if (!UseOprofile) {
2551     return;
2552   }
2553 
2554   char buf[PATH_MAX+1];
2555   int num = Atomic::add(1, &cnt);
2556 
2557   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2558            os::get_temp_directory(), os::current_process_id(), num);
2559   unlink(buf);
2560 
2561   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2562 
2563   if (fd != -1) {
2564     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2565     if (rv != (off_t)-1) {
2566       if (::write(fd, "", 1) == 1) {
2567         mmap(base, size,
2568              PROT_READ|PROT_WRITE|PROT_EXEC,
2569              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2570       }
2571     }
2572     ::close(fd);
2573     unlink(buf);
2574   }
2575 }
2576 
2577 static bool recoverable_mmap_error(int err) {
2578   // See if the error is one we can let the caller handle. This
2579   // list of errno values comes from JBS-6843484. I can't find a
2580   // Linux man page that documents this specific set of errno
2581   // values so while this list currently matches Solaris, it may
2582   // change as we gain experience with this failure mode.
2583   switch (err) {
2584   case EBADF:
2585   case EINVAL:
2586   case ENOTSUP:
2587     // let the caller deal with these errors
2588     return true;
2589 
2590   default:
2591     // Any remaining errors on this OS can cause our reserved mapping
2592     // to be lost. That can cause confusion where different data
2593     // structures think they have the same memory mapped. The worst
2594     // scenario is if both the VM and a library think they have the
2595     // same memory mapped.
2596     return false;
2597   }
2598 }
2599 
2600 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2601                                     int err) {
2602   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2603           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2604           strerror(err), err);
2605 }
2606 
2607 static void warn_fail_commit_memory(char* addr, size_t size,
2608                                     size_t alignment_hint, bool exec,
2609                                     int err) {
2610   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2611           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2612           alignment_hint, exec, strerror(err), err);
2613 }
2614 
2615 // NOTE: Linux kernel does not really reserve the pages for us.
2616 //       All it does is to check if there are enough free pages
2617 //       left at the time of mmap(). This could be a potential
2618 //       problem.
2619 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2620   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2621   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2622                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2623   if (res != (uintptr_t) MAP_FAILED) {
2624     if (UseNUMAInterleaving) {
2625       numa_make_global(addr, size);
2626     }
2627     return 0;
2628   }
2629 
2630   int err = errno;  // save errno from mmap() call above
2631 
2632   if (!recoverable_mmap_error(err)) {
2633     warn_fail_commit_memory(addr, size, exec, err);
2634     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2635   }
2636 
2637   return err;
2638 }
2639 
2640 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2641   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2642 }
2643 
2644 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2645                                   const char* mesg) {
2646   assert(mesg != NULL, "mesg must be specified");
2647   int err = os::Linux::commit_memory_impl(addr, size, exec);
2648   if (err != 0) {
2649     // the caller wants all commit errors to exit with the specified mesg:
2650     warn_fail_commit_memory(addr, size, exec, err);
2651     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2652   }
2653 }
2654 
2655 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2656 #ifndef MAP_HUGETLB
2657   #define MAP_HUGETLB 0x40000
2658 #endif
2659 
2660 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2661 #ifndef MADV_HUGEPAGE
2662   #define MADV_HUGEPAGE 14
2663 #endif
2664 
2665 int os::Linux::commit_memory_impl(char* addr, size_t size,
2666                                   size_t alignment_hint, bool exec) {
2667   int err = os::Linux::commit_memory_impl(addr, size, exec);
2668   if (err == 0) {
2669     realign_memory(addr, size, alignment_hint);
2670   }
2671   return err;
2672 }
2673 
2674 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2675                           bool exec) {
2676   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2677 }
2678 
2679 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2680                                   size_t alignment_hint, bool exec,
2681                                   const char* mesg) {
2682   assert(mesg != NULL, "mesg must be specified");
2683   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2684   if (err != 0) {
2685     // the caller wants all commit errors to exit with the specified mesg:
2686     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2687     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2688   }
2689 }
2690 
2691 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2692   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2693     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2694     // be supported or the memory may already be backed by huge pages.
2695     ::madvise(addr, bytes, MADV_HUGEPAGE);
2696   }
2697 }
2698 
2699 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2700   // This method works by doing an mmap over an existing mmaping and effectively discarding
2701   // the existing pages. However it won't work for SHM-based large pages that cannot be
2702   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2703   // small pages on top of the SHM segment. This method always works for small pages, so we
2704   // allow that in any case.
2705   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2706     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2707   }
2708 }
2709 
2710 void os::numa_make_global(char *addr, size_t bytes) {
2711   Linux::numa_interleave_memory(addr, bytes);
2712 }
2713 
2714 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2715 // bind policy to MPOL_PREFERRED for the current thread.
2716 #define USE_MPOL_PREFERRED 0
2717 
2718 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2719   // To make NUMA and large pages more robust when both enabled, we need to ease
2720   // the requirements on where the memory should be allocated. MPOL_BIND is the
2721   // default policy and it will force memory to be allocated on the specified
2722   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2723   // the specified node, but will not force it. Using this policy will prevent
2724   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2725   // free large pages.
2726   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2727   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2728 }
2729 
2730 bool os::numa_topology_changed() { return false; }
2731 
2732 size_t os::numa_get_groups_num() {
2733   int max_node = Linux::numa_max_node();
2734   return max_node > 0 ? max_node + 1 : 1;
2735 }
2736 
2737 int os::numa_get_group_id() {
2738   int cpu_id = Linux::sched_getcpu();
2739   if (cpu_id != -1) {
2740     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2741     if (lgrp_id != -1) {
2742       return lgrp_id;
2743     }
2744   }
2745   return 0;
2746 }
2747 
2748 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2749   for (size_t i = 0; i < size; i++) {
2750     ids[i] = i;
2751   }
2752   return size;
2753 }
2754 
2755 bool os::get_page_info(char *start, page_info* info) {
2756   return false;
2757 }
2758 
2759 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2760                      page_info* page_found) {
2761   return end;
2762 }
2763 
2764 
2765 int os::Linux::sched_getcpu_syscall(void) {
2766   unsigned int cpu = 0;
2767   int retval = -1;
2768 
2769 #if defined(IA32)
2770   #ifndef SYS_getcpu
2771     #define SYS_getcpu 318
2772   #endif
2773   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2774 #elif defined(AMD64)
2775 // Unfortunately we have to bring all these macros here from vsyscall.h
2776 // to be able to compile on old linuxes.
2777   #define __NR_vgetcpu 2
2778   #define VSYSCALL_START (-10UL << 20)
2779   #define VSYSCALL_SIZE 1024
2780   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2781   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2782   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2783   retval = vgetcpu(&cpu, NULL, NULL);
2784 #endif
2785 
2786   return (retval == -1) ? retval : cpu;
2787 }
2788 
2789 // Something to do with the numa-aware allocator needs these symbols
2790 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2791 extern "C" JNIEXPORT void numa_error(char *where) { }
2792 
2793 
2794 // If we are running with libnuma version > 2, then we should
2795 // be trying to use symbols with versions 1.1
2796 // If we are running with earlier version, which did not have symbol versions,
2797 // we should use the base version.
2798 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2799   void *f = dlvsym(handle, name, "libnuma_1.1");
2800   if (f == NULL) {
2801     f = dlsym(handle, name);
2802   }
2803   return f;
2804 }
2805 
2806 bool os::Linux::libnuma_init() {
2807   // sched_getcpu() should be in libc.
2808   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2809                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2810 
2811   // If it's not, try a direct syscall.
2812   if (sched_getcpu() == -1) {
2813     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2814                                     (void*)&sched_getcpu_syscall));
2815   }
2816 
2817   if (sched_getcpu() != -1) { // Does it work?
2818     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2819     if (handle != NULL) {
2820       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2821                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2822       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2823                                        libnuma_dlsym(handle, "numa_max_node")));
2824       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2825                                         libnuma_dlsym(handle, "numa_available")));
2826       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2827                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2828       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2829                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2830       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2831                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2832 
2833 
2834       if (numa_available() != -1) {
2835         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2836         // Create a cpu -> node mapping
2837         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2838         rebuild_cpu_to_node_map();
2839         return true;
2840       }
2841     }
2842   }
2843   return false;
2844 }
2845 
2846 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2847 // The table is later used in get_node_by_cpu().
2848 void os::Linux::rebuild_cpu_to_node_map() {
2849   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2850                               // in libnuma (possible values are starting from 16,
2851                               // and continuing up with every other power of 2, but less
2852                               // than the maximum number of CPUs supported by kernel), and
2853                               // is a subject to change (in libnuma version 2 the requirements
2854                               // are more reasonable) we'll just hardcode the number they use
2855                               // in the library.
2856   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2857 
2858   size_t cpu_num = os::active_processor_count();
2859   size_t cpu_map_size = NCPUS / BitsPerCLong;
2860   size_t cpu_map_valid_size =
2861     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2862 
2863   cpu_to_node()->clear();
2864   cpu_to_node()->at_grow(cpu_num - 1);
2865   size_t node_num = numa_get_groups_num();
2866 
2867   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2868   for (size_t i = 0; i < node_num; i++) {
2869     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2870       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2871         if (cpu_map[j] != 0) {
2872           for (size_t k = 0; k < BitsPerCLong; k++) {
2873             if (cpu_map[j] & (1UL << k)) {
2874               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2875             }
2876           }
2877         }
2878       }
2879     }
2880   }
2881   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2882 }
2883 
2884 int os::Linux::get_node_by_cpu(int cpu_id) {
2885   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2886     return cpu_to_node()->at(cpu_id);
2887   }
2888   return -1;
2889 }
2890 
2891 GrowableArray<int>* os::Linux::_cpu_to_node;
2892 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2893 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2894 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2895 os::Linux::numa_available_func_t os::Linux::_numa_available;
2896 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2897 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2898 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2899 unsigned long* os::Linux::_numa_all_nodes;
2900 
2901 bool os::pd_uncommit_memory(char* addr, size_t size) {
2902   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2903                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2904   return res  != (uintptr_t) MAP_FAILED;
2905 }
2906 
2907 static address get_stack_commited_bottom(address bottom, size_t size) {
2908   address nbot = bottom;
2909   address ntop = bottom + size;
2910 
2911   size_t page_sz = os::vm_page_size();
2912   unsigned pages = size / page_sz;
2913 
2914   unsigned char vec[1];
2915   unsigned imin = 1, imax = pages + 1, imid;
2916   int mincore_return_value = 0;
2917 
2918   assert(imin <= imax, "Unexpected page size");
2919 
2920   while (imin < imax) {
2921     imid = (imax + imin) / 2;
2922     nbot = ntop - (imid * page_sz);
2923 
2924     // Use a trick with mincore to check whether the page is mapped or not.
2925     // mincore sets vec to 1 if page resides in memory and to 0 if page
2926     // is swapped output but if page we are asking for is unmapped
2927     // it returns -1,ENOMEM
2928     mincore_return_value = mincore(nbot, page_sz, vec);
2929 
2930     if (mincore_return_value == -1) {
2931       // Page is not mapped go up
2932       // to find first mapped page
2933       if (errno != EAGAIN) {
2934         assert(errno == ENOMEM, "Unexpected mincore errno");
2935         imax = imid;
2936       }
2937     } else {
2938       // Page is mapped go down
2939       // to find first not mapped page
2940       imin = imid + 1;
2941     }
2942   }
2943 
2944   nbot = nbot + page_sz;
2945 
2946   // Adjust stack bottom one page up if last checked page is not mapped
2947   if (mincore_return_value == -1) {
2948     nbot = nbot + page_sz;
2949   }
2950 
2951   return nbot;
2952 }
2953 
2954 
2955 // Linux uses a growable mapping for the stack, and if the mapping for
2956 // the stack guard pages is not removed when we detach a thread the
2957 // stack cannot grow beyond the pages where the stack guard was
2958 // mapped.  If at some point later in the process the stack expands to
2959 // that point, the Linux kernel cannot expand the stack any further
2960 // because the guard pages are in the way, and a segfault occurs.
2961 //
2962 // However, it's essential not to split the stack region by unmapping
2963 // a region (leaving a hole) that's already part of the stack mapping,
2964 // so if the stack mapping has already grown beyond the guard pages at
2965 // the time we create them, we have to truncate the stack mapping.
2966 // So, we need to know the extent of the stack mapping when
2967 // create_stack_guard_pages() is called.
2968 
2969 // We only need this for stacks that are growable: at the time of
2970 // writing thread stacks don't use growable mappings (i.e. those
2971 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2972 // only applies to the main thread.
2973 
2974 // If the (growable) stack mapping already extends beyond the point
2975 // where we're going to put our guard pages, truncate the mapping at
2976 // that point by munmap()ping it.  This ensures that when we later
2977 // munmap() the guard pages we don't leave a hole in the stack
2978 // mapping. This only affects the main/initial thread
2979 
2980 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2981   if (os::Linux::is_initial_thread()) {
2982     // As we manually grow stack up to bottom inside create_attached_thread(),
2983     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2984     // we don't need to do anything special.
2985     // Check it first, before calling heavy function.
2986     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
2987     unsigned char vec[1];
2988 
2989     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
2990       // Fallback to slow path on all errors, including EAGAIN
2991       stack_extent = (uintptr_t) get_stack_commited_bottom(
2992                                                            os::Linux::initial_thread_stack_bottom(),
2993                                                            (size_t)addr - stack_extent);
2994     }
2995 
2996     if (stack_extent < (uintptr_t)addr) {
2997       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
2998     }
2999   }
3000 
3001   return os::commit_memory(addr, size, !ExecMem);
3002 }
3003 
3004 // If this is a growable mapping, remove the guard pages entirely by
3005 // munmap()ping them.  If not, just call uncommit_memory(). This only
3006 // affects the main/initial thread, but guard against future OS changes
3007 // It's safe to always unmap guard pages for initial thread because we
3008 // always place it right after end of the mapped region
3009 
3010 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3011   uintptr_t stack_extent, stack_base;
3012 
3013   if (os::Linux::is_initial_thread()) {
3014     return ::munmap(addr, size) == 0;
3015   }
3016 
3017   return os::uncommit_memory(addr, size);
3018 }
3019 
3020 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3021 // at 'requested_addr'. If there are existing memory mappings at the same
3022 // location, however, they will be overwritten. If 'fixed' is false,
3023 // 'requested_addr' is only treated as a hint, the return value may or
3024 // may not start from the requested address. Unlike Linux mmap(), this
3025 // function returns NULL to indicate failure.
3026 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3027   char * addr;
3028   int flags;
3029 
3030   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3031   if (fixed) {
3032     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3033     flags |= MAP_FIXED;
3034   }
3035 
3036   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3037   // touch an uncommitted page. Otherwise, the read/write might
3038   // succeed if we have enough swap space to back the physical page.
3039   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3040                        flags, -1, 0);
3041 
3042   return addr == MAP_FAILED ? NULL : addr;
3043 }
3044 
3045 static int anon_munmap(char * addr, size_t size) {
3046   return ::munmap(addr, size) == 0;
3047 }
3048 
3049 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3050                             size_t alignment_hint) {
3051   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3052 }
3053 
3054 bool os::pd_release_memory(char* addr, size_t size) {
3055   return anon_munmap(addr, size);
3056 }
3057 
3058 static bool linux_mprotect(char* addr, size_t size, int prot) {
3059   // Linux wants the mprotect address argument to be page aligned.
3060   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3061 
3062   // According to SUSv3, mprotect() should only be used with mappings
3063   // established by mmap(), and mmap() always maps whole pages. Unaligned
3064   // 'addr' likely indicates problem in the VM (e.g. trying to change
3065   // protection of malloc'ed or statically allocated memory). Check the
3066   // caller if you hit this assert.
3067   assert(addr == bottom, "sanity check");
3068 
3069   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3070   return ::mprotect(bottom, size, prot) == 0;
3071 }
3072 
3073 // Set protections specified
3074 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3075                         bool is_committed) {
3076   unsigned int p = 0;
3077   switch (prot) {
3078   case MEM_PROT_NONE: p = PROT_NONE; break;
3079   case MEM_PROT_READ: p = PROT_READ; break;
3080   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3081   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3082   default:
3083     ShouldNotReachHere();
3084   }
3085   // is_committed is unused.
3086   return linux_mprotect(addr, bytes, p);
3087 }
3088 
3089 bool os::guard_memory(char* addr, size_t size) {
3090   return linux_mprotect(addr, size, PROT_NONE);
3091 }
3092 
3093 bool os::unguard_memory(char* addr, size_t size) {
3094   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3095 }
3096 
3097 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3098                                                     size_t page_size) {
3099   bool result = false;
3100   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3101                  MAP_ANONYMOUS|MAP_PRIVATE,
3102                  -1, 0);
3103   if (p != MAP_FAILED) {
3104     void *aligned_p = align_ptr_up(p, page_size);
3105 
3106     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3107 
3108     munmap(p, page_size * 2);
3109   }
3110 
3111   if (warn && !result) {
3112     warning("TransparentHugePages is not supported by the operating system.");
3113   }
3114 
3115   return result;
3116 }
3117 
3118 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3119   bool result = false;
3120   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3121                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3122                  -1, 0);
3123 
3124   if (p != MAP_FAILED) {
3125     // We don't know if this really is a huge page or not.
3126     FILE *fp = fopen("/proc/self/maps", "r");
3127     if (fp) {
3128       while (!feof(fp)) {
3129         char chars[257];
3130         long x = 0;
3131         if (fgets(chars, sizeof(chars), fp)) {
3132           if (sscanf(chars, "%lx-%*x", &x) == 1
3133               && x == (long)p) {
3134             if (strstr (chars, "hugepage")) {
3135               result = true;
3136               break;
3137             }
3138           }
3139         }
3140       }
3141       fclose(fp);
3142     }
3143     munmap(p, page_size);
3144   }
3145 
3146   if (warn && !result) {
3147     warning("HugeTLBFS is not supported by the operating system.");
3148   }
3149 
3150   return result;
3151 }
3152 
3153 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3154 //
3155 // From the coredump_filter documentation:
3156 //
3157 // - (bit 0) anonymous private memory
3158 // - (bit 1) anonymous shared memory
3159 // - (bit 2) file-backed private memory
3160 // - (bit 3) file-backed shared memory
3161 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3162 //           effective only if the bit 2 is cleared)
3163 // - (bit 5) hugetlb private memory
3164 // - (bit 6) hugetlb shared memory
3165 //
3166 static void set_coredump_filter(void) {
3167   FILE *f;
3168   long cdm;
3169 
3170   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3171     return;
3172   }
3173 
3174   if (fscanf(f, "%lx", &cdm) != 1) {
3175     fclose(f);
3176     return;
3177   }
3178 
3179   rewind(f);
3180 
3181   if ((cdm & LARGEPAGES_BIT) == 0) {
3182     cdm |= LARGEPAGES_BIT;
3183     fprintf(f, "%#lx", cdm);
3184   }
3185 
3186   fclose(f);
3187 }
3188 
3189 // Large page support
3190 
3191 static size_t _large_page_size = 0;
3192 
3193 size_t os::Linux::find_large_page_size() {
3194   size_t large_page_size = 0;
3195 
3196   // large_page_size on Linux is used to round up heap size. x86 uses either
3197   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3198   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3199   // page as large as 256M.
3200   //
3201   // Here we try to figure out page size by parsing /proc/meminfo and looking
3202   // for a line with the following format:
3203   //    Hugepagesize:     2048 kB
3204   //
3205   // If we can't determine the value (e.g. /proc is not mounted, or the text
3206   // format has been changed), we'll use the largest page size supported by
3207   // the processor.
3208 
3209 #ifndef ZERO
3210   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3211                      ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3212 #endif // ZERO
3213 
3214   FILE *fp = fopen("/proc/meminfo", "r");
3215   if (fp) {
3216     while (!feof(fp)) {
3217       int x = 0;
3218       char buf[16];
3219       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3220         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3221           large_page_size = x * K;
3222           break;
3223         }
3224       } else {
3225         // skip to next line
3226         for (;;) {
3227           int ch = fgetc(fp);
3228           if (ch == EOF || ch == (int)'\n') break;
3229         }
3230       }
3231     }
3232     fclose(fp);
3233   }
3234 
3235   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3236     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3237             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3238             proper_unit_for_byte_size(large_page_size));
3239   }
3240 
3241   return large_page_size;
3242 }
3243 
3244 size_t os::Linux::setup_large_page_size() {
3245   _large_page_size = Linux::find_large_page_size();
3246   const size_t default_page_size = (size_t)Linux::page_size();
3247   if (_large_page_size > default_page_size) {
3248     _page_sizes[0] = _large_page_size;
3249     _page_sizes[1] = default_page_size;
3250     _page_sizes[2] = 0;
3251   }
3252 
3253   return _large_page_size;
3254 }
3255 
3256 bool os::Linux::setup_large_page_type(size_t page_size) {
3257   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3258       FLAG_IS_DEFAULT(UseSHM) &&
3259       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3260 
3261     // The type of large pages has not been specified by the user.
3262 
3263     // Try UseHugeTLBFS and then UseSHM.
3264     UseHugeTLBFS = UseSHM = true;
3265 
3266     // Don't try UseTransparentHugePages since there are known
3267     // performance issues with it turned on. This might change in the future.
3268     UseTransparentHugePages = false;
3269   }
3270 
3271   if (UseTransparentHugePages) {
3272     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3273     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3274       UseHugeTLBFS = false;
3275       UseSHM = false;
3276       return true;
3277     }
3278     UseTransparentHugePages = false;
3279   }
3280 
3281   if (UseHugeTLBFS) {
3282     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3283     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3284       UseSHM = false;
3285       return true;
3286     }
3287     UseHugeTLBFS = false;
3288   }
3289 
3290   return UseSHM;
3291 }
3292 
3293 void os::large_page_init() {
3294   if (!UseLargePages &&
3295       !UseTransparentHugePages &&
3296       !UseHugeTLBFS &&
3297       !UseSHM) {
3298     // Not using large pages.
3299     return;
3300   }
3301 
3302   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3303     // The user explicitly turned off large pages.
3304     // Ignore the rest of the large pages flags.
3305     UseTransparentHugePages = false;
3306     UseHugeTLBFS = false;
3307     UseSHM = false;
3308     return;
3309   }
3310 
3311   size_t large_page_size = Linux::setup_large_page_size();
3312   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3313 
3314   set_coredump_filter();
3315 }
3316 
3317 #ifndef SHM_HUGETLB
3318   #define SHM_HUGETLB 04000
3319 #endif
3320 
3321 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3322                                             char* req_addr, bool exec) {
3323   // "exec" is passed in but not used.  Creating the shared image for
3324   // the code cache doesn't have an SHM_X executable permission to check.
3325   assert(UseLargePages && UseSHM, "only for SHM large pages");
3326   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3327 
3328   if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3329     return NULL; // Fallback to small pages.
3330   }
3331 
3332   key_t key = IPC_PRIVATE;
3333   char *addr;
3334 
3335   bool warn_on_failure = UseLargePages &&
3336                         (!FLAG_IS_DEFAULT(UseLargePages) ||
3337                          !FLAG_IS_DEFAULT(UseSHM) ||
3338                          !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3339   char msg[128];
3340 
3341   // Create a large shared memory region to attach to based on size.
3342   // Currently, size is the total size of the heap
3343   int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3344   if (shmid == -1) {
3345     // Possible reasons for shmget failure:
3346     // 1. shmmax is too small for Java heap.
3347     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3348     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3349     // 2. not enough large page memory.
3350     //    > check available large pages: cat /proc/meminfo
3351     //    > increase amount of large pages:
3352     //          echo new_value > /proc/sys/vm/nr_hugepages
3353     //      Note 1: different Linux may use different name for this property,
3354     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3355     //      Note 2: it's possible there's enough physical memory available but
3356     //            they are so fragmented after a long run that they can't
3357     //            coalesce into large pages. Try to reserve large pages when
3358     //            the system is still "fresh".
3359     if (warn_on_failure) {
3360       jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3361       warning("%s", msg);
3362     }
3363     return NULL;
3364   }
3365 
3366   // attach to the region
3367   addr = (char*)shmat(shmid, req_addr, 0);
3368   int err = errno;
3369 
3370   // Remove shmid. If shmat() is successful, the actual shared memory segment
3371   // will be deleted when it's detached by shmdt() or when the process
3372   // terminates. If shmat() is not successful this will remove the shared
3373   // segment immediately.
3374   shmctl(shmid, IPC_RMID, NULL);
3375 
3376   if ((intptr_t)addr == -1) {
3377     if (warn_on_failure) {
3378       jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3379       warning("%s", msg);
3380     }
3381     return NULL;
3382   }
3383 
3384   return addr;
3385 }
3386 
3387 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3388                                         int error) {
3389   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3390 
3391   bool warn_on_failure = UseLargePages &&
3392       (!FLAG_IS_DEFAULT(UseLargePages) ||
3393        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3394        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3395 
3396   if (warn_on_failure) {
3397     char msg[128];
3398     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3399                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3400     warning("%s", msg);
3401   }
3402 }
3403 
3404 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3405                                                         char* req_addr,
3406                                                         bool exec) {
3407   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3408   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3409   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3410 
3411   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3412   char* addr = (char*)::mmap(req_addr, bytes, prot,
3413                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3414                              -1, 0);
3415 
3416   if (addr == MAP_FAILED) {
3417     warn_on_large_pages_failure(req_addr, bytes, errno);
3418     return NULL;
3419   }
3420 
3421   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3422 
3423   return addr;
3424 }
3425 
3426 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3427 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3428 //   (req_addr != NULL) or with a given alignment.
3429 //  - bytes shall be a multiple of alignment.
3430 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3431 //  - alignment sets the alignment at which memory shall be allocated.
3432 //     It must be a multiple of allocation granularity.
3433 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3434 //  req_addr or NULL.
3435 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3436 
3437   size_t extra_size = bytes;
3438   if (req_addr == NULL && alignment > 0) {
3439     extra_size += alignment;
3440   }
3441 
3442   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3443     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3444     -1, 0);
3445   if (start == MAP_FAILED) {
3446     start = NULL;
3447   } else {
3448     if (req_addr != NULL) {
3449       if (start != req_addr) {
3450         ::munmap(start, extra_size);
3451         start = NULL;
3452       }
3453     } else {
3454       char* const start_aligned = (char*) align_ptr_up(start, alignment);
3455       char* const end_aligned = start_aligned + bytes;
3456       char* const end = start + extra_size;
3457       if (start_aligned > start) {
3458         ::munmap(start, start_aligned - start);
3459       }
3460       if (end_aligned < end) {
3461         ::munmap(end_aligned, end - end_aligned);
3462       }
3463       start = start_aligned;
3464     }
3465   }
3466   return start;
3467 
3468 }
3469 
3470 // Reserve memory using mmap(MAP_HUGETLB).
3471 //  - bytes shall be a multiple of alignment.
3472 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3473 //  - alignment sets the alignment at which memory shall be allocated.
3474 //     It must be a multiple of allocation granularity.
3475 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3476 //  req_addr or NULL.
3477 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3478                                                          size_t alignment,
3479                                                          char* req_addr,
3480                                                          bool exec) {
3481   size_t large_page_size = os::large_page_size();
3482   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3483 
3484   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3485   assert(is_size_aligned(bytes, alignment), "Must be");
3486 
3487   // First reserve - but not commit - the address range in small pages.
3488   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3489 
3490   if (start == NULL) {
3491     return NULL;
3492   }
3493 
3494   assert(is_ptr_aligned(start, alignment), "Must be");
3495 
3496   char* end = start + bytes;
3497 
3498   // Find the regions of the allocated chunk that can be promoted to large pages.
3499   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3500   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3501 
3502   size_t lp_bytes = lp_end - lp_start;
3503 
3504   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3505 
3506   if (lp_bytes == 0) {
3507     // The mapped region doesn't even span the start and the end of a large page.
3508     // Fall back to allocate a non-special area.
3509     ::munmap(start, end - start);
3510     return NULL;
3511   }
3512 
3513   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3514 
3515   void* result;
3516 
3517   // Commit small-paged leading area.
3518   if (start != lp_start) {
3519     result = ::mmap(start, lp_start - start, prot,
3520                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3521                     -1, 0);
3522     if (result == MAP_FAILED) {
3523       ::munmap(lp_start, end - lp_start);
3524       return NULL;
3525     }
3526   }
3527 
3528   // Commit large-paged area.
3529   result = ::mmap(lp_start, lp_bytes, prot,
3530                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3531                   -1, 0);
3532   if (result == MAP_FAILED) {
3533     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3534     // If the mmap above fails, the large pages region will be unmapped and we
3535     // have regions before and after with small pages. Release these regions.
3536     //
3537     // |  mapped  |  unmapped  |  mapped  |
3538     // ^          ^            ^          ^
3539     // start      lp_start     lp_end     end
3540     //
3541     ::munmap(start, lp_start - start);
3542     ::munmap(lp_end, end - lp_end);
3543     return NULL;
3544   }
3545 
3546   // Commit small-paged trailing area.
3547   if (lp_end != end) {
3548     result = ::mmap(lp_end, end - lp_end, prot,
3549                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3550                     -1, 0);
3551     if (result == MAP_FAILED) {
3552       ::munmap(start, lp_end - start);
3553       return NULL;
3554     }
3555   }
3556 
3557   return start;
3558 }
3559 
3560 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3561                                                    size_t alignment,
3562                                                    char* req_addr,
3563                                                    bool exec) {
3564   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3565   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3566   assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3567   assert(is_power_of_2(os::large_page_size()), "Must be");
3568   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3569 
3570   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3571     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3572   } else {
3573     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3574   }
3575 }
3576 
3577 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3578                                  char* req_addr, bool exec) {
3579   assert(UseLargePages, "only for large pages");
3580 
3581   char* addr;
3582   if (UseSHM) {
3583     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3584   } else {
3585     assert(UseHugeTLBFS, "must be");
3586     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3587   }
3588 
3589   if (addr != NULL) {
3590     if (UseNUMAInterleaving) {
3591       numa_make_global(addr, bytes);
3592     }
3593 
3594     // The memory is committed
3595     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3596   }
3597 
3598   return addr;
3599 }
3600 
3601 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3602   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3603   return shmdt(base) == 0;
3604 }
3605 
3606 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3607   return pd_release_memory(base, bytes);
3608 }
3609 
3610 bool os::release_memory_special(char* base, size_t bytes) {
3611   bool res;
3612   if (MemTracker::tracking_level() > NMT_minimal) {
3613     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3614     res = os::Linux::release_memory_special_impl(base, bytes);
3615     if (res) {
3616       tkr.record((address)base, bytes);
3617     }
3618 
3619   } else {
3620     res = os::Linux::release_memory_special_impl(base, bytes);
3621   }
3622   return res;
3623 }
3624 
3625 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3626   assert(UseLargePages, "only for large pages");
3627   bool res;
3628 
3629   if (UseSHM) {
3630     res = os::Linux::release_memory_special_shm(base, bytes);
3631   } else {
3632     assert(UseHugeTLBFS, "must be");
3633     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3634   }
3635   return res;
3636 }
3637 
3638 size_t os::large_page_size() {
3639   return _large_page_size;
3640 }
3641 
3642 // With SysV SHM the entire memory region must be allocated as shared
3643 // memory.
3644 // HugeTLBFS allows application to commit large page memory on demand.
3645 // However, when committing memory with HugeTLBFS fails, the region
3646 // that was supposed to be committed will lose the old reservation
3647 // and allow other threads to steal that memory region. Because of this
3648 // behavior we can't commit HugeTLBFS memory.
3649 bool os::can_commit_large_page_memory() {
3650   return UseTransparentHugePages;
3651 }
3652 
3653 bool os::can_execute_large_page_memory() {
3654   return UseTransparentHugePages || UseHugeTLBFS;
3655 }
3656 
3657 // Reserve memory at an arbitrary address, only if that area is
3658 // available (and not reserved for something else).
3659 
3660 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3661   const int max_tries = 10;
3662   char* base[max_tries];
3663   size_t size[max_tries];
3664   const size_t gap = 0x000000;
3665 
3666   // Assert only that the size is a multiple of the page size, since
3667   // that's all that mmap requires, and since that's all we really know
3668   // about at this low abstraction level.  If we need higher alignment,
3669   // we can either pass an alignment to this method or verify alignment
3670   // in one of the methods further up the call chain.  See bug 5044738.
3671   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3672 
3673   // Repeatedly allocate blocks until the block is allocated at the
3674   // right spot.
3675 
3676   // Linux mmap allows caller to pass an address as hint; give it a try first,
3677   // if kernel honors the hint then we can return immediately.
3678   char * addr = anon_mmap(requested_addr, bytes, false);
3679   if (addr == requested_addr) {
3680     return requested_addr;
3681   }
3682 
3683   if (addr != NULL) {
3684     // mmap() is successful but it fails to reserve at the requested address
3685     anon_munmap(addr, bytes);
3686   }
3687 
3688   int i;
3689   for (i = 0; i < max_tries; ++i) {
3690     base[i] = reserve_memory(bytes);
3691 
3692     if (base[i] != NULL) {
3693       // Is this the block we wanted?
3694       if (base[i] == requested_addr) {
3695         size[i] = bytes;
3696         break;
3697       }
3698 
3699       // Does this overlap the block we wanted? Give back the overlapped
3700       // parts and try again.
3701 
3702       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3703       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3704         unmap_memory(base[i], top_overlap);
3705         base[i] += top_overlap;
3706         size[i] = bytes - top_overlap;
3707       } else {
3708         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3709         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3710           unmap_memory(requested_addr, bottom_overlap);
3711           size[i] = bytes - bottom_overlap;
3712         } else {
3713           size[i] = bytes;
3714         }
3715       }
3716     }
3717   }
3718 
3719   // Give back the unused reserved pieces.
3720 
3721   for (int j = 0; j < i; ++j) {
3722     if (base[j] != NULL) {
3723       unmap_memory(base[j], size[j]);
3724     }
3725   }
3726 
3727   if (i < max_tries) {
3728     return requested_addr;
3729   } else {
3730     return NULL;
3731   }
3732 }
3733 
3734 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3735   return ::read(fd, buf, nBytes);
3736 }
3737 
3738 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3739   return ::pread(fd, buf, nBytes, offset);
3740 }
3741 
3742 // Short sleep, direct OS call.
3743 //
3744 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3745 // sched_yield(2) will actually give up the CPU:
3746 //
3747 //   * Alone on this pariticular CPU, keeps running.
3748 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3749 //     (pre 2.6.39).
3750 //
3751 // So calling this with 0 is an alternative.
3752 //
3753 void os::naked_short_sleep(jlong ms) {
3754   struct timespec req;
3755 
3756   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3757   req.tv_sec = 0;
3758   if (ms > 0) {
3759     req.tv_nsec = (ms % 1000) * 1000000;
3760   } else {
3761     req.tv_nsec = 1;
3762   }
3763 
3764   nanosleep(&req, NULL);
3765 
3766   return;
3767 }
3768 
3769 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3770 void os::infinite_sleep() {
3771   while (true) {    // sleep forever ...
3772     ::sleep(100);   // ... 100 seconds at a time
3773   }
3774 }
3775 
3776 // Used to convert frequent JVM_Yield() to nops
3777 bool os::dont_yield() {
3778   return DontYieldALot;
3779 }
3780 
3781 void os::naked_yield() {
3782   sched_yield();
3783 }
3784 
3785 ////////////////////////////////////////////////////////////////////////////////
3786 // thread priority support
3787 
3788 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3789 // only supports dynamic priority, static priority must be zero. For real-time
3790 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3791 // However, for large multi-threaded applications, SCHED_RR is not only slower
3792 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3793 // of 5 runs - Sep 2005).
3794 //
3795 // The following code actually changes the niceness of kernel-thread/LWP. It
3796 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3797 // not the entire user process, and user level threads are 1:1 mapped to kernel
3798 // threads. It has always been the case, but could change in the future. For
3799 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3800 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3801 
3802 int os::java_to_os_priority[CriticalPriority + 1] = {
3803   19,              // 0 Entry should never be used
3804 
3805    4,              // 1 MinPriority
3806    3,              // 2
3807    2,              // 3
3808 
3809    1,              // 4
3810    0,              // 5 NormPriority
3811   -1,              // 6
3812 
3813   -2,              // 7
3814   -3,              // 8
3815   -4,              // 9 NearMaxPriority
3816 
3817   -5,              // 10 MaxPriority
3818 
3819   -5               // 11 CriticalPriority
3820 };
3821 
3822 static int prio_init() {
3823   if (ThreadPriorityPolicy == 1) {
3824     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3825     // if effective uid is not root. Perhaps, a more elegant way of doing
3826     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3827     if (geteuid() != 0) {
3828       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3829         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3830       }
3831       ThreadPriorityPolicy = 0;
3832     }
3833   }
3834   if (UseCriticalJavaThreadPriority) {
3835     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3836   }
3837   return 0;
3838 }
3839 
3840 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3841   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3842 
3843   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3844   return (ret == 0) ? OS_OK : OS_ERR;
3845 }
3846 
3847 OSReturn os::get_native_priority(const Thread* const thread,
3848                                  int *priority_ptr) {
3849   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3850     *priority_ptr = java_to_os_priority[NormPriority];
3851     return OS_OK;
3852   }
3853 
3854   errno = 0;
3855   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3856   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3857 }
3858 
3859 // Hint to the underlying OS that a task switch would not be good.
3860 // Void return because it's a hint and can fail.
3861 void os::hint_no_preempt() {}
3862 
3863 ////////////////////////////////////////////////////////////////////////////////
3864 // suspend/resume support
3865 
3866 //  the low-level signal-based suspend/resume support is a remnant from the
3867 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3868 //  within hotspot. Now there is a single use-case for this:
3869 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3870 //      that runs in the watcher thread.
3871 //  The remaining code is greatly simplified from the more general suspension
3872 //  code that used to be used.
3873 //
3874 //  The protocol is quite simple:
3875 //  - suspend:
3876 //      - sends a signal to the target thread
3877 //      - polls the suspend state of the osthread using a yield loop
3878 //      - target thread signal handler (SR_handler) sets suspend state
3879 //        and blocks in sigsuspend until continued
3880 //  - resume:
3881 //      - sets target osthread state to continue
3882 //      - sends signal to end the sigsuspend loop in the SR_handler
3883 //
3884 //  Note that the SR_lock plays no role in this suspend/resume protocol.
3885 
3886 static void resume_clear_context(OSThread *osthread) {
3887   osthread->set_ucontext(NULL);
3888   osthread->set_siginfo(NULL);
3889 }
3890 
3891 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3892                                  ucontext_t* context) {
3893   osthread->set_ucontext(context);
3894   osthread->set_siginfo(siginfo);
3895 }
3896 
3897 // Handler function invoked when a thread's execution is suspended or
3898 // resumed. We have to be careful that only async-safe functions are
3899 // called here (Note: most pthread functions are not async safe and
3900 // should be avoided.)
3901 //
3902 // Note: sigwait() is a more natural fit than sigsuspend() from an
3903 // interface point of view, but sigwait() prevents the signal hander
3904 // from being run. libpthread would get very confused by not having
3905 // its signal handlers run and prevents sigwait()'s use with the
3906 // mutex granting granting signal.
3907 //
3908 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3909 //
3910 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3911   // Save and restore errno to avoid confusing native code with EINTR
3912   // after sigsuspend.
3913   int old_errno = errno;
3914 
3915   Thread* thread = Thread::current();
3916   OSThread* osthread = thread->osthread();
3917   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3918 
3919   os::SuspendResume::State current = osthread->sr.state();
3920   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3921     suspend_save_context(osthread, siginfo, context);
3922 
3923     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3924     os::SuspendResume::State state = osthread->sr.suspended();
3925     if (state == os::SuspendResume::SR_SUSPENDED) {
3926       sigset_t suspend_set;  // signals for sigsuspend()
3927 
3928       // get current set of blocked signals and unblock resume signal
3929       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3930       sigdelset(&suspend_set, SR_signum);
3931 
3932       sr_semaphore.signal();
3933       // wait here until we are resumed
3934       while (1) {
3935         sigsuspend(&suspend_set);
3936 
3937         os::SuspendResume::State result = osthread->sr.running();
3938         if (result == os::SuspendResume::SR_RUNNING) {
3939           sr_semaphore.signal();
3940           break;
3941         }
3942       }
3943 
3944     } else if (state == os::SuspendResume::SR_RUNNING) {
3945       // request was cancelled, continue
3946     } else {
3947       ShouldNotReachHere();
3948     }
3949 
3950     resume_clear_context(osthread);
3951   } else if (current == os::SuspendResume::SR_RUNNING) {
3952     // request was cancelled, continue
3953   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3954     // ignore
3955   } else {
3956     // ignore
3957   }
3958 
3959   errno = old_errno;
3960 }
3961 
3962 static int SR_initialize() {
3963   struct sigaction act;
3964   char *s;
3965 
3966   // Get signal number to use for suspend/resume
3967   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3968     int sig = ::strtol(s, 0, 10);
3969     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
3970         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
3971       SR_signum = sig;
3972     } else {
3973       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
3974               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
3975     }
3976   }
3977 
3978   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3979          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3980 
3981   sigemptyset(&SR_sigset);
3982   sigaddset(&SR_sigset, SR_signum);
3983 
3984   // Set up signal handler for suspend/resume
3985   act.sa_flags = SA_RESTART|SA_SIGINFO;
3986   act.sa_handler = (void (*)(int)) SR_handler;
3987 
3988   // SR_signum is blocked by default.
3989   // 4528190 - We also need to block pthread restart signal (32 on all
3990   // supported Linux platforms). Note that LinuxThreads need to block
3991   // this signal for all threads to work properly. So we don't have
3992   // to use hard-coded signal number when setting up the mask.
3993   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3994 
3995   if (sigaction(SR_signum, &act, 0) == -1) {
3996     return -1;
3997   }
3998 
3999   // Save signal flag
4000   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4001   return 0;
4002 }
4003 
4004 static int sr_notify(OSThread* osthread) {
4005   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4006   assert_status(status == 0, status, "pthread_kill");
4007   return status;
4008 }
4009 
4010 // "Randomly" selected value for how long we want to spin
4011 // before bailing out on suspending a thread, also how often
4012 // we send a signal to a thread we want to resume
4013 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4014 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4015 
4016 // returns true on success and false on error - really an error is fatal
4017 // but this seems the normal response to library errors
4018 static bool do_suspend(OSThread* osthread) {
4019   assert(osthread->sr.is_running(), "thread should be running");
4020   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4021 
4022   // mark as suspended and send signal
4023   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4024     // failed to switch, state wasn't running?
4025     ShouldNotReachHere();
4026     return false;
4027   }
4028 
4029   if (sr_notify(osthread) != 0) {
4030     ShouldNotReachHere();
4031   }
4032 
4033   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4034   while (true) {
4035     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4036       break;
4037     } else {
4038       // timeout
4039       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4040       if (cancelled == os::SuspendResume::SR_RUNNING) {
4041         return false;
4042       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4043         // make sure that we consume the signal on the semaphore as well
4044         sr_semaphore.wait();
4045         break;
4046       } else {
4047         ShouldNotReachHere();
4048         return false;
4049       }
4050     }
4051   }
4052 
4053   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4054   return true;
4055 }
4056 
4057 static void do_resume(OSThread* osthread) {
4058   assert(osthread->sr.is_suspended(), "thread should be suspended");
4059   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4060 
4061   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4062     // failed to switch to WAKEUP_REQUEST
4063     ShouldNotReachHere();
4064     return;
4065   }
4066 
4067   while (true) {
4068     if (sr_notify(osthread) == 0) {
4069       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4070         if (osthread->sr.is_running()) {
4071           return;
4072         }
4073       }
4074     } else {
4075       ShouldNotReachHere();
4076     }
4077   }
4078 
4079   guarantee(osthread->sr.is_running(), "Must be running!");
4080 }
4081 
4082 ///////////////////////////////////////////////////////////////////////////////////
4083 // signal handling (except suspend/resume)
4084 
4085 // This routine may be used by user applications as a "hook" to catch signals.
4086 // The user-defined signal handler must pass unrecognized signals to this
4087 // routine, and if it returns true (non-zero), then the signal handler must
4088 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4089 // routine will never retun false (zero), but instead will execute a VM panic
4090 // routine kill the process.
4091 //
4092 // If this routine returns false, it is OK to call it again.  This allows
4093 // the user-defined signal handler to perform checks either before or after
4094 // the VM performs its own checks.  Naturally, the user code would be making
4095 // a serious error if it tried to handle an exception (such as a null check
4096 // or breakpoint) that the VM was generating for its own correct operation.
4097 //
4098 // This routine may recognize any of the following kinds of signals:
4099 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4100 // It should be consulted by handlers for any of those signals.
4101 //
4102 // The caller of this routine must pass in the three arguments supplied
4103 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4104 // field of the structure passed to sigaction().  This routine assumes that
4105 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4106 //
4107 // Note that the VM will print warnings if it detects conflicting signal
4108 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4109 //
4110 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4111                                                  siginfo_t* siginfo,
4112                                                  void* ucontext,
4113                                                  int abort_if_unrecognized);
4114 
4115 void signalHandler(int sig, siginfo_t* info, void* uc) {
4116   assert(info != NULL && uc != NULL, "it must be old kernel");
4117   int orig_errno = errno;  // Preserve errno value over signal handler.
4118   JVM_handle_linux_signal(sig, info, uc, true);
4119   errno = orig_errno;
4120 }
4121 
4122 
4123 // This boolean allows users to forward their own non-matching signals
4124 // to JVM_handle_linux_signal, harmlessly.
4125 bool os::Linux::signal_handlers_are_installed = false;
4126 
4127 // For signal-chaining
4128 struct sigaction sigact[NSIG];
4129 uint64_t sigs = 0;
4130 #if (64 < NSIG-1)
4131 #error "Not all signals can be encoded in sigs. Adapt its type!"
4132 #endif
4133 bool os::Linux::libjsig_is_loaded = false;
4134 typedef struct sigaction *(*get_signal_t)(int);
4135 get_signal_t os::Linux::get_signal_action = NULL;
4136 
4137 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4138   struct sigaction *actp = NULL;
4139 
4140   if (libjsig_is_loaded) {
4141     // Retrieve the old signal handler from libjsig
4142     actp = (*get_signal_action)(sig);
4143   }
4144   if (actp == NULL) {
4145     // Retrieve the preinstalled signal handler from jvm
4146     actp = get_preinstalled_handler(sig);
4147   }
4148 
4149   return actp;
4150 }
4151 
4152 static bool call_chained_handler(struct sigaction *actp, int sig,
4153                                  siginfo_t *siginfo, void *context) {
4154   // Call the old signal handler
4155   if (actp->sa_handler == SIG_DFL) {
4156     // It's more reasonable to let jvm treat it as an unexpected exception
4157     // instead of taking the default action.
4158     return false;
4159   } else if (actp->sa_handler != SIG_IGN) {
4160     if ((actp->sa_flags & SA_NODEFER) == 0) {
4161       // automaticlly block the signal
4162       sigaddset(&(actp->sa_mask), sig);
4163     }
4164 
4165     sa_handler_t hand = NULL;
4166     sa_sigaction_t sa = NULL;
4167     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4168     // retrieve the chained handler
4169     if (siginfo_flag_set) {
4170       sa = actp->sa_sigaction;
4171     } else {
4172       hand = actp->sa_handler;
4173     }
4174 
4175     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4176       actp->sa_handler = SIG_DFL;
4177     }
4178 
4179     // try to honor the signal mask
4180     sigset_t oset;
4181     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4182 
4183     // call into the chained handler
4184     if (siginfo_flag_set) {
4185       (*sa)(sig, siginfo, context);
4186     } else {
4187       (*hand)(sig);
4188     }
4189 
4190     // restore the signal mask
4191     pthread_sigmask(SIG_SETMASK, &oset, 0);
4192   }
4193   // Tell jvm's signal handler the signal is taken care of.
4194   return true;
4195 }
4196 
4197 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4198   bool chained = false;
4199   // signal-chaining
4200   if (UseSignalChaining) {
4201     struct sigaction *actp = get_chained_signal_action(sig);
4202     if (actp != NULL) {
4203       chained = call_chained_handler(actp, sig, siginfo, context);
4204     }
4205   }
4206   return chained;
4207 }
4208 
4209 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4210   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4211     return &sigact[sig];
4212   }
4213   return NULL;
4214 }
4215 
4216 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4217   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4218   sigact[sig] = oldAct;
4219   sigs |= (uint64_t)1 << (sig-1);
4220 }
4221 
4222 // for diagnostic
4223 int sigflags[NSIG];
4224 
4225 int os::Linux::get_our_sigflags(int sig) {
4226   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4227   return sigflags[sig];
4228 }
4229 
4230 void os::Linux::set_our_sigflags(int sig, int flags) {
4231   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4232   if (sig > 0 && sig < NSIG) {
4233     sigflags[sig] = flags;
4234   }
4235 }
4236 
4237 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4238   // Check for overwrite.
4239   struct sigaction oldAct;
4240   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4241 
4242   void* oldhand = oldAct.sa_sigaction
4243                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4244                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4245   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4246       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4247       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4248     if (AllowUserSignalHandlers || !set_installed) {
4249       // Do not overwrite; user takes responsibility to forward to us.
4250       return;
4251     } else if (UseSignalChaining) {
4252       // save the old handler in jvm
4253       save_preinstalled_handler(sig, oldAct);
4254       // libjsig also interposes the sigaction() call below and saves the
4255       // old sigaction on it own.
4256     } else {
4257       fatal("Encountered unexpected pre-existing sigaction handler "
4258             "%#lx for signal %d.", (long)oldhand, sig);
4259     }
4260   }
4261 
4262   struct sigaction sigAct;
4263   sigfillset(&(sigAct.sa_mask));
4264   sigAct.sa_handler = SIG_DFL;
4265   if (!set_installed) {
4266     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4267   } else {
4268     sigAct.sa_sigaction = signalHandler;
4269     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4270   }
4271   // Save flags, which are set by ours
4272   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4273   sigflags[sig] = sigAct.sa_flags;
4274 
4275   int ret = sigaction(sig, &sigAct, &oldAct);
4276   assert(ret == 0, "check");
4277 
4278   void* oldhand2  = oldAct.sa_sigaction
4279                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4280                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4281   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4282 }
4283 
4284 // install signal handlers for signals that HotSpot needs to
4285 // handle in order to support Java-level exception handling.
4286 
4287 void os::Linux::install_signal_handlers() {
4288   if (!signal_handlers_are_installed) {
4289     signal_handlers_are_installed = true;
4290 
4291     // signal-chaining
4292     typedef void (*signal_setting_t)();
4293     signal_setting_t begin_signal_setting = NULL;
4294     signal_setting_t end_signal_setting = NULL;
4295     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4296                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4297     if (begin_signal_setting != NULL) {
4298       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4299                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4300       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4301                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4302       libjsig_is_loaded = true;
4303       assert(UseSignalChaining, "should enable signal-chaining");
4304     }
4305     if (libjsig_is_loaded) {
4306       // Tell libjsig jvm is setting signal handlers
4307       (*begin_signal_setting)();
4308     }
4309 
4310     set_signal_handler(SIGSEGV, true);
4311     set_signal_handler(SIGPIPE, true);
4312     set_signal_handler(SIGBUS, true);
4313     set_signal_handler(SIGILL, true);
4314     set_signal_handler(SIGFPE, true);
4315 #if defined(PPC64)
4316     set_signal_handler(SIGTRAP, true);
4317 #endif
4318     set_signal_handler(SIGXFSZ, true);
4319 
4320     if (libjsig_is_loaded) {
4321       // Tell libjsig jvm finishes setting signal handlers
4322       (*end_signal_setting)();
4323     }
4324 
4325     // We don't activate signal checker if libjsig is in place, we trust ourselves
4326     // and if UserSignalHandler is installed all bets are off.
4327     // Log that signal checking is off only if -verbose:jni is specified.
4328     if (CheckJNICalls) {
4329       if (libjsig_is_loaded) {
4330         if (PrintJNIResolving) {
4331           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4332         }
4333         check_signals = false;
4334       }
4335       if (AllowUserSignalHandlers) {
4336         if (PrintJNIResolving) {
4337           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4338         }
4339         check_signals = false;
4340       }
4341     }
4342   }
4343 }
4344 
4345 // This is the fastest way to get thread cpu time on Linux.
4346 // Returns cpu time (user+sys) for any thread, not only for current.
4347 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4348 // It might work on 2.6.10+ with a special kernel/glibc patch.
4349 // For reference, please, see IEEE Std 1003.1-2004:
4350 //   http://www.unix.org/single_unix_specification
4351 
4352 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4353   struct timespec tp;
4354   int rc = os::Linux::clock_gettime(clockid, &tp);
4355   assert(rc == 0, "clock_gettime is expected to return 0 code");
4356 
4357   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4358 }
4359 
4360 void os::Linux::initialize_os_info() {
4361   assert(_os_version == 0, "OS info already initialized");
4362 
4363   struct utsname _uname;
4364 
4365   uint32_t major;
4366   uint32_t minor;
4367   uint32_t fix;
4368 
4369   int rc;
4370 
4371   // Kernel version is unknown if
4372   // verification below fails.
4373   _os_version = 0x01000000;
4374 
4375   rc = uname(&_uname);
4376   if (rc != -1) {
4377 
4378     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4379     if (rc == 3) {
4380 
4381       if (major < 256 && minor < 256 && fix < 256) {
4382         // Kernel version format is as expected,
4383         // set it overriding unknown state.
4384         _os_version = (major << 16) |
4385                       (minor << 8 ) |
4386                       (fix   << 0 ) ;
4387       }
4388     }
4389   }
4390 }
4391 
4392 uint32_t os::Linux::os_version() {
4393   assert(_os_version != 0, "not initialized");
4394   return _os_version & 0x00FFFFFF;
4395 }
4396 
4397 bool os::Linux::os_version_is_known() {
4398   assert(_os_version != 0, "not initialized");
4399   return _os_version & 0x01000000 ? false : true;
4400 }
4401 
4402 /////
4403 // glibc on Linux platform uses non-documented flag
4404 // to indicate, that some special sort of signal
4405 // trampoline is used.
4406 // We will never set this flag, and we should
4407 // ignore this flag in our diagnostic
4408 #ifdef SIGNIFICANT_SIGNAL_MASK
4409   #undef SIGNIFICANT_SIGNAL_MASK
4410 #endif
4411 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4412 
4413 static const char* get_signal_handler_name(address handler,
4414                                            char* buf, int buflen) {
4415   int offset = 0;
4416   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4417   if (found) {
4418     // skip directory names
4419     const char *p1, *p2;
4420     p1 = buf;
4421     size_t len = strlen(os::file_separator());
4422     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4423     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4424   } else {
4425     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4426   }
4427   return buf;
4428 }
4429 
4430 static void print_signal_handler(outputStream* st, int sig,
4431                                  char* buf, size_t buflen) {
4432   struct sigaction sa;
4433 
4434   sigaction(sig, NULL, &sa);
4435 
4436   // See comment for SIGNIFICANT_SIGNAL_MASK define
4437   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4438 
4439   st->print("%s: ", os::exception_name(sig, buf, buflen));
4440 
4441   address handler = (sa.sa_flags & SA_SIGINFO)
4442     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4443     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4444 
4445   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4446     st->print("SIG_DFL");
4447   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4448     st->print("SIG_IGN");
4449   } else {
4450     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4451   }
4452 
4453   st->print(", sa_mask[0]=");
4454   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4455 
4456   address rh = VMError::get_resetted_sighandler(sig);
4457   // May be, handler was resetted by VMError?
4458   if (rh != NULL) {
4459     handler = rh;
4460     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4461   }
4462 
4463   st->print(", sa_flags=");
4464   os::Posix::print_sa_flags(st, sa.sa_flags);
4465 
4466   // Check: is it our handler?
4467   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4468       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4469     // It is our signal handler
4470     // check for flags, reset system-used one!
4471     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4472       st->print(
4473                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4474                 os::Linux::get_our_sigflags(sig));
4475     }
4476   }
4477   st->cr();
4478 }
4479 
4480 
4481 #define DO_SIGNAL_CHECK(sig)                      \
4482   do {                                            \
4483     if (!sigismember(&check_signal_done, sig)) {  \
4484       os::Linux::check_signal_handler(sig);       \
4485     }                                             \
4486   } while (0)
4487 
4488 // This method is a periodic task to check for misbehaving JNI applications
4489 // under CheckJNI, we can add any periodic checks here
4490 
4491 void os::run_periodic_checks() {
4492   if (check_signals == false) return;
4493 
4494   // SEGV and BUS if overridden could potentially prevent
4495   // generation of hs*.log in the event of a crash, debugging
4496   // such a case can be very challenging, so we absolutely
4497   // check the following for a good measure:
4498   DO_SIGNAL_CHECK(SIGSEGV);
4499   DO_SIGNAL_CHECK(SIGILL);
4500   DO_SIGNAL_CHECK(SIGFPE);
4501   DO_SIGNAL_CHECK(SIGBUS);
4502   DO_SIGNAL_CHECK(SIGPIPE);
4503   DO_SIGNAL_CHECK(SIGXFSZ);
4504 #if defined(PPC64)
4505   DO_SIGNAL_CHECK(SIGTRAP);
4506 #endif
4507 
4508   // ReduceSignalUsage allows the user to override these handlers
4509   // see comments at the very top and jvm_solaris.h
4510   if (!ReduceSignalUsage) {
4511     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4512     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4513     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4514     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4515   }
4516 
4517   DO_SIGNAL_CHECK(SR_signum);
4518 }
4519 
4520 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4521 
4522 static os_sigaction_t os_sigaction = NULL;
4523 
4524 void os::Linux::check_signal_handler(int sig) {
4525   char buf[O_BUFLEN];
4526   address jvmHandler = NULL;
4527 
4528 
4529   struct sigaction act;
4530   if (os_sigaction == NULL) {
4531     // only trust the default sigaction, in case it has been interposed
4532     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4533     if (os_sigaction == NULL) return;
4534   }
4535 
4536   os_sigaction(sig, (struct sigaction*)NULL, &act);
4537 
4538 
4539   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4540 
4541   address thisHandler = (act.sa_flags & SA_SIGINFO)
4542     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4543     : CAST_FROM_FN_PTR(address, act.sa_handler);
4544 
4545 
4546   switch (sig) {
4547   case SIGSEGV:
4548   case SIGBUS:
4549   case SIGFPE:
4550   case SIGPIPE:
4551   case SIGILL:
4552   case SIGXFSZ:
4553     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4554     break;
4555 
4556   case SHUTDOWN1_SIGNAL:
4557   case SHUTDOWN2_SIGNAL:
4558   case SHUTDOWN3_SIGNAL:
4559   case BREAK_SIGNAL:
4560     jvmHandler = (address)user_handler();
4561     break;
4562 
4563   default:
4564     if (sig == SR_signum) {
4565       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4566     } else {
4567       return;
4568     }
4569     break;
4570   }
4571 
4572   if (thisHandler != jvmHandler) {
4573     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4574     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4575     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4576     // No need to check this sig any longer
4577     sigaddset(&check_signal_done, sig);
4578     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4579     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4580       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4581                     exception_name(sig, buf, O_BUFLEN));
4582     }
4583   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4584     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4585     tty->print("expected:");
4586     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4587     tty->cr();
4588     tty->print("  found:");
4589     os::Posix::print_sa_flags(tty, act.sa_flags);
4590     tty->cr();
4591     // No need to check this sig any longer
4592     sigaddset(&check_signal_done, sig);
4593   }
4594 
4595   // Dump all the signal
4596   if (sigismember(&check_signal_done, sig)) {
4597     print_signal_handlers(tty, buf, O_BUFLEN);
4598   }
4599 }
4600 
4601 extern void report_error(char* file_name, int line_no, char* title,
4602                          char* format, ...);
4603 
4604 // this is called _before_ the most of global arguments have been parsed
4605 void os::init(void) {
4606   char dummy;   // used to get a guess on initial stack address
4607 //  first_hrtime = gethrtime();
4608 
4609   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4610 
4611   init_random(1234567);
4612 
4613   ThreadCritical::initialize();
4614 
4615   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4616   if (Linux::page_size() == -1) {
4617     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4618           strerror(errno));
4619   }
4620   init_page_sizes((size_t) Linux::page_size());
4621 
4622   Linux::initialize_system_info();
4623 
4624   Linux::initialize_os_info();
4625 
4626   // main_thread points to the aboriginal thread
4627   Linux::_main_thread = pthread_self();
4628 
4629   Linux::clock_init();
4630   initial_time_count = javaTimeNanos();
4631 
4632   // pthread_condattr initialization for monotonic clock
4633   int status;
4634   pthread_condattr_t* _condattr = os::Linux::condAttr();
4635   if ((status = pthread_condattr_init(_condattr)) != 0) {
4636     fatal("pthread_condattr_init: %s", strerror(status));
4637   }
4638   // Only set the clock if CLOCK_MONOTONIC is available
4639   if (os::supports_monotonic_clock()) {
4640     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4641       if (status == EINVAL) {
4642         warning("Unable to use monotonic clock with relative timed-waits" \
4643                 " - changes to the time-of-day clock may have adverse affects");
4644       } else {
4645         fatal("pthread_condattr_setclock: %s", strerror(status));
4646       }
4647     }
4648   }
4649   // else it defaults to CLOCK_REALTIME
4650 
4651   // retrieve entry point for pthread_setname_np
4652   Linux::_pthread_setname_np =
4653     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4654 
4655 }
4656 
4657 // To install functions for atexit system call
4658 extern "C" {
4659   static void perfMemory_exit_helper() {
4660     perfMemory_exit();
4661   }
4662 }
4663 
4664 // this is called _after_ the global arguments have been parsed
4665 jint os::init_2(void) {
4666   Linux::fast_thread_clock_init();
4667 
4668   // Allocate a single page and mark it as readable for safepoint polling
4669   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4670   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4671 
4672   os::set_polling_page(polling_page);
4673 
4674 #ifndef PRODUCT
4675   if (Verbose && PrintMiscellaneous) {
4676     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4677                (intptr_t)polling_page);
4678   }
4679 #endif
4680 
4681   if (!UseMembar) {
4682     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4683     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4684     os::set_memory_serialize_page(mem_serialize_page);
4685 
4686 #ifndef PRODUCT
4687     if (Verbose && PrintMiscellaneous) {
4688       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n",
4689                  (intptr_t)mem_serialize_page);
4690     }
4691 #endif
4692   }
4693 
4694   // initialize suspend/resume support - must do this before signal_sets_init()
4695   if (SR_initialize() != 0) {
4696     perror("SR_initialize failed");
4697     return JNI_ERR;
4698   }
4699 
4700   Linux::signal_sets_init();
4701   Linux::install_signal_handlers();
4702 
4703   // Check minimum allowable stack size for thread creation and to initialize
4704   // the java system classes, including StackOverflowError - depends on page
4705   // size.  Add a page for compiler2 recursion in main thread.
4706   // Add in 2*BytesPerWord times page size to account for VM stack during
4707   // class initialization depending on 32 or 64 bit VM.
4708   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4709                                       JavaThread::stack_guard_zone_size() +
4710                                       JavaThread::stack_shadow_zone_size() +
4711                                       (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4712 
4713   size_t threadStackSizeInBytes = ThreadStackSize * K;
4714   if (threadStackSizeInBytes != 0 &&
4715       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4716     tty->print_cr("\nThe stack size specified is too small, "
4717                   "Specify at least " SIZE_FORMAT "k",
4718                   os::Linux::min_stack_allowed/ K);
4719     return JNI_ERR;
4720   }
4721 
4722   // Make the stack size a multiple of the page size so that
4723   // the yellow/red zones can be guarded.
4724   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4725                                                 vm_page_size()));
4726 
4727   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4728 
4729 #if defined(IA32)
4730   workaround_expand_exec_shield_cs_limit();
4731 #endif
4732 
4733   Linux::libpthread_init();
4734   if (PrintMiscellaneous && (Verbose || WizardMode)) {
4735     tty->print_cr("[HotSpot is running with %s, %s]\n",
4736                   Linux::glibc_version(), Linux::libpthread_version());
4737   }
4738 
4739   if (UseNUMA) {
4740     if (!Linux::libnuma_init()) {
4741       UseNUMA = false;
4742     } else {
4743       if ((Linux::numa_max_node() < 1)) {
4744         // There's only one node(they start from 0), disable NUMA.
4745         UseNUMA = false;
4746       }
4747     }
4748     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4749     // we can make the adaptive lgrp chunk resizing work. If the user specified
4750     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4751     // disable adaptive resizing.
4752     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4753       if (FLAG_IS_DEFAULT(UseNUMA)) {
4754         UseNUMA = false;
4755       } else {
4756         if (FLAG_IS_DEFAULT(UseLargePages) &&
4757             FLAG_IS_DEFAULT(UseSHM) &&
4758             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4759           UseLargePages = false;
4760         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4761           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4762           UseAdaptiveSizePolicy = false;
4763           UseAdaptiveNUMAChunkSizing = false;
4764         }
4765       }
4766     }
4767     if (!UseNUMA && ForceNUMA) {
4768       UseNUMA = true;
4769     }
4770   }
4771 
4772   if (MaxFDLimit) {
4773     // set the number of file descriptors to max. print out error
4774     // if getrlimit/setrlimit fails but continue regardless.
4775     struct rlimit nbr_files;
4776     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4777     if (status != 0) {
4778       if (PrintMiscellaneous && (Verbose || WizardMode)) {
4779         perror("os::init_2 getrlimit failed");
4780       }
4781     } else {
4782       nbr_files.rlim_cur = nbr_files.rlim_max;
4783       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4784       if (status != 0) {
4785         if (PrintMiscellaneous && (Verbose || WizardMode)) {
4786           perror("os::init_2 setrlimit failed");
4787         }
4788       }
4789     }
4790   }
4791 
4792   // Initialize lock used to serialize thread creation (see os::create_thread)
4793   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4794 
4795   // at-exit methods are called in the reverse order of their registration.
4796   // atexit functions are called on return from main or as a result of a
4797   // call to exit(3C). There can be only 32 of these functions registered
4798   // and atexit() does not set errno.
4799 
4800   if (PerfAllowAtExitRegistration) {
4801     // only register atexit functions if PerfAllowAtExitRegistration is set.
4802     // atexit functions can be delayed until process exit time, which
4803     // can be problematic for embedded VM situations. Embedded VMs should
4804     // call DestroyJavaVM() to assure that VM resources are released.
4805 
4806     // note: perfMemory_exit_helper atexit function may be removed in
4807     // the future if the appropriate cleanup code can be added to the
4808     // VM_Exit VMOperation's doit method.
4809     if (atexit(perfMemory_exit_helper) != 0) {
4810       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4811     }
4812   }
4813 
4814   // initialize thread priority policy
4815   prio_init();
4816 
4817   return JNI_OK;
4818 }
4819 
4820 // Mark the polling page as unreadable
4821 void os::make_polling_page_unreadable(void) {
4822   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4823     fatal("Could not disable polling page");
4824   }
4825 }
4826 
4827 // Mark the polling page as readable
4828 void os::make_polling_page_readable(void) {
4829   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4830     fatal("Could not enable polling page");
4831   }
4832 }
4833 
4834 // older glibc versions don't have this macro (which expands to
4835 // an optimized bit-counting function) so we have to roll our own
4836 #ifndef CPU_COUNT
4837 
4838 static int _cpu_count(const cpu_set_t* cpus) {
4839   int count = 0;
4840   // only look up to the number of configured processors
4841   for (int i = 0; i < os::processor_count(); i++) {
4842     if (CPU_ISSET(i, cpus)) {
4843       count++;
4844     }
4845   }
4846   return count;
4847 }
4848 
4849 #define CPU_COUNT(cpus) _cpu_count(cpus)
4850 
4851 #endif // CPU_COUNT
4852 
4853 // Get the current number of available processors for this process.
4854 // This value can change at any time during a process's lifetime.
4855 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4856 // If it appears there may be more than 1024 processors then we do a
4857 // dynamic check - see 6515172 for details.
4858 // If anything goes wrong we fallback to returning the number of online
4859 // processors - which can be greater than the number available to the process.
4860 int os::active_processor_count() {
4861   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
4862   cpu_set_t* cpus_p = &cpus;
4863   int cpus_size = sizeof(cpu_set_t);
4864 
4865   int configured_cpus = processor_count();  // upper bound on available cpus
4866   int cpu_count = 0;
4867 
4868 // old build platforms may not support dynamic cpu sets
4869 #ifdef CPU_ALLOC
4870 
4871   // To enable easy testing of the dynamic path on different platforms we
4872   // introduce a diagnostic flag: UseCpuAllocPath
4873   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4874     // kernel may use a mask bigger than cpu_set_t
4875     log_trace(os)("active_processor_count: using dynamic path %s"
4876                   "- configured processors: %d",
4877                   UseCpuAllocPath ? "(forced) " : "",
4878                   configured_cpus);
4879     cpus_p = CPU_ALLOC(configured_cpus);
4880     if (cpus_p != NULL) {
4881       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4882       // zero it just to be safe
4883       CPU_ZERO_S(cpus_size, cpus_p);
4884     }
4885     else {
4886        // failed to allocate so fallback to online cpus
4887        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4888        log_trace(os)("active_processor_count: "
4889                      "CPU_ALLOC failed (%s) - using "
4890                      "online processor count: %d",
4891                      strerror(errno), online_cpus);
4892        return online_cpus;
4893     }
4894   }
4895   else {
4896     log_trace(os)("active_processor_count: using static path - configured processors: %d",
4897                   configured_cpus);
4898   }
4899 #else // CPU_ALLOC
4900 // these stubs won't be executed
4901 #define CPU_COUNT_S(size, cpus) -1
4902 #define CPU_FREE(cpus)
4903 
4904   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
4905                 configured_cpus);
4906 #endif // CPU_ALLOC
4907 
4908   // pid 0 means the current thread - which we have to assume represents the process
4909   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
4910     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4911       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
4912     }
4913     else {
4914       cpu_count = CPU_COUNT(cpus_p);
4915     }
4916     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
4917   }
4918   else {
4919     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
4920     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
4921             "which may exceed available processors", strerror(errno), cpu_count);
4922   }
4923 
4924   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4925     CPU_FREE(cpus_p);
4926   }
4927 
4928   assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
4929   return cpu_count;
4930 }
4931 
4932 void os::set_native_thread_name(const char *name) {
4933   if (Linux::_pthread_setname_np) {
4934     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4935     snprintf(buf, sizeof(buf), "%s", name);
4936     buf[sizeof(buf) - 1] = '\0';
4937     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4938     // ERANGE should not happen; all other errors should just be ignored.
4939     assert(rc != ERANGE, "pthread_setname_np failed");
4940   }
4941 }
4942 
4943 bool os::distribute_processes(uint length, uint* distribution) {
4944   // Not yet implemented.
4945   return false;
4946 }
4947 
4948 bool os::bind_to_processor(uint processor_id) {
4949   // Not yet implemented.
4950   return false;
4951 }
4952 
4953 ///
4954 
4955 void os::SuspendedThreadTask::internal_do_task() {
4956   if (do_suspend(_thread->osthread())) {
4957     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4958     do_task(context);
4959     do_resume(_thread->osthread());
4960   }
4961 }
4962 
4963 class PcFetcher : public os::SuspendedThreadTask {
4964  public:
4965   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4966   ExtendedPC result();
4967  protected:
4968   void do_task(const os::SuspendedThreadTaskContext& context);
4969  private:
4970   ExtendedPC _epc;
4971 };
4972 
4973 ExtendedPC PcFetcher::result() {
4974   guarantee(is_done(), "task is not done yet.");
4975   return _epc;
4976 }
4977 
4978 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4979   Thread* thread = context.thread();
4980   OSThread* osthread = thread->osthread();
4981   if (osthread->ucontext() != NULL) {
4982     _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
4983   } else {
4984     // NULL context is unexpected, double-check this is the VMThread
4985     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4986   }
4987 }
4988 
4989 // Suspends the target using the signal mechanism and then grabs the PC before
4990 // resuming the target. Used by the flat-profiler only
4991 ExtendedPC os::get_thread_pc(Thread* thread) {
4992   // Make sure that it is called by the watcher for the VMThread
4993   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4994   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4995 
4996   PcFetcher fetcher(thread);
4997   fetcher.run();
4998   return fetcher.result();
4999 }
5000 
5001 ////////////////////////////////////////////////////////////////////////////////
5002 // debug support
5003 
5004 bool os::find(address addr, outputStream* st) {
5005   Dl_info dlinfo;
5006   memset(&dlinfo, 0, sizeof(dlinfo));
5007   if (dladdr(addr, &dlinfo) != 0) {
5008     st->print(PTR_FORMAT ": ", p2i(addr));
5009     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5010       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5011                 p2i(addr) - p2i(dlinfo.dli_saddr));
5012     } else if (dlinfo.dli_fbase != NULL) {
5013       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5014     } else {
5015       st->print("<absolute address>");
5016     }
5017     if (dlinfo.dli_fname != NULL) {
5018       st->print(" in %s", dlinfo.dli_fname);
5019     }
5020     if (dlinfo.dli_fbase != NULL) {
5021       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5022     }
5023     st->cr();
5024 
5025     if (Verbose) {
5026       // decode some bytes around the PC
5027       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5028       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5029       address       lowest = (address) dlinfo.dli_sname;
5030       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5031       if (begin < lowest)  begin = lowest;
5032       Dl_info dlinfo2;
5033       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5034           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5035         end = (address) dlinfo2.dli_saddr;
5036       }
5037       Disassembler::decode(begin, end, st);
5038     }
5039     return true;
5040   }
5041   return false;
5042 }
5043 
5044 ////////////////////////////////////////////////////////////////////////////////
5045 // misc
5046 
5047 // This does not do anything on Linux. This is basically a hook for being
5048 // able to use structured exception handling (thread-local exception filters)
5049 // on, e.g., Win32.
5050 void
5051 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5052                          JavaCallArguments* args, Thread* thread) {
5053   f(value, method, args, thread);
5054 }
5055 
5056 void os::print_statistics() {
5057 }
5058 
5059 bool os::message_box(const char* title, const char* message) {
5060   int i;
5061   fdStream err(defaultStream::error_fd());
5062   for (i = 0; i < 78; i++) err.print_raw("=");
5063   err.cr();
5064   err.print_raw_cr(title);
5065   for (i = 0; i < 78; i++) err.print_raw("-");
5066   err.cr();
5067   err.print_raw_cr(message);
5068   for (i = 0; i < 78; i++) err.print_raw("=");
5069   err.cr();
5070 
5071   char buf[16];
5072   // Prevent process from exiting upon "read error" without consuming all CPU
5073   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5074 
5075   return buf[0] == 'y' || buf[0] == 'Y';
5076 }
5077 
5078 int os::stat(const char *path, struct stat *sbuf) {
5079   char pathbuf[MAX_PATH];
5080   if (strlen(path) > MAX_PATH - 1) {
5081     errno = ENAMETOOLONG;
5082     return -1;
5083   }
5084   os::native_path(strcpy(pathbuf, path));
5085   return ::stat(pathbuf, sbuf);
5086 }
5087 
5088 bool os::check_heap(bool force) {
5089   return true;
5090 }
5091 
5092 // Is a (classpath) directory empty?
5093 bool os::dir_is_empty(const char* path) {
5094   DIR *dir = NULL;
5095   struct dirent *ptr;
5096 
5097   dir = opendir(path);
5098   if (dir == NULL) return true;
5099 
5100   // Scan the directory
5101   bool result = true;
5102   char buf[sizeof(struct dirent) + MAX_PATH];
5103   while (result && (ptr = ::readdir(dir)) != NULL) {
5104     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5105       result = false;
5106     }
5107   }
5108   closedir(dir);
5109   return result;
5110 }
5111 
5112 // This code originates from JDK's sysOpen and open64_w
5113 // from src/solaris/hpi/src/system_md.c
5114 
5115 int os::open(const char *path, int oflag, int mode) {
5116   if (strlen(path) > MAX_PATH - 1) {
5117     errno = ENAMETOOLONG;
5118     return -1;
5119   }
5120 
5121   // All file descriptors that are opened in the Java process and not
5122   // specifically destined for a subprocess should have the close-on-exec
5123   // flag set.  If we don't set it, then careless 3rd party native code
5124   // might fork and exec without closing all appropriate file descriptors
5125   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5126   // turn might:
5127   //
5128   // - cause end-of-file to fail to be detected on some file
5129   //   descriptors, resulting in mysterious hangs, or
5130   //
5131   // - might cause an fopen in the subprocess to fail on a system
5132   //   suffering from bug 1085341.
5133   //
5134   // (Yes, the default setting of the close-on-exec flag is a Unix
5135   // design flaw)
5136   //
5137   // See:
5138   // 1085341: 32-bit stdio routines should support file descriptors >255
5139   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5140   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5141   //
5142   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5143   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5144   // because it saves a system call and removes a small window where the flag
5145   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5146   // and we fall back to using FD_CLOEXEC (see below).
5147 #ifdef O_CLOEXEC
5148   oflag |= O_CLOEXEC;
5149 #endif
5150 
5151   int fd = ::open64(path, oflag, mode);
5152   if (fd == -1) return -1;
5153 
5154   //If the open succeeded, the file might still be a directory
5155   {
5156     struct stat64 buf64;
5157     int ret = ::fstat64(fd, &buf64);
5158     int st_mode = buf64.st_mode;
5159 
5160     if (ret != -1) {
5161       if ((st_mode & S_IFMT) == S_IFDIR) {
5162         errno = EISDIR;
5163         ::close(fd);
5164         return -1;
5165       }
5166     } else {
5167       ::close(fd);
5168       return -1;
5169     }
5170   }
5171 
5172 #ifdef FD_CLOEXEC
5173   // Validate that the use of the O_CLOEXEC flag on open above worked.
5174   // With recent kernels, we will perform this check exactly once.
5175   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5176   if (!O_CLOEXEC_is_known_to_work) {
5177     int flags = ::fcntl(fd, F_GETFD);
5178     if (flags != -1) {
5179       if ((flags & FD_CLOEXEC) != 0)
5180         O_CLOEXEC_is_known_to_work = 1;
5181       else
5182         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5183     }
5184   }
5185 #endif
5186 
5187   return fd;
5188 }
5189 
5190 
5191 // create binary file, rewriting existing file if required
5192 int os::create_binary_file(const char* path, bool rewrite_existing) {
5193   int oflags = O_WRONLY | O_CREAT;
5194   if (!rewrite_existing) {
5195     oflags |= O_EXCL;
5196   }
5197   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5198 }
5199 
5200 // return current position of file pointer
5201 jlong os::current_file_offset(int fd) {
5202   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5203 }
5204 
5205 // move file pointer to the specified offset
5206 jlong os::seek_to_file_offset(int fd, jlong offset) {
5207   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5208 }
5209 
5210 // This code originates from JDK's sysAvailable
5211 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5212 
5213 int os::available(int fd, jlong *bytes) {
5214   jlong cur, end;
5215   int mode;
5216   struct stat64 buf64;
5217 
5218   if (::fstat64(fd, &buf64) >= 0) {
5219     mode = buf64.st_mode;
5220     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5221       int n;
5222       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5223         *bytes = n;
5224         return 1;
5225       }
5226     }
5227   }
5228   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5229     return 0;
5230   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5231     return 0;
5232   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5233     return 0;
5234   }
5235   *bytes = end - cur;
5236   return 1;
5237 }
5238 
5239 // Map a block of memory.
5240 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5241                         char *addr, size_t bytes, bool read_only,
5242                         bool allow_exec) {
5243   int prot;
5244   int flags = MAP_PRIVATE;
5245 
5246   if (read_only) {
5247     prot = PROT_READ;
5248   } else {
5249     prot = PROT_READ | PROT_WRITE;
5250   }
5251 
5252   if (allow_exec) {
5253     prot |= PROT_EXEC;
5254   }
5255 
5256   if (addr != NULL) {
5257     flags |= MAP_FIXED;
5258   }
5259 
5260   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5261                                      fd, file_offset);
5262   if (mapped_address == MAP_FAILED) {
5263     return NULL;
5264   }
5265   return mapped_address;
5266 }
5267 
5268 
5269 // Remap a block of memory.
5270 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5271                           char *addr, size_t bytes, bool read_only,
5272                           bool allow_exec) {
5273   // same as map_memory() on this OS
5274   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5275                         allow_exec);
5276 }
5277 
5278 
5279 // Unmap a block of memory.
5280 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5281   return munmap(addr, bytes) == 0;
5282 }
5283 
5284 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5285 
5286 static clockid_t thread_cpu_clockid(Thread* thread) {
5287   pthread_t tid = thread->osthread()->pthread_id();
5288   clockid_t clockid;
5289 
5290   // Get thread clockid
5291   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5292   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5293   return clockid;
5294 }
5295 
5296 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5297 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5298 // of a thread.
5299 //
5300 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5301 // the fast estimate available on the platform.
5302 
5303 jlong os::current_thread_cpu_time() {
5304   if (os::Linux::supports_fast_thread_cpu_time()) {
5305     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5306   } else {
5307     // return user + sys since the cost is the same
5308     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5309   }
5310 }
5311 
5312 jlong os::thread_cpu_time(Thread* thread) {
5313   // consistent with what current_thread_cpu_time() returns
5314   if (os::Linux::supports_fast_thread_cpu_time()) {
5315     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5316   } else {
5317     return slow_thread_cpu_time(thread, true /* user + sys */);
5318   }
5319 }
5320 
5321 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5322   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5323     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5324   } else {
5325     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5326   }
5327 }
5328 
5329 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5330   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5331     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5332   } else {
5333     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5334   }
5335 }
5336 
5337 //  -1 on error.
5338 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5339   pid_t  tid = thread->osthread()->thread_id();
5340   char *s;
5341   char stat[2048];
5342   int statlen;
5343   char proc_name[64];
5344   int count;
5345   long sys_time, user_time;
5346   char cdummy;
5347   int idummy;
5348   long ldummy;
5349   FILE *fp;
5350 
5351   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5352   fp = fopen(proc_name, "r");
5353   if (fp == NULL) return -1;
5354   statlen = fread(stat, 1, 2047, fp);
5355   stat[statlen] = '\0';
5356   fclose(fp);
5357 
5358   // Skip pid and the command string. Note that we could be dealing with
5359   // weird command names, e.g. user could decide to rename java launcher
5360   // to "java 1.4.2 :)", then the stat file would look like
5361   //                1234 (java 1.4.2 :)) R ... ...
5362   // We don't really need to know the command string, just find the last
5363   // occurrence of ")" and then start parsing from there. See bug 4726580.
5364   s = strrchr(stat, ')');
5365   if (s == NULL) return -1;
5366 
5367   // Skip blank chars
5368   do { s++; } while (s && isspace(*s));
5369 
5370   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5371                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5372                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5373                  &user_time, &sys_time);
5374   if (count != 13) return -1;
5375   if (user_sys_cpu_time) {
5376     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5377   } else {
5378     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5379   }
5380 }
5381 
5382 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5383   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5384   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5385   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5386   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5387 }
5388 
5389 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5390   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5391   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5392   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5393   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5394 }
5395 
5396 bool os::is_thread_cpu_time_supported() {
5397   return true;
5398 }
5399 
5400 // System loadavg support.  Returns -1 if load average cannot be obtained.
5401 // Linux doesn't yet have a (official) notion of processor sets,
5402 // so just return the system wide load average.
5403 int os::loadavg(double loadavg[], int nelem) {
5404   return ::getloadavg(loadavg, nelem);
5405 }
5406 
5407 void os::pause() {
5408   char filename[MAX_PATH];
5409   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5410     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5411   } else {
5412     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5413   }
5414 
5415   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5416   if (fd != -1) {
5417     struct stat buf;
5418     ::close(fd);
5419     while (::stat(filename, &buf) == 0) {
5420       (void)::poll(NULL, 0, 100);
5421     }
5422   } else {
5423     jio_fprintf(stderr,
5424                 "Could not open pause file '%s', continuing immediately.\n", filename);
5425   }
5426 }
5427 
5428 
5429 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5430 // comment paragraphs are worth repeating here:
5431 //
5432 // Assumption:
5433 //    Only one parker can exist on an event, which is why we allocate
5434 //    them per-thread. Multiple unparkers can coexist.
5435 //
5436 // _Event serves as a restricted-range semaphore.
5437 //   -1 : thread is blocked, i.e. there is a waiter
5438 //    0 : neutral: thread is running or ready,
5439 //        could have been signaled after a wait started
5440 //    1 : signaled - thread is running or ready
5441 //
5442 
5443 // utility to compute the abstime argument to timedwait:
5444 // millis is the relative timeout time
5445 // abstime will be the absolute timeout time
5446 // TODO: replace compute_abstime() with unpackTime()
5447 
5448 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5449   if (millis < 0)  millis = 0;
5450 
5451   jlong seconds = millis / 1000;
5452   millis %= 1000;
5453   if (seconds > 50000000) { // see man cond_timedwait(3T)
5454     seconds = 50000000;
5455   }
5456 
5457   if (os::supports_monotonic_clock()) {
5458     struct timespec now;
5459     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5460     assert_status(status == 0, status, "clock_gettime");
5461     abstime->tv_sec = now.tv_sec  + seconds;
5462     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5463     if (nanos >= NANOSECS_PER_SEC) {
5464       abstime->tv_sec += 1;
5465       nanos -= NANOSECS_PER_SEC;
5466     }
5467     abstime->tv_nsec = nanos;
5468   } else {
5469     struct timeval now;
5470     int status = gettimeofday(&now, NULL);
5471     assert(status == 0, "gettimeofday");
5472     abstime->tv_sec = now.tv_sec  + seconds;
5473     long usec = now.tv_usec + millis * 1000;
5474     if (usec >= 1000000) {
5475       abstime->tv_sec += 1;
5476       usec -= 1000000;
5477     }
5478     abstime->tv_nsec = usec * 1000;
5479   }
5480   return abstime;
5481 }
5482 
5483 void os::PlatformEvent::park() {       // AKA "down()"
5484   // Transitions for _Event:
5485   //   -1 => -1 : illegal
5486   //    1 =>  0 : pass - return immediately
5487   //    0 => -1 : block; then set _Event to 0 before returning
5488 
5489   // Invariant: Only the thread associated with the Event/PlatformEvent
5490   // may call park().
5491   // TODO: assert that _Assoc != NULL or _Assoc == Self
5492   assert(_nParked == 0, "invariant");
5493 
5494   int v;
5495   for (;;) {
5496     v = _Event;
5497     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5498   }
5499   guarantee(v >= 0, "invariant");
5500   if (v == 0) {
5501     // Do this the hard way by blocking ...
5502     int status = pthread_mutex_lock(_mutex);
5503     assert_status(status == 0, status, "mutex_lock");
5504     guarantee(_nParked == 0, "invariant");
5505     ++_nParked;
5506     while (_Event < 0) {
5507       status = pthread_cond_wait(_cond, _mutex);
5508       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5509       // Treat this the same as if the wait was interrupted
5510       if (status == ETIME) { status = EINTR; }
5511       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5512     }
5513     --_nParked;
5514 
5515     _Event = 0;
5516     status = pthread_mutex_unlock(_mutex);
5517     assert_status(status == 0, status, "mutex_unlock");
5518     // Paranoia to ensure our locked and lock-free paths interact
5519     // correctly with each other.
5520     OrderAccess::fence();
5521   }
5522   guarantee(_Event >= 0, "invariant");
5523 }
5524 
5525 int os::PlatformEvent::park(jlong millis) {
5526   // Transitions for _Event:
5527   //   -1 => -1 : illegal
5528   //    1 =>  0 : pass - return immediately
5529   //    0 => -1 : block; then set _Event to 0 before returning
5530 
5531   guarantee(_nParked == 0, "invariant");
5532 
5533   int v;
5534   for (;;) {
5535     v = _Event;
5536     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5537   }
5538   guarantee(v >= 0, "invariant");
5539   if (v != 0) return OS_OK;
5540 
5541   // We do this the hard way, by blocking the thread.
5542   // Consider enforcing a minimum timeout value.
5543   struct timespec abst;
5544   compute_abstime(&abst, millis);
5545 
5546   int ret = OS_TIMEOUT;
5547   int status = pthread_mutex_lock(_mutex);
5548   assert_status(status == 0, status, "mutex_lock");
5549   guarantee(_nParked == 0, "invariant");
5550   ++_nParked;
5551 
5552   // Object.wait(timo) will return because of
5553   // (a) notification
5554   // (b) timeout
5555   // (c) thread.interrupt
5556   //
5557   // Thread.interrupt and object.notify{All} both call Event::set.
5558   // That is, we treat thread.interrupt as a special case of notification.
5559   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5560   // We assume all ETIME returns are valid.
5561   //
5562   // TODO: properly differentiate simultaneous notify+interrupt.
5563   // In that case, we should propagate the notify to another waiter.
5564 
5565   while (_Event < 0) {
5566     status = pthread_cond_timedwait(_cond, _mutex, &abst);
5567     assert_status(status == 0 || status == EINTR ||
5568                   status == ETIME || status == ETIMEDOUT,
5569                   status, "cond_timedwait");
5570     if (!FilterSpuriousWakeups) break;                 // previous semantics
5571     if (status == ETIME || status == ETIMEDOUT) break;
5572     // We consume and ignore EINTR and spurious wakeups.
5573   }
5574   --_nParked;
5575   if (_Event >= 0) {
5576     ret = OS_OK;
5577   }
5578   _Event = 0;
5579   status = pthread_mutex_unlock(_mutex);
5580   assert_status(status == 0, status, "mutex_unlock");
5581   assert(_nParked == 0, "invariant");
5582   // Paranoia to ensure our locked and lock-free paths interact
5583   // correctly with each other.
5584   OrderAccess::fence();
5585   return ret;
5586 }
5587 
5588 void os::PlatformEvent::unpark() {
5589   // Transitions for _Event:
5590   //    0 => 1 : just return
5591   //    1 => 1 : just return
5592   //   -1 => either 0 or 1; must signal target thread
5593   //         That is, we can safely transition _Event from -1 to either
5594   //         0 or 1.
5595   // See also: "Semaphores in Plan 9" by Mullender & Cox
5596   //
5597   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5598   // that it will take two back-to-back park() calls for the owning
5599   // thread to block. This has the benefit of forcing a spurious return
5600   // from the first park() call after an unpark() call which will help
5601   // shake out uses of park() and unpark() without condition variables.
5602 
5603   if (Atomic::xchg(1, &_Event) >= 0) return;
5604 
5605   // Wait for the thread associated with the event to vacate
5606   int status = pthread_mutex_lock(_mutex);
5607   assert_status(status == 0, status, "mutex_lock");
5608   int AnyWaiters = _nParked;
5609   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5610   status = pthread_mutex_unlock(_mutex);
5611   assert_status(status == 0, status, "mutex_unlock");
5612   if (AnyWaiters != 0) {
5613     // Note that we signal() *after* dropping the lock for "immortal" Events.
5614     // This is safe and avoids a common class of  futile wakeups.  In rare
5615     // circumstances this can cause a thread to return prematurely from
5616     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5617     // will simply re-test the condition and re-park itself.
5618     // This provides particular benefit if the underlying platform does not
5619     // provide wait morphing.
5620     status = pthread_cond_signal(_cond);
5621     assert_status(status == 0, status, "cond_signal");
5622   }
5623 }
5624 
5625 
5626 // JSR166
5627 // -------------------------------------------------------
5628 
5629 // The solaris and linux implementations of park/unpark are fairly
5630 // conservative for now, but can be improved. They currently use a
5631 // mutex/condvar pair, plus a a count.
5632 // Park decrements count if > 0, else does a condvar wait.  Unpark
5633 // sets count to 1 and signals condvar.  Only one thread ever waits
5634 // on the condvar. Contention seen when trying to park implies that someone
5635 // is unparking you, so don't wait. And spurious returns are fine, so there
5636 // is no need to track notifications.
5637 
5638 // This code is common to linux and solaris and will be moved to a
5639 // common place in dolphin.
5640 //
5641 // The passed in time value is either a relative time in nanoseconds
5642 // or an absolute time in milliseconds. Either way it has to be unpacked
5643 // into suitable seconds and nanoseconds components and stored in the
5644 // given timespec structure.
5645 // Given time is a 64-bit value and the time_t used in the timespec is only
5646 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5647 // overflow if times way in the future are given. Further on Solaris versions
5648 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5649 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5650 // As it will be 28 years before "now + 100000000" will overflow we can
5651 // ignore overflow and just impose a hard-limit on seconds using the value
5652 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5653 // years from "now".
5654 
5655 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5656   assert(time > 0, "convertTime");
5657   time_t max_secs = 0;
5658 
5659   if (!os::supports_monotonic_clock() || isAbsolute) {
5660     struct timeval now;
5661     int status = gettimeofday(&now, NULL);
5662     assert(status == 0, "gettimeofday");
5663 
5664     max_secs = now.tv_sec + MAX_SECS;
5665 
5666     if (isAbsolute) {
5667       jlong secs = time / 1000;
5668       if (secs > max_secs) {
5669         absTime->tv_sec = max_secs;
5670       } else {
5671         absTime->tv_sec = secs;
5672       }
5673       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5674     } else {
5675       jlong secs = time / NANOSECS_PER_SEC;
5676       if (secs >= MAX_SECS) {
5677         absTime->tv_sec = max_secs;
5678         absTime->tv_nsec = 0;
5679       } else {
5680         absTime->tv_sec = now.tv_sec + secs;
5681         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5682         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5683           absTime->tv_nsec -= NANOSECS_PER_SEC;
5684           ++absTime->tv_sec; // note: this must be <= max_secs
5685         }
5686       }
5687     }
5688   } else {
5689     // must be relative using monotonic clock
5690     struct timespec now;
5691     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5692     assert_status(status == 0, status, "clock_gettime");
5693     max_secs = now.tv_sec + MAX_SECS;
5694     jlong secs = time / NANOSECS_PER_SEC;
5695     if (secs >= MAX_SECS) {
5696       absTime->tv_sec = max_secs;
5697       absTime->tv_nsec = 0;
5698     } else {
5699       absTime->tv_sec = now.tv_sec + secs;
5700       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5701       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5702         absTime->tv_nsec -= NANOSECS_PER_SEC;
5703         ++absTime->tv_sec; // note: this must be <= max_secs
5704       }
5705     }
5706   }
5707   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5708   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5709   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5710   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5711 }
5712 
5713 void Parker::park(bool isAbsolute, jlong time) {
5714   // Ideally we'd do something useful while spinning, such
5715   // as calling unpackTime().
5716 
5717   // Optional fast-path check:
5718   // Return immediately if a permit is available.
5719   // We depend on Atomic::xchg() having full barrier semantics
5720   // since we are doing a lock-free update to _counter.
5721   if (Atomic::xchg(0, &_counter) > 0) return;
5722 
5723   Thread* thread = Thread::current();
5724   assert(thread->is_Java_thread(), "Must be JavaThread");
5725   JavaThread *jt = (JavaThread *)thread;
5726 
5727   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5728   // Check interrupt before trying to wait
5729   if (Thread::is_interrupted(thread, false)) {
5730     return;
5731   }
5732 
5733   // Next, demultiplex/decode time arguments
5734   timespec absTime;
5735   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5736     return;
5737   }
5738   if (time > 0) {
5739     unpackTime(&absTime, isAbsolute, time);
5740   }
5741 
5742 
5743   // Enter safepoint region
5744   // Beware of deadlocks such as 6317397.
5745   // The per-thread Parker:: mutex is a classic leaf-lock.
5746   // In particular a thread must never block on the Threads_lock while
5747   // holding the Parker:: mutex.  If safepoints are pending both the
5748   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5749   ThreadBlockInVM tbivm(jt);
5750 
5751   // Don't wait if cannot get lock since interference arises from
5752   // unblocking.  Also. check interrupt before trying wait
5753   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5754     return;
5755   }
5756 
5757   int status;
5758   if (_counter > 0)  { // no wait needed
5759     _counter = 0;
5760     status = pthread_mutex_unlock(_mutex);
5761     assert_status(status == 0, status, "invariant");
5762     // Paranoia to ensure our locked and lock-free paths interact
5763     // correctly with each other and Java-level accesses.
5764     OrderAccess::fence();
5765     return;
5766   }
5767 
5768 #ifdef ASSERT
5769   // Don't catch signals while blocked; let the running threads have the signals.
5770   // (This allows a debugger to break into the running thread.)
5771   sigset_t oldsigs;
5772   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5773   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5774 #endif
5775 
5776   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5777   jt->set_suspend_equivalent();
5778   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5779 
5780   assert(_cur_index == -1, "invariant");
5781   if (time == 0) {
5782     _cur_index = REL_INDEX; // arbitrary choice when not timed
5783     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5784   } else {
5785     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5786     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5787   }
5788   _cur_index = -1;
5789   assert_status(status == 0 || status == EINTR ||
5790                 status == ETIME || status == ETIMEDOUT,
5791                 status, "cond_timedwait");
5792 
5793 #ifdef ASSERT
5794   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5795 #endif
5796 
5797   _counter = 0;
5798   status = pthread_mutex_unlock(_mutex);
5799   assert_status(status == 0, status, "invariant");
5800   // Paranoia to ensure our locked and lock-free paths interact
5801   // correctly with each other and Java-level accesses.
5802   OrderAccess::fence();
5803 
5804   // If externally suspended while waiting, re-suspend
5805   if (jt->handle_special_suspend_equivalent_condition()) {
5806     jt->java_suspend_self();
5807   }
5808 }
5809 
5810 void Parker::unpark() {
5811   int status = pthread_mutex_lock(_mutex);
5812   assert_status(status == 0, status, "invariant");
5813   const int s = _counter;
5814   _counter = 1;
5815   // must capture correct index before unlocking
5816   int index = _cur_index;
5817   status = pthread_mutex_unlock(_mutex);
5818   assert_status(status == 0, status, "invariant");
5819   if (s < 1 && index != -1) {
5820     // thread is definitely parked
5821     status = pthread_cond_signal(&_cond[index]);
5822     assert_status(status == 0, status, "invariant");
5823   }
5824 }
5825 
5826 
5827 extern char** environ;
5828 
5829 // Run the specified command in a separate process. Return its exit value,
5830 // or -1 on failure (e.g. can't fork a new process).
5831 // Unlike system(), this function can be called from signal handler. It
5832 // doesn't block SIGINT et al.
5833 int os::fork_and_exec(char* cmd) {
5834   const char * argv[4] = {"sh", "-c", cmd, NULL};
5835 
5836   pid_t pid = fork();
5837 
5838   if (pid < 0) {
5839     // fork failed
5840     return -1;
5841 
5842   } else if (pid == 0) {
5843     // child process
5844 
5845     execve("/bin/sh", (char* const*)argv, environ);
5846 
5847     // execve failed
5848     _exit(-1);
5849 
5850   } else  {
5851     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5852     // care about the actual exit code, for now.
5853 
5854     int status;
5855 
5856     // Wait for the child process to exit.  This returns immediately if
5857     // the child has already exited. */
5858     while (waitpid(pid, &status, 0) < 0) {
5859       switch (errno) {
5860       case ECHILD: return 0;
5861       case EINTR: break;
5862       default: return -1;
5863       }
5864     }
5865 
5866     if (WIFEXITED(status)) {
5867       // The child exited normally; get its exit code.
5868       return WEXITSTATUS(status);
5869     } else if (WIFSIGNALED(status)) {
5870       // The child exited because of a signal
5871       // The best value to return is 0x80 + signal number,
5872       // because that is what all Unix shells do, and because
5873       // it allows callers to distinguish between process exit and
5874       // process death by signal.
5875       return 0x80 + WTERMSIG(status);
5876     } else {
5877       // Unknown exit code; pass it through
5878       return status;
5879     }
5880   }
5881 }
5882 
5883 // is_headless_jre()
5884 //
5885 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5886 // in order to report if we are running in a headless jre
5887 //
5888 // Since JDK8 xawt/libmawt.so was moved into the same directory
5889 // as libawt.so, and renamed libawt_xawt.so
5890 //
5891 bool os::is_headless_jre() {
5892   struct stat statbuf;
5893   char buf[MAXPATHLEN];
5894   char libmawtpath[MAXPATHLEN];
5895   const char *xawtstr  = "/xawt/libmawt.so";
5896   const char *new_xawtstr = "/libawt_xawt.so";
5897   char *p;
5898 
5899   // Get path to libjvm.so
5900   os::jvm_path(buf, sizeof(buf));
5901 
5902   // Get rid of libjvm.so
5903   p = strrchr(buf, '/');
5904   if (p == NULL) {
5905     return false;
5906   } else {
5907     *p = '\0';
5908   }
5909 
5910   // Get rid of client or server
5911   p = strrchr(buf, '/');
5912   if (p == NULL) {
5913     return false;
5914   } else {
5915     *p = '\0';
5916   }
5917 
5918   // check xawt/libmawt.so
5919   strcpy(libmawtpath, buf);
5920   strcat(libmawtpath, xawtstr);
5921   if (::stat(libmawtpath, &statbuf) == 0) return false;
5922 
5923   // check libawt_xawt.so
5924   strcpy(libmawtpath, buf);
5925   strcat(libmawtpath, new_xawtstr);
5926   if (::stat(libmawtpath, &statbuf) == 0) return false;
5927 
5928   return true;
5929 }
5930 
5931 // Get the default path to the core file
5932 // Returns the length of the string
5933 int os::get_core_path(char* buffer, size_t bufferSize) {
5934   /*
5935    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5936    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5937    */
5938   const int core_pattern_len = 129;
5939   char core_pattern[core_pattern_len] = {0};
5940 
5941   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5942   if (core_pattern_file == -1) {
5943     return -1;
5944   }
5945 
5946   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5947   ::close(core_pattern_file);
5948   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5949     return -1;
5950   }
5951   if (core_pattern[ret-1] == '\n') {
5952     core_pattern[ret-1] = '\0';
5953   } else {
5954     core_pattern[ret] = '\0';
5955   }
5956 
5957   char *pid_pos = strstr(core_pattern, "%p");
5958   int written;
5959 
5960   if (core_pattern[0] == '/') {
5961     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5962   } else {
5963     char cwd[PATH_MAX];
5964 
5965     const char* p = get_current_directory(cwd, PATH_MAX);
5966     if (p == NULL) {
5967       return -1;
5968     }
5969 
5970     if (core_pattern[0] == '|') {
5971       written = jio_snprintf(buffer, bufferSize,
5972                         "\"%s\" (or dumping to %s/core.%d)",
5973                                      &core_pattern[1], p, current_process_id());
5974     } else {
5975       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5976     }
5977   }
5978 
5979   if (written < 0) {
5980     return -1;
5981   }
5982 
5983   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5984     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5985 
5986     if (core_uses_pid_file != -1) {
5987       char core_uses_pid = 0;
5988       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5989       ::close(core_uses_pid_file);
5990 
5991       if (core_uses_pid == '1') {
5992         jio_snprintf(buffer + written, bufferSize - written,
5993                                           ".%d", current_process_id());
5994       }
5995     }
5996   }
5997 
5998   return strlen(buffer);
5999 }
6000 
6001 bool os::start_debugging(char *buf, int buflen) {
6002   int len = (int)strlen(buf);
6003   char *p = &buf[len];
6004 
6005   jio_snprintf(p, buflen-len,
6006              "\n\n"
6007              "Do you want to debug the problem?\n\n"
6008              "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6009              "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6010              "Otherwise, press RETURN to abort...",
6011              os::current_process_id(), os::current_process_id(),
6012              os::current_thread_id(), os::current_thread_id());
6013 
6014   bool yes = os::message_box("Unexpected Error", buf);
6015 
6016   if (yes) {
6017     // yes, user asked VM to launch debugger
6018     jio_snprintf(buf, sizeof(buf), "gdb /proc/%d/exe %d",
6019                      os::current_process_id(), os::current_process_id());
6020 
6021     os::fork_and_exec(buf);
6022     yes = false;
6023   }
6024   return yes;
6025 }
6026 
6027 
6028 
6029 /////////////// Unit tests ///////////////
6030 
6031 #ifndef PRODUCT
6032 
6033 #define test_log(...)              \
6034   do {                             \
6035     if (VerboseInternalVMTests) {  \
6036       tty->print_cr(__VA_ARGS__);  \
6037       tty->flush();                \
6038     }                              \
6039   } while (false)
6040 
6041 class TestReserveMemorySpecial : AllStatic {
6042  public:
6043   static void small_page_write(void* addr, size_t size) {
6044     size_t page_size = os::vm_page_size();
6045 
6046     char* end = (char*)addr + size;
6047     for (char* p = (char*)addr; p < end; p += page_size) {
6048       *p = 1;
6049     }
6050   }
6051 
6052   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6053     if (!UseHugeTLBFS) {
6054       return;
6055     }
6056 
6057     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6058 
6059     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6060 
6061     if (addr != NULL) {
6062       small_page_write(addr, size);
6063 
6064       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6065     }
6066   }
6067 
6068   static void test_reserve_memory_special_huge_tlbfs_only() {
6069     if (!UseHugeTLBFS) {
6070       return;
6071     }
6072 
6073     size_t lp = os::large_page_size();
6074 
6075     for (size_t size = lp; size <= lp * 10; size += lp) {
6076       test_reserve_memory_special_huge_tlbfs_only(size);
6077     }
6078   }
6079 
6080   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6081     size_t lp = os::large_page_size();
6082     size_t ag = os::vm_allocation_granularity();
6083 
6084     // sizes to test
6085     const size_t sizes[] = {
6086       lp, lp + ag, lp + lp / 2, lp * 2,
6087       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6088       lp * 10, lp * 10 + lp / 2
6089     };
6090     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6091 
6092     // For each size/alignment combination, we test three scenarios:
6093     // 1) with req_addr == NULL
6094     // 2) with a non-null req_addr at which we expect to successfully allocate
6095     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6096     //    expect the allocation to either fail or to ignore req_addr
6097 
6098     // Pre-allocate two areas; they shall be as large as the largest allocation
6099     //  and aligned to the largest alignment we will be testing.
6100     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6101     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6102       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6103       -1, 0);
6104     assert(mapping1 != MAP_FAILED, "should work");
6105 
6106     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6107       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6108       -1, 0);
6109     assert(mapping2 != MAP_FAILED, "should work");
6110 
6111     // Unmap the first mapping, but leave the second mapping intact: the first
6112     // mapping will serve as a value for a "good" req_addr (case 2). The second
6113     // mapping, still intact, as "bad" req_addr (case 3).
6114     ::munmap(mapping1, mapping_size);
6115 
6116     // Case 1
6117     test_log("%s, req_addr NULL:", __FUNCTION__);
6118     test_log("size            align           result");
6119 
6120     for (int i = 0; i < num_sizes; i++) {
6121       const size_t size = sizes[i];
6122       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6123         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6124         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6125                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6126         if (p != NULL) {
6127           assert(is_ptr_aligned(p, alignment), "must be");
6128           small_page_write(p, size);
6129           os::Linux::release_memory_special_huge_tlbfs(p, size);
6130         }
6131       }
6132     }
6133 
6134     // Case 2
6135     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6136     test_log("size            align           req_addr         result");
6137 
6138     for (int i = 0; i < num_sizes; i++) {
6139       const size_t size = sizes[i];
6140       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6141         char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6142         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6143         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6144                  size, alignment, p2i(req_addr), p2i(p),
6145                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6146         if (p != NULL) {
6147           assert(p == req_addr, "must be");
6148           small_page_write(p, size);
6149           os::Linux::release_memory_special_huge_tlbfs(p, size);
6150         }
6151       }
6152     }
6153 
6154     // Case 3
6155     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6156     test_log("size            align           req_addr         result");
6157 
6158     for (int i = 0; i < num_sizes; i++) {
6159       const size_t size = sizes[i];
6160       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6161         char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6162         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6163         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6164                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6165         // as the area around req_addr contains already existing mappings, the API should always
6166         // return NULL (as per contract, it cannot return another address)
6167         assert(p == NULL, "must be");
6168       }
6169     }
6170 
6171     ::munmap(mapping2, mapping_size);
6172 
6173   }
6174 
6175   static void test_reserve_memory_special_huge_tlbfs() {
6176     if (!UseHugeTLBFS) {
6177       return;
6178     }
6179 
6180     test_reserve_memory_special_huge_tlbfs_only();
6181     test_reserve_memory_special_huge_tlbfs_mixed();
6182   }
6183 
6184   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6185     if (!UseSHM) {
6186       return;
6187     }
6188 
6189     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6190 
6191     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6192 
6193     if (addr != NULL) {
6194       assert(is_ptr_aligned(addr, alignment), "Check");
6195       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6196 
6197       small_page_write(addr, size);
6198 
6199       os::Linux::release_memory_special_shm(addr, size);
6200     }
6201   }
6202 
6203   static void test_reserve_memory_special_shm() {
6204     size_t lp = os::large_page_size();
6205     size_t ag = os::vm_allocation_granularity();
6206 
6207     for (size_t size = ag; size < lp * 3; size += ag) {
6208       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6209         test_reserve_memory_special_shm(size, alignment);
6210       }
6211     }
6212   }
6213 
6214   static void test() {
6215     test_reserve_memory_special_huge_tlbfs();
6216     test_reserve_memory_special_shm();
6217   }
6218 };
6219 
6220 void TestReserveMemorySpecial_test() {
6221   TestReserveMemorySpecial::test();
6222 }
6223 
6224 #endif