1 /*
   2  * Copyright (c) 1999, 2017, 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 "oops/oop.inline.hpp"
  39 #include "os_linux.inline.hpp"
  40 #include "os_share_linux.hpp"
  41 #include "prims/jniFastGetField.hpp"
  42 #include "prims/jvm.h"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.hpp"
  49 #include "runtime/init.hpp"
  50 #include "runtime/java.hpp"
  51 #include "runtime/javaCalls.hpp"
  52 #include "runtime/mutexLocker.hpp"
  53 #include "runtime/objectMonitor.hpp"
  54 #include "runtime/orderAccess.inline.hpp"
  55 #include "runtime/osThread.hpp"
  56 #include "runtime/perfMemory.hpp"
  57 #include "runtime/sharedRuntime.hpp"
  58 #include "runtime/statSampler.hpp"
  59 #include "runtime/stubRoutines.hpp"
  60 #include "runtime/thread.inline.hpp"
  61 #include "runtime/threadCritical.hpp"
  62 #include "runtime/timer.hpp"
  63 #include "semaphore_posix.hpp"
  64 #include "services/attachListener.hpp"
  65 #include "services/memTracker.hpp"
  66 #include "services/runtimeService.hpp"
  67 #include "utilities/decoder.hpp"
  68 #include "utilities/defaultStream.hpp"
  69 #include "utilities/events.hpp"
  70 #include "utilities/elfFile.hpp"
  71 #include "utilities/growableArray.hpp"
  72 #include "utilities/macros.hpp"
  73 #include "utilities/vmError.hpp"
  74 
  75 // put OS-includes here
  76 # include <sys/types.h>
  77 # include <sys/mman.h>
  78 # include <sys/stat.h>
  79 # include <sys/select.h>
  80 # include <pthread.h>
  81 # include <signal.h>
  82 # include <errno.h>
  83 # include <dlfcn.h>
  84 # include <stdio.h>
  85 # include <unistd.h>
  86 # include <sys/resource.h>
  87 # include <pthread.h>
  88 # include <sys/stat.h>
  89 # include <sys/time.h>
  90 # include <sys/times.h>
  91 # include <sys/utsname.h>
  92 # include <sys/socket.h>
  93 # include <sys/wait.h>
  94 # include <pwd.h>
  95 # include <poll.h>
  96 # include <semaphore.h>
  97 # include <fcntl.h>
  98 # include <string.h>
  99 # include <syscall.h>
 100 # include <sys/sysinfo.h>
 101 # include <gnu/libc-version.h>
 102 # include <sys/ipc.h>
 103 # include <sys/shm.h>
 104 # include <link.h>
 105 # include <stdint.h>
 106 # include <inttypes.h>
 107 # include <sys/ioctl.h>
 108 
 109 #ifndef _GNU_SOURCE
 110   #define _GNU_SOURCE
 111   #include <sched.h>
 112   #undef _GNU_SOURCE
 113 #else
 114   #include <sched.h>
 115 #endif
 116 
 117 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 118 // getrusage() is prepared to handle the associated failure.
 119 #ifndef RUSAGE_THREAD
 120   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 121 #endif
 122 
 123 #define MAX_PATH    (2 * K)
 124 
 125 #define MAX_SECS 100000000
 126 
 127 // for timer info max values which include all bits
 128 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 129 
 130 #define LARGEPAGES_BIT (1 << 6)
 131 ////////////////////////////////////////////////////////////////////////////////
 132 // global variables
 133 julong os::Linux::_physical_memory = 0;
 134 
 135 address   os::Linux::_initial_thread_stack_bottom = NULL;
 136 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 137 
 138 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 139 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 140 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 141 Mutex* os::Linux::_createThread_lock = NULL;
 142 pthread_t os::Linux::_main_thread;
 143 int os::Linux::_page_size = -1;
 144 bool os::Linux::_supports_fast_thread_cpu_time = false;
 145 uint32_t os::Linux::_os_version = 0;
 146 const char * os::Linux::_glibc_version = NULL;
 147 const char * os::Linux::_libpthread_version = NULL;
 148 
 149 static jlong initial_time_count=0;
 150 
 151 static int clock_tics_per_sec = 100;
 152 
 153 // For diagnostics to print a message once. see run_periodic_checks
 154 static sigset_t check_signal_done;
 155 static bool check_signals = true;
 156 
 157 // Signal number used to suspend/resume a thread
 158 
 159 // do not use any signal number less than SIGSEGV, see 4355769
 160 static int SR_signum = SIGUSR2;
 161 sigset_t SR_sigset;
 162 
 163 // utility functions
 164 
 165 static int SR_initialize();
 166 
 167 julong os::available_memory() {
 168   return Linux::available_memory();
 169 }
 170 
 171 julong os::Linux::available_memory() {
 172   // values in struct sysinfo are "unsigned long"
 173   struct sysinfo si;
 174   sysinfo(&si);
 175 
 176   return (julong)si.freeram * si.mem_unit;
 177 }
 178 
 179 julong os::physical_memory() {
 180   return Linux::physical_memory();
 181 }
 182 
 183 // Return true if user is running as root.
 184 
 185 bool os::have_special_privileges() {
 186   static bool init = false;
 187   static bool privileges = false;
 188   if (!init) {
 189     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 190     init = true;
 191   }
 192   return privileges;
 193 }
 194 
 195 
 196 #ifndef SYS_gettid
 197 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 198   #ifdef __ia64__
 199     #define SYS_gettid 1105
 200   #else
 201     #ifdef __i386__
 202       #define SYS_gettid 224
 203     #else
 204       #ifdef __amd64__
 205         #define SYS_gettid 186
 206       #else
 207         #ifdef __sparc__
 208           #define SYS_gettid 143
 209         #else
 210           #error define gettid for the arch
 211         #endif
 212       #endif
 213     #endif
 214   #endif
 215 #endif
 216 
 217 
 218 // pid_t gettid()
 219 //
 220 // Returns the kernel thread id of the currently running thread. Kernel
 221 // thread id is used to access /proc.
 222 pid_t os::Linux::gettid() {
 223   int rslt = syscall(SYS_gettid);
 224   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 225   return (pid_t)rslt;
 226 }
 227 
 228 // Most versions of linux have a bug where the number of processors are
 229 // determined by looking at the /proc file system.  In a chroot environment,
 230 // the system call returns 1.  This causes the VM to act as if it is
 231 // a single processor and elide locking (see is_MP() call).
 232 static bool unsafe_chroot_detected = false;
 233 static const char *unstable_chroot_error = "/proc file system not found.\n"
 234                      "Java may be unstable running multithreaded in a chroot "
 235                      "environment on Linux when /proc filesystem is not mounted.";
 236 
 237 void os::Linux::initialize_system_info() {
 238   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 239   if (processor_count() == 1) {
 240     pid_t pid = os::Linux::gettid();
 241     char fname[32];
 242     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 243     FILE *fp = fopen(fname, "r");
 244     if (fp == NULL) {
 245       unsafe_chroot_detected = true;
 246     } else {
 247       fclose(fp);
 248     }
 249   }
 250   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 251   assert(processor_count() > 0, "linux error");
 252 }
 253 
 254 void os::init_system_properties_values() {
 255   // The next steps are taken in the product version:
 256   //
 257   // Obtain the JAVA_HOME value from the location of libjvm.so.
 258   // This library should be located at:
 259   // <JAVA_HOME>/lib/{client|server}/libjvm.so.
 260   //
 261   // If "/jre/lib/" appears at the right place in the path, then we
 262   // assume libjvm.so is installed in a JDK and we use this path.
 263   //
 264   // Otherwise exit with message: "Could not create the Java virtual machine."
 265   //
 266   // The following extra steps are taken in the debugging version:
 267   //
 268   // If "/jre/lib/" does NOT appear at the right place in the path
 269   // instead of exit check for $JAVA_HOME environment variable.
 270   //
 271   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 272   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 273   // it looks like libjvm.so is installed there
 274   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 275   //
 276   // Otherwise exit.
 277   //
 278   // Important note: if the location of libjvm.so changes this
 279   // code needs to be changed accordingly.
 280 
 281   // See ld(1):
 282   //      The linker uses the following search paths to locate required
 283   //      shared libraries:
 284   //        1: ...
 285   //        ...
 286   //        7: The default directories, normally /lib and /usr/lib.
 287 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
 288   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 289 #else
 290   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 291 #endif
 292 
 293 // Base path of extensions installed on the system.
 294 #define SYS_EXT_DIR     "/usr/java/packages"
 295 #define EXTENSIONS_DIR  "/lib/ext"
 296 
 297   // Buffer that fits several sprintfs.
 298   // Note that the space for the colon and the trailing null are provided
 299   // by the nulls included by the sizeof operator.
 300   const size_t bufsize =
 301     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 302          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 303   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 304 
 305   // sysclasspath, java_home, dll_dir
 306   {
 307     char *pslash;
 308     os::jvm_path(buf, bufsize);
 309 
 310     // Found the full path to libjvm.so.
 311     // Now cut the path to <java_home>/jre if we can.
 312     pslash = strrchr(buf, '/');
 313     if (pslash != NULL) {
 314       *pslash = '\0';            // Get rid of /libjvm.so.
 315     }
 316     pslash = strrchr(buf, '/');
 317     if (pslash != NULL) {
 318       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 319     }
 320     Arguments::set_dll_dir(buf);
 321 
 322     if (pslash != NULL) {
 323       pslash = strrchr(buf, '/');
 324       if (pslash != NULL) {
 325         *pslash = '\0';        // Get rid of /lib.
 326       }
 327     }
 328     Arguments::set_java_home(buf);
 329     set_boot_path('/', ':');
 330   }
 331 
 332   // Where to look for native libraries.
 333   //
 334   // Note: Due to a legacy implementation, most of the library path
 335   // is set in the launcher. This was to accomodate linking restrictions
 336   // on legacy Linux implementations (which are no longer supported).
 337   // Eventually, all the library path setting will be done here.
 338   //
 339   // However, to prevent the proliferation of improperly built native
 340   // libraries, the new path component /usr/java/packages is added here.
 341   // Eventually, all the library path setting will be done here.
 342   {
 343     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 344     // should always exist (until the legacy problem cited above is
 345     // addressed).
 346     const char *v = ::getenv("LD_LIBRARY_PATH");
 347     const char *v_colon = ":";
 348     if (v == NULL) { v = ""; v_colon = ""; }
 349     // That's +1 for the colon and +1 for the trailing '\0'.
 350     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 351                                                      strlen(v) + 1 +
 352                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
 353                                                      mtInternal);
 354     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
 355     Arguments::set_library_path(ld_library_path);
 356     FREE_C_HEAP_ARRAY(char, ld_library_path);
 357   }
 358 
 359   // Extensions directories.
 360   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 361   Arguments::set_ext_dirs(buf);
 362 
 363   FREE_C_HEAP_ARRAY(char, buf);
 364 
 365 #undef DEFAULT_LIBPATH
 366 #undef SYS_EXT_DIR
 367 #undef EXTENSIONS_DIR
 368 }
 369 
 370 ////////////////////////////////////////////////////////////////////////////////
 371 // breakpoint support
 372 
 373 void os::breakpoint() {
 374   BREAKPOINT;
 375 }
 376 
 377 extern "C" void breakpoint() {
 378   // use debugger to set breakpoint here
 379 }
 380 
 381 ////////////////////////////////////////////////////////////////////////////////
 382 // signal support
 383 
 384 debug_only(static bool signal_sets_initialized = false);
 385 static sigset_t unblocked_sigs, vm_sigs;
 386 
 387 bool os::Linux::is_sig_ignored(int sig) {
 388   struct sigaction oact;
 389   sigaction(sig, (struct sigaction*)NULL, &oact);
 390   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 391                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 392   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 393     return true;
 394   } else {
 395     return false;
 396   }
 397 }
 398 
 399 void os::Linux::signal_sets_init() {
 400   // Should also have an assertion stating we are still single-threaded.
 401   assert(!signal_sets_initialized, "Already initialized");
 402   // Fill in signals that are necessarily unblocked for all threads in
 403   // the VM. Currently, we unblock the following signals:
 404   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 405   //                         by -Xrs (=ReduceSignalUsage));
 406   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 407   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 408   // the dispositions or masks wrt these signals.
 409   // Programs embedding the VM that want to use the above signals for their
 410   // own purposes must, at this time, use the "-Xrs" option to prevent
 411   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 412   // (See bug 4345157, and other related bugs).
 413   // In reality, though, unblocking these signals is really a nop, since
 414   // these signals are not blocked by default.
 415   sigemptyset(&unblocked_sigs);
 416   sigaddset(&unblocked_sigs, SIGILL);
 417   sigaddset(&unblocked_sigs, SIGSEGV);
 418   sigaddset(&unblocked_sigs, SIGBUS);
 419   sigaddset(&unblocked_sigs, SIGFPE);
 420 #if defined(PPC64)
 421   sigaddset(&unblocked_sigs, SIGTRAP);
 422 #endif
 423   sigaddset(&unblocked_sigs, SR_signum);
 424 
 425   if (!ReduceSignalUsage) {
 426     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 427       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 428     }
 429     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 430       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 431     }
 432     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 433       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 434     }
 435   }
 436   // Fill in signals that are blocked by all but the VM thread.
 437   sigemptyset(&vm_sigs);
 438   if (!ReduceSignalUsage) {
 439     sigaddset(&vm_sigs, BREAK_SIGNAL);
 440   }
 441   debug_only(signal_sets_initialized = true);
 442 
 443 }
 444 
 445 // These are signals that are unblocked while a thread is running Java.
 446 // (For some reason, they get blocked by default.)
 447 sigset_t* os::Linux::unblocked_signals() {
 448   assert(signal_sets_initialized, "Not initialized");
 449   return &unblocked_sigs;
 450 }
 451 
 452 // These are the signals that are blocked while a (non-VM) thread is
 453 // running Java. Only the VM thread handles these signals.
 454 sigset_t* os::Linux::vm_signals() {
 455   assert(signal_sets_initialized, "Not initialized");
 456   return &vm_sigs;
 457 }
 458 
 459 void os::Linux::hotspot_sigmask(Thread* thread) {
 460 
 461   //Save caller's signal mask before setting VM signal mask
 462   sigset_t caller_sigmask;
 463   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 464 
 465   OSThread* osthread = thread->osthread();
 466   osthread->set_caller_sigmask(caller_sigmask);
 467 
 468   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 469 
 470   if (!ReduceSignalUsage) {
 471     if (thread->is_VM_thread()) {
 472       // Only the VM thread handles BREAK_SIGNAL ...
 473       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 474     } else {
 475       // ... all other threads block BREAK_SIGNAL
 476       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 477     }
 478   }
 479 }
 480 
 481 //////////////////////////////////////////////////////////////////////////////
 482 // detecting pthread library
 483 
 484 void os::Linux::libpthread_init() {
 485   // Save glibc and pthread version strings.
 486 #if !defined(_CS_GNU_LIBC_VERSION) || \
 487     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 488   #error "glibc too old (< 2.3.2)"
 489 #endif
 490 
 491   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 492   assert(n > 0, "cannot retrieve glibc version");
 493   char *str = (char *)malloc(n, mtInternal);
 494   confstr(_CS_GNU_LIBC_VERSION, str, n);
 495   os::Linux::set_glibc_version(str);
 496 
 497   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 498   assert(n > 0, "cannot retrieve pthread version");
 499   str = (char *)malloc(n, mtInternal);
 500   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 501   os::Linux::set_libpthread_version(str);
 502 }
 503 
 504 /////////////////////////////////////////////////////////////////////////////
 505 // thread stack expansion
 506 
 507 // os::Linux::manually_expand_stack() takes care of expanding the thread
 508 // stack. Note that this is normally not needed: pthread stacks allocate
 509 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 510 // committed. Therefore it is not necessary to expand the stack manually.
 511 //
 512 // Manually expanding the stack was historically needed on LinuxThreads
 513 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 514 // it is kept to deal with very rare corner cases:
 515 //
 516 // For one, user may run the VM on an own implementation of threads
 517 // whose stacks are - like the old LinuxThreads - implemented using
 518 // mmap(MAP_GROWSDOWN).
 519 //
 520 // Also, this coding may be needed if the VM is running on the primordial
 521 // thread. Normally we avoid running on the primordial thread; however,
 522 // user may still invoke the VM on the primordial thread.
 523 //
 524 // The following historical comment describes the details about running
 525 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 526 
 527 
 528 // Force Linux kernel to expand current thread stack. If "bottom" is close
 529 // to the stack guard, caller should block all signals.
 530 //
 531 // MAP_GROWSDOWN:
 532 //   A special mmap() flag that is used to implement thread stacks. It tells
 533 //   kernel that the memory region should extend downwards when needed. This
 534 //   allows early versions of LinuxThreads to only mmap the first few pages
 535 //   when creating a new thread. Linux kernel will automatically expand thread
 536 //   stack as needed (on page faults).
 537 //
 538 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 539 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 540 //   region, it's hard to tell if the fault is due to a legitimate stack
 541 //   access or because of reading/writing non-exist memory (e.g. buffer
 542 //   overrun). As a rule, if the fault happens below current stack pointer,
 543 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 544 //   application (see Linux kernel fault.c).
 545 //
 546 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 547 //   stack overflow detection.
 548 //
 549 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 550 //   not use MAP_GROWSDOWN.
 551 //
 552 // To get around the problem and allow stack banging on Linux, we need to
 553 // manually expand thread stack after receiving the SIGSEGV.
 554 //
 555 // There are two ways to expand thread stack to address "bottom", we used
 556 // both of them in JVM before 1.5:
 557 //   1. adjust stack pointer first so that it is below "bottom", and then
 558 //      touch "bottom"
 559 //   2. mmap() the page in question
 560 //
 561 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 562 // if current sp is already near the lower end of page 101, and we need to
 563 // call mmap() to map page 100, it is possible that part of the mmap() frame
 564 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 565 // That will destroy the mmap() frame and cause VM to crash.
 566 //
 567 // The following code works by adjusting sp first, then accessing the "bottom"
 568 // page to force a page fault. Linux kernel will then automatically expand the
 569 // stack mapping.
 570 //
 571 // _expand_stack_to() assumes its frame size is less than page size, which
 572 // should always be true if the function is not inlined.
 573 
 574 static void NOINLINE _expand_stack_to(address bottom) {
 575   address sp;
 576   size_t size;
 577   volatile char *p;
 578 
 579   // Adjust bottom to point to the largest address within the same page, it
 580   // gives us a one-page buffer if alloca() allocates slightly more memory.
 581   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 582   bottom += os::Linux::page_size() - 1;
 583 
 584   // sp might be slightly above current stack pointer; if that's the case, we
 585   // will alloca() a little more space than necessary, which is OK. Don't use
 586   // os::current_stack_pointer(), as its result can be slightly below current
 587   // stack pointer, causing us to not alloca enough to reach "bottom".
 588   sp = (address)&sp;
 589 
 590   if (sp > bottom) {
 591     size = sp - bottom;
 592     p = (volatile char *)alloca(size);
 593     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 594     p[0] = '\0';
 595   }
 596 }
 597 
 598 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 599   assert(t!=NULL, "just checking");
 600   assert(t->osthread()->expanding_stack(), "expand should be set");
 601   assert(t->stack_base() != NULL, "stack_base was not initialized");
 602 
 603   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 604     sigset_t mask_all, old_sigset;
 605     sigfillset(&mask_all);
 606     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 607     _expand_stack_to(addr);
 608     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 609     return true;
 610   }
 611   return false;
 612 }
 613 
 614 //////////////////////////////////////////////////////////////////////////////
 615 // create new thread
 616 
 617 // Thread start routine for all newly created threads
 618 static void *thread_native_entry(Thread *thread) {
 619   // Try to randomize the cache line index of hot stack frames.
 620   // This helps when threads of the same stack traces evict each other's
 621   // cache lines. The threads can be either from the same JVM instance, or
 622   // from different JVM instances. The benefit is especially true for
 623   // processors with hyperthreading technology.
 624   static int counter = 0;
 625   int pid = os::current_process_id();
 626   alloca(((pid ^ counter++) & 7) * 128);
 627 
 628   thread->initialize_thread_current();
 629 
 630   OSThread* osthread = thread->osthread();
 631   Monitor* sync = osthread->startThread_lock();
 632 
 633   osthread->set_thread_id(os::current_thread_id());
 634 
 635   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 636     os::current_thread_id(), (uintx) pthread_self());
 637 
 638   if (UseNUMA) {
 639     int lgrp_id = os::numa_get_group_id();
 640     if (lgrp_id != -1) {
 641       thread->set_lgrp_id(lgrp_id);
 642     }
 643   }
 644   // initialize signal mask for this thread
 645   os::Linux::hotspot_sigmask(thread);
 646 
 647   // initialize floating point control register
 648   os::Linux::init_thread_fpu_state();
 649 
 650   // handshaking with parent thread
 651   {
 652     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 653 
 654     // notify parent thread
 655     osthread->set_state(INITIALIZED);
 656     sync->notify_all();
 657 
 658     // wait until os::start_thread()
 659     while (osthread->get_state() == INITIALIZED) {
 660       sync->wait(Mutex::_no_safepoint_check_flag);
 661     }
 662   }
 663 
 664   // call one more level start routine
 665   thread->run();
 666 
 667   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 668     os::current_thread_id(), (uintx) pthread_self());
 669 
 670   // If a thread has not deleted itself ("delete this") as part of its
 671   // termination sequence, we have to ensure thread-local-storage is
 672   // cleared before we actually terminate. No threads should ever be
 673   // deleted asynchronously with respect to their termination.
 674   if (Thread::current_or_null_safe() != NULL) {
 675     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 676     thread->clear_thread_current();
 677   }
 678 
 679   return 0;
 680 }
 681 
 682 bool os::create_thread(Thread* thread, ThreadType thr_type,
 683                        size_t req_stack_size) {
 684   assert(thread->osthread() == NULL, "caller responsible");
 685 
 686   // Allocate the OSThread object
 687   OSThread* osthread = new OSThread(NULL, NULL);
 688   if (osthread == NULL) {
 689     return false;
 690   }
 691 
 692   // set the correct thread state
 693   osthread->set_thread_type(thr_type);
 694 
 695   // Initial state is ALLOCATED but not INITIALIZED
 696   osthread->set_state(ALLOCATED);
 697 
 698   thread->set_osthread(osthread);
 699 
 700   // init thread attributes
 701   pthread_attr_t attr;
 702   pthread_attr_init(&attr);
 703   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 704 
 705   // Calculate stack size if it's not specified by caller.
 706   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 707   // In the Linux NPTL pthread implementation the guard size mechanism
 708   // is not implemented properly. The posix standard requires adding
 709   // the size of the guard pages to the stack size, instead Linux
 710   // takes the space out of 'stacksize'. Thus we adapt the requested
 711   // stack_size by the size of the guard pages to mimick proper
 712   // behaviour. However, be careful not to end up with a size
 713   // of zero due to overflow. Don't add the guard page in that case.
 714   size_t guard_size = os::Linux::default_guard_size(thr_type);
 715   if (stack_size <= SIZE_MAX - guard_size) {
 716     stack_size += guard_size;
 717   }
 718   assert(is_size_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
 719 
 720   int status = pthread_attr_setstacksize(&attr, stack_size);
 721   assert_status(status == 0, status, "pthread_attr_setstacksize");
 722 
 723   // Configure glibc guard page.
 724   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 725 
 726   ThreadState state;
 727 
 728   {
 729     pthread_t tid;
 730     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 731 
 732     char buf[64];
 733     if (ret == 0) {
 734       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 735         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 736     } else {
 737       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 738         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 739     }
 740 
 741     pthread_attr_destroy(&attr);
 742 
 743     if (ret != 0) {
 744       // Need to clean up stuff we've allocated so far
 745       thread->set_osthread(NULL);
 746       delete osthread;
 747       return false;
 748     }
 749 
 750     // Store pthread info into the OSThread
 751     osthread->set_pthread_id(tid);
 752 
 753     // Wait until child thread is either initialized or aborted
 754     {
 755       Monitor* sync_with_child = osthread->startThread_lock();
 756       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 757       while ((state = osthread->get_state()) == ALLOCATED) {
 758         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 759       }
 760     }
 761   }
 762 
 763   // Aborted due to thread limit being reached
 764   if (state == ZOMBIE) {
 765     thread->set_osthread(NULL);
 766     delete osthread;
 767     return false;
 768   }
 769 
 770   // The thread is returned suspended (in state INITIALIZED),
 771   // and is started higher up in the call chain
 772   assert(state == INITIALIZED, "race condition");
 773   return true;
 774 }
 775 
 776 /////////////////////////////////////////////////////////////////////////////
 777 // attach existing thread
 778 
 779 // bootstrap the main thread
 780 bool os::create_main_thread(JavaThread* thread) {
 781   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 782   return create_attached_thread(thread);
 783 }
 784 
 785 bool os::create_attached_thread(JavaThread* thread) {
 786 #ifdef ASSERT
 787   thread->verify_not_published();
 788 #endif
 789 
 790   // Allocate the OSThread object
 791   OSThread* osthread = new OSThread(NULL, NULL);
 792 
 793   if (osthread == NULL) {
 794     return false;
 795   }
 796 
 797   // Store pthread info into the OSThread
 798   osthread->set_thread_id(os::Linux::gettid());
 799   osthread->set_pthread_id(::pthread_self());
 800 
 801   // initialize floating point control register
 802   os::Linux::init_thread_fpu_state();
 803 
 804   // Initial thread state is RUNNABLE
 805   osthread->set_state(RUNNABLE);
 806 
 807   thread->set_osthread(osthread);
 808 
 809   if (UseNUMA) {
 810     int lgrp_id = os::numa_get_group_id();
 811     if (lgrp_id != -1) {
 812       thread->set_lgrp_id(lgrp_id);
 813     }
 814   }
 815 
 816   if (os::Linux::is_initial_thread()) {
 817     // If current thread is initial thread, its stack is mapped on demand,
 818     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 819     // the entire stack region to avoid SEGV in stack banging.
 820     // It is also useful to get around the heap-stack-gap problem on SuSE
 821     // kernel (see 4821821 for details). We first expand stack to the top
 822     // of yellow zone, then enable stack yellow zone (order is significant,
 823     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 824     // is no gap between the last two virtual memory regions.
 825 
 826     JavaThread *jt = (JavaThread *)thread;
 827     address addr = jt->stack_reserved_zone_base();
 828     assert(addr != NULL, "initialization problem?");
 829     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 830 
 831     osthread->set_expanding_stack();
 832     os::Linux::manually_expand_stack(jt, addr);
 833     osthread->clear_expanding_stack();
 834   }
 835 
 836   // initialize signal mask for this thread
 837   // and save the caller's signal mask
 838   os::Linux::hotspot_sigmask(thread);
 839 
 840   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 841     os::current_thread_id(), (uintx) pthread_self());
 842 
 843   return true;
 844 }
 845 
 846 void os::pd_start_thread(Thread* thread) {
 847   OSThread * osthread = thread->osthread();
 848   assert(osthread->get_state() != INITIALIZED, "just checking");
 849   Monitor* sync_with_child = osthread->startThread_lock();
 850   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 851   sync_with_child->notify();
 852 }
 853 
 854 // Free Linux resources related to the OSThread
 855 void os::free_thread(OSThread* osthread) {
 856   assert(osthread != NULL, "osthread not set");
 857 
 858   // We are told to free resources of the argument thread,
 859   // but we can only really operate on the current thread.
 860   assert(Thread::current()->osthread() == osthread,
 861          "os::free_thread but not current thread");
 862 
 863 #ifdef ASSERT
 864   sigset_t current;
 865   sigemptyset(&current);
 866   pthread_sigmask(SIG_SETMASK, NULL, &current);
 867   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 868 #endif
 869 
 870   // Restore caller's signal mask
 871   sigset_t sigmask = osthread->caller_sigmask();
 872   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 873 
 874   delete osthread;
 875 }
 876 
 877 //////////////////////////////////////////////////////////////////////////////
 878 // initial thread
 879 
 880 // Check if current thread is the initial thread, similar to Solaris thr_main.
 881 bool os::Linux::is_initial_thread(void) {
 882   char dummy;
 883   // If called before init complete, thread stack bottom will be null.
 884   // Can be called if fatal error occurs before initialization.
 885   if (initial_thread_stack_bottom() == NULL) return false;
 886   assert(initial_thread_stack_bottom() != NULL &&
 887          initial_thread_stack_size()   != 0,
 888          "os::init did not locate initial thread's stack region");
 889   if ((address)&dummy >= initial_thread_stack_bottom() &&
 890       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 891     return true;
 892   } else {
 893     return false;
 894   }
 895 }
 896 
 897 // Find the virtual memory area that contains addr
 898 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 899   FILE *fp = fopen("/proc/self/maps", "r");
 900   if (fp) {
 901     address low, high;
 902     while (!feof(fp)) {
 903       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 904         if (low <= addr && addr < high) {
 905           if (vma_low)  *vma_low  = low;
 906           if (vma_high) *vma_high = high;
 907           fclose(fp);
 908           return true;
 909         }
 910       }
 911       for (;;) {
 912         int ch = fgetc(fp);
 913         if (ch == EOF || ch == (int)'\n') break;
 914       }
 915     }
 916     fclose(fp);
 917   }
 918   return false;
 919 }
 920 
 921 // Locate initial thread stack. This special handling of initial thread stack
 922 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 923 // bogus value for the primordial process thread. While the launcher has created
 924 // the VM in a new thread since JDK 6, we still have to allow for the use of the
 925 // JNI invocation API from a primordial thread.
 926 void os::Linux::capture_initial_stack(size_t max_size) {
 927 
 928   // max_size is either 0 (which means accept OS default for thread stacks) or
 929   // a user-specified value known to be at least the minimum needed. If we
 930   // are actually on the primordial thread we can make it appear that we have a
 931   // smaller max_size stack by inserting the guard pages at that location. But we
 932   // cannot do anything to emulate a larger stack than what has been provided by
 933   // the OS or threading library. In fact if we try to use a stack greater than
 934   // what is set by rlimit then we will crash the hosting process.
 935 
 936   // Maximum stack size is the easy part, get it from RLIMIT_STACK.
 937   // If this is "unlimited" then it will be a huge value.
 938   struct rlimit rlim;
 939   getrlimit(RLIMIT_STACK, &rlim);
 940   size_t stack_size = rlim.rlim_cur;
 941 
 942   // 6308388: a bug in ld.so will relocate its own .data section to the
 943   //   lower end of primordial stack; reduce ulimit -s value a little bit
 944   //   so we won't install guard page on ld.so's data section.
 945   stack_size -= 2 * page_size();
 946 
 947   // Try to figure out where the stack base (top) is. This is harder.
 948   //
 949   // When an application is started, glibc saves the initial stack pointer in
 950   // a global variable "__libc_stack_end", which is then used by system
 951   // libraries. __libc_stack_end should be pretty close to stack top. The
 952   // variable is available since the very early days. However, because it is
 953   // a private interface, it could disappear in the future.
 954   //
 955   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 956   // to __libc_stack_end, it is very close to stack top, but isn't the real
 957   // stack top. Note that /proc may not exist if VM is running as a chroot
 958   // program, so reading /proc/<pid>/stat could fail. Also the contents of
 959   // /proc/<pid>/stat could change in the future (though unlikely).
 960   //
 961   // We try __libc_stack_end first. If that doesn't work, look for
 962   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
 963   // as a hint, which should work well in most cases.
 964 
 965   uintptr_t stack_start;
 966 
 967   // try __libc_stack_end first
 968   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
 969   if (p && *p) {
 970     stack_start = *p;
 971   } else {
 972     // see if we can get the start_stack field from /proc/self/stat
 973     FILE *fp;
 974     int pid;
 975     char state;
 976     int ppid;
 977     int pgrp;
 978     int session;
 979     int nr;
 980     int tpgrp;
 981     unsigned long flags;
 982     unsigned long minflt;
 983     unsigned long cminflt;
 984     unsigned long majflt;
 985     unsigned long cmajflt;
 986     unsigned long utime;
 987     unsigned long stime;
 988     long cutime;
 989     long cstime;
 990     long prio;
 991     long nice;
 992     long junk;
 993     long it_real;
 994     uintptr_t start;
 995     uintptr_t vsize;
 996     intptr_t rss;
 997     uintptr_t rsslim;
 998     uintptr_t scodes;
 999     uintptr_t ecode;
1000     int i;
1001 
1002     // Figure what the primordial thread stack base is. Code is inspired
1003     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1004     // followed by command name surrounded by parentheses, state, etc.
1005     char stat[2048];
1006     int statlen;
1007 
1008     fp = fopen("/proc/self/stat", "r");
1009     if (fp) {
1010       statlen = fread(stat, 1, 2047, fp);
1011       stat[statlen] = '\0';
1012       fclose(fp);
1013 
1014       // Skip pid and the command string. Note that we could be dealing with
1015       // weird command names, e.g. user could decide to rename java launcher
1016       // to "java 1.4.2 :)", then the stat file would look like
1017       //                1234 (java 1.4.2 :)) R ... ...
1018       // We don't really need to know the command string, just find the last
1019       // occurrence of ")" and then start parsing from there. See bug 4726580.
1020       char * s = strrchr(stat, ')');
1021 
1022       i = 0;
1023       if (s) {
1024         // Skip blank chars
1025         do { s++; } while (s && isspace(*s));
1026 
1027 #define _UFM UINTX_FORMAT
1028 #define _DFM INTX_FORMAT
1029 
1030         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1031         //              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
1032         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,
1033                    &state,          // 3  %c
1034                    &ppid,           // 4  %d
1035                    &pgrp,           // 5  %d
1036                    &session,        // 6  %d
1037                    &nr,             // 7  %d
1038                    &tpgrp,          // 8  %d
1039                    &flags,          // 9  %lu
1040                    &minflt,         // 10 %lu
1041                    &cminflt,        // 11 %lu
1042                    &majflt,         // 12 %lu
1043                    &cmajflt,        // 13 %lu
1044                    &utime,          // 14 %lu
1045                    &stime,          // 15 %lu
1046                    &cutime,         // 16 %ld
1047                    &cstime,         // 17 %ld
1048                    &prio,           // 18 %ld
1049                    &nice,           // 19 %ld
1050                    &junk,           // 20 %ld
1051                    &it_real,        // 21 %ld
1052                    &start,          // 22 UINTX_FORMAT
1053                    &vsize,          // 23 UINTX_FORMAT
1054                    &rss,            // 24 INTX_FORMAT
1055                    &rsslim,         // 25 UINTX_FORMAT
1056                    &scodes,         // 26 UINTX_FORMAT
1057                    &ecode,          // 27 UINTX_FORMAT
1058                    &stack_start);   // 28 UINTX_FORMAT
1059       }
1060 
1061 #undef _UFM
1062 #undef _DFM
1063 
1064       if (i != 28 - 2) {
1065         assert(false, "Bad conversion from /proc/self/stat");
1066         // product mode - assume we are the initial thread, good luck in the
1067         // embedded case.
1068         warning("Can't detect initial thread stack location - bad conversion");
1069         stack_start = (uintptr_t) &rlim;
1070       }
1071     } else {
1072       // For some reason we can't open /proc/self/stat (for example, running on
1073       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1074       // most cases, so don't abort:
1075       warning("Can't detect initial thread stack location - no /proc/self/stat");
1076       stack_start = (uintptr_t) &rlim;
1077     }
1078   }
1079 
1080   // Now we have a pointer (stack_start) very close to the stack top, the
1081   // next thing to do is to figure out the exact location of stack top. We
1082   // can find out the virtual memory area that contains stack_start by
1083   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1084   // and its upper limit is the real stack top. (again, this would fail if
1085   // running inside chroot, because /proc may not exist.)
1086 
1087   uintptr_t stack_top;
1088   address low, high;
1089   if (find_vma((address)stack_start, &low, &high)) {
1090     // success, "high" is the true stack top. (ignore "low", because initial
1091     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1092     stack_top = (uintptr_t)high;
1093   } else {
1094     // failed, likely because /proc/self/maps does not exist
1095     warning("Can't detect initial thread stack location - find_vma failed");
1096     // best effort: stack_start is normally within a few pages below the real
1097     // stack top, use it as stack top, and reduce stack size so we won't put
1098     // guard page outside stack.
1099     stack_top = stack_start;
1100     stack_size -= 16 * page_size();
1101   }
1102 
1103   // stack_top could be partially down the page so align it
1104   stack_top = align_size_up(stack_top, page_size());
1105 
1106   // Allowed stack value is minimum of max_size and what we derived from rlimit
1107   if (max_size > 0) {
1108     _initial_thread_stack_size = MIN2(max_size, stack_size);
1109   } else {
1110     // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1111     // clamp it at 8MB as we do on Solaris
1112     _initial_thread_stack_size = MIN2(stack_size, 8*M);
1113   }
1114   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1115   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1116 
1117   assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1118 
1119   if (log_is_enabled(Info, os, thread)) {
1120     // See if we seem to be on primordial process thread
1121     bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1122                       uintptr_t(&rlim) < stack_top;
1123 
1124     log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1125                          SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1126                          primordial ? "primordial" : "user", max_size / K,  _initial_thread_stack_size / K,
1127                          stack_top, intptr_t(_initial_thread_stack_bottom));
1128   }
1129 }
1130 
1131 ////////////////////////////////////////////////////////////////////////////////
1132 // time support
1133 
1134 // Time since start-up in seconds to a fine granularity.
1135 // Used by VMSelfDestructTimer and the MemProfiler.
1136 double os::elapsedTime() {
1137 
1138   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1139 }
1140 
1141 jlong os::elapsed_counter() {
1142   return javaTimeNanos() - initial_time_count;
1143 }
1144 
1145 jlong os::elapsed_frequency() {
1146   return NANOSECS_PER_SEC; // nanosecond resolution
1147 }
1148 
1149 bool os::supports_vtime() { return true; }
1150 bool os::enable_vtime()   { return false; }
1151 bool os::vtime_enabled()  { return false; }
1152 
1153 double os::elapsedVTime() {
1154   struct rusage usage;
1155   int retval = getrusage(RUSAGE_THREAD, &usage);
1156   if (retval == 0) {
1157     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);
1158   } else {
1159     // better than nothing, but not much
1160     return elapsedTime();
1161   }
1162 }
1163 
1164 jlong os::javaTimeMillis() {
1165   timeval time;
1166   int status = gettimeofday(&time, NULL);
1167   assert(status != -1, "linux error");
1168   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1169 }
1170 
1171 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1172   timeval time;
1173   int status = gettimeofday(&time, NULL);
1174   assert(status != -1, "linux error");
1175   seconds = jlong(time.tv_sec);
1176   nanos = jlong(time.tv_usec) * 1000;
1177 }
1178 
1179 
1180 #ifndef CLOCK_MONOTONIC
1181   #define CLOCK_MONOTONIC (1)
1182 #endif
1183 
1184 void os::Linux::clock_init() {
1185   // we do dlopen's in this particular order due to bug in linux
1186   // dynamical loader (see 6348968) leading to crash on exit
1187   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1188   if (handle == NULL) {
1189     handle = dlopen("librt.so", RTLD_LAZY);
1190   }
1191 
1192   if (handle) {
1193     int (*clock_getres_func)(clockid_t, struct timespec*) =
1194            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1195     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1196            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1197     if (clock_getres_func && clock_gettime_func) {
1198       // See if monotonic clock is supported by the kernel. Note that some
1199       // early implementations simply return kernel jiffies (updated every
1200       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1201       // for nano time (though the monotonic property is still nice to have).
1202       // It's fixed in newer kernels, however clock_getres() still returns
1203       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1204       // resolution for now. Hopefully as people move to new kernels, this
1205       // won't be a problem.
1206       struct timespec res;
1207       struct timespec tp;
1208       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1209           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1210         // yes, monotonic clock is supported
1211         _clock_gettime = clock_gettime_func;
1212         return;
1213       } else {
1214         // close librt if there is no monotonic clock
1215         dlclose(handle);
1216       }
1217     }
1218   }
1219   warning("No monotonic clock was available - timed services may " \
1220           "be adversely affected if the time-of-day clock changes");
1221 }
1222 
1223 #ifndef SYS_clock_getres
1224   #if defined(X86) || defined(PPC64) || defined(S390)
1225     #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1226     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1227   #else
1228     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1229     #define sys_clock_getres(x,y)  -1
1230   #endif
1231 #else
1232   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1233 #endif
1234 
1235 void os::Linux::fast_thread_clock_init() {
1236   if (!UseLinuxPosixThreadCPUClocks) {
1237     return;
1238   }
1239   clockid_t clockid;
1240   struct timespec tp;
1241   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1242       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1243 
1244   // Switch to using fast clocks for thread cpu time if
1245   // the sys_clock_getres() returns 0 error code.
1246   // Note, that some kernels may support the current thread
1247   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1248   // returned by the pthread_getcpuclockid().
1249   // If the fast Posix clocks are supported then the sys_clock_getres()
1250   // must return at least tp.tv_sec == 0 which means a resolution
1251   // better than 1 sec. This is extra check for reliability.
1252 
1253   if (pthread_getcpuclockid_func &&
1254       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1255       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1256     _supports_fast_thread_cpu_time = true;
1257     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1258   }
1259 }
1260 
1261 jlong os::javaTimeNanos() {
1262   if (os::supports_monotonic_clock()) {
1263     struct timespec tp;
1264     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1265     assert(status == 0, "gettime error");
1266     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1267     return result;
1268   } else {
1269     timeval time;
1270     int status = gettimeofday(&time, NULL);
1271     assert(status != -1, "linux error");
1272     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1273     return 1000 * usecs;
1274   }
1275 }
1276 
1277 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1278   if (os::supports_monotonic_clock()) {
1279     info_ptr->max_value = ALL_64_BITS;
1280 
1281     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1282     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1283     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1284   } else {
1285     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1286     info_ptr->max_value = ALL_64_BITS;
1287 
1288     // gettimeofday is a real time clock so it skips
1289     info_ptr->may_skip_backward = true;
1290     info_ptr->may_skip_forward = true;
1291   }
1292 
1293   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1294 }
1295 
1296 // Return the real, user, and system times in seconds from an
1297 // arbitrary fixed point in the past.
1298 bool os::getTimesSecs(double* process_real_time,
1299                       double* process_user_time,
1300                       double* process_system_time) {
1301   struct tms ticks;
1302   clock_t real_ticks = times(&ticks);
1303 
1304   if (real_ticks == (clock_t) (-1)) {
1305     return false;
1306   } else {
1307     double ticks_per_second = (double) clock_tics_per_sec;
1308     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1309     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1310     *process_real_time = ((double) real_ticks) / ticks_per_second;
1311 
1312     return true;
1313   }
1314 }
1315 
1316 
1317 char * os::local_time_string(char *buf, size_t buflen) {
1318   struct tm t;
1319   time_t long_time;
1320   time(&long_time);
1321   localtime_r(&long_time, &t);
1322   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1323                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1324                t.tm_hour, t.tm_min, t.tm_sec);
1325   return buf;
1326 }
1327 
1328 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1329   return localtime_r(clock, res);
1330 }
1331 
1332 ////////////////////////////////////////////////////////////////////////////////
1333 // runtime exit support
1334 
1335 // Note: os::shutdown() might be called very early during initialization, or
1336 // called from signal handler. Before adding something to os::shutdown(), make
1337 // sure it is async-safe and can handle partially initialized VM.
1338 void os::shutdown() {
1339 
1340   // allow PerfMemory to attempt cleanup of any persistent resources
1341   perfMemory_exit();
1342 
1343   // needs to remove object in file system
1344   AttachListener::abort();
1345 
1346   // flush buffered output, finish log files
1347   ostream_abort();
1348 
1349   // Check for abort hook
1350   abort_hook_t abort_hook = Arguments::abort_hook();
1351   if (abort_hook != NULL) {
1352     abort_hook();
1353   }
1354 
1355 }
1356 
1357 // Note: os::abort() might be called very early during initialization, or
1358 // called from signal handler. Before adding something to os::abort(), make
1359 // sure it is async-safe and can handle partially initialized VM.
1360 void os::abort(bool dump_core, void* siginfo, const void* context) {
1361   os::shutdown();
1362   if (dump_core) {
1363 #ifndef PRODUCT
1364     fdStream out(defaultStream::output_fd());
1365     out.print_raw("Current thread is ");
1366     char buf[16];
1367     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1368     out.print_raw_cr(buf);
1369     out.print_raw_cr("Dumping core ...");
1370 #endif
1371     ::abort(); // dump core
1372   }
1373 
1374   ::exit(1);
1375 }
1376 
1377 // Die immediately, no exit hook, no abort hook, no cleanup.
1378 void os::die() {
1379   ::abort();
1380 }
1381 
1382 
1383 // This method is a copy of JDK's sysGetLastErrorString
1384 // from src/solaris/hpi/src/system_md.c
1385 
1386 size_t os::lasterror(char *buf, size_t len) {
1387   if (errno == 0)  return 0;
1388 
1389   const char *s = os::strerror(errno);
1390   size_t n = ::strlen(s);
1391   if (n >= len) {
1392     n = len - 1;
1393   }
1394   ::strncpy(buf, s, n);
1395   buf[n] = '\0';
1396   return n;
1397 }
1398 
1399 // thread_id is kernel thread id (similar to Solaris LWP id)
1400 intx os::current_thread_id() { return os::Linux::gettid(); }
1401 int os::current_process_id() {
1402   return ::getpid();
1403 }
1404 
1405 // DLL functions
1406 
1407 const char* os::dll_file_extension() { return ".so"; }
1408 
1409 // This must be hard coded because it's the system's temporary
1410 // directory not the java application's temp directory, ala java.io.tmpdir.
1411 const char* os::get_temp_directory() { return "/tmp"; }
1412 
1413 static bool file_exists(const char* filename) {
1414   struct stat statbuf;
1415   if (filename == NULL || strlen(filename) == 0) {
1416     return false;
1417   }
1418   return os::stat(filename, &statbuf) == 0;
1419 }
1420 
1421 bool os::dll_build_name(char* buffer, size_t buflen,
1422                         const char* pname, const char* fname) {
1423   bool retval = false;
1424   // Copied from libhpi
1425   const size_t pnamelen = pname ? strlen(pname) : 0;
1426 
1427   // Return error on buffer overflow.
1428   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1429     return retval;
1430   }
1431 
1432   if (pnamelen == 0) {
1433     snprintf(buffer, buflen, "lib%s.so", fname);
1434     retval = true;
1435   } else if (strchr(pname, *os::path_separator()) != NULL) {
1436     int n;
1437     char** pelements = split_path(pname, &n);
1438     if (pelements == NULL) {
1439       return false;
1440     }
1441     for (int i = 0; i < n; i++) {
1442       // Really shouldn't be NULL, but check can't hurt
1443       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1444         continue; // skip the empty path values
1445       }
1446       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1447       if (file_exists(buffer)) {
1448         retval = true;
1449         break;
1450       }
1451     }
1452     // release the storage
1453     for (int i = 0; i < n; i++) {
1454       if (pelements[i] != NULL) {
1455         FREE_C_HEAP_ARRAY(char, pelements[i]);
1456       }
1457     }
1458     if (pelements != NULL) {
1459       FREE_C_HEAP_ARRAY(char*, pelements);
1460     }
1461   } else {
1462     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1463     retval = true;
1464   }
1465   return retval;
1466 }
1467 
1468 // check if addr is inside libjvm.so
1469 bool os::address_is_in_vm(address addr) {
1470   static address libjvm_base_addr;
1471   Dl_info dlinfo;
1472 
1473   if (libjvm_base_addr == NULL) {
1474     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1475       libjvm_base_addr = (address)dlinfo.dli_fbase;
1476     }
1477     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1478   }
1479 
1480   if (dladdr((void *)addr, &dlinfo) != 0) {
1481     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1482   }
1483 
1484   return false;
1485 }
1486 
1487 bool os::dll_address_to_function_name(address addr, char *buf,
1488                                       int buflen, int *offset,
1489                                       bool demangle) {
1490   // buf is not optional, but offset is optional
1491   assert(buf != NULL, "sanity check");
1492 
1493   Dl_info dlinfo;
1494 
1495   if (dladdr((void*)addr, &dlinfo) != 0) {
1496     // see if we have a matching symbol
1497     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1498       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1499         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1500       }
1501       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1502       return true;
1503     }
1504     // no matching symbol so try for just file info
1505     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1506       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1507                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1508         return true;
1509       }
1510     }
1511   }
1512 
1513   buf[0] = '\0';
1514   if (offset != NULL) *offset = -1;
1515   return false;
1516 }
1517 
1518 struct _address_to_library_name {
1519   address addr;          // input : memory address
1520   size_t  buflen;        //         size of fname
1521   char*   fname;         // output: library name
1522   address base;          //         library base addr
1523 };
1524 
1525 static int address_to_library_name_callback(struct dl_phdr_info *info,
1526                                             size_t size, void *data) {
1527   int i;
1528   bool found = false;
1529   address libbase = NULL;
1530   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1531 
1532   // iterate through all loadable segments
1533   for (i = 0; i < info->dlpi_phnum; i++) {
1534     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1535     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1536       // base address of a library is the lowest address of its loaded
1537       // segments.
1538       if (libbase == NULL || libbase > segbase) {
1539         libbase = segbase;
1540       }
1541       // see if 'addr' is within current segment
1542       if (segbase <= d->addr &&
1543           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1544         found = true;
1545       }
1546     }
1547   }
1548 
1549   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1550   // so dll_address_to_library_name() can fall through to use dladdr() which
1551   // can figure out executable name from argv[0].
1552   if (found && info->dlpi_name && info->dlpi_name[0]) {
1553     d->base = libbase;
1554     if (d->fname) {
1555       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1556     }
1557     return 1;
1558   }
1559   return 0;
1560 }
1561 
1562 bool os::dll_address_to_library_name(address addr, char* buf,
1563                                      int buflen, int* offset) {
1564   // buf is not optional, but offset is optional
1565   assert(buf != NULL, "sanity check");
1566 
1567   Dl_info dlinfo;
1568   struct _address_to_library_name data;
1569 
1570   // There is a bug in old glibc dladdr() implementation that it could resolve
1571   // to wrong library name if the .so file has a base address != NULL. Here
1572   // we iterate through the program headers of all loaded libraries to find
1573   // out which library 'addr' really belongs to. This workaround can be
1574   // removed once the minimum requirement for glibc is moved to 2.3.x.
1575   data.addr = addr;
1576   data.fname = buf;
1577   data.buflen = buflen;
1578   data.base = NULL;
1579   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1580 
1581   if (rslt) {
1582     // buf already contains library name
1583     if (offset) *offset = addr - data.base;
1584     return true;
1585   }
1586   if (dladdr((void*)addr, &dlinfo) != 0) {
1587     if (dlinfo.dli_fname != NULL) {
1588       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1589     }
1590     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1591       *offset = addr - (address)dlinfo.dli_fbase;
1592     }
1593     return true;
1594   }
1595 
1596   buf[0] = '\0';
1597   if (offset) *offset = -1;
1598   return false;
1599 }
1600 
1601 // Loads .dll/.so and
1602 // in case of error it checks if .dll/.so was built for the
1603 // same architecture as Hotspot is running on
1604 
1605 
1606 // Remember the stack's state. The Linux dynamic linker will change
1607 // the stack to 'executable' at most once, so we must safepoint only once.
1608 bool os::Linux::_stack_is_executable = false;
1609 
1610 // VM operation that loads a library.  This is necessary if stack protection
1611 // of the Java stacks can be lost during loading the library.  If we
1612 // do not stop the Java threads, they can stack overflow before the stacks
1613 // are protected again.
1614 class VM_LinuxDllLoad: public VM_Operation {
1615  private:
1616   const char *_filename;
1617   char *_ebuf;
1618   int _ebuflen;
1619   void *_lib;
1620  public:
1621   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1622     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1623   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1624   void doit() {
1625     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1626     os::Linux::_stack_is_executable = true;
1627   }
1628   void* loaded_library() { return _lib; }
1629 };
1630 
1631 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1632   void * result = NULL;
1633   bool load_attempted = false;
1634 
1635   // Check whether the library to load might change execution rights
1636   // of the stack. If they are changed, the protection of the stack
1637   // guard pages will be lost. We need a safepoint to fix this.
1638   //
1639   // See Linux man page execstack(8) for more info.
1640   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1641     ElfFile ef(filename);
1642     if (!ef.specifies_noexecstack()) {
1643       if (!is_init_completed()) {
1644         os::Linux::_stack_is_executable = true;
1645         // This is OK - No Java threads have been created yet, and hence no
1646         // stack guard pages to fix.
1647         //
1648         // This should happen only when you are building JDK7 using a very
1649         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1650         //
1651         // Dynamic loader will make all stacks executable after
1652         // this function returns, and will not do that again.
1653         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1654       } else {
1655         warning("You have loaded library %s which might have disabled stack guard. "
1656                 "The VM will try to fix the stack guard now.\n"
1657                 "It's highly recommended that you fix the library with "
1658                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1659                 filename);
1660 
1661         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1662         JavaThread *jt = JavaThread::current();
1663         if (jt->thread_state() != _thread_in_native) {
1664           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1665           // that requires ExecStack. Cannot enter safe point. Let's give up.
1666           warning("Unable to fix stack guard. Giving up.");
1667         } else {
1668           if (!LoadExecStackDllInVMThread) {
1669             // This is for the case where the DLL has an static
1670             // constructor function that executes JNI code. We cannot
1671             // load such DLLs in the VMThread.
1672             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1673           }
1674 
1675           ThreadInVMfromNative tiv(jt);
1676           debug_only(VMNativeEntryWrapper vew;)
1677 
1678           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1679           VMThread::execute(&op);
1680           if (LoadExecStackDllInVMThread) {
1681             result = op.loaded_library();
1682           }
1683           load_attempted = true;
1684         }
1685       }
1686     }
1687   }
1688 
1689   if (!load_attempted) {
1690     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1691   }
1692 
1693   if (result != NULL) {
1694     // Successful loading
1695     return result;
1696   }
1697 
1698   Elf32_Ehdr elf_head;
1699   int diag_msg_max_length=ebuflen-strlen(ebuf);
1700   char* diag_msg_buf=ebuf+strlen(ebuf);
1701 
1702   if (diag_msg_max_length==0) {
1703     // No more space in ebuf for additional diagnostics message
1704     return NULL;
1705   }
1706 
1707 
1708   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1709 
1710   if (file_descriptor < 0) {
1711     // Can't open library, report dlerror() message
1712     return NULL;
1713   }
1714 
1715   bool failed_to_read_elf_head=
1716     (sizeof(elf_head)!=
1717      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1718 
1719   ::close(file_descriptor);
1720   if (failed_to_read_elf_head) {
1721     // file i/o error - report dlerror() msg
1722     return NULL;
1723   }
1724 
1725   typedef struct {
1726     Elf32_Half    code;         // Actual value as defined in elf.h
1727     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1728     unsigned char elf_class;    // 32 or 64 bit
1729     unsigned char endianess;    // MSB or LSB
1730     char*         name;         // String representation
1731   } arch_t;
1732 
1733 #ifndef EM_486
1734   #define EM_486          6               /* Intel 80486 */
1735 #endif
1736 #ifndef EM_AARCH64
1737   #define EM_AARCH64    183               /* ARM AARCH64 */
1738 #endif
1739 
1740   static const arch_t arch_array[]={
1741     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1742     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1743     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1744     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1745     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1746     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1747     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1748     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1749 #if defined(VM_LITTLE_ENDIAN)
1750     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1751 #else
1752     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1753 #endif
1754     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1755     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1756     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1757     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1758     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1759     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1760     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1761     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1762   };
1763 
1764 #if  (defined IA32)
1765   static  Elf32_Half running_arch_code=EM_386;
1766 #elif   (defined AMD64)
1767   static  Elf32_Half running_arch_code=EM_X86_64;
1768 #elif  (defined IA64)
1769   static  Elf32_Half running_arch_code=EM_IA_64;
1770 #elif  (defined __sparc) && (defined _LP64)
1771   static  Elf32_Half running_arch_code=EM_SPARCV9;
1772 #elif  (defined __sparc) && (!defined _LP64)
1773   static  Elf32_Half running_arch_code=EM_SPARC;
1774 #elif  (defined __powerpc64__)
1775   static  Elf32_Half running_arch_code=EM_PPC64;
1776 #elif  (defined __powerpc__)
1777   static  Elf32_Half running_arch_code=EM_PPC;
1778 #elif  (defined AARCH64)
1779   static  Elf32_Half running_arch_code=EM_AARCH64;
1780 #elif  (defined ARM)
1781   static  Elf32_Half running_arch_code=EM_ARM;
1782 #elif  (defined S390)
1783   static  Elf32_Half running_arch_code=EM_S390;
1784 #elif  (defined ALPHA)
1785   static  Elf32_Half running_arch_code=EM_ALPHA;
1786 #elif  (defined MIPSEL)
1787   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1788 #elif  (defined PARISC)
1789   static  Elf32_Half running_arch_code=EM_PARISC;
1790 #elif  (defined MIPS)
1791   static  Elf32_Half running_arch_code=EM_MIPS;
1792 #elif  (defined M68K)
1793   static  Elf32_Half running_arch_code=EM_68K;
1794 #else
1795     #error Method os::dll_load requires that one of following is defined:\
1796         AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, __sparc
1797 #endif
1798 
1799   // Identify compatability class for VM's architecture and library's architecture
1800   // Obtain string descriptions for architectures
1801 
1802   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1803   int running_arch_index=-1;
1804 
1805   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1806     if (running_arch_code == arch_array[i].code) {
1807       running_arch_index    = i;
1808     }
1809     if (lib_arch.code == arch_array[i].code) {
1810       lib_arch.compat_class = arch_array[i].compat_class;
1811       lib_arch.name         = arch_array[i].name;
1812     }
1813   }
1814 
1815   assert(running_arch_index != -1,
1816          "Didn't find running architecture code (running_arch_code) in arch_array");
1817   if (running_arch_index == -1) {
1818     // Even though running architecture detection failed
1819     // we may still continue with reporting dlerror() message
1820     return NULL;
1821   }
1822 
1823   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1824     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1825     return NULL;
1826   }
1827 
1828 #ifndef S390
1829   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1830     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1831     return NULL;
1832   }
1833 #endif // !S390
1834 
1835   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1836     if (lib_arch.name!=NULL) {
1837       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1838                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1839                  lib_arch.name, arch_array[running_arch_index].name);
1840     } else {
1841       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1842                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1843                  lib_arch.code,
1844                  arch_array[running_arch_index].name);
1845     }
1846   }
1847 
1848   return NULL;
1849 }
1850 
1851 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1852                                 int ebuflen) {
1853   void * result = ::dlopen(filename, RTLD_LAZY);
1854   if (result == NULL) {
1855     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1856     ebuf[ebuflen-1] = '\0';
1857   }
1858   return result;
1859 }
1860 
1861 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1862                                        int ebuflen) {
1863   void * result = NULL;
1864   if (LoadExecStackDllInVMThread) {
1865     result = dlopen_helper(filename, ebuf, ebuflen);
1866   }
1867 
1868   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1869   // library that requires an executable stack, or which does not have this
1870   // stack attribute set, dlopen changes the stack attribute to executable. The
1871   // read protection of the guard pages gets lost.
1872   //
1873   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1874   // may have been queued at the same time.
1875 
1876   if (!_stack_is_executable) {
1877     JavaThread *jt = Threads::first();
1878 
1879     while (jt) {
1880       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1881           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1882         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1883           warning("Attempt to reguard stack yellow zone failed.");
1884         }
1885       }
1886       jt = jt->next();
1887     }
1888   }
1889 
1890   return result;
1891 }
1892 
1893 void* os::dll_lookup(void* handle, const char* name) {
1894   void* res = dlsym(handle, name);
1895   return res;
1896 }
1897 
1898 void* os::get_default_process_handle() {
1899   return (void*)::dlopen(NULL, RTLD_LAZY);
1900 }
1901 
1902 static bool _print_ascii_file(const char* filename, outputStream* st) {
1903   int fd = ::open(filename, O_RDONLY);
1904   if (fd == -1) {
1905     return false;
1906   }
1907 
1908   char buf[33];
1909   int bytes;
1910   buf[32] = '\0';
1911   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1912     st->print_raw(buf, bytes);
1913   }
1914 
1915   ::close(fd);
1916 
1917   return true;
1918 }
1919 
1920 void os::print_dll_info(outputStream *st) {
1921   st->print_cr("Dynamic libraries:");
1922 
1923   char fname[32];
1924   pid_t pid = os::Linux::gettid();
1925 
1926   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1927 
1928   if (!_print_ascii_file(fname, st)) {
1929     st->print("Can not get library information for pid = %d\n", pid);
1930   }
1931 }
1932 
1933 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1934   FILE *procmapsFile = NULL;
1935 
1936   // Open the procfs maps file for the current process
1937   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1938     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1939     char line[PATH_MAX + 100];
1940 
1941     // Read line by line from 'file'
1942     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1943       u8 base, top, offset, inode;
1944       char permissions[5];
1945       char device[6];
1946       char name[PATH_MAX + 1];
1947 
1948       // Parse fields from line
1949       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1950              &base, &top, permissions, &offset, device, &inode, name);
1951 
1952       // Filter by device id '00:00' so that we only get file system mapped files.
1953       if (strcmp(device, "00:00") != 0) {
1954 
1955         // Call callback with the fields of interest
1956         if(callback(name, (address)base, (address)top, param)) {
1957           // Oops abort, callback aborted
1958           fclose(procmapsFile);
1959           return 1;
1960         }
1961       }
1962     }
1963     fclose(procmapsFile);
1964   }
1965   return 0;
1966 }
1967 
1968 void os::print_os_info_brief(outputStream* st) {
1969   os::Linux::print_distro_info(st);
1970 
1971   os::Posix::print_uname_info(st);
1972 
1973   os::Linux::print_libversion_info(st);
1974 
1975 }
1976 
1977 void os::print_os_info(outputStream* st) {
1978   st->print("OS:");
1979 
1980   os::Linux::print_distro_info(st);
1981 
1982   os::Posix::print_uname_info(st);
1983 
1984   // Print warning if unsafe chroot environment detected
1985   if (unsafe_chroot_detected) {
1986     st->print("WARNING!! ");
1987     st->print_cr("%s", unstable_chroot_error);
1988   }
1989 
1990   os::Linux::print_libversion_info(st);
1991 
1992   os::Posix::print_rlimit_info(st);
1993 
1994   os::Posix::print_load_average(st);
1995 
1996   os::Linux::print_full_memory_info(st);
1997 }
1998 
1999 // Try to identify popular distros.
2000 // Most Linux distributions have a /etc/XXX-release file, which contains
2001 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2002 // file that also contains the OS version string. Some have more than one
2003 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2004 // /etc/redhat-release.), so the order is important.
2005 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2006 // their own specific XXX-release file as well as a redhat-release file.
2007 // Because of this the XXX-release file needs to be searched for before the
2008 // redhat-release file.
2009 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2010 // search for redhat-release / SuSE-release needs to be before lsb-release.
2011 // Since the lsb-release file is the new standard it needs to be searched
2012 // before the older style release files.
2013 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2014 // next to last resort.  The os-release file is a new standard that contains
2015 // distribution information and the system-release file seems to be an old
2016 // standard that has been replaced by the lsb-release and os-release files.
2017 // Searching for the debian_version file is the last resort.  It contains
2018 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2019 // "Debian " is printed before the contents of the debian_version file.
2020 
2021 const char* distro_files[] = {
2022   "/etc/oracle-release",
2023   "/etc/mandriva-release",
2024   "/etc/mandrake-release",
2025   "/etc/sun-release",
2026   "/etc/redhat-release",
2027   "/etc/SuSE-release",
2028   "/etc/lsb-release",
2029   "/etc/turbolinux-release",
2030   "/etc/gentoo-release",
2031   "/etc/ltib-release",
2032   "/etc/angstrom-version",
2033   "/etc/system-release",
2034   "/etc/os-release",
2035   NULL };
2036 
2037 void os::Linux::print_distro_info(outputStream* st) {
2038   for (int i = 0;; i++) {
2039     const char* file = distro_files[i];
2040     if (file == NULL) {
2041       break;  // done
2042     }
2043     // If file prints, we found it.
2044     if (_print_ascii_file(file, st)) {
2045       return;
2046     }
2047   }
2048 
2049   if (file_exists("/etc/debian_version")) {
2050     st->print("Debian ");
2051     _print_ascii_file("/etc/debian_version", st);
2052   } else {
2053     st->print("Linux");
2054   }
2055   st->cr();
2056 }
2057 
2058 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2059   char buf[256];
2060   while (fgets(buf, sizeof(buf), fp)) {
2061     // Edit out extra stuff in expected format
2062     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2063       char* ptr = strstr(buf, "\"");  // the name is in quotes
2064       if (ptr != NULL) {
2065         ptr++; // go beyond first quote
2066         char* nl = strchr(ptr, '\"');
2067         if (nl != NULL) *nl = '\0';
2068         strncpy(distro, ptr, length);
2069       } else {
2070         ptr = strstr(buf, "=");
2071         ptr++; // go beyond equals then
2072         char* nl = strchr(ptr, '\n');
2073         if (nl != NULL) *nl = '\0';
2074         strncpy(distro, ptr, length);
2075       }
2076       return;
2077     } else if (get_first_line) {
2078       char* nl = strchr(buf, '\n');
2079       if (nl != NULL) *nl = '\0';
2080       strncpy(distro, buf, length);
2081       return;
2082     }
2083   }
2084   // print last line and close
2085   char* nl = strchr(buf, '\n');
2086   if (nl != NULL) *nl = '\0';
2087   strncpy(distro, buf, length);
2088 }
2089 
2090 static void parse_os_info(char* distro, size_t length, const char* file) {
2091   FILE* fp = fopen(file, "r");
2092   if (fp != NULL) {
2093     // if suse format, print out first line
2094     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2095     parse_os_info_helper(fp, distro, length, get_first_line);
2096     fclose(fp);
2097   }
2098 }
2099 
2100 void os::get_summary_os_info(char* buf, size_t buflen) {
2101   for (int i = 0;; i++) {
2102     const char* file = distro_files[i];
2103     if (file == NULL) {
2104       break; // ran out of distro_files
2105     }
2106     if (file_exists(file)) {
2107       parse_os_info(buf, buflen, file);
2108       return;
2109     }
2110   }
2111   // special case for debian
2112   if (file_exists("/etc/debian_version")) {
2113     strncpy(buf, "Debian ", buflen);
2114     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2115   } else {
2116     strncpy(buf, "Linux", buflen);
2117   }
2118 }
2119 
2120 void os::Linux::print_libversion_info(outputStream* st) {
2121   // libc, pthread
2122   st->print("libc:");
2123   st->print("%s ", os::Linux::glibc_version());
2124   st->print("%s ", os::Linux::libpthread_version());
2125   st->cr();
2126 }
2127 
2128 void os::Linux::print_full_memory_info(outputStream* st) {
2129   st->print("\n/proc/meminfo:\n");
2130   _print_ascii_file("/proc/meminfo", st);
2131   st->cr();
2132 }
2133 
2134 void os::print_memory_info(outputStream* st) {
2135 
2136   st->print("Memory:");
2137   st->print(" %dk page", os::vm_page_size()>>10);
2138 
2139   // values in struct sysinfo are "unsigned long"
2140   struct sysinfo si;
2141   sysinfo(&si);
2142 
2143   st->print(", physical " UINT64_FORMAT "k",
2144             os::physical_memory() >> 10);
2145   st->print("(" UINT64_FORMAT "k free)",
2146             os::available_memory() >> 10);
2147   st->print(", swap " UINT64_FORMAT "k",
2148             ((jlong)si.totalswap * si.mem_unit) >> 10);
2149   st->print("(" UINT64_FORMAT "k free)",
2150             ((jlong)si.freeswap * si.mem_unit) >> 10);
2151   st->cr();
2152 }
2153 
2154 // Print the first "model name" line and the first "flags" line
2155 // that we find and nothing more. We assume "model name" comes
2156 // before "flags" so if we find a second "model name", then the
2157 // "flags" field is considered missing.
2158 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2159 #if defined(IA32) || defined(AMD64)
2160   // Other platforms have less repetitive cpuinfo files
2161   FILE *fp = fopen("/proc/cpuinfo", "r");
2162   if (fp) {
2163     while (!feof(fp)) {
2164       if (fgets(buf, buflen, fp)) {
2165         // Assume model name comes before flags
2166         bool model_name_printed = false;
2167         if (strstr(buf, "model name") != NULL) {
2168           if (!model_name_printed) {
2169             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2170             st->print_raw(buf);
2171             model_name_printed = true;
2172           } else {
2173             // model name printed but not flags?  Odd, just return
2174             fclose(fp);
2175             return true;
2176           }
2177         }
2178         // print the flags line too
2179         if (strstr(buf, "flags") != NULL) {
2180           st->print_raw(buf);
2181           fclose(fp);
2182           return true;
2183         }
2184       }
2185     }
2186     fclose(fp);
2187   }
2188 #endif // x86 platforms
2189   return false;
2190 }
2191 
2192 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2193   // Only print the model name if the platform provides this as a summary
2194   if (!print_model_name_and_flags(st, buf, buflen)) {
2195     st->print("\n/proc/cpuinfo:\n");
2196     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2197       st->print_cr("  <Not Available>");
2198     }
2199   }
2200 }
2201 
2202 #if defined(AMD64) || defined(IA32) || defined(X32)
2203 const char* search_string = "model name";
2204 #elif defined(PPC64)
2205 const char* search_string = "cpu";
2206 #elif defined(S390)
2207 const char* search_string = "processor";
2208 #elif defined(SPARC)
2209 const char* search_string = "cpu";
2210 #else
2211 const char* search_string = "Processor";
2212 #endif
2213 
2214 // Parses the cpuinfo file for string representing the model name.
2215 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2216   FILE* fp = fopen("/proc/cpuinfo", "r");
2217   if (fp != NULL) {
2218     while (!feof(fp)) {
2219       char buf[256];
2220       if (fgets(buf, sizeof(buf), fp)) {
2221         char* start = strstr(buf, search_string);
2222         if (start != NULL) {
2223           char *ptr = start + strlen(search_string);
2224           char *end = buf + strlen(buf);
2225           while (ptr != end) {
2226              // skip whitespace and colon for the rest of the name.
2227              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2228                break;
2229              }
2230              ptr++;
2231           }
2232           if (ptr != end) {
2233             // reasonable string, get rid of newline and keep the rest
2234             char* nl = strchr(buf, '\n');
2235             if (nl != NULL) *nl = '\0';
2236             strncpy(cpuinfo, ptr, length);
2237             fclose(fp);
2238             return;
2239           }
2240         }
2241       }
2242     }
2243     fclose(fp);
2244   }
2245   // cpuinfo not found or parsing failed, just print generic string.  The entire
2246   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2247 #if   defined(AARCH64)
2248   strncpy(cpuinfo, "AArch64", length);
2249 #elif defined(AMD64)
2250   strncpy(cpuinfo, "x86_64", length);
2251 #elif defined(ARM)  // Order wrt. AARCH64 is relevant!
2252   strncpy(cpuinfo, "ARM", length);
2253 #elif defined(IA32)
2254   strncpy(cpuinfo, "x86_32", length);
2255 #elif defined(IA64)
2256   strncpy(cpuinfo, "IA64", length);
2257 #elif defined(PPC)
2258   strncpy(cpuinfo, "PPC64", length);
2259 #elif defined(S390)
2260   strncpy(cpuinfo, "S390", length);
2261 #elif defined(SPARC)
2262   strncpy(cpuinfo, "sparcv9", length);
2263 #elif defined(ZERO_LIBARCH)
2264   strncpy(cpuinfo, ZERO_LIBARCH, length);
2265 #else
2266   strncpy(cpuinfo, "unknown", length);
2267 #endif
2268 }
2269 
2270 static void print_signal_handler(outputStream* st, int sig,
2271                                  char* buf, size_t buflen);
2272 
2273 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2274   st->print_cr("Signal Handlers:");
2275   print_signal_handler(st, SIGSEGV, buf, buflen);
2276   print_signal_handler(st, SIGBUS , buf, buflen);
2277   print_signal_handler(st, SIGFPE , buf, buflen);
2278   print_signal_handler(st, SIGPIPE, buf, buflen);
2279   print_signal_handler(st, SIGXFSZ, buf, buflen);
2280   print_signal_handler(st, SIGILL , buf, buflen);
2281   print_signal_handler(st, SR_signum, buf, buflen);
2282   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2283   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2284   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2285   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2286 #if defined(PPC64)
2287   print_signal_handler(st, SIGTRAP, buf, buflen);
2288 #endif
2289 }
2290 
2291 static char saved_jvm_path[MAXPATHLEN] = {0};
2292 
2293 // Find the full path to the current module, libjvm.so
2294 void os::jvm_path(char *buf, jint buflen) {
2295   // Error checking.
2296   if (buflen < MAXPATHLEN) {
2297     assert(false, "must use a large-enough buffer");
2298     buf[0] = '\0';
2299     return;
2300   }
2301   // Lazy resolve the path to current module.
2302   if (saved_jvm_path[0] != 0) {
2303     strcpy(buf, saved_jvm_path);
2304     return;
2305   }
2306 
2307   char dli_fname[MAXPATHLEN];
2308   bool ret = dll_address_to_library_name(
2309                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2310                                          dli_fname, sizeof(dli_fname), NULL);
2311   assert(ret, "cannot locate libjvm");
2312   char *rp = NULL;
2313   if (ret && dli_fname[0] != '\0') {
2314     rp = os::Posix::realpath(dli_fname, buf, buflen);
2315   }
2316   if (rp == NULL) {
2317     return;
2318   }
2319 
2320   if (Arguments::sun_java_launcher_is_altjvm()) {
2321     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2322     // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2323     // If "/jre/lib/" appears at the right place in the string, then
2324     // assume we are installed in a JDK and we're done. Otherwise, check
2325     // for a JAVA_HOME environment variable and fix up the path so it
2326     // looks like libjvm.so is installed there (append a fake suffix
2327     // hotspot/libjvm.so).
2328     const char *p = buf + strlen(buf) - 1;
2329     for (int count = 0; p > buf && count < 5; ++count) {
2330       for (--p; p > buf && *p != '/'; --p)
2331         /* empty */ ;
2332     }
2333 
2334     if (strncmp(p, "/jre/lib/", 9) != 0) {
2335       // Look for JAVA_HOME in the environment.
2336       char* java_home_var = ::getenv("JAVA_HOME");
2337       if (java_home_var != NULL && java_home_var[0] != 0) {
2338         char* jrelib_p;
2339         int len;
2340 
2341         // Check the current module name "libjvm.so".
2342         p = strrchr(buf, '/');
2343         if (p == NULL) {
2344           return;
2345         }
2346         assert(strstr(p, "/libjvm") == p, "invalid library name");
2347 
2348         rp = os::Posix::realpath(java_home_var, buf, buflen);
2349         if (rp == NULL) {
2350           return;
2351         }
2352 
2353         // determine if this is a legacy image or modules image
2354         // modules image doesn't have "jre" subdirectory
2355         len = strlen(buf);
2356         assert(len < buflen, "Ran out of buffer room");
2357         jrelib_p = buf + len;
2358         snprintf(jrelib_p, buflen-len, "/jre/lib");
2359         if (0 != access(buf, F_OK)) {
2360           snprintf(jrelib_p, buflen-len, "/lib");
2361         }
2362 
2363         if (0 == access(buf, F_OK)) {
2364           // Use current module name "libjvm.so"
2365           len = strlen(buf);
2366           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2367         } else {
2368           // Go back to path of .so
2369           rp = os::Posix::realpath(dli_fname, buf, buflen);
2370           if (rp == NULL) {
2371             return;
2372           }
2373         }
2374       }
2375     }
2376   }
2377 
2378   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2379   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2380 }
2381 
2382 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2383   // no prefix required, not even "_"
2384 }
2385 
2386 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2387   // no suffix required
2388 }
2389 
2390 ////////////////////////////////////////////////////////////////////////////////
2391 // sun.misc.Signal support
2392 
2393 static volatile jint sigint_count = 0;
2394 
2395 static void UserHandler(int sig, void *siginfo, void *context) {
2396   // 4511530 - sem_post is serialized and handled by the manager thread. When
2397   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2398   // don't want to flood the manager thread with sem_post requests.
2399   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2400     return;
2401   }
2402 
2403   // Ctrl-C is pressed during error reporting, likely because the error
2404   // handler fails to abort. Let VM die immediately.
2405   if (sig == SIGINT && is_error_reported()) {
2406     os::die();
2407   }
2408 
2409   os::signal_notify(sig);
2410 }
2411 
2412 void* os::user_handler() {
2413   return CAST_FROM_FN_PTR(void*, UserHandler);
2414 }
2415 
2416 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2417   struct timespec ts;
2418   // Semaphore's are always associated with CLOCK_REALTIME
2419   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2420   // see unpackTime for discussion on overflow checking
2421   if (sec >= MAX_SECS) {
2422     ts.tv_sec += MAX_SECS;
2423     ts.tv_nsec = 0;
2424   } else {
2425     ts.tv_sec += sec;
2426     ts.tv_nsec += nsec;
2427     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2428       ts.tv_nsec -= NANOSECS_PER_SEC;
2429       ++ts.tv_sec; // note: this must be <= max_secs
2430     }
2431   }
2432 
2433   return ts;
2434 }
2435 
2436 extern "C" {
2437   typedef void (*sa_handler_t)(int);
2438   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2439 }
2440 
2441 void* os::signal(int signal_number, void* handler) {
2442   struct sigaction sigAct, oldSigAct;
2443 
2444   sigfillset(&(sigAct.sa_mask));
2445   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2446   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2447 
2448   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2449     // -1 means registration failed
2450     return (void *)-1;
2451   }
2452 
2453   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2454 }
2455 
2456 void os::signal_raise(int signal_number) {
2457   ::raise(signal_number);
2458 }
2459 
2460 // The following code is moved from os.cpp for making this
2461 // code platform specific, which it is by its very nature.
2462 
2463 // Will be modified when max signal is changed to be dynamic
2464 int os::sigexitnum_pd() {
2465   return NSIG;
2466 }
2467 
2468 // a counter for each possible signal value
2469 static volatile jint pending_signals[NSIG+1] = { 0 };
2470 
2471 // Linux(POSIX) specific hand shaking semaphore.
2472 static sem_t sig_sem;
2473 static PosixSemaphore sr_semaphore;
2474 
2475 void os::signal_init_pd() {
2476   // Initialize signal structures
2477   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2478 
2479   // Initialize signal semaphore
2480   ::sem_init(&sig_sem, 0, 0);
2481 }
2482 
2483 void os::signal_notify(int sig) {
2484   Atomic::inc(&pending_signals[sig]);
2485   ::sem_post(&sig_sem);
2486 }
2487 
2488 static int check_pending_signals(bool wait) {
2489   Atomic::store(0, &sigint_count);
2490   for (;;) {
2491     for (int i = 0; i < NSIG + 1; i++) {
2492       jint n = pending_signals[i];
2493       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2494         return i;
2495       }
2496     }
2497     if (!wait) {
2498       return -1;
2499     }
2500     JavaThread *thread = JavaThread::current();
2501     ThreadBlockInVM tbivm(thread);
2502 
2503     bool threadIsSuspended;
2504     do {
2505       thread->set_suspend_equivalent();
2506       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2507       ::sem_wait(&sig_sem);
2508 
2509       // were we externally suspended while we were waiting?
2510       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2511       if (threadIsSuspended) {
2512         // The semaphore has been incremented, but while we were waiting
2513         // another thread suspended us. We don't want to continue running
2514         // while suspended because that would surprise the thread that
2515         // suspended us.
2516         ::sem_post(&sig_sem);
2517 
2518         thread->java_suspend_self();
2519       }
2520     } while (threadIsSuspended);
2521   }
2522 }
2523 
2524 int os::signal_lookup() {
2525   return check_pending_signals(false);
2526 }
2527 
2528 int os::signal_wait() {
2529   return check_pending_signals(true);
2530 }
2531 
2532 ////////////////////////////////////////////////////////////////////////////////
2533 // Virtual Memory
2534 
2535 int os::vm_page_size() {
2536   // Seems redundant as all get out
2537   assert(os::Linux::page_size() != -1, "must call os::init");
2538   return os::Linux::page_size();
2539 }
2540 
2541 // Solaris allocates memory by pages.
2542 int os::vm_allocation_granularity() {
2543   assert(os::Linux::page_size() != -1, "must call os::init");
2544   return os::Linux::page_size();
2545 }
2546 
2547 // Rationale behind this function:
2548 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2549 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2550 //  samples for JITted code. Here we create private executable mapping over the code cache
2551 //  and then we can use standard (well, almost, as mapping can change) way to provide
2552 //  info for the reporting script by storing timestamp and location of symbol
2553 void linux_wrap_code(char* base, size_t size) {
2554   static volatile jint cnt = 0;
2555 
2556   if (!UseOprofile) {
2557     return;
2558   }
2559 
2560   char buf[PATH_MAX+1];
2561   int num = Atomic::add(1, &cnt);
2562 
2563   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2564            os::get_temp_directory(), os::current_process_id(), num);
2565   unlink(buf);
2566 
2567   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2568 
2569   if (fd != -1) {
2570     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2571     if (rv != (off_t)-1) {
2572       if (::write(fd, "", 1) == 1) {
2573         mmap(base, size,
2574              PROT_READ|PROT_WRITE|PROT_EXEC,
2575              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2576       }
2577     }
2578     ::close(fd);
2579     unlink(buf);
2580   }
2581 }
2582 
2583 static bool recoverable_mmap_error(int err) {
2584   // See if the error is one we can let the caller handle. This
2585   // list of errno values comes from JBS-6843484. I can't find a
2586   // Linux man page that documents this specific set of errno
2587   // values so while this list currently matches Solaris, it may
2588   // change as we gain experience with this failure mode.
2589   switch (err) {
2590   case EBADF:
2591   case EINVAL:
2592   case ENOTSUP:
2593     // let the caller deal with these errors
2594     return true;
2595 
2596   default:
2597     // Any remaining errors on this OS can cause our reserved mapping
2598     // to be lost. That can cause confusion where different data
2599     // structures think they have the same memory mapped. The worst
2600     // scenario is if both the VM and a library think they have the
2601     // same memory mapped.
2602     return false;
2603   }
2604 }
2605 
2606 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2607                                     int err) {
2608   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2609           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2610           os::strerror(err), err);
2611 }
2612 
2613 static void warn_fail_commit_memory(char* addr, size_t size,
2614                                     size_t alignment_hint, bool exec,
2615                                     int err) {
2616   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2617           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2618           alignment_hint, exec, os::strerror(err), err);
2619 }
2620 
2621 // NOTE: Linux kernel does not really reserve the pages for us.
2622 //       All it does is to check if there are enough free pages
2623 //       left at the time of mmap(). This could be a potential
2624 //       problem.
2625 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2626   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2627   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2628                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2629   if (res != (uintptr_t) MAP_FAILED) {
2630     if (UseNUMAInterleaving) {
2631       numa_make_global(addr, size);
2632     }
2633     return 0;
2634   }
2635 
2636   int err = errno;  // save errno from mmap() call above
2637 
2638   if (!recoverable_mmap_error(err)) {
2639     warn_fail_commit_memory(addr, size, exec, err);
2640     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2641   }
2642 
2643   return err;
2644 }
2645 
2646 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2647   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2648 }
2649 
2650 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2651                                   const char* mesg) {
2652   assert(mesg != NULL, "mesg must be specified");
2653   int err = os::Linux::commit_memory_impl(addr, size, exec);
2654   if (err != 0) {
2655     // the caller wants all commit errors to exit with the specified mesg:
2656     warn_fail_commit_memory(addr, size, exec, err);
2657     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2658   }
2659 }
2660 
2661 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2662 #ifndef MAP_HUGETLB
2663   #define MAP_HUGETLB 0x40000
2664 #endif
2665 
2666 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2667 #ifndef MADV_HUGEPAGE
2668   #define MADV_HUGEPAGE 14
2669 #endif
2670 
2671 int os::Linux::commit_memory_impl(char* addr, size_t size,
2672                                   size_t alignment_hint, bool exec) {
2673   int err = os::Linux::commit_memory_impl(addr, size, exec);
2674   if (err == 0) {
2675     realign_memory(addr, size, alignment_hint);
2676   }
2677   return err;
2678 }
2679 
2680 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2681                           bool exec) {
2682   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2683 }
2684 
2685 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2686                                   size_t alignment_hint, bool exec,
2687                                   const char* mesg) {
2688   assert(mesg != NULL, "mesg must be specified");
2689   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2690   if (err != 0) {
2691     // the caller wants all commit errors to exit with the specified mesg:
2692     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2693     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2694   }
2695 }
2696 
2697 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2698   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2699     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2700     // be supported or the memory may already be backed by huge pages.
2701     ::madvise(addr, bytes, MADV_HUGEPAGE);
2702   }
2703 }
2704 
2705 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2706   // This method works by doing an mmap over an existing mmaping and effectively discarding
2707   // the existing pages. However it won't work for SHM-based large pages that cannot be
2708   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2709   // small pages on top of the SHM segment. This method always works for small pages, so we
2710   // allow that in any case.
2711   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2712     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2713   }
2714 }
2715 
2716 void os::numa_make_global(char *addr, size_t bytes) {
2717   Linux::numa_interleave_memory(addr, bytes);
2718 }
2719 
2720 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2721 // bind policy to MPOL_PREFERRED for the current thread.
2722 #define USE_MPOL_PREFERRED 0
2723 
2724 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2725   // To make NUMA and large pages more robust when both enabled, we need to ease
2726   // the requirements on where the memory should be allocated. MPOL_BIND is the
2727   // default policy and it will force memory to be allocated on the specified
2728   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2729   // the specified node, but will not force it. Using this policy will prevent
2730   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2731   // free large pages.
2732   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2733   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2734 }
2735 
2736 bool os::numa_topology_changed() { return false; }
2737 
2738 size_t os::numa_get_groups_num() {
2739   // Return just the number of nodes in which it's possible to allocate memory
2740   // (in numa terminology, configured nodes).
2741   return Linux::numa_num_configured_nodes();
2742 }
2743 
2744 int os::numa_get_group_id() {
2745   int cpu_id = Linux::sched_getcpu();
2746   if (cpu_id != -1) {
2747     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2748     if (lgrp_id != -1) {
2749       return lgrp_id;
2750     }
2751   }
2752   return 0;
2753 }
2754 
2755 int os::Linux::get_existing_num_nodes() {
2756   size_t node;
2757   size_t highest_node_number = Linux::numa_max_node();
2758   int num_nodes = 0;
2759 
2760   // Get the total number of nodes in the system including nodes without memory.
2761   for (node = 0; node <= highest_node_number; node++) {
2762     if (isnode_in_existing_nodes(node)) {
2763       num_nodes++;
2764     }
2765   }
2766   return num_nodes;
2767 }
2768 
2769 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2770   size_t highest_node_number = Linux::numa_max_node();
2771   size_t i = 0;
2772 
2773   // Map all node ids in which is possible to allocate memory. Also nodes are
2774   // not always consecutively available, i.e. available from 0 to the highest
2775   // node number.
2776   for (size_t node = 0; node <= highest_node_number; node++) {
2777     if (Linux::isnode_in_configured_nodes(node)) {
2778       ids[i++] = node;
2779     }
2780   }
2781   return i;
2782 }
2783 
2784 bool os::get_page_info(char *start, page_info* info) {
2785   return false;
2786 }
2787 
2788 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2789                      page_info* page_found) {
2790   return end;
2791 }
2792 
2793 
2794 int os::Linux::sched_getcpu_syscall(void) {
2795   unsigned int cpu = 0;
2796   int retval = -1;
2797 
2798 #if defined(IA32)
2799   #ifndef SYS_getcpu
2800     #define SYS_getcpu 318
2801   #endif
2802   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2803 #elif defined(AMD64)
2804 // Unfortunately we have to bring all these macros here from vsyscall.h
2805 // to be able to compile on old linuxes.
2806   #define __NR_vgetcpu 2
2807   #define VSYSCALL_START (-10UL << 20)
2808   #define VSYSCALL_SIZE 1024
2809   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2810   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2811   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2812   retval = vgetcpu(&cpu, NULL, NULL);
2813 #endif
2814 
2815   return (retval == -1) ? retval : cpu;
2816 }
2817 
2818 void os::Linux::sched_getcpu_init() {
2819   // sched_getcpu() should be in libc.
2820   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2821                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2822 
2823   // If it's not, try a direct syscall.
2824   if (sched_getcpu() == -1) {
2825     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2826                                     (void*)&sched_getcpu_syscall));
2827   }
2828 }
2829 
2830 // Something to do with the numa-aware allocator needs these symbols
2831 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2832 extern "C" JNIEXPORT void numa_error(char *where) { }
2833 
2834 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2835 // load symbol from base version instead.
2836 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2837   void *f = dlvsym(handle, name, "libnuma_1.1");
2838   if (f == NULL) {
2839     f = dlsym(handle, name);
2840   }
2841   return f;
2842 }
2843 
2844 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2845 // Return NULL if the symbol is not defined in this particular version.
2846 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2847   return dlvsym(handle, name, "libnuma_1.2");
2848 }
2849 
2850 bool os::Linux::libnuma_init() {
2851   if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2852     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2853     if (handle != NULL) {
2854       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2855                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2856       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2857                                        libnuma_dlsym(handle, "numa_max_node")));
2858       set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2859                                                    libnuma_dlsym(handle, "numa_num_configured_nodes")));
2860       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2861                                         libnuma_dlsym(handle, "numa_available")));
2862       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2863                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2864       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2865                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2866       set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2867                                                 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2868       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2869                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2870       set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2871                                                libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2872       set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2873                                        libnuma_dlsym(handle, "numa_distance")));
2874 
2875       if (numa_available() != -1) {
2876         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2877         set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2878         set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2879         // Create an index -> node mapping, since nodes are not always consecutive
2880         _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2881         rebuild_nindex_to_node_map();
2882         // Create a cpu -> node mapping
2883         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2884         rebuild_cpu_to_node_map();
2885         return true;
2886       }
2887     }
2888   }
2889   return false;
2890 }
2891 
2892 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2893   // Creating guard page is very expensive. Java thread has HotSpot
2894   // guard pages, only enable glibc guard page for non-Java threads.
2895   // (Remember: compiler thread is a Java thread, too!)
2896   return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2897 }
2898 
2899 void os::Linux::rebuild_nindex_to_node_map() {
2900   int highest_node_number = Linux::numa_max_node();
2901 
2902   nindex_to_node()->clear();
2903   for (int node = 0; node <= highest_node_number; node++) {
2904     if (Linux::isnode_in_existing_nodes(node)) {
2905       nindex_to_node()->append(node);
2906     }
2907   }
2908 }
2909 
2910 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2911 // The table is later used in get_node_by_cpu().
2912 void os::Linux::rebuild_cpu_to_node_map() {
2913   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2914                               // in libnuma (possible values are starting from 16,
2915                               // and continuing up with every other power of 2, but less
2916                               // than the maximum number of CPUs supported by kernel), and
2917                               // is a subject to change (in libnuma version 2 the requirements
2918                               // are more reasonable) we'll just hardcode the number they use
2919                               // in the library.
2920   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2921 
2922   size_t cpu_num = processor_count();
2923   size_t cpu_map_size = NCPUS / BitsPerCLong;
2924   size_t cpu_map_valid_size =
2925     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2926 
2927   cpu_to_node()->clear();
2928   cpu_to_node()->at_grow(cpu_num - 1);
2929 
2930   size_t node_num = get_existing_num_nodes();
2931 
2932   int distance = 0;
2933   int closest_distance = INT_MAX;
2934   int closest_node = 0;
2935   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2936   for (size_t i = 0; i < node_num; i++) {
2937     // Check if node is configured (not a memory-less node). If it is not, find
2938     // the closest configured node.
2939     if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
2940       closest_distance = INT_MAX;
2941       // Check distance from all remaining nodes in the system. Ignore distance
2942       // from itself and from another non-configured node.
2943       for (size_t m = 0; m < node_num; m++) {
2944         if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
2945           distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
2946           // If a closest node is found, update. There is always at least one
2947           // configured node in the system so there is always at least one node
2948           // close.
2949           if (distance != 0 && distance < closest_distance) {
2950             closest_distance = distance;
2951             closest_node = nindex_to_node()->at(m);
2952           }
2953         }
2954       }
2955      } else {
2956        // Current node is already a configured node.
2957        closest_node = nindex_to_node()->at(i);
2958      }
2959 
2960     // Get cpus from the original node and map them to the closest node. If node
2961     // is a configured node (not a memory-less node), then original node and
2962     // closest node are the same.
2963     if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2964       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2965         if (cpu_map[j] != 0) {
2966           for (size_t k = 0; k < BitsPerCLong; k++) {
2967             if (cpu_map[j] & (1UL << k)) {
2968               cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
2969             }
2970           }
2971         }
2972       }
2973     }
2974   }
2975   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2976 }
2977 
2978 int os::Linux::get_node_by_cpu(int cpu_id) {
2979   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2980     return cpu_to_node()->at(cpu_id);
2981   }
2982   return -1;
2983 }
2984 
2985 GrowableArray<int>* os::Linux::_cpu_to_node;
2986 GrowableArray<int>* os::Linux::_nindex_to_node;
2987 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2988 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2989 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2990 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
2991 os::Linux::numa_available_func_t os::Linux::_numa_available;
2992 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2993 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2994 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
2995 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2996 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
2997 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
2998 unsigned long* os::Linux::_numa_all_nodes;
2999 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3000 struct bitmask* os::Linux::_numa_nodes_ptr;
3001 
3002 bool os::pd_uncommit_memory(char* addr, size_t size) {
3003   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3004                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3005   return res  != (uintptr_t) MAP_FAILED;
3006 }
3007 
3008 static address get_stack_commited_bottom(address bottom, size_t size) {
3009   address nbot = bottom;
3010   address ntop = bottom + size;
3011 
3012   size_t page_sz = os::vm_page_size();
3013   unsigned pages = size / page_sz;
3014 
3015   unsigned char vec[1];
3016   unsigned imin = 1, imax = pages + 1, imid;
3017   int mincore_return_value = 0;
3018 
3019   assert(imin <= imax, "Unexpected page size");
3020 
3021   while (imin < imax) {
3022     imid = (imax + imin) / 2;
3023     nbot = ntop - (imid * page_sz);
3024 
3025     // Use a trick with mincore to check whether the page is mapped or not.
3026     // mincore sets vec to 1 if page resides in memory and to 0 if page
3027     // is swapped output but if page we are asking for is unmapped
3028     // it returns -1,ENOMEM
3029     mincore_return_value = mincore(nbot, page_sz, vec);
3030 
3031     if (mincore_return_value == -1) {
3032       // Page is not mapped go up
3033       // to find first mapped page
3034       if (errno != EAGAIN) {
3035         assert(errno == ENOMEM, "Unexpected mincore errno");
3036         imax = imid;
3037       }
3038     } else {
3039       // Page is mapped go down
3040       // to find first not mapped page
3041       imin = imid + 1;
3042     }
3043   }
3044 
3045   nbot = nbot + page_sz;
3046 
3047   // Adjust stack bottom one page up if last checked page is not mapped
3048   if (mincore_return_value == -1) {
3049     nbot = nbot + page_sz;
3050   }
3051 
3052   return nbot;
3053 }
3054 
3055 
3056 // Linux uses a growable mapping for the stack, and if the mapping for
3057 // the stack guard pages is not removed when we detach a thread the
3058 // stack cannot grow beyond the pages where the stack guard was
3059 // mapped.  If at some point later in the process the stack expands to
3060 // that point, the Linux kernel cannot expand the stack any further
3061 // because the guard pages are in the way, and a segfault occurs.
3062 //
3063 // However, it's essential not to split the stack region by unmapping
3064 // a region (leaving a hole) that's already part of the stack mapping,
3065 // so if the stack mapping has already grown beyond the guard pages at
3066 // the time we create them, we have to truncate the stack mapping.
3067 // So, we need to know the extent of the stack mapping when
3068 // create_stack_guard_pages() is called.
3069 
3070 // We only need this for stacks that are growable: at the time of
3071 // writing thread stacks don't use growable mappings (i.e. those
3072 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3073 // only applies to the main thread.
3074 
3075 // If the (growable) stack mapping already extends beyond the point
3076 // where we're going to put our guard pages, truncate the mapping at
3077 // that point by munmap()ping it.  This ensures that when we later
3078 // munmap() the guard pages we don't leave a hole in the stack
3079 // mapping. This only affects the main/initial thread
3080 
3081 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3082   if (os::Linux::is_initial_thread()) {
3083     // As we manually grow stack up to bottom inside create_attached_thread(),
3084     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3085     // we don't need to do anything special.
3086     // Check it first, before calling heavy function.
3087     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3088     unsigned char vec[1];
3089 
3090     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3091       // Fallback to slow path on all errors, including EAGAIN
3092       stack_extent = (uintptr_t) get_stack_commited_bottom(
3093                                                            os::Linux::initial_thread_stack_bottom(),
3094                                                            (size_t)addr - stack_extent);
3095     }
3096 
3097     if (stack_extent < (uintptr_t)addr) {
3098       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3099     }
3100   }
3101 
3102   return os::commit_memory(addr, size, !ExecMem);
3103 }
3104 
3105 // If this is a growable mapping, remove the guard pages entirely by
3106 // munmap()ping them.  If not, just call uncommit_memory(). This only
3107 // affects the main/initial thread, but guard against future OS changes
3108 // It's safe to always unmap guard pages for initial thread because we
3109 // always place it right after end of the mapped region
3110 
3111 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3112   uintptr_t stack_extent, stack_base;
3113 
3114   if (os::Linux::is_initial_thread()) {
3115     return ::munmap(addr, size) == 0;
3116   }
3117 
3118   return os::uncommit_memory(addr, size);
3119 }
3120 
3121 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3122 // at 'requested_addr'. If there are existing memory mappings at the same
3123 // location, however, they will be overwritten. If 'fixed' is false,
3124 // 'requested_addr' is only treated as a hint, the return value may or
3125 // may not start from the requested address. Unlike Linux mmap(), this
3126 // function returns NULL to indicate failure.
3127 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3128   char * addr;
3129   int flags;
3130 
3131   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3132   if (fixed) {
3133     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3134     flags |= MAP_FIXED;
3135   }
3136 
3137   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3138   // touch an uncommitted page. Otherwise, the read/write might
3139   // succeed if we have enough swap space to back the physical page.
3140   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3141                        flags, -1, 0);
3142 
3143   return addr == MAP_FAILED ? NULL : addr;
3144 }
3145 
3146 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3147 //   (req_addr != NULL) or with a given alignment.
3148 //  - bytes shall be a multiple of alignment.
3149 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3150 //  - alignment sets the alignment at which memory shall be allocated.
3151 //     It must be a multiple of allocation granularity.
3152 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3153 //  req_addr or NULL.
3154 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3155 
3156   size_t extra_size = bytes;
3157   if (req_addr == NULL && alignment > 0) {
3158     extra_size += alignment;
3159   }
3160 
3161   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3162     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3163     -1, 0);
3164   if (start == MAP_FAILED) {
3165     start = NULL;
3166   } else {
3167     if (req_addr != NULL) {
3168       if (start != req_addr) {
3169         ::munmap(start, extra_size);
3170         start = NULL;
3171       }
3172     } else {
3173       char* const start_aligned = align_ptr_up(start, alignment);
3174       char* const end_aligned = start_aligned + bytes;
3175       char* const end = start + extra_size;
3176       if (start_aligned > start) {
3177         ::munmap(start, start_aligned - start);
3178       }
3179       if (end_aligned < end) {
3180         ::munmap(end_aligned, end - end_aligned);
3181       }
3182       start = start_aligned;
3183     }
3184   }
3185   return start;
3186 }
3187 
3188 static int anon_munmap(char * addr, size_t size) {
3189   return ::munmap(addr, size) == 0;
3190 }
3191 
3192 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3193                             size_t alignment_hint) {
3194   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3195 }
3196 
3197 bool os::pd_release_memory(char* addr, size_t size) {
3198   return anon_munmap(addr, size);
3199 }
3200 
3201 static bool linux_mprotect(char* addr, size_t size, int prot) {
3202   // Linux wants the mprotect address argument to be page aligned.
3203   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3204 
3205   // According to SUSv3, mprotect() should only be used with mappings
3206   // established by mmap(), and mmap() always maps whole pages. Unaligned
3207   // 'addr' likely indicates problem in the VM (e.g. trying to change
3208   // protection of malloc'ed or statically allocated memory). Check the
3209   // caller if you hit this assert.
3210   assert(addr == bottom, "sanity check");
3211 
3212   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3213   return ::mprotect(bottom, size, prot) == 0;
3214 }
3215 
3216 // Set protections specified
3217 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3218                         bool is_committed) {
3219   unsigned int p = 0;
3220   switch (prot) {
3221   case MEM_PROT_NONE: p = PROT_NONE; break;
3222   case MEM_PROT_READ: p = PROT_READ; break;
3223   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3224   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3225   default:
3226     ShouldNotReachHere();
3227   }
3228   // is_committed is unused.
3229   return linux_mprotect(addr, bytes, p);
3230 }
3231 
3232 bool os::guard_memory(char* addr, size_t size) {
3233   return linux_mprotect(addr, size, PROT_NONE);
3234 }
3235 
3236 bool os::unguard_memory(char* addr, size_t size) {
3237   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3238 }
3239 
3240 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3241                                                     size_t page_size) {
3242   bool result = false;
3243   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3244                  MAP_ANONYMOUS|MAP_PRIVATE,
3245                  -1, 0);
3246   if (p != MAP_FAILED) {
3247     void *aligned_p = align_ptr_up(p, page_size);
3248 
3249     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3250 
3251     munmap(p, page_size * 2);
3252   }
3253 
3254   if (warn && !result) {
3255     warning("TransparentHugePages is not supported by the operating system.");
3256   }
3257 
3258   return result;
3259 }
3260 
3261 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3262   bool result = false;
3263   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3264                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3265                  -1, 0);
3266 
3267   if (p != MAP_FAILED) {
3268     // We don't know if this really is a huge page or not.
3269     FILE *fp = fopen("/proc/self/maps", "r");
3270     if (fp) {
3271       while (!feof(fp)) {
3272         char chars[257];
3273         long x = 0;
3274         if (fgets(chars, sizeof(chars), fp)) {
3275           if (sscanf(chars, "%lx-%*x", &x) == 1
3276               && x == (long)p) {
3277             if (strstr (chars, "hugepage")) {
3278               result = true;
3279               break;
3280             }
3281           }
3282         }
3283       }
3284       fclose(fp);
3285     }
3286     munmap(p, page_size);
3287   }
3288 
3289   if (warn && !result) {
3290     warning("HugeTLBFS is not supported by the operating system.");
3291   }
3292 
3293   return result;
3294 }
3295 
3296 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3297 //
3298 // From the coredump_filter documentation:
3299 //
3300 // - (bit 0) anonymous private memory
3301 // - (bit 1) anonymous shared memory
3302 // - (bit 2) file-backed private memory
3303 // - (bit 3) file-backed shared memory
3304 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3305 //           effective only if the bit 2 is cleared)
3306 // - (bit 5) hugetlb private memory
3307 // - (bit 6) hugetlb shared memory
3308 //
3309 static void set_coredump_filter(void) {
3310   FILE *f;
3311   long cdm;
3312 
3313   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3314     return;
3315   }
3316 
3317   if (fscanf(f, "%lx", &cdm) != 1) {
3318     fclose(f);
3319     return;
3320   }
3321 
3322   rewind(f);
3323 
3324   if ((cdm & LARGEPAGES_BIT) == 0) {
3325     cdm |= LARGEPAGES_BIT;
3326     fprintf(f, "%#lx", cdm);
3327   }
3328 
3329   fclose(f);
3330 }
3331 
3332 // Large page support
3333 
3334 static size_t _large_page_size = 0;
3335 
3336 size_t os::Linux::find_large_page_size() {
3337   size_t large_page_size = 0;
3338 
3339   // large_page_size on Linux is used to round up heap size. x86 uses either
3340   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3341   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3342   // page as large as 256M.
3343   //
3344   // Here we try to figure out page size by parsing /proc/meminfo and looking
3345   // for a line with the following format:
3346   //    Hugepagesize:     2048 kB
3347   //
3348   // If we can't determine the value (e.g. /proc is not mounted, or the text
3349   // format has been changed), we'll use the largest page size supported by
3350   // the processor.
3351 
3352 #ifndef ZERO
3353   large_page_size =
3354     AARCH64_ONLY(2 * M)
3355     AMD64_ONLY(2 * M)
3356     ARM32_ONLY(2 * M)
3357     IA32_ONLY(4 * M)
3358     IA64_ONLY(256 * M)
3359     PPC_ONLY(4 * M)
3360     S390_ONLY(1 * M)
3361     SPARC_ONLY(4 * M);
3362 #endif // ZERO
3363 
3364   FILE *fp = fopen("/proc/meminfo", "r");
3365   if (fp) {
3366     while (!feof(fp)) {
3367       int x = 0;
3368       char buf[16];
3369       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3370         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3371           large_page_size = x * K;
3372           break;
3373         }
3374       } else {
3375         // skip to next line
3376         for (;;) {
3377           int ch = fgetc(fp);
3378           if (ch == EOF || ch == (int)'\n') break;
3379         }
3380       }
3381     }
3382     fclose(fp);
3383   }
3384 
3385   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3386     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3387             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3388             proper_unit_for_byte_size(large_page_size));
3389   }
3390 
3391   return large_page_size;
3392 }
3393 
3394 size_t os::Linux::setup_large_page_size() {
3395   _large_page_size = Linux::find_large_page_size();
3396   const size_t default_page_size = (size_t)Linux::page_size();
3397   if (_large_page_size > default_page_size) {
3398     _page_sizes[0] = _large_page_size;
3399     _page_sizes[1] = default_page_size;
3400     _page_sizes[2] = 0;
3401   }
3402 
3403   return _large_page_size;
3404 }
3405 
3406 bool os::Linux::setup_large_page_type(size_t page_size) {
3407   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3408       FLAG_IS_DEFAULT(UseSHM) &&
3409       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3410 
3411     // The type of large pages has not been specified by the user.
3412 
3413     // Try UseHugeTLBFS and then UseSHM.
3414     UseHugeTLBFS = UseSHM = true;
3415 
3416     // Don't try UseTransparentHugePages since there are known
3417     // performance issues with it turned on. This might change in the future.
3418     UseTransparentHugePages = false;
3419   }
3420 
3421   if (UseTransparentHugePages) {
3422     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3423     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3424       UseHugeTLBFS = false;
3425       UseSHM = false;
3426       return true;
3427     }
3428     UseTransparentHugePages = false;
3429   }
3430 
3431   if (UseHugeTLBFS) {
3432     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3433     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3434       UseSHM = false;
3435       return true;
3436     }
3437     UseHugeTLBFS = false;
3438   }
3439 
3440   return UseSHM;
3441 }
3442 
3443 void os::large_page_init() {
3444   if (!UseLargePages &&
3445       !UseTransparentHugePages &&
3446       !UseHugeTLBFS &&
3447       !UseSHM) {
3448     // Not using large pages.
3449     return;
3450   }
3451 
3452   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3453     // The user explicitly turned off large pages.
3454     // Ignore the rest of the large pages flags.
3455     UseTransparentHugePages = false;
3456     UseHugeTLBFS = false;
3457     UseSHM = false;
3458     return;
3459   }
3460 
3461   size_t large_page_size = Linux::setup_large_page_size();
3462   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3463 
3464   set_coredump_filter();
3465 }
3466 
3467 #ifndef SHM_HUGETLB
3468   #define SHM_HUGETLB 04000
3469 #endif
3470 
3471 #define shm_warning_format(format, ...)              \
3472   do {                                               \
3473     if (UseLargePages &&                             \
3474         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3475          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3476          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3477       warning(format, __VA_ARGS__);                  \
3478     }                                                \
3479   } while (0)
3480 
3481 #define shm_warning(str) shm_warning_format("%s", str)
3482 
3483 #define shm_warning_with_errno(str)                \
3484   do {                                             \
3485     int err = errno;                               \
3486     shm_warning_format(str " (error = %d)", err);  \
3487   } while (0)
3488 
3489 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3490   assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3491 
3492   if (!is_size_aligned(alignment, SHMLBA)) {
3493     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3494     return NULL;
3495   }
3496 
3497   // To ensure that we get 'alignment' aligned memory from shmat,
3498   // we pre-reserve aligned virtual memory and then attach to that.
3499 
3500   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3501   if (pre_reserved_addr == NULL) {
3502     // Couldn't pre-reserve aligned memory.
3503     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3504     return NULL;
3505   }
3506 
3507   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3508   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3509 
3510   if ((intptr_t)addr == -1) {
3511     int err = errno;
3512     shm_warning_with_errno("Failed to attach shared memory.");
3513 
3514     assert(err != EACCES, "Unexpected error");
3515     assert(err != EIDRM,  "Unexpected error");
3516     assert(err != EINVAL, "Unexpected error");
3517 
3518     // Since we don't know if the kernel unmapped the pre-reserved memory area
3519     // we can't unmap it, since that would potentially unmap memory that was
3520     // mapped from other threads.
3521     return NULL;
3522   }
3523 
3524   return addr;
3525 }
3526 
3527 static char* shmat_at_address(int shmid, char* req_addr) {
3528   if (!is_ptr_aligned(req_addr, SHMLBA)) {
3529     assert(false, "Requested address needs to be SHMLBA aligned");
3530     return NULL;
3531   }
3532 
3533   char* addr = (char*)shmat(shmid, req_addr, 0);
3534 
3535   if ((intptr_t)addr == -1) {
3536     shm_warning_with_errno("Failed to attach shared memory.");
3537     return NULL;
3538   }
3539 
3540   return addr;
3541 }
3542 
3543 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3544   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3545   if (req_addr != NULL) {
3546     assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3547     assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3548     return shmat_at_address(shmid, req_addr);
3549   }
3550 
3551   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3552   // return large page size aligned memory addresses when req_addr == NULL.
3553   // However, if the alignment is larger than the large page size, we have
3554   // to manually ensure that the memory returned is 'alignment' aligned.
3555   if (alignment > os::large_page_size()) {
3556     assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3557     return shmat_with_alignment(shmid, bytes, alignment);
3558   } else {
3559     return shmat_at_address(shmid, NULL);
3560   }
3561 }
3562 
3563 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3564                                             char* req_addr, bool exec) {
3565   // "exec" is passed in but not used.  Creating the shared image for
3566   // the code cache doesn't have an SHM_X executable permission to check.
3567   assert(UseLargePages && UseSHM, "only for SHM large pages");
3568   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3569   assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3570 
3571   if (!is_size_aligned(bytes, os::large_page_size())) {
3572     return NULL; // Fallback to small pages.
3573   }
3574 
3575   // Create a large shared memory region to attach to based on size.
3576   // Currently, size is the total size of the heap.
3577   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3578   if (shmid == -1) {
3579     // Possible reasons for shmget failure:
3580     // 1. shmmax is too small for Java heap.
3581     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3582     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3583     // 2. not enough large page memory.
3584     //    > check available large pages: cat /proc/meminfo
3585     //    > increase amount of large pages:
3586     //          echo new_value > /proc/sys/vm/nr_hugepages
3587     //      Note 1: different Linux may use different name for this property,
3588     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3589     //      Note 2: it's possible there's enough physical memory available but
3590     //            they are so fragmented after a long run that they can't
3591     //            coalesce into large pages. Try to reserve large pages when
3592     //            the system is still "fresh".
3593     shm_warning_with_errno("Failed to reserve shared memory.");
3594     return NULL;
3595   }
3596 
3597   // Attach to the region.
3598   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3599 
3600   // Remove shmid. If shmat() is successful, the actual shared memory segment
3601   // will be deleted when it's detached by shmdt() or when the process
3602   // terminates. If shmat() is not successful this will remove the shared
3603   // segment immediately.
3604   shmctl(shmid, IPC_RMID, NULL);
3605 
3606   return addr;
3607 }
3608 
3609 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3610                                         int error) {
3611   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3612 
3613   bool warn_on_failure = UseLargePages &&
3614       (!FLAG_IS_DEFAULT(UseLargePages) ||
3615        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3616        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3617 
3618   if (warn_on_failure) {
3619     char msg[128];
3620     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3621                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3622     warning("%s", msg);
3623   }
3624 }
3625 
3626 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3627                                                         char* req_addr,
3628                                                         bool exec) {
3629   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3630   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3631   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3632 
3633   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3634   char* addr = (char*)::mmap(req_addr, bytes, prot,
3635                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3636                              -1, 0);
3637 
3638   if (addr == MAP_FAILED) {
3639     warn_on_large_pages_failure(req_addr, bytes, errno);
3640     return NULL;
3641   }
3642 
3643   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3644 
3645   return addr;
3646 }
3647 
3648 // Reserve memory using mmap(MAP_HUGETLB).
3649 //  - bytes shall be a multiple of alignment.
3650 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3651 //  - alignment sets the alignment at which memory shall be allocated.
3652 //     It must be a multiple of allocation granularity.
3653 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3654 //  req_addr or NULL.
3655 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3656                                                          size_t alignment,
3657                                                          char* req_addr,
3658                                                          bool exec) {
3659   size_t large_page_size = os::large_page_size();
3660   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3661 
3662   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3663   assert(is_size_aligned(bytes, alignment), "Must be");
3664 
3665   // First reserve - but not commit - the address range in small pages.
3666   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3667 
3668   if (start == NULL) {
3669     return NULL;
3670   }
3671 
3672   assert(is_ptr_aligned(start, alignment), "Must be");
3673 
3674   char* end = start + bytes;
3675 
3676   // Find the regions of the allocated chunk that can be promoted to large pages.
3677   char* lp_start = align_ptr_up(start, large_page_size);
3678   char* lp_end   = align_ptr_down(end, large_page_size);
3679 
3680   size_t lp_bytes = lp_end - lp_start;
3681 
3682   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3683 
3684   if (lp_bytes == 0) {
3685     // The mapped region doesn't even span the start and the end of a large page.
3686     // Fall back to allocate a non-special area.
3687     ::munmap(start, end - start);
3688     return NULL;
3689   }
3690 
3691   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3692 
3693   void* result;
3694 
3695   // Commit small-paged leading area.
3696   if (start != lp_start) {
3697     result = ::mmap(start, lp_start - start, prot,
3698                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3699                     -1, 0);
3700     if (result == MAP_FAILED) {
3701       ::munmap(lp_start, end - lp_start);
3702       return NULL;
3703     }
3704   }
3705 
3706   // Commit large-paged area.
3707   result = ::mmap(lp_start, lp_bytes, prot,
3708                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3709                   -1, 0);
3710   if (result == MAP_FAILED) {
3711     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3712     // If the mmap above fails, the large pages region will be unmapped and we
3713     // have regions before and after with small pages. Release these regions.
3714     //
3715     // |  mapped  |  unmapped  |  mapped  |
3716     // ^          ^            ^          ^
3717     // start      lp_start     lp_end     end
3718     //
3719     ::munmap(start, lp_start - start);
3720     ::munmap(lp_end, end - lp_end);
3721     return NULL;
3722   }
3723 
3724   // Commit small-paged trailing area.
3725   if (lp_end != end) {
3726     result = ::mmap(lp_end, end - lp_end, prot,
3727                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3728                     -1, 0);
3729     if (result == MAP_FAILED) {
3730       ::munmap(start, lp_end - start);
3731       return NULL;
3732     }
3733   }
3734 
3735   return start;
3736 }
3737 
3738 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3739                                                    size_t alignment,
3740                                                    char* req_addr,
3741                                                    bool exec) {
3742   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3743   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3744   assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3745   assert(is_power_of_2(os::large_page_size()), "Must be");
3746   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3747 
3748   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3749     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3750   } else {
3751     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3752   }
3753 }
3754 
3755 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3756                                  char* req_addr, bool exec) {
3757   assert(UseLargePages, "only for large pages");
3758 
3759   char* addr;
3760   if (UseSHM) {
3761     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3762   } else {
3763     assert(UseHugeTLBFS, "must be");
3764     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3765   }
3766 
3767   if (addr != NULL) {
3768     if (UseNUMAInterleaving) {
3769       numa_make_global(addr, bytes);
3770     }
3771 
3772     // The memory is committed
3773     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3774   }
3775 
3776   return addr;
3777 }
3778 
3779 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3780   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3781   return shmdt(base) == 0;
3782 }
3783 
3784 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3785   return pd_release_memory(base, bytes);
3786 }
3787 
3788 bool os::release_memory_special(char* base, size_t bytes) {
3789   bool res;
3790   if (MemTracker::tracking_level() > NMT_minimal) {
3791     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3792     res = os::Linux::release_memory_special_impl(base, bytes);
3793     if (res) {
3794       tkr.record((address)base, bytes);
3795     }
3796 
3797   } else {
3798     res = os::Linux::release_memory_special_impl(base, bytes);
3799   }
3800   return res;
3801 }
3802 
3803 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3804   assert(UseLargePages, "only for large pages");
3805   bool res;
3806 
3807   if (UseSHM) {
3808     res = os::Linux::release_memory_special_shm(base, bytes);
3809   } else {
3810     assert(UseHugeTLBFS, "must be");
3811     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3812   }
3813   return res;
3814 }
3815 
3816 size_t os::large_page_size() {
3817   return _large_page_size;
3818 }
3819 
3820 // With SysV SHM the entire memory region must be allocated as shared
3821 // memory.
3822 // HugeTLBFS allows application to commit large page memory on demand.
3823 // However, when committing memory with HugeTLBFS fails, the region
3824 // that was supposed to be committed will lose the old reservation
3825 // and allow other threads to steal that memory region. Because of this
3826 // behavior we can't commit HugeTLBFS memory.
3827 bool os::can_commit_large_page_memory() {
3828   return UseTransparentHugePages;
3829 }
3830 
3831 bool os::can_execute_large_page_memory() {
3832   return UseTransparentHugePages || UseHugeTLBFS;
3833 }
3834 
3835 // Reserve memory at an arbitrary address, only if that area is
3836 // available (and not reserved for something else).
3837 
3838 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3839   const int max_tries = 10;
3840   char* base[max_tries];
3841   size_t size[max_tries];
3842   const size_t gap = 0x000000;
3843 
3844   // Assert only that the size is a multiple of the page size, since
3845   // that's all that mmap requires, and since that's all we really know
3846   // about at this low abstraction level.  If we need higher alignment,
3847   // we can either pass an alignment to this method or verify alignment
3848   // in one of the methods further up the call chain.  See bug 5044738.
3849   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3850 
3851   // Repeatedly allocate blocks until the block is allocated at the
3852   // right spot.
3853 
3854   // Linux mmap allows caller to pass an address as hint; give it a try first,
3855   // if kernel honors the hint then we can return immediately.
3856   char * addr = anon_mmap(requested_addr, bytes, false);
3857   if (addr == requested_addr) {
3858     return requested_addr;
3859   }
3860 
3861   if (addr != NULL) {
3862     // mmap() is successful but it fails to reserve at the requested address
3863     anon_munmap(addr, bytes);
3864   }
3865 
3866   int i;
3867   for (i = 0; i < max_tries; ++i) {
3868     base[i] = reserve_memory(bytes);
3869 
3870     if (base[i] != NULL) {
3871       // Is this the block we wanted?
3872       if (base[i] == requested_addr) {
3873         size[i] = bytes;
3874         break;
3875       }
3876 
3877       // Does this overlap the block we wanted? Give back the overlapped
3878       // parts and try again.
3879 
3880       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3881       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3882         unmap_memory(base[i], top_overlap);
3883         base[i] += top_overlap;
3884         size[i] = bytes - top_overlap;
3885       } else {
3886         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3887         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3888           unmap_memory(requested_addr, bottom_overlap);
3889           size[i] = bytes - bottom_overlap;
3890         } else {
3891           size[i] = bytes;
3892         }
3893       }
3894     }
3895   }
3896 
3897   // Give back the unused reserved pieces.
3898 
3899   for (int j = 0; j < i; ++j) {
3900     if (base[j] != NULL) {
3901       unmap_memory(base[j], size[j]);
3902     }
3903   }
3904 
3905   if (i < max_tries) {
3906     return requested_addr;
3907   } else {
3908     return NULL;
3909   }
3910 }
3911 
3912 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3913   return ::read(fd, buf, nBytes);
3914 }
3915 
3916 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3917   return ::pread(fd, buf, nBytes, offset);
3918 }
3919 
3920 // Short sleep, direct OS call.
3921 //
3922 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3923 // sched_yield(2) will actually give up the CPU:
3924 //
3925 //   * Alone on this pariticular CPU, keeps running.
3926 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3927 //     (pre 2.6.39).
3928 //
3929 // So calling this with 0 is an alternative.
3930 //
3931 void os::naked_short_sleep(jlong ms) {
3932   struct timespec req;
3933 
3934   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3935   req.tv_sec = 0;
3936   if (ms > 0) {
3937     req.tv_nsec = (ms % 1000) * 1000000;
3938   } else {
3939     req.tv_nsec = 1;
3940   }
3941 
3942   nanosleep(&req, NULL);
3943 
3944   return;
3945 }
3946 
3947 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3948 void os::infinite_sleep() {
3949   while (true) {    // sleep forever ...
3950     ::sleep(100);   // ... 100 seconds at a time
3951   }
3952 }
3953 
3954 // Used to convert frequent JVM_Yield() to nops
3955 bool os::dont_yield() {
3956   return DontYieldALot;
3957 }
3958 
3959 void os::naked_yield() {
3960   sched_yield();
3961 }
3962 
3963 ////////////////////////////////////////////////////////////////////////////////
3964 // thread priority support
3965 
3966 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3967 // only supports dynamic priority, static priority must be zero. For real-time
3968 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3969 // However, for large multi-threaded applications, SCHED_RR is not only slower
3970 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3971 // of 5 runs - Sep 2005).
3972 //
3973 // The following code actually changes the niceness of kernel-thread/LWP. It
3974 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3975 // not the entire user process, and user level threads are 1:1 mapped to kernel
3976 // threads. It has always been the case, but could change in the future. For
3977 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3978 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3979 
3980 int os::java_to_os_priority[CriticalPriority + 1] = {
3981   19,              // 0 Entry should never be used
3982 
3983    4,              // 1 MinPriority
3984    3,              // 2
3985    2,              // 3
3986 
3987    1,              // 4
3988    0,              // 5 NormPriority
3989   -1,              // 6
3990 
3991   -2,              // 7
3992   -3,              // 8
3993   -4,              // 9 NearMaxPriority
3994 
3995   -5,              // 10 MaxPriority
3996 
3997   -5               // 11 CriticalPriority
3998 };
3999 
4000 static int prio_init() {
4001   if (ThreadPriorityPolicy == 1) {
4002     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4003     // if effective uid is not root. Perhaps, a more elegant way of doing
4004     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4005     if (geteuid() != 0) {
4006       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4007         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4008       }
4009       ThreadPriorityPolicy = 0;
4010     }
4011   }
4012   if (UseCriticalJavaThreadPriority) {
4013     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4014   }
4015   return 0;
4016 }
4017 
4018 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4019   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4020 
4021   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4022   return (ret == 0) ? OS_OK : OS_ERR;
4023 }
4024 
4025 OSReturn os::get_native_priority(const Thread* const thread,
4026                                  int *priority_ptr) {
4027   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4028     *priority_ptr = java_to_os_priority[NormPriority];
4029     return OS_OK;
4030   }
4031 
4032   errno = 0;
4033   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4034   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4035 }
4036 
4037 // Hint to the underlying OS that a task switch would not be good.
4038 // Void return because it's a hint and can fail.
4039 void os::hint_no_preempt() {}
4040 
4041 ////////////////////////////////////////////////////////////////////////////////
4042 // suspend/resume support
4043 
4044 //  the low-level signal-based suspend/resume support is a remnant from the
4045 //  old VM-suspension that used to be for java-suspension, safepoints etc,
4046 //  within hotspot. Now there is a single use-case for this:
4047 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
4048 //      that runs in the watcher thread.
4049 //  The remaining code is greatly simplified from the more general suspension
4050 //  code that used to be used.
4051 //
4052 //  The protocol is quite simple:
4053 //  - suspend:
4054 //      - sends a signal to the target thread
4055 //      - polls the suspend state of the osthread using a yield loop
4056 //      - target thread signal handler (SR_handler) sets suspend state
4057 //        and blocks in sigsuspend until continued
4058 //  - resume:
4059 //      - sets target osthread state to continue
4060 //      - sends signal to end the sigsuspend loop in the SR_handler
4061 //
4062 //  Note that the SR_lock plays no role in this suspend/resume protocol,
4063 //  but is checked for NULL in SR_handler as a thread termination indicator.
4064 
4065 static void resume_clear_context(OSThread *osthread) {
4066   osthread->set_ucontext(NULL);
4067   osthread->set_siginfo(NULL);
4068 }
4069 
4070 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4071                                  ucontext_t* context) {
4072   osthread->set_ucontext(context);
4073   osthread->set_siginfo(siginfo);
4074 }
4075 
4076 // Handler function invoked when a thread's execution is suspended or
4077 // resumed. We have to be careful that only async-safe functions are
4078 // called here (Note: most pthread functions are not async safe and
4079 // should be avoided.)
4080 //
4081 // Note: sigwait() is a more natural fit than sigsuspend() from an
4082 // interface point of view, but sigwait() prevents the signal hander
4083 // from being run. libpthread would get very confused by not having
4084 // its signal handlers run and prevents sigwait()'s use with the
4085 // mutex granting granting signal.
4086 //
4087 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4088 //
4089 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4090   // Save and restore errno to avoid confusing native code with EINTR
4091   // after sigsuspend.
4092   int old_errno = errno;
4093 
4094   Thread* thread = Thread::current_or_null_safe();
4095   assert(thread != NULL, "Missing current thread in SR_handler");
4096 
4097   // On some systems we have seen signal delivery get "stuck" until the signal
4098   // mask is changed as part of thread termination. Check that the current thread
4099   // has not already terminated (via SR_lock()) - else the following assertion
4100   // will fail because the thread is no longer a JavaThread as the ~JavaThread
4101   // destructor has completed.
4102 
4103   if (thread->SR_lock() == NULL) {
4104     return;
4105   }
4106 
4107   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4108 
4109   OSThread* osthread = thread->osthread();
4110 
4111   os::SuspendResume::State current = osthread->sr.state();
4112   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4113     suspend_save_context(osthread, siginfo, context);
4114 
4115     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4116     os::SuspendResume::State state = osthread->sr.suspended();
4117     if (state == os::SuspendResume::SR_SUSPENDED) {
4118       sigset_t suspend_set;  // signals for sigsuspend()
4119       sigemptyset(&suspend_set);
4120       // get current set of blocked signals and unblock resume signal
4121       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4122       sigdelset(&suspend_set, SR_signum);
4123 
4124       sr_semaphore.signal();
4125       // wait here until we are resumed
4126       while (1) {
4127         sigsuspend(&suspend_set);
4128 
4129         os::SuspendResume::State result = osthread->sr.running();
4130         if (result == os::SuspendResume::SR_RUNNING) {
4131           sr_semaphore.signal();
4132           break;
4133         }
4134       }
4135 
4136     } else if (state == os::SuspendResume::SR_RUNNING) {
4137       // request was cancelled, continue
4138     } else {
4139       ShouldNotReachHere();
4140     }
4141 
4142     resume_clear_context(osthread);
4143   } else if (current == os::SuspendResume::SR_RUNNING) {
4144     // request was cancelled, continue
4145   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4146     // ignore
4147   } else {
4148     // ignore
4149   }
4150 
4151   errno = old_errno;
4152 }
4153 
4154 static int SR_initialize() {
4155   struct sigaction act;
4156   char *s;
4157 
4158   // Get signal number to use for suspend/resume
4159   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4160     int sig = ::strtol(s, 0, 10);
4161     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4162         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4163       SR_signum = sig;
4164     } else {
4165       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4166               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4167     }
4168   }
4169 
4170   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4171          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4172 
4173   sigemptyset(&SR_sigset);
4174   sigaddset(&SR_sigset, SR_signum);
4175 
4176   // Set up signal handler for suspend/resume
4177   act.sa_flags = SA_RESTART|SA_SIGINFO;
4178   act.sa_handler = (void (*)(int)) SR_handler;
4179 
4180   // SR_signum is blocked by default.
4181   // 4528190 - We also need to block pthread restart signal (32 on all
4182   // supported Linux platforms). Note that LinuxThreads need to block
4183   // this signal for all threads to work properly. So we don't have
4184   // to use hard-coded signal number when setting up the mask.
4185   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4186 
4187   if (sigaction(SR_signum, &act, 0) == -1) {
4188     return -1;
4189   }
4190 
4191   // Save signal flag
4192   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4193   return 0;
4194 }
4195 
4196 static int sr_notify(OSThread* osthread) {
4197   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4198   assert_status(status == 0, status, "pthread_kill");
4199   return status;
4200 }
4201 
4202 // "Randomly" selected value for how long we want to spin
4203 // before bailing out on suspending a thread, also how often
4204 // we send a signal to a thread we want to resume
4205 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4206 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4207 
4208 // returns true on success and false on error - really an error is fatal
4209 // but this seems the normal response to library errors
4210 static bool do_suspend(OSThread* osthread) {
4211   assert(osthread->sr.is_running(), "thread should be running");
4212   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4213 
4214   // mark as suspended and send signal
4215   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4216     // failed to switch, state wasn't running?
4217     ShouldNotReachHere();
4218     return false;
4219   }
4220 
4221   if (sr_notify(osthread) != 0) {
4222     ShouldNotReachHere();
4223   }
4224 
4225   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4226   while (true) {
4227     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4228       break;
4229     } else {
4230       // timeout
4231       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4232       if (cancelled == os::SuspendResume::SR_RUNNING) {
4233         return false;
4234       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4235         // make sure that we consume the signal on the semaphore as well
4236         sr_semaphore.wait();
4237         break;
4238       } else {
4239         ShouldNotReachHere();
4240         return false;
4241       }
4242     }
4243   }
4244 
4245   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4246   return true;
4247 }
4248 
4249 static void do_resume(OSThread* osthread) {
4250   assert(osthread->sr.is_suspended(), "thread should be suspended");
4251   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4252 
4253   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4254     // failed to switch to WAKEUP_REQUEST
4255     ShouldNotReachHere();
4256     return;
4257   }
4258 
4259   while (true) {
4260     if (sr_notify(osthread) == 0) {
4261       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4262         if (osthread->sr.is_running()) {
4263           return;
4264         }
4265       }
4266     } else {
4267       ShouldNotReachHere();
4268     }
4269   }
4270 
4271   guarantee(osthread->sr.is_running(), "Must be running!");
4272 }
4273 
4274 ///////////////////////////////////////////////////////////////////////////////////
4275 // signal handling (except suspend/resume)
4276 
4277 // This routine may be used by user applications as a "hook" to catch signals.
4278 // The user-defined signal handler must pass unrecognized signals to this
4279 // routine, and if it returns true (non-zero), then the signal handler must
4280 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4281 // routine will never retun false (zero), but instead will execute a VM panic
4282 // routine kill the process.
4283 //
4284 // If this routine returns false, it is OK to call it again.  This allows
4285 // the user-defined signal handler to perform checks either before or after
4286 // the VM performs its own checks.  Naturally, the user code would be making
4287 // a serious error if it tried to handle an exception (such as a null check
4288 // or breakpoint) that the VM was generating for its own correct operation.
4289 //
4290 // This routine may recognize any of the following kinds of signals:
4291 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4292 // It should be consulted by handlers for any of those signals.
4293 //
4294 // The caller of this routine must pass in the three arguments supplied
4295 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4296 // field of the structure passed to sigaction().  This routine assumes that
4297 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4298 //
4299 // Note that the VM will print warnings if it detects conflicting signal
4300 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4301 //
4302 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4303                                                  siginfo_t* siginfo,
4304                                                  void* ucontext,
4305                                                  int abort_if_unrecognized);
4306 
4307 void signalHandler(int sig, siginfo_t* info, void* uc) {
4308   assert(info != NULL && uc != NULL, "it must be old kernel");
4309   int orig_errno = errno;  // Preserve errno value over signal handler.
4310   JVM_handle_linux_signal(sig, info, uc, true);
4311   errno = orig_errno;
4312 }
4313 
4314 
4315 // This boolean allows users to forward their own non-matching signals
4316 // to JVM_handle_linux_signal, harmlessly.
4317 bool os::Linux::signal_handlers_are_installed = false;
4318 
4319 // For signal-chaining
4320 struct sigaction sigact[NSIG];
4321 uint64_t sigs = 0;
4322 #if (64 < NSIG-1)
4323 #error "Not all signals can be encoded in sigs. Adapt its type!"
4324 #endif
4325 bool os::Linux::libjsig_is_loaded = false;
4326 typedef struct sigaction *(*get_signal_t)(int);
4327 get_signal_t os::Linux::get_signal_action = NULL;
4328 
4329 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4330   struct sigaction *actp = NULL;
4331 
4332   if (libjsig_is_loaded) {
4333     // Retrieve the old signal handler from libjsig
4334     actp = (*get_signal_action)(sig);
4335   }
4336   if (actp == NULL) {
4337     // Retrieve the preinstalled signal handler from jvm
4338     actp = get_preinstalled_handler(sig);
4339   }
4340 
4341   return actp;
4342 }
4343 
4344 static bool call_chained_handler(struct sigaction *actp, int sig,
4345                                  siginfo_t *siginfo, void *context) {
4346   // Call the old signal handler
4347   if (actp->sa_handler == SIG_DFL) {
4348     // It's more reasonable to let jvm treat it as an unexpected exception
4349     // instead of taking the default action.
4350     return false;
4351   } else if (actp->sa_handler != SIG_IGN) {
4352     if ((actp->sa_flags & SA_NODEFER) == 0) {
4353       // automaticlly block the signal
4354       sigaddset(&(actp->sa_mask), sig);
4355     }
4356 
4357     sa_handler_t hand = NULL;
4358     sa_sigaction_t sa = NULL;
4359     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4360     // retrieve the chained handler
4361     if (siginfo_flag_set) {
4362       sa = actp->sa_sigaction;
4363     } else {
4364       hand = actp->sa_handler;
4365     }
4366 
4367     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4368       actp->sa_handler = SIG_DFL;
4369     }
4370 
4371     // try to honor the signal mask
4372     sigset_t oset;
4373     sigemptyset(&oset);
4374     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4375 
4376     // call into the chained handler
4377     if (siginfo_flag_set) {
4378       (*sa)(sig, siginfo, context);
4379     } else {
4380       (*hand)(sig);
4381     }
4382 
4383     // restore the signal mask
4384     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4385   }
4386   // Tell jvm's signal handler the signal is taken care of.
4387   return true;
4388 }
4389 
4390 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4391   bool chained = false;
4392   // signal-chaining
4393   if (UseSignalChaining) {
4394     struct sigaction *actp = get_chained_signal_action(sig);
4395     if (actp != NULL) {
4396       chained = call_chained_handler(actp, sig, siginfo, context);
4397     }
4398   }
4399   return chained;
4400 }
4401 
4402 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4403   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4404     return &sigact[sig];
4405   }
4406   return NULL;
4407 }
4408 
4409 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4410   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4411   sigact[sig] = oldAct;
4412   sigs |= (uint64_t)1 << (sig-1);
4413 }
4414 
4415 // for diagnostic
4416 int sigflags[NSIG];
4417 
4418 int os::Linux::get_our_sigflags(int sig) {
4419   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4420   return sigflags[sig];
4421 }
4422 
4423 void os::Linux::set_our_sigflags(int sig, int flags) {
4424   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4425   if (sig > 0 && sig < NSIG) {
4426     sigflags[sig] = flags;
4427   }
4428 }
4429 
4430 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4431   // Check for overwrite.
4432   struct sigaction oldAct;
4433   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4434 
4435   void* oldhand = oldAct.sa_sigaction
4436                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4437                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4438   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4439       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4440       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4441     if (AllowUserSignalHandlers || !set_installed) {
4442       // Do not overwrite; user takes responsibility to forward to us.
4443       return;
4444     } else if (UseSignalChaining) {
4445       // save the old handler in jvm
4446       save_preinstalled_handler(sig, oldAct);
4447       // libjsig also interposes the sigaction() call below and saves the
4448       // old sigaction on it own.
4449     } else {
4450       fatal("Encountered unexpected pre-existing sigaction handler "
4451             "%#lx for signal %d.", (long)oldhand, sig);
4452     }
4453   }
4454 
4455   struct sigaction sigAct;
4456   sigfillset(&(sigAct.sa_mask));
4457   sigAct.sa_handler = SIG_DFL;
4458   if (!set_installed) {
4459     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4460   } else {
4461     sigAct.sa_sigaction = signalHandler;
4462     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4463   }
4464   // Save flags, which are set by ours
4465   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4466   sigflags[sig] = sigAct.sa_flags;
4467 
4468   int ret = sigaction(sig, &sigAct, &oldAct);
4469   assert(ret == 0, "check");
4470 
4471   void* oldhand2  = oldAct.sa_sigaction
4472                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4473                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4474   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4475 }
4476 
4477 // install signal handlers for signals that HotSpot needs to
4478 // handle in order to support Java-level exception handling.
4479 
4480 void os::Linux::install_signal_handlers() {
4481   if (!signal_handlers_are_installed) {
4482     signal_handlers_are_installed = true;
4483 
4484     // signal-chaining
4485     typedef void (*signal_setting_t)();
4486     signal_setting_t begin_signal_setting = NULL;
4487     signal_setting_t end_signal_setting = NULL;
4488     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4489                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4490     if (begin_signal_setting != NULL) {
4491       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4492                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4493       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4494                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4495       libjsig_is_loaded = true;
4496       assert(UseSignalChaining, "should enable signal-chaining");
4497     }
4498     if (libjsig_is_loaded) {
4499       // Tell libjsig jvm is setting signal handlers
4500       (*begin_signal_setting)();
4501     }
4502 
4503     set_signal_handler(SIGSEGV, true);
4504     set_signal_handler(SIGPIPE, true);
4505     set_signal_handler(SIGBUS, true);
4506     set_signal_handler(SIGILL, true);
4507     set_signal_handler(SIGFPE, true);
4508 #if defined(PPC64)
4509     set_signal_handler(SIGTRAP, true);
4510 #endif
4511     set_signal_handler(SIGXFSZ, true);
4512 
4513     if (libjsig_is_loaded) {
4514       // Tell libjsig jvm finishes setting signal handlers
4515       (*end_signal_setting)();
4516     }
4517 
4518     // We don't activate signal checker if libjsig is in place, we trust ourselves
4519     // and if UserSignalHandler is installed all bets are off.
4520     // Log that signal checking is off only if -verbose:jni is specified.
4521     if (CheckJNICalls) {
4522       if (libjsig_is_loaded) {
4523         if (PrintJNIResolving) {
4524           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4525         }
4526         check_signals = false;
4527       }
4528       if (AllowUserSignalHandlers) {
4529         if (PrintJNIResolving) {
4530           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4531         }
4532         check_signals = false;
4533       }
4534     }
4535   }
4536 }
4537 
4538 // This is the fastest way to get thread cpu time on Linux.
4539 // Returns cpu time (user+sys) for any thread, not only for current.
4540 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4541 // It might work on 2.6.10+ with a special kernel/glibc patch.
4542 // For reference, please, see IEEE Std 1003.1-2004:
4543 //   http://www.unix.org/single_unix_specification
4544 
4545 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4546   struct timespec tp;
4547   int rc = os::Linux::clock_gettime(clockid, &tp);
4548   assert(rc == 0, "clock_gettime is expected to return 0 code");
4549 
4550   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4551 }
4552 
4553 void os::Linux::initialize_os_info() {
4554   assert(_os_version == 0, "OS info already initialized");
4555 
4556   struct utsname _uname;
4557 
4558   uint32_t major;
4559   uint32_t minor;
4560   uint32_t fix;
4561 
4562   int rc;
4563 
4564   // Kernel version is unknown if
4565   // verification below fails.
4566   _os_version = 0x01000000;
4567 
4568   rc = uname(&_uname);
4569   if (rc != -1) {
4570 
4571     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4572     if (rc == 3) {
4573 
4574       if (major < 256 && minor < 256 && fix < 256) {
4575         // Kernel version format is as expected,
4576         // set it overriding unknown state.
4577         _os_version = (major << 16) |
4578                       (minor << 8 ) |
4579                       (fix   << 0 ) ;
4580       }
4581     }
4582   }
4583 }
4584 
4585 uint32_t os::Linux::os_version() {
4586   assert(_os_version != 0, "not initialized");
4587   return _os_version & 0x00FFFFFF;
4588 }
4589 
4590 bool os::Linux::os_version_is_known() {
4591   assert(_os_version != 0, "not initialized");
4592   return _os_version & 0x01000000 ? false : true;
4593 }
4594 
4595 /////
4596 // glibc on Linux platform uses non-documented flag
4597 // to indicate, that some special sort of signal
4598 // trampoline is used.
4599 // We will never set this flag, and we should
4600 // ignore this flag in our diagnostic
4601 #ifdef SIGNIFICANT_SIGNAL_MASK
4602   #undef SIGNIFICANT_SIGNAL_MASK
4603 #endif
4604 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4605 
4606 static const char* get_signal_handler_name(address handler,
4607                                            char* buf, int buflen) {
4608   int offset = 0;
4609   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4610   if (found) {
4611     // skip directory names
4612     const char *p1, *p2;
4613     p1 = buf;
4614     size_t len = strlen(os::file_separator());
4615     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4616     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4617   } else {
4618     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4619   }
4620   return buf;
4621 }
4622 
4623 static void print_signal_handler(outputStream* st, int sig,
4624                                  char* buf, size_t buflen) {
4625   struct sigaction sa;
4626 
4627   sigaction(sig, NULL, &sa);
4628 
4629   // See comment for SIGNIFICANT_SIGNAL_MASK define
4630   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4631 
4632   st->print("%s: ", os::exception_name(sig, buf, buflen));
4633 
4634   address handler = (sa.sa_flags & SA_SIGINFO)
4635     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4636     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4637 
4638   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4639     st->print("SIG_DFL");
4640   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4641     st->print("SIG_IGN");
4642   } else {
4643     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4644   }
4645 
4646   st->print(", sa_mask[0]=");
4647   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4648 
4649   address rh = VMError::get_resetted_sighandler(sig);
4650   // May be, handler was resetted by VMError?
4651   if (rh != NULL) {
4652     handler = rh;
4653     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4654   }
4655 
4656   st->print(", sa_flags=");
4657   os::Posix::print_sa_flags(st, sa.sa_flags);
4658 
4659   // Check: is it our handler?
4660   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4661       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4662     // It is our signal handler
4663     // check for flags, reset system-used one!
4664     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4665       st->print(
4666                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4667                 os::Linux::get_our_sigflags(sig));
4668     }
4669   }
4670   st->cr();
4671 }
4672 
4673 
4674 #define DO_SIGNAL_CHECK(sig)                      \
4675   do {                                            \
4676     if (!sigismember(&check_signal_done, sig)) {  \
4677       os::Linux::check_signal_handler(sig);       \
4678     }                                             \
4679   } while (0)
4680 
4681 // This method is a periodic task to check for misbehaving JNI applications
4682 // under CheckJNI, we can add any periodic checks here
4683 
4684 void os::run_periodic_checks() {
4685   if (check_signals == false) return;
4686 
4687   // SEGV and BUS if overridden could potentially prevent
4688   // generation of hs*.log in the event of a crash, debugging
4689   // such a case can be very challenging, so we absolutely
4690   // check the following for a good measure:
4691   DO_SIGNAL_CHECK(SIGSEGV);
4692   DO_SIGNAL_CHECK(SIGILL);
4693   DO_SIGNAL_CHECK(SIGFPE);
4694   DO_SIGNAL_CHECK(SIGBUS);
4695   DO_SIGNAL_CHECK(SIGPIPE);
4696   DO_SIGNAL_CHECK(SIGXFSZ);
4697 #if defined(PPC64)
4698   DO_SIGNAL_CHECK(SIGTRAP);
4699 #endif
4700 
4701   // ReduceSignalUsage allows the user to override these handlers
4702   // see comments at the very top and jvm_solaris.h
4703   if (!ReduceSignalUsage) {
4704     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4705     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4706     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4707     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4708   }
4709 
4710   DO_SIGNAL_CHECK(SR_signum);
4711 }
4712 
4713 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4714 
4715 static os_sigaction_t os_sigaction = NULL;
4716 
4717 void os::Linux::check_signal_handler(int sig) {
4718   char buf[O_BUFLEN];
4719   address jvmHandler = NULL;
4720 
4721 
4722   struct sigaction act;
4723   if (os_sigaction == NULL) {
4724     // only trust the default sigaction, in case it has been interposed
4725     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4726     if (os_sigaction == NULL) return;
4727   }
4728 
4729   os_sigaction(sig, (struct sigaction*)NULL, &act);
4730 
4731 
4732   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4733 
4734   address thisHandler = (act.sa_flags & SA_SIGINFO)
4735     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4736     : CAST_FROM_FN_PTR(address, act.sa_handler);
4737 
4738 
4739   switch (sig) {
4740   case SIGSEGV:
4741   case SIGBUS:
4742   case SIGFPE:
4743   case SIGPIPE:
4744   case SIGILL:
4745   case SIGXFSZ:
4746     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4747     break;
4748 
4749   case SHUTDOWN1_SIGNAL:
4750   case SHUTDOWN2_SIGNAL:
4751   case SHUTDOWN3_SIGNAL:
4752   case BREAK_SIGNAL:
4753     jvmHandler = (address)user_handler();
4754     break;
4755 
4756   default:
4757     if (sig == SR_signum) {
4758       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4759     } else {
4760       return;
4761     }
4762     break;
4763   }
4764 
4765   if (thisHandler != jvmHandler) {
4766     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4767     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4768     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4769     // No need to check this sig any longer
4770     sigaddset(&check_signal_done, sig);
4771     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4772     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4773       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4774                     exception_name(sig, buf, O_BUFLEN));
4775     }
4776   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4777     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4778     tty->print("expected:");
4779     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4780     tty->cr();
4781     tty->print("  found:");
4782     os::Posix::print_sa_flags(tty, act.sa_flags);
4783     tty->cr();
4784     // No need to check this sig any longer
4785     sigaddset(&check_signal_done, sig);
4786   }
4787 
4788   // Dump all the signal
4789   if (sigismember(&check_signal_done, sig)) {
4790     print_signal_handlers(tty, buf, O_BUFLEN);
4791   }
4792 }
4793 
4794 extern void report_error(char* file_name, int line_no, char* title,
4795                          char* format, ...);
4796 
4797 // this is called _before_ the most of global arguments have been parsed
4798 void os::init(void) {
4799   char dummy;   // used to get a guess on initial stack address
4800 //  first_hrtime = gethrtime();
4801 
4802   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4803 
4804   init_random(1234567);
4805 
4806   ThreadCritical::initialize();
4807 
4808   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4809   if (Linux::page_size() == -1) {
4810     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4811           os::strerror(errno));
4812   }
4813   init_page_sizes((size_t) Linux::page_size());
4814 
4815   Linux::initialize_system_info();
4816 
4817   Linux::initialize_os_info();
4818 
4819   // main_thread points to the aboriginal thread
4820   Linux::_main_thread = pthread_self();
4821 
4822   Linux::clock_init();
4823   initial_time_count = javaTimeNanos();
4824 
4825   // retrieve entry point for pthread_setname_np
4826   Linux::_pthread_setname_np =
4827     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4828 
4829   os::Posix::init();
4830 }
4831 
4832 // To install functions for atexit system call
4833 extern "C" {
4834   static void perfMemory_exit_helper() {
4835     perfMemory_exit();
4836   }
4837 }
4838 
4839 // this is called _after_ the global arguments have been parsed
4840 jint os::init_2(void) {
4841 
4842   os::Posix::init_2();
4843 
4844   Linux::fast_thread_clock_init();
4845 
4846   // Allocate a single page and mark it as readable for safepoint polling
4847   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4848   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4849 
4850   os::set_polling_page(polling_page);
4851   log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4852 
4853   if (!UseMembar) {
4854     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4855     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4856     os::set_memory_serialize_page(mem_serialize_page);
4857     log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4858   }
4859 
4860   // initialize suspend/resume support - must do this before signal_sets_init()
4861   if (SR_initialize() != 0) {
4862     perror("SR_initialize failed");
4863     return JNI_ERR;
4864   }
4865 
4866   Linux::signal_sets_init();
4867   Linux::install_signal_handlers();
4868 
4869   // Check and sets minimum stack sizes against command line options
4870   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4871     return JNI_ERR;
4872   }
4873   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4874 
4875 #if defined(IA32)
4876   workaround_expand_exec_shield_cs_limit();
4877 #endif
4878 
4879   Linux::libpthread_init();
4880   Linux::sched_getcpu_init();
4881   log_info(os)("HotSpot is running with %s, %s",
4882                Linux::glibc_version(), Linux::libpthread_version());
4883 
4884   if (UseNUMA) {
4885     if (!Linux::libnuma_init()) {
4886       UseNUMA = false;
4887     } else {
4888       if ((Linux::numa_max_node() < 1)) {
4889         // There's only one node(they start from 0), disable NUMA.
4890         UseNUMA = false;
4891       }
4892     }
4893     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4894     // we can make the adaptive lgrp chunk resizing work. If the user specified
4895     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4896     // disable adaptive resizing.
4897     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4898       if (FLAG_IS_DEFAULT(UseNUMA)) {
4899         UseNUMA = false;
4900       } else {
4901         if (FLAG_IS_DEFAULT(UseLargePages) &&
4902             FLAG_IS_DEFAULT(UseSHM) &&
4903             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4904           UseLargePages = false;
4905         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4906           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4907           UseAdaptiveSizePolicy = false;
4908           UseAdaptiveNUMAChunkSizing = false;
4909         }
4910       }
4911     }
4912     if (!UseNUMA && ForceNUMA) {
4913       UseNUMA = true;
4914     }
4915   }
4916 
4917   if (MaxFDLimit) {
4918     // set the number of file descriptors to max. print out error
4919     // if getrlimit/setrlimit fails but continue regardless.
4920     struct rlimit nbr_files;
4921     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4922     if (status != 0) {
4923       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4924     } else {
4925       nbr_files.rlim_cur = nbr_files.rlim_max;
4926       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4927       if (status != 0) {
4928         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4929       }
4930     }
4931   }
4932 
4933   // Initialize lock used to serialize thread creation (see os::create_thread)
4934   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4935 
4936   // at-exit methods are called in the reverse order of their registration.
4937   // atexit functions are called on return from main or as a result of a
4938   // call to exit(3C). There can be only 32 of these functions registered
4939   // and atexit() does not set errno.
4940 
4941   if (PerfAllowAtExitRegistration) {
4942     // only register atexit functions if PerfAllowAtExitRegistration is set.
4943     // atexit functions can be delayed until process exit time, which
4944     // can be problematic for embedded VM situations. Embedded VMs should
4945     // call DestroyJavaVM() to assure that VM resources are released.
4946 
4947     // note: perfMemory_exit_helper atexit function may be removed in
4948     // the future if the appropriate cleanup code can be added to the
4949     // VM_Exit VMOperation's doit method.
4950     if (atexit(perfMemory_exit_helper) != 0) {
4951       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4952     }
4953   }
4954 
4955   // initialize thread priority policy
4956   prio_init();
4957 
4958   return JNI_OK;
4959 }
4960 
4961 // Mark the polling page as unreadable
4962 void os::make_polling_page_unreadable(void) {
4963   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4964     fatal("Could not disable polling page");
4965   }
4966 }
4967 
4968 // Mark the polling page as readable
4969 void os::make_polling_page_readable(void) {
4970   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4971     fatal("Could not enable polling page");
4972   }
4973 }
4974 
4975 // older glibc versions don't have this macro (which expands to
4976 // an optimized bit-counting function) so we have to roll our own
4977 #ifndef CPU_COUNT
4978 
4979 static int _cpu_count(const cpu_set_t* cpus) {
4980   int count = 0;
4981   // only look up to the number of configured processors
4982   for (int i = 0; i < os::processor_count(); i++) {
4983     if (CPU_ISSET(i, cpus)) {
4984       count++;
4985     }
4986   }
4987   return count;
4988 }
4989 
4990 #define CPU_COUNT(cpus) _cpu_count(cpus)
4991 
4992 #endif // CPU_COUNT
4993 
4994 // Get the current number of available processors for this process.
4995 // This value can change at any time during a process's lifetime.
4996 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4997 // If it appears there may be more than 1024 processors then we do a
4998 // dynamic check - see 6515172 for details.
4999 // If anything goes wrong we fallback to returning the number of online
5000 // processors - which can be greater than the number available to the process.
5001 int os::active_processor_count() {
5002   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
5003   cpu_set_t* cpus_p = &cpus;
5004   int cpus_size = sizeof(cpu_set_t);
5005 
5006   int configured_cpus = processor_count();  // upper bound on available cpus
5007   int cpu_count = 0;
5008 
5009 // old build platforms may not support dynamic cpu sets
5010 #ifdef CPU_ALLOC
5011 
5012   // To enable easy testing of the dynamic path on different platforms we
5013   // introduce a diagnostic flag: UseCpuAllocPath
5014   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5015     // kernel may use a mask bigger than cpu_set_t
5016     log_trace(os)("active_processor_count: using dynamic path %s"
5017                   "- configured processors: %d",
5018                   UseCpuAllocPath ? "(forced) " : "",
5019                   configured_cpus);
5020     cpus_p = CPU_ALLOC(configured_cpus);
5021     if (cpus_p != NULL) {
5022       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5023       // zero it just to be safe
5024       CPU_ZERO_S(cpus_size, cpus_p);
5025     }
5026     else {
5027        // failed to allocate so fallback to online cpus
5028        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5029        log_trace(os)("active_processor_count: "
5030                      "CPU_ALLOC failed (%s) - using "
5031                      "online processor count: %d",
5032                      os::strerror(errno), online_cpus);
5033        return online_cpus;
5034     }
5035   }
5036   else {
5037     log_trace(os)("active_processor_count: using static path - configured processors: %d",
5038                   configured_cpus);
5039   }
5040 #else // CPU_ALLOC
5041 // these stubs won't be executed
5042 #define CPU_COUNT_S(size, cpus) -1
5043 #define CPU_FREE(cpus)
5044 
5045   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5046                 configured_cpus);
5047 #endif // CPU_ALLOC
5048 
5049   // pid 0 means the current thread - which we have to assume represents the process
5050   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5051     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5052       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5053     }
5054     else {
5055       cpu_count = CPU_COUNT(cpus_p);
5056     }
5057     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5058   }
5059   else {
5060     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5061     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5062             "which may exceed available processors", os::strerror(errno), cpu_count);
5063   }
5064 
5065   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5066     CPU_FREE(cpus_p);
5067   }
5068 
5069   assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
5070   return cpu_count;
5071 }
5072 
5073 void os::set_native_thread_name(const char *name) {
5074   if (Linux::_pthread_setname_np) {
5075     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5076     snprintf(buf, sizeof(buf), "%s", name);
5077     buf[sizeof(buf) - 1] = '\0';
5078     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5079     // ERANGE should not happen; all other errors should just be ignored.
5080     assert(rc != ERANGE, "pthread_setname_np failed");
5081   }
5082 }
5083 
5084 bool os::distribute_processes(uint length, uint* distribution) {
5085   // Not yet implemented.
5086   return false;
5087 }
5088 
5089 bool os::bind_to_processor(uint processor_id) {
5090   // Not yet implemented.
5091   return false;
5092 }
5093 
5094 ///
5095 
5096 void os::SuspendedThreadTask::internal_do_task() {
5097   if (do_suspend(_thread->osthread())) {
5098     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5099     do_task(context);
5100     do_resume(_thread->osthread());
5101   }
5102 }
5103 
5104 class PcFetcher : public os::SuspendedThreadTask {
5105  public:
5106   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5107   ExtendedPC result();
5108  protected:
5109   void do_task(const os::SuspendedThreadTaskContext& context);
5110  private:
5111   ExtendedPC _epc;
5112 };
5113 
5114 ExtendedPC PcFetcher::result() {
5115   guarantee(is_done(), "task is not done yet.");
5116   return _epc;
5117 }
5118 
5119 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5120   Thread* thread = context.thread();
5121   OSThread* osthread = thread->osthread();
5122   if (osthread->ucontext() != NULL) {
5123     _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
5124   } else {
5125     // NULL context is unexpected, double-check this is the VMThread
5126     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5127   }
5128 }
5129 
5130 // Suspends the target using the signal mechanism and then grabs the PC before
5131 // resuming the target. Used by the flat-profiler only
5132 ExtendedPC os::get_thread_pc(Thread* thread) {
5133   // Make sure that it is called by the watcher for the VMThread
5134   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5135   assert(thread->is_VM_thread(), "Can only be called for VMThread");
5136 
5137   PcFetcher fetcher(thread);
5138   fetcher.run();
5139   return fetcher.result();
5140 }
5141 
5142 ////////////////////////////////////////////////////////////////////////////////
5143 // debug support
5144 
5145 bool os::find(address addr, outputStream* st) {
5146   Dl_info dlinfo;
5147   memset(&dlinfo, 0, sizeof(dlinfo));
5148   if (dladdr(addr, &dlinfo) != 0) {
5149     st->print(PTR_FORMAT ": ", p2i(addr));
5150     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5151       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5152                 p2i(addr) - p2i(dlinfo.dli_saddr));
5153     } else if (dlinfo.dli_fbase != NULL) {
5154       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5155     } else {
5156       st->print("<absolute address>");
5157     }
5158     if (dlinfo.dli_fname != NULL) {
5159       st->print(" in %s", dlinfo.dli_fname);
5160     }
5161     if (dlinfo.dli_fbase != NULL) {
5162       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5163     }
5164     st->cr();
5165 
5166     if (Verbose) {
5167       // decode some bytes around the PC
5168       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5169       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5170       address       lowest = (address) dlinfo.dli_sname;
5171       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5172       if (begin < lowest)  begin = lowest;
5173       Dl_info dlinfo2;
5174       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5175           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5176         end = (address) dlinfo2.dli_saddr;
5177       }
5178       Disassembler::decode(begin, end, st);
5179     }
5180     return true;
5181   }
5182   return false;
5183 }
5184 
5185 ////////////////////////////////////////////////////////////////////////////////
5186 // misc
5187 
5188 // This does not do anything on Linux. This is basically a hook for being
5189 // able to use structured exception handling (thread-local exception filters)
5190 // on, e.g., Win32.
5191 void
5192 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5193                          JavaCallArguments* args, Thread* thread) {
5194   f(value, method, args, thread);
5195 }
5196 
5197 void os::print_statistics() {
5198 }
5199 
5200 bool os::message_box(const char* title, const char* message) {
5201   int i;
5202   fdStream err(defaultStream::error_fd());
5203   for (i = 0; i < 78; i++) err.print_raw("=");
5204   err.cr();
5205   err.print_raw_cr(title);
5206   for (i = 0; i < 78; i++) err.print_raw("-");
5207   err.cr();
5208   err.print_raw_cr(message);
5209   for (i = 0; i < 78; i++) err.print_raw("=");
5210   err.cr();
5211 
5212   char buf[16];
5213   // Prevent process from exiting upon "read error" without consuming all CPU
5214   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5215 
5216   return buf[0] == 'y' || buf[0] == 'Y';
5217 }
5218 
5219 int os::stat(const char *path, struct stat *sbuf) {
5220   char pathbuf[MAX_PATH];
5221   if (strlen(path) > MAX_PATH - 1) {
5222     errno = ENAMETOOLONG;
5223     return -1;
5224   }
5225   os::native_path(strcpy(pathbuf, path));
5226   return ::stat(pathbuf, sbuf);
5227 }
5228 
5229 // Is a (classpath) directory empty?
5230 bool os::dir_is_empty(const char* path) {
5231   DIR *dir = NULL;
5232   struct dirent *ptr;
5233 
5234   dir = opendir(path);
5235   if (dir == NULL) return true;
5236 
5237   // Scan the directory
5238   bool result = true;
5239   char buf[sizeof(struct dirent) + MAX_PATH];
5240   while (result && (ptr = ::readdir(dir)) != NULL) {
5241     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5242       result = false;
5243     }
5244   }
5245   closedir(dir);
5246   return result;
5247 }
5248 
5249 // This code originates from JDK's sysOpen and open64_w
5250 // from src/solaris/hpi/src/system_md.c
5251 
5252 int os::open(const char *path, int oflag, int mode) {
5253   if (strlen(path) > MAX_PATH - 1) {
5254     errno = ENAMETOOLONG;
5255     return -1;
5256   }
5257 
5258   // All file descriptors that are opened in the Java process and not
5259   // specifically destined for a subprocess should have the close-on-exec
5260   // flag set.  If we don't set it, then careless 3rd party native code
5261   // might fork and exec without closing all appropriate file descriptors
5262   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5263   // turn might:
5264   //
5265   // - cause end-of-file to fail to be detected on some file
5266   //   descriptors, resulting in mysterious hangs, or
5267   //
5268   // - might cause an fopen in the subprocess to fail on a system
5269   //   suffering from bug 1085341.
5270   //
5271   // (Yes, the default setting of the close-on-exec flag is a Unix
5272   // design flaw)
5273   //
5274   // See:
5275   // 1085341: 32-bit stdio routines should support file descriptors >255
5276   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5277   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5278   //
5279   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5280   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5281   // because it saves a system call and removes a small window where the flag
5282   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5283   // and we fall back to using FD_CLOEXEC (see below).
5284 #ifdef O_CLOEXEC
5285   oflag |= O_CLOEXEC;
5286 #endif
5287 
5288   int fd = ::open64(path, oflag, mode);
5289   if (fd == -1) return -1;
5290 
5291   //If the open succeeded, the file might still be a directory
5292   {
5293     struct stat64 buf64;
5294     int ret = ::fstat64(fd, &buf64);
5295     int st_mode = buf64.st_mode;
5296 
5297     if (ret != -1) {
5298       if ((st_mode & S_IFMT) == S_IFDIR) {
5299         errno = EISDIR;
5300         ::close(fd);
5301         return -1;
5302       }
5303     } else {
5304       ::close(fd);
5305       return -1;
5306     }
5307   }
5308 
5309 #ifdef FD_CLOEXEC
5310   // Validate that the use of the O_CLOEXEC flag on open above worked.
5311   // With recent kernels, we will perform this check exactly once.
5312   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5313   if (!O_CLOEXEC_is_known_to_work) {
5314     int flags = ::fcntl(fd, F_GETFD);
5315     if (flags != -1) {
5316       if ((flags & FD_CLOEXEC) != 0)
5317         O_CLOEXEC_is_known_to_work = 1;
5318       else
5319         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5320     }
5321   }
5322 #endif
5323 
5324   return fd;
5325 }
5326 
5327 
5328 // create binary file, rewriting existing file if required
5329 int os::create_binary_file(const char* path, bool rewrite_existing) {
5330   int oflags = O_WRONLY | O_CREAT;
5331   if (!rewrite_existing) {
5332     oflags |= O_EXCL;
5333   }
5334   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5335 }
5336 
5337 // return current position of file pointer
5338 jlong os::current_file_offset(int fd) {
5339   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5340 }
5341 
5342 // move file pointer to the specified offset
5343 jlong os::seek_to_file_offset(int fd, jlong offset) {
5344   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5345 }
5346 
5347 // This code originates from JDK's sysAvailable
5348 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5349 
5350 int os::available(int fd, jlong *bytes) {
5351   jlong cur, end;
5352   int mode;
5353   struct stat64 buf64;
5354 
5355   if (::fstat64(fd, &buf64) >= 0) {
5356     mode = buf64.st_mode;
5357     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5358       int n;
5359       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5360         *bytes = n;
5361         return 1;
5362       }
5363     }
5364   }
5365   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5366     return 0;
5367   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5368     return 0;
5369   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5370     return 0;
5371   }
5372   *bytes = end - cur;
5373   return 1;
5374 }
5375 
5376 // Map a block of memory.
5377 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5378                         char *addr, size_t bytes, bool read_only,
5379                         bool allow_exec) {
5380   int prot;
5381   int flags = MAP_PRIVATE;
5382 
5383   if (read_only) {
5384     prot = PROT_READ;
5385   } else {
5386     prot = PROT_READ | PROT_WRITE;
5387   }
5388 
5389   if (allow_exec) {
5390     prot |= PROT_EXEC;
5391   }
5392 
5393   if (addr != NULL) {
5394     flags |= MAP_FIXED;
5395   }
5396 
5397   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5398                                      fd, file_offset);
5399   if (mapped_address == MAP_FAILED) {
5400     return NULL;
5401   }
5402   return mapped_address;
5403 }
5404 
5405 
5406 // Remap a block of memory.
5407 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5408                           char *addr, size_t bytes, bool read_only,
5409                           bool allow_exec) {
5410   // same as map_memory() on this OS
5411   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5412                         allow_exec);
5413 }
5414 
5415 
5416 // Unmap a block of memory.
5417 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5418   return munmap(addr, bytes) == 0;
5419 }
5420 
5421 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5422 
5423 static clockid_t thread_cpu_clockid(Thread* thread) {
5424   pthread_t tid = thread->osthread()->pthread_id();
5425   clockid_t clockid;
5426 
5427   // Get thread clockid
5428   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5429   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5430   return clockid;
5431 }
5432 
5433 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5434 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5435 // of a thread.
5436 //
5437 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5438 // the fast estimate available on the platform.
5439 
5440 jlong os::current_thread_cpu_time() {
5441   if (os::Linux::supports_fast_thread_cpu_time()) {
5442     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5443   } else {
5444     // return user + sys since the cost is the same
5445     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5446   }
5447 }
5448 
5449 jlong os::thread_cpu_time(Thread* thread) {
5450   // consistent with what current_thread_cpu_time() returns
5451   if (os::Linux::supports_fast_thread_cpu_time()) {
5452     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5453   } else {
5454     return slow_thread_cpu_time(thread, true /* user + sys */);
5455   }
5456 }
5457 
5458 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5459   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5460     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5461   } else {
5462     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5463   }
5464 }
5465 
5466 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5467   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5468     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5469   } else {
5470     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5471   }
5472 }
5473 
5474 //  -1 on error.
5475 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5476   pid_t  tid = thread->osthread()->thread_id();
5477   char *s;
5478   char stat[2048];
5479   int statlen;
5480   char proc_name[64];
5481   int count;
5482   long sys_time, user_time;
5483   char cdummy;
5484   int idummy;
5485   long ldummy;
5486   FILE *fp;
5487 
5488   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5489   fp = fopen(proc_name, "r");
5490   if (fp == NULL) return -1;
5491   statlen = fread(stat, 1, 2047, fp);
5492   stat[statlen] = '\0';
5493   fclose(fp);
5494 
5495   // Skip pid and the command string. Note that we could be dealing with
5496   // weird command names, e.g. user could decide to rename java launcher
5497   // to "java 1.4.2 :)", then the stat file would look like
5498   //                1234 (java 1.4.2 :)) R ... ...
5499   // We don't really need to know the command string, just find the last
5500   // occurrence of ")" and then start parsing from there. See bug 4726580.
5501   s = strrchr(stat, ')');
5502   if (s == NULL) return -1;
5503 
5504   // Skip blank chars
5505   do { s++; } while (s && isspace(*s));
5506 
5507   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5508                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5509                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5510                  &user_time, &sys_time);
5511   if (count != 13) return -1;
5512   if (user_sys_cpu_time) {
5513     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5514   } else {
5515     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5516   }
5517 }
5518 
5519 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5520   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5521   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5522   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5523   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5524 }
5525 
5526 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5527   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5528   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5529   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5530   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5531 }
5532 
5533 bool os::is_thread_cpu_time_supported() {
5534   return true;
5535 }
5536 
5537 // System loadavg support.  Returns -1 if load average cannot be obtained.
5538 // Linux doesn't yet have a (official) notion of processor sets,
5539 // so just return the system wide load average.
5540 int os::loadavg(double loadavg[], int nelem) {
5541   return ::getloadavg(loadavg, nelem);
5542 }
5543 
5544 void os::pause() {
5545   char filename[MAX_PATH];
5546   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5547     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5548   } else {
5549     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5550   }
5551 
5552   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5553   if (fd != -1) {
5554     struct stat buf;
5555     ::close(fd);
5556     while (::stat(filename, &buf) == 0) {
5557       (void)::poll(NULL, 0, 100);
5558     }
5559   } else {
5560     jio_fprintf(stderr,
5561                 "Could not open pause file '%s', continuing immediately.\n", filename);
5562   }
5563 }
5564 
5565 extern char** environ;
5566 
5567 // Run the specified command in a separate process. Return its exit value,
5568 // or -1 on failure (e.g. can't fork a new process).
5569 // Unlike system(), this function can be called from signal handler. It
5570 // doesn't block SIGINT et al.
5571 int os::fork_and_exec(char* cmd) {
5572   const char * argv[4] = {"sh", "-c", cmd, NULL};
5573 
5574   pid_t pid = fork();
5575 
5576   if (pid < 0) {
5577     // fork failed
5578     return -1;
5579 
5580   } else if (pid == 0) {
5581     // child process
5582 
5583     execve("/bin/sh", (char* const*)argv, environ);
5584 
5585     // execve failed
5586     _exit(-1);
5587 
5588   } else  {
5589     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5590     // care about the actual exit code, for now.
5591 
5592     int status;
5593 
5594     // Wait for the child process to exit.  This returns immediately if
5595     // the child has already exited. */
5596     while (waitpid(pid, &status, 0) < 0) {
5597       switch (errno) {
5598       case ECHILD: return 0;
5599       case EINTR: break;
5600       default: return -1;
5601       }
5602     }
5603 
5604     if (WIFEXITED(status)) {
5605       // The child exited normally; get its exit code.
5606       return WEXITSTATUS(status);
5607     } else if (WIFSIGNALED(status)) {
5608       // The child exited because of a signal
5609       // The best value to return is 0x80 + signal number,
5610       // because that is what all Unix shells do, and because
5611       // it allows callers to distinguish between process exit and
5612       // process death by signal.
5613       return 0x80 + WTERMSIG(status);
5614     } else {
5615       // Unknown exit code; pass it through
5616       return status;
5617     }
5618   }
5619 }
5620 
5621 // is_headless_jre()
5622 //
5623 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5624 // in order to report if we are running in a headless jre
5625 //
5626 // Since JDK8 xawt/libmawt.so was moved into the same directory
5627 // as libawt.so, and renamed libawt_xawt.so
5628 //
5629 bool os::is_headless_jre() {
5630   struct stat statbuf;
5631   char buf[MAXPATHLEN];
5632   char libmawtpath[MAXPATHLEN];
5633   const char *xawtstr  = "/xawt/libmawt.so";
5634   const char *new_xawtstr = "/libawt_xawt.so";
5635   char *p;
5636 
5637   // Get path to libjvm.so
5638   os::jvm_path(buf, sizeof(buf));
5639 
5640   // Get rid of libjvm.so
5641   p = strrchr(buf, '/');
5642   if (p == NULL) {
5643     return false;
5644   } else {
5645     *p = '\0';
5646   }
5647 
5648   // Get rid of client or server
5649   p = strrchr(buf, '/');
5650   if (p == NULL) {
5651     return false;
5652   } else {
5653     *p = '\0';
5654   }
5655 
5656   // check xawt/libmawt.so
5657   strcpy(libmawtpath, buf);
5658   strcat(libmawtpath, xawtstr);
5659   if (::stat(libmawtpath, &statbuf) == 0) return false;
5660 
5661   // check libawt_xawt.so
5662   strcpy(libmawtpath, buf);
5663   strcat(libmawtpath, new_xawtstr);
5664   if (::stat(libmawtpath, &statbuf) == 0) return false;
5665 
5666   return true;
5667 }
5668 
5669 // Get the default path to the core file
5670 // Returns the length of the string
5671 int os::get_core_path(char* buffer, size_t bufferSize) {
5672   /*
5673    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5674    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5675    */
5676   const int core_pattern_len = 129;
5677   char core_pattern[core_pattern_len] = {0};
5678 
5679   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5680   if (core_pattern_file == -1) {
5681     return -1;
5682   }
5683 
5684   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5685   ::close(core_pattern_file);
5686   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5687     return -1;
5688   }
5689   if (core_pattern[ret-1] == '\n') {
5690     core_pattern[ret-1] = '\0';
5691   } else {
5692     core_pattern[ret] = '\0';
5693   }
5694 
5695   char *pid_pos = strstr(core_pattern, "%p");
5696   int written;
5697 
5698   if (core_pattern[0] == '/') {
5699     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5700   } else {
5701     char cwd[PATH_MAX];
5702 
5703     const char* p = get_current_directory(cwd, PATH_MAX);
5704     if (p == NULL) {
5705       return -1;
5706     }
5707 
5708     if (core_pattern[0] == '|') {
5709       written = jio_snprintf(buffer, bufferSize,
5710                              "\"%s\" (or dumping to %s/core.%d)",
5711                              &core_pattern[1], p, current_process_id());
5712     } else {
5713       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5714     }
5715   }
5716 
5717   if (written < 0) {
5718     return -1;
5719   }
5720 
5721   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5722     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5723 
5724     if (core_uses_pid_file != -1) {
5725       char core_uses_pid = 0;
5726       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5727       ::close(core_uses_pid_file);
5728 
5729       if (core_uses_pid == '1') {
5730         jio_snprintf(buffer + written, bufferSize - written,
5731                                           ".%d", current_process_id());
5732       }
5733     }
5734   }
5735 
5736   return strlen(buffer);
5737 }
5738 
5739 bool os::start_debugging(char *buf, int buflen) {
5740   int len = (int)strlen(buf);
5741   char *p = &buf[len];
5742 
5743   jio_snprintf(p, buflen-len,
5744                "\n\n"
5745                "Do you want to debug the problem?\n\n"
5746                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5747                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5748                "Otherwise, press RETURN to abort...",
5749                os::current_process_id(), os::current_process_id(),
5750                os::current_thread_id(), os::current_thread_id());
5751 
5752   bool yes = os::message_box("Unexpected Error", buf);
5753 
5754   if (yes) {
5755     // yes, user asked VM to launch debugger
5756     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5757                  os::current_process_id(), os::current_process_id());
5758 
5759     os::fork_and_exec(buf);
5760     yes = false;
5761   }
5762   return yes;
5763 }
5764 
5765 
5766 // Java/Compiler thread:
5767 //
5768 //   Low memory addresses
5769 // P0 +------------------------+
5770 //    |                        |\  Java thread created by VM does not have glibc
5771 //    |    glibc guard page    | - guard page, attached Java thread usually has
5772 //    |                        |/  1 glibc guard page.
5773 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5774 //    |                        |\
5775 //    |  HotSpot Guard Pages   | - red, yellow and reserved pages
5776 //    |                        |/
5777 //    +------------------------+ JavaThread::stack_reserved_zone_base()
5778 //    |                        |\
5779 //    |      Normal Stack      | -
5780 //    |                        |/
5781 // P2 +------------------------+ Thread::stack_base()
5782 //
5783 // Non-Java thread:
5784 //
5785 //   Low memory addresses
5786 // P0 +------------------------+
5787 //    |                        |\
5788 //    |  glibc guard page      | - usually 1 page
5789 //    |                        |/
5790 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5791 //    |                        |\
5792 //    |      Normal Stack      | -
5793 //    |                        |/
5794 // P2 +------------------------+ Thread::stack_base()
5795 //
5796 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5797 //    returned from pthread_attr_getstack().
5798 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5799 //    of the stack size given in pthread_attr. We work around this for
5800 //    threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5801 //
5802 #ifndef ZERO
5803 static void current_stack_region(address * bottom, size_t * size) {
5804   if (os::Linux::is_initial_thread()) {
5805     // initial thread needs special handling because pthread_getattr_np()
5806     // may return bogus value.
5807     *bottom = os::Linux::initial_thread_stack_bottom();
5808     *size   = os::Linux::initial_thread_stack_size();
5809   } else {
5810     pthread_attr_t attr;
5811 
5812     int rslt = pthread_getattr_np(pthread_self(), &attr);
5813 
5814     // JVM needs to know exact stack location, abort if it fails
5815     if (rslt != 0) {
5816       if (rslt == ENOMEM) {
5817         vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5818       } else {
5819         fatal("pthread_getattr_np failed with error = %d", rslt);
5820       }
5821     }
5822 
5823     if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5824       fatal("Cannot locate current stack attributes!");
5825     }
5826 
5827     // Work around NPTL stack guard error.
5828     size_t guard_size = 0;
5829     rslt = pthread_attr_getguardsize(&attr, &guard_size);
5830     if (rslt != 0) {
5831       fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5832     }
5833     *bottom += guard_size;
5834     *size   -= guard_size;
5835 
5836     pthread_attr_destroy(&attr);
5837 
5838   }
5839   assert(os::current_stack_pointer() >= *bottom &&
5840          os::current_stack_pointer() < *bottom + *size, "just checking");
5841 }
5842 
5843 address os::current_stack_base() {
5844   address bottom;
5845   size_t size;
5846   current_stack_region(&bottom, &size);
5847   return (bottom + size);
5848 }
5849 
5850 size_t os::current_stack_size() {
5851   // This stack size includes the usable stack and HotSpot guard pages
5852   // (for the threads that have Hotspot guard pages).
5853   address bottom;
5854   size_t size;
5855   current_stack_region(&bottom, &size);
5856   return size;
5857 }
5858 #endif
5859 
5860 static inline struct timespec get_mtime(const char* filename) {
5861   struct stat st;
5862   int ret = os::stat(filename, &st);
5863   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5864   return st.st_mtim;
5865 }
5866 
5867 int os::compare_file_modified_times(const char* file1, const char* file2) {
5868   struct timespec filetime1 = get_mtime(file1);
5869   struct timespec filetime2 = get_mtime(file2);
5870   int diff = filetime1.tv_sec - filetime2.tv_sec;
5871   if (diff == 0) {
5872     return filetime1.tv_nsec - filetime2.tv_nsec;
5873   }
5874   return diff;
5875 }
5876 
5877 /////////////// Unit tests ///////////////
5878 
5879 #ifndef PRODUCT
5880 
5881 #define test_log(...)              \
5882   do {                             \
5883     if (VerboseInternalVMTests) {  \
5884       tty->print_cr(__VA_ARGS__);  \
5885       tty->flush();                \
5886     }                              \
5887   } while (false)
5888 
5889 class TestReserveMemorySpecial : AllStatic {
5890  public:
5891   static void small_page_write(void* addr, size_t size) {
5892     size_t page_size = os::vm_page_size();
5893 
5894     char* end = (char*)addr + size;
5895     for (char* p = (char*)addr; p < end; p += page_size) {
5896       *p = 1;
5897     }
5898   }
5899 
5900   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5901     if (!UseHugeTLBFS) {
5902       return;
5903     }
5904 
5905     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5906 
5907     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5908 
5909     if (addr != NULL) {
5910       small_page_write(addr, size);
5911 
5912       os::Linux::release_memory_special_huge_tlbfs(addr, size);
5913     }
5914   }
5915 
5916   static void test_reserve_memory_special_huge_tlbfs_only() {
5917     if (!UseHugeTLBFS) {
5918       return;
5919     }
5920 
5921     size_t lp = os::large_page_size();
5922 
5923     for (size_t size = lp; size <= lp * 10; size += lp) {
5924       test_reserve_memory_special_huge_tlbfs_only(size);
5925     }
5926   }
5927 
5928   static void test_reserve_memory_special_huge_tlbfs_mixed() {
5929     size_t lp = os::large_page_size();
5930     size_t ag = os::vm_allocation_granularity();
5931 
5932     // sizes to test
5933     const size_t sizes[] = {
5934       lp, lp + ag, lp + lp / 2, lp * 2,
5935       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
5936       lp * 10, lp * 10 + lp / 2
5937     };
5938     const int num_sizes = sizeof(sizes) / sizeof(size_t);
5939 
5940     // For each size/alignment combination, we test three scenarios:
5941     // 1) with req_addr == NULL
5942     // 2) with a non-null req_addr at which we expect to successfully allocate
5943     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
5944     //    expect the allocation to either fail or to ignore req_addr
5945 
5946     // Pre-allocate two areas; they shall be as large as the largest allocation
5947     //  and aligned to the largest alignment we will be testing.
5948     const size_t mapping_size = sizes[num_sizes - 1] * 2;
5949     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
5950       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5951       -1, 0);
5952     assert(mapping1 != MAP_FAILED, "should work");
5953 
5954     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
5955       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5956       -1, 0);
5957     assert(mapping2 != MAP_FAILED, "should work");
5958 
5959     // Unmap the first mapping, but leave the second mapping intact: the first
5960     // mapping will serve as a value for a "good" req_addr (case 2). The second
5961     // mapping, still intact, as "bad" req_addr (case 3).
5962     ::munmap(mapping1, mapping_size);
5963 
5964     // Case 1
5965     test_log("%s, req_addr NULL:", __FUNCTION__);
5966     test_log("size            align           result");
5967 
5968     for (int i = 0; i < num_sizes; i++) {
5969       const size_t size = sizes[i];
5970       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
5971         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
5972         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
5973                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
5974         if (p != NULL) {
5975           assert(is_ptr_aligned(p, alignment), "must be");
5976           small_page_write(p, size);
5977           os::Linux::release_memory_special_huge_tlbfs(p, size);
5978         }
5979       }
5980     }
5981 
5982     // Case 2
5983     test_log("%s, req_addr non-NULL:", __FUNCTION__);
5984     test_log("size            align           req_addr         result");
5985 
5986     for (int i = 0; i < num_sizes; i++) {
5987       const size_t size = sizes[i];
5988       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
5989         char* const req_addr = align_ptr_up(mapping1, alignment);
5990         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
5991         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
5992                  size, alignment, p2i(req_addr), p2i(p),
5993                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
5994         if (p != NULL) {
5995           assert(p == req_addr, "must be");
5996           small_page_write(p, size);
5997           os::Linux::release_memory_special_huge_tlbfs(p, size);
5998         }
5999       }
6000     }
6001 
6002     // Case 3
6003     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6004     test_log("size            align           req_addr         result");
6005 
6006     for (int i = 0; i < num_sizes; i++) {
6007       const size_t size = sizes[i];
6008       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6009         char* const req_addr = align_ptr_up(mapping2, alignment);
6010         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6011         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6012                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6013         // as the area around req_addr contains already existing mappings, the API should always
6014         // return NULL (as per contract, it cannot return another address)
6015         assert(p == NULL, "must be");
6016       }
6017     }
6018 
6019     ::munmap(mapping2, mapping_size);
6020 
6021   }
6022 
6023   static void test_reserve_memory_special_huge_tlbfs() {
6024     if (!UseHugeTLBFS) {
6025       return;
6026     }
6027 
6028     test_reserve_memory_special_huge_tlbfs_only();
6029     test_reserve_memory_special_huge_tlbfs_mixed();
6030   }
6031 
6032   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6033     if (!UseSHM) {
6034       return;
6035     }
6036 
6037     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6038 
6039     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6040 
6041     if (addr != NULL) {
6042       assert(is_ptr_aligned(addr, alignment), "Check");
6043       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6044 
6045       small_page_write(addr, size);
6046 
6047       os::Linux::release_memory_special_shm(addr, size);
6048     }
6049   }
6050 
6051   static void test_reserve_memory_special_shm() {
6052     size_t lp = os::large_page_size();
6053     size_t ag = os::vm_allocation_granularity();
6054 
6055     for (size_t size = ag; size < lp * 3; size += ag) {
6056       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6057         test_reserve_memory_special_shm(size, alignment);
6058       }
6059     }
6060   }
6061 
6062   static void test() {
6063     test_reserve_memory_special_huge_tlbfs();
6064     test_reserve_memory_special_shm();
6065   }
6066 };
6067 
6068 void TestReserveMemorySpecial_test() {
6069   TestReserveMemorySpecial::test();
6070 }
6071 
6072 #endif