1 /*
   2  * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_linux.h"
  35 #include "logging/log.hpp"
  36 #include "memory/allocation.inline.hpp"
  37 #include "memory/filemap.hpp"
  38 #include "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 pthread_condattr_t os::Linux::_condattr[1];
 149 
 150 static jlong initial_time_count=0;
 151 
 152 static int clock_tics_per_sec = 100;
 153 
 154 // For diagnostics to print a message once. see run_periodic_checks
 155 static sigset_t check_signal_done;
 156 static bool check_signals = true;
 157 
 158 // Signal number used to suspend/resume a thread
 159 
 160 // do not use any signal number less than SIGSEGV, see 4355769
 161 static int SR_signum = SIGUSR2;
 162 sigset_t SR_sigset;
 163 
 164 // Declarations
 165 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 166 
 167 // utility functions
 168 
 169 static int SR_initialize();
 170 
 171 julong os::available_memory() {
 172   return Linux::available_memory();
 173 }
 174 
 175 julong os::Linux::available_memory() {
 176   // values in struct sysinfo are "unsigned long"
 177   struct sysinfo si;
 178   sysinfo(&si);
 179 
 180   return (julong)si.freeram * si.mem_unit;
 181 }
 182 
 183 julong os::physical_memory() {
 184   return Linux::physical_memory();
 185 }
 186 
 187 // Return true if user is running as root.
 188 
 189 bool os::have_special_privileges() {
 190   static bool init = false;
 191   static bool privileges = false;
 192   if (!init) {
 193     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 194     init = true;
 195   }
 196   return privileges;
 197 }
 198 
 199 
 200 #ifndef SYS_gettid
 201 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 202   #ifdef __ia64__
 203     #define SYS_gettid 1105
 204   #else
 205     #ifdef __i386__
 206       #define SYS_gettid 224
 207     #else
 208       #ifdef __amd64__
 209         #define SYS_gettid 186
 210       #else
 211         #ifdef __sparc__
 212           #define SYS_gettid 143
 213         #else
 214           #error define gettid for the arch
 215         #endif
 216       #endif
 217     #endif
 218   #endif
 219 #endif
 220 
 221 
 222 // pid_t gettid()
 223 //
 224 // Returns the kernel thread id of the currently running thread. Kernel
 225 // thread id is used to access /proc.
 226 pid_t os::Linux::gettid() {
 227   int rslt = syscall(SYS_gettid);
 228   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 229   return (pid_t)rslt;
 230 }
 231 
 232 // Most versions of linux have a bug where the number of processors are
 233 // determined by looking at the /proc file system.  In a chroot environment,
 234 // the system call returns 1.  This causes the VM to act as if it is
 235 // a single processor and elide locking (see is_MP() call).
 236 static bool unsafe_chroot_detected = false;
 237 static const char *unstable_chroot_error = "/proc file system not found.\n"
 238                      "Java may be unstable running multithreaded in a chroot "
 239                      "environment on Linux when /proc filesystem is not mounted.";
 240 
 241 void os::Linux::initialize_system_info() {
 242   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 243   if (processor_count() == 1) {
 244     pid_t pid = os::Linux::gettid();
 245     char fname[32];
 246     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 247     FILE *fp = fopen(fname, "r");
 248     if (fp == NULL) {
 249       unsafe_chroot_detected = true;
 250     } else {
 251       fclose(fp);
 252     }
 253   }
 254   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 255   assert(processor_count() > 0, "linux error");
 256 }
 257 
 258 void os::init_system_properties_values() {
 259   // The next steps are taken in the product version:
 260   //
 261   // Obtain the JAVA_HOME value from the location of libjvm.so.
 262   // This library should be located at:
 263   // <JAVA_HOME>/lib/{client|server}/libjvm.so.
 264   //
 265   // If "/jre/lib/" appears at the right place in the path, then we
 266   // assume libjvm.so is installed in a JDK and we use this path.
 267   //
 268   // Otherwise exit with message: "Could not create the Java virtual machine."
 269   //
 270   // The following extra steps are taken in the debugging version:
 271   //
 272   // If "/jre/lib/" does NOT appear at the right place in the path
 273   // instead of exit check for $JAVA_HOME environment variable.
 274   //
 275   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 276   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 277   // it looks like libjvm.so is installed there
 278   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 279   //
 280   // Otherwise exit.
 281   //
 282   // Important note: if the location of libjvm.so changes this
 283   // code needs to be changed accordingly.
 284 
 285   // See ld(1):
 286   //      The linker uses the following search paths to locate required
 287   //      shared libraries:
 288   //        1: ...
 289   //        ...
 290   //        7: The default directories, normally /lib and /usr/lib.
 291 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
 292   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 293 #else
 294   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 295 #endif
 296 
 297 // Base path of extensions installed on the system.
 298 #define SYS_EXT_DIR     "/usr/java/packages"
 299 #define EXTENSIONS_DIR  "/lib/ext"
 300 
 301   // Buffer that fits several sprintfs.
 302   // Note that the space for the colon and the trailing null are provided
 303   // by the nulls included by the sizeof operator.
 304   const size_t bufsize =
 305     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 306          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 307   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 308 
 309   // sysclasspath, java_home, dll_dir
 310   {
 311     char *pslash;
 312     os::jvm_path(buf, bufsize);
 313 
 314     // Found the full path to libjvm.so.
 315     // Now cut the path to <java_home>/jre if we can.
 316     pslash = strrchr(buf, '/');
 317     if (pslash != NULL) {
 318       *pslash = '\0';            // Get rid of /libjvm.so.
 319     }
 320     pslash = strrchr(buf, '/');
 321     if (pslash != NULL) {
 322       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 323     }
 324     Arguments::set_dll_dir(buf);
 325 
 326     if (pslash != NULL) {
 327       pslash = strrchr(buf, '/');
 328       if (pslash != NULL) {
 329         *pslash = '\0';        // Get rid of /lib.
 330       }
 331     }
 332     Arguments::set_java_home(buf);
 333     set_boot_path('/', ':');
 334   }
 335 
 336   // Where to look for native libraries.
 337   //
 338   // Note: Due to a legacy implementation, most of the library path
 339   // is set in the launcher. This was to accomodate linking restrictions
 340   // on legacy Linux implementations (which are no longer supported).
 341   // Eventually, all the library path setting will be done here.
 342   //
 343   // However, to prevent the proliferation of improperly built native
 344   // libraries, the new path component /usr/java/packages is added here.
 345   // Eventually, all the library path setting will be done here.
 346   {
 347     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 348     // should always exist (until the legacy problem cited above is
 349     // addressed).
 350     const char *v = ::getenv("LD_LIBRARY_PATH");
 351     const char *v_colon = ":";
 352     if (v == NULL) { v = ""; v_colon = ""; }
 353     // That's +1 for the colon and +1 for the trailing '\0'.
 354     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 355                                                      strlen(v) + 1 +
 356                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
 357                                                      mtInternal);
 358     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
 359     Arguments::set_library_path(ld_library_path);
 360     FREE_C_HEAP_ARRAY(char, ld_library_path);
 361   }
 362 
 363   // Extensions directories.
 364   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 365   Arguments::set_ext_dirs(buf);
 366 
 367   FREE_C_HEAP_ARRAY(char, buf);
 368 
 369 #undef DEFAULT_LIBPATH
 370 #undef SYS_EXT_DIR
 371 #undef EXTENSIONS_DIR
 372 }
 373 
 374 ////////////////////////////////////////////////////////////////////////////////
 375 // breakpoint support
 376 
 377 void os::breakpoint() {
 378   BREAKPOINT;
 379 }
 380 
 381 extern "C" void breakpoint() {
 382   // use debugger to set breakpoint here
 383 }
 384 
 385 ////////////////////////////////////////////////////////////////////////////////
 386 // signal support
 387 
 388 debug_only(static bool signal_sets_initialized = false);
 389 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 390 
 391 bool os::Linux::is_sig_ignored(int sig) {
 392   struct sigaction oact;
 393   sigaction(sig, (struct sigaction*)NULL, &oact);
 394   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 395                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 396   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 397     return true;
 398   } else {
 399     return false;
 400   }
 401 }
 402 
 403 void os::Linux::signal_sets_init() {
 404   // Should also have an assertion stating we are still single-threaded.
 405   assert(!signal_sets_initialized, "Already initialized");
 406   // Fill in signals that are necessarily unblocked for all threads in
 407   // the VM. Currently, we unblock the following signals:
 408   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 409   //                         by -Xrs (=ReduceSignalUsage));
 410   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 411   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 412   // the dispositions or masks wrt these signals.
 413   // Programs embedding the VM that want to use the above signals for their
 414   // own purposes must, at this time, use the "-Xrs" option to prevent
 415   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 416   // (See bug 4345157, and other related bugs).
 417   // In reality, though, unblocking these signals is really a nop, since
 418   // these signals are not blocked by default.
 419   sigemptyset(&unblocked_sigs);
 420   sigemptyset(&allowdebug_blocked_sigs);
 421   sigaddset(&unblocked_sigs, SIGILL);
 422   sigaddset(&unblocked_sigs, SIGSEGV);
 423   sigaddset(&unblocked_sigs, SIGBUS);
 424   sigaddset(&unblocked_sigs, SIGFPE);
 425 #if defined(PPC64)
 426   sigaddset(&unblocked_sigs, SIGTRAP);
 427 #endif
 428   sigaddset(&unblocked_sigs, SR_signum);
 429 
 430   if (!ReduceSignalUsage) {
 431     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 432       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 433       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 434     }
 435     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 436       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 437       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 438     }
 439     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 440       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 441       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 442     }
 443   }
 444   // Fill in signals that are blocked by all but the VM thread.
 445   sigemptyset(&vm_sigs);
 446   if (!ReduceSignalUsage) {
 447     sigaddset(&vm_sigs, BREAK_SIGNAL);
 448   }
 449   debug_only(signal_sets_initialized = true);
 450 
 451 }
 452 
 453 // These are signals that are unblocked while a thread is running Java.
 454 // (For some reason, they get blocked by default.)
 455 sigset_t* os::Linux::unblocked_signals() {
 456   assert(signal_sets_initialized, "Not initialized");
 457   return &unblocked_sigs;
 458 }
 459 
 460 // These are the signals that are blocked while a (non-VM) thread is
 461 // running Java. Only the VM thread handles these signals.
 462 sigset_t* os::Linux::vm_signals() {
 463   assert(signal_sets_initialized, "Not initialized");
 464   return &vm_sigs;
 465 }
 466 
 467 // These are signals that are blocked during cond_wait to allow debugger in
 468 sigset_t* os::Linux::allowdebug_blocked_signals() {
 469   assert(signal_sets_initialized, "Not initialized");
 470   return &allowdebug_blocked_sigs;
 471 }
 472 
 473 void os::Linux::hotspot_sigmask(Thread* thread) {
 474 
 475   //Save caller's signal mask before setting VM signal mask
 476   sigset_t caller_sigmask;
 477   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 478 
 479   OSThread* osthread = thread->osthread();
 480   osthread->set_caller_sigmask(caller_sigmask);
 481 
 482   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 483 
 484   if (!ReduceSignalUsage) {
 485     if (thread->is_VM_thread()) {
 486       // Only the VM thread handles BREAK_SIGNAL ...
 487       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 488     } else {
 489       // ... all other threads block BREAK_SIGNAL
 490       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 491     }
 492   }
 493 }
 494 
 495 //////////////////////////////////////////////////////////////////////////////
 496 // detecting pthread library
 497 
 498 void os::Linux::libpthread_init() {
 499   // Save glibc and pthread version strings.
 500 #if !defined(_CS_GNU_LIBC_VERSION) || \
 501     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 502   #error "glibc too old (< 2.3.2)"
 503 #endif
 504 
 505   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 506   assert(n > 0, "cannot retrieve glibc version");
 507   char *str = (char *)malloc(n, mtInternal);
 508   confstr(_CS_GNU_LIBC_VERSION, str, n);
 509   os::Linux::set_glibc_version(str);
 510 
 511   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 512   assert(n > 0, "cannot retrieve pthread version");
 513   str = (char *)malloc(n, mtInternal);
 514   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 515   os::Linux::set_libpthread_version(str);
 516 }
 517 
 518 /////////////////////////////////////////////////////////////////////////////
 519 // thread stack expansion
 520 
 521 // os::Linux::manually_expand_stack() takes care of expanding the thread
 522 // stack. Note that this is normally not needed: pthread stacks allocate
 523 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 524 // committed. Therefore it is not necessary to expand the stack manually.
 525 //
 526 // Manually expanding the stack was historically needed on LinuxThreads
 527 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 528 // it is kept to deal with very rare corner cases:
 529 //
 530 // For one, user may run the VM on an own implementation of threads
 531 // whose stacks are - like the old LinuxThreads - implemented using
 532 // mmap(MAP_GROWSDOWN).
 533 //
 534 // Also, this coding may be needed if the VM is running on the primordial
 535 // thread. Normally we avoid running on the primordial thread; however,
 536 // user may still invoke the VM on the primordial thread.
 537 //
 538 // The following historical comment describes the details about running
 539 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 540 
 541 
 542 // Force Linux kernel to expand current thread stack. If "bottom" is close
 543 // to the stack guard, caller should block all signals.
 544 //
 545 // MAP_GROWSDOWN:
 546 //   A special mmap() flag that is used to implement thread stacks. It tells
 547 //   kernel that the memory region should extend downwards when needed. This
 548 //   allows early versions of LinuxThreads to only mmap the first few pages
 549 //   when creating a new thread. Linux kernel will automatically expand thread
 550 //   stack as needed (on page faults).
 551 //
 552 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 553 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 554 //   region, it's hard to tell if the fault is due to a legitimate stack
 555 //   access or because of reading/writing non-exist memory (e.g. buffer
 556 //   overrun). As a rule, if the fault happens below current stack pointer,
 557 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 558 //   application (see Linux kernel fault.c).
 559 //
 560 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 561 //   stack overflow detection.
 562 //
 563 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 564 //   not use MAP_GROWSDOWN.
 565 //
 566 // To get around the problem and allow stack banging on Linux, we need to
 567 // manually expand thread stack after receiving the SIGSEGV.
 568 //
 569 // There are two ways to expand thread stack to address "bottom", we used
 570 // both of them in JVM before 1.5:
 571 //   1. adjust stack pointer first so that it is below "bottom", and then
 572 //      touch "bottom"
 573 //   2. mmap() the page in question
 574 //
 575 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 576 // if current sp is already near the lower end of page 101, and we need to
 577 // call mmap() to map page 100, it is possible that part of the mmap() frame
 578 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 579 // That will destroy the mmap() frame and cause VM to crash.
 580 //
 581 // The following code works by adjusting sp first, then accessing the "bottom"
 582 // page to force a page fault. Linux kernel will then automatically expand the
 583 // stack mapping.
 584 //
 585 // _expand_stack_to() assumes its frame size is less than page size, which
 586 // should always be true if the function is not inlined.
 587 
 588 static void NOINLINE _expand_stack_to(address bottom) {
 589   address sp;
 590   size_t size;
 591   volatile char *p;
 592 
 593   // Adjust bottom to point to the largest address within the same page, it
 594   // gives us a one-page buffer if alloca() allocates slightly more memory.
 595   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 596   bottom += os::Linux::page_size() - 1;
 597 
 598   // sp might be slightly above current stack pointer; if that's the case, we
 599   // will alloca() a little more space than necessary, which is OK. Don't use
 600   // os::current_stack_pointer(), as its result can be slightly below current
 601   // stack pointer, causing us to not alloca enough to reach "bottom".
 602   sp = (address)&sp;
 603 
 604   if (sp > bottom) {
 605     size = sp - bottom;
 606     p = (volatile char *)alloca(size);
 607     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 608     p[0] = '\0';
 609   }
 610 }
 611 
 612 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 613   assert(t!=NULL, "just checking");
 614   assert(t->osthread()->expanding_stack(), "expand should be set");
 615   assert(t->stack_base() != NULL, "stack_base was not initialized");
 616 
 617   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 618     sigset_t mask_all, old_sigset;
 619     sigfillset(&mask_all);
 620     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 621     _expand_stack_to(addr);
 622     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 623     return true;
 624   }
 625   return false;
 626 }
 627 
 628 //////////////////////////////////////////////////////////////////////////////
 629 // create new thread
 630 
 631 // Thread start routine for all newly created threads
 632 static void *thread_native_entry(Thread *thread) {
 633   // Try to randomize the cache line index of hot stack frames.
 634   // This helps when threads of the same stack traces evict each other's
 635   // cache lines. The threads can be either from the same JVM instance, or
 636   // from different JVM instances. The benefit is especially true for
 637   // processors with hyperthreading technology.
 638   static int counter = 0;
 639   int pid = os::current_process_id();
 640   alloca(((pid ^ counter++) & 7) * 128);
 641 
 642   thread->initialize_thread_current();
 643 
 644   OSThread* osthread = thread->osthread();
 645   Monitor* sync = osthread->startThread_lock();
 646 
 647   osthread->set_thread_id(os::current_thread_id());
 648 
 649   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 650     os::current_thread_id(), (uintx) pthread_self());
 651 
 652   if (UseNUMA) {
 653     int lgrp_id = os::numa_get_group_id();
 654     if (lgrp_id != -1) {
 655       thread->set_lgrp_id(lgrp_id);
 656     }
 657   }
 658   // initialize signal mask for this thread
 659   os::Linux::hotspot_sigmask(thread);
 660 
 661   // initialize floating point control register
 662   os::Linux::init_thread_fpu_state();
 663 
 664   // handshaking with parent thread
 665   {
 666     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 667 
 668     // notify parent thread
 669     osthread->set_state(INITIALIZED);
 670     sync->notify_all();
 671 
 672     // wait until os::start_thread()
 673     while (osthread->get_state() == INITIALIZED) {
 674       sync->wait(Mutex::_no_safepoint_check_flag);
 675     }
 676   }
 677 
 678   // call one more level start routine
 679   thread->run();
 680 
 681   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 682     os::current_thread_id(), (uintx) pthread_self());
 683 
 684   // If a thread has not deleted itself ("delete this") as part of its
 685   // termination sequence, we have to ensure thread-local-storage is
 686   // cleared before we actually terminate. No threads should ever be
 687   // deleted asynchronously with respect to their termination.
 688   if (Thread::current_or_null_safe() != NULL) {
 689     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 690     thread->clear_thread_current();
 691   }
 692 
 693   return 0;
 694 }
 695 
 696 bool os::create_thread(Thread* thread, ThreadType thr_type,
 697                        size_t req_stack_size) {
 698   assert(thread->osthread() == NULL, "caller responsible");
 699 
 700   // Allocate the OSThread object
 701   OSThread* osthread = new OSThread(NULL, NULL);
 702   if (osthread == NULL) {
 703     return false;
 704   }
 705 
 706   // set the correct thread state
 707   osthread->set_thread_type(thr_type);
 708 
 709   // Initial state is ALLOCATED but not INITIALIZED
 710   osthread->set_state(ALLOCATED);
 711 
 712   thread->set_osthread(osthread);
 713 
 714   // init thread attributes
 715   pthread_attr_t attr;
 716   pthread_attr_init(&attr);
 717   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 718 
 719   // Calculate stack size if it's not specified by caller.
 720   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 721   // In the Linux NPTL pthread implementation the guard size mechanism
 722   // is not implemented properly. The posix standard requires adding
 723   // the size of the guard pages to the stack size, instead Linux
 724   // takes the space out of 'stacksize'. Thus we adapt the requested
 725   // stack_size by the size of the guard pages to mimick proper
 726   // behaviour.
 727   stack_size = align_size_up(stack_size + os::Linux::default_guard_size(thr_type), vm_page_size());
 728   pthread_attr_setstacksize(&attr, stack_size);
 729 
 730   // Configure glibc guard page.
 731   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 732 
 733   ThreadState state;
 734 
 735   {
 736     pthread_t tid;
 737     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 738 
 739     char buf[64];
 740     if (ret == 0) {
 741       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 742         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 743     } else {
 744       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 745         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 746     }
 747 
 748     pthread_attr_destroy(&attr);
 749 
 750     if (ret != 0) {
 751       // Need to clean up stuff we've allocated so far
 752       thread->set_osthread(NULL);
 753       delete osthread;
 754       return false;
 755     }
 756 
 757     // Store pthread info into the OSThread
 758     osthread->set_pthread_id(tid);
 759 
 760     // Wait until child thread is either initialized or aborted
 761     {
 762       Monitor* sync_with_child = osthread->startThread_lock();
 763       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 764       while ((state = osthread->get_state()) == ALLOCATED) {
 765         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 766       }
 767     }
 768   }
 769 
 770   // Aborted due to thread limit being reached
 771   if (state == ZOMBIE) {
 772     thread->set_osthread(NULL);
 773     delete osthread;
 774     return false;
 775   }
 776 
 777   // The thread is returned suspended (in state INITIALIZED),
 778   // and is started higher up in the call chain
 779   assert(state == INITIALIZED, "race condition");
 780   return true;
 781 }
 782 
 783 /////////////////////////////////////////////////////////////////////////////
 784 // attach existing thread
 785 
 786 // bootstrap the main thread
 787 bool os::create_main_thread(JavaThread* thread) {
 788   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 789   return create_attached_thread(thread);
 790 }
 791 
 792 bool os::create_attached_thread(JavaThread* thread) {
 793 #ifdef ASSERT
 794   thread->verify_not_published();
 795 #endif
 796 
 797   // Allocate the OSThread object
 798   OSThread* osthread = new OSThread(NULL, NULL);
 799 
 800   if (osthread == NULL) {
 801     return false;
 802   }
 803 
 804   // Store pthread info into the OSThread
 805   osthread->set_thread_id(os::Linux::gettid());
 806   osthread->set_pthread_id(::pthread_self());
 807 
 808   // initialize floating point control register
 809   os::Linux::init_thread_fpu_state();
 810 
 811   // Initial thread state is RUNNABLE
 812   osthread->set_state(RUNNABLE);
 813 
 814   thread->set_osthread(osthread);
 815 
 816   if (UseNUMA) {
 817     int lgrp_id = os::numa_get_group_id();
 818     if (lgrp_id != -1) {
 819       thread->set_lgrp_id(lgrp_id);
 820     }
 821   }
 822 
 823   if (os::Linux::is_initial_thread()) {
 824     // If current thread is initial thread, its stack is mapped on demand,
 825     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 826     // the entire stack region to avoid SEGV in stack banging.
 827     // It is also useful to get around the heap-stack-gap problem on SuSE
 828     // kernel (see 4821821 for details). We first expand stack to the top
 829     // of yellow zone, then enable stack yellow zone (order is significant,
 830     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 831     // is no gap between the last two virtual memory regions.
 832 
 833     JavaThread *jt = (JavaThread *)thread;
 834     address addr = jt->stack_reserved_zone_base();
 835     assert(addr != NULL, "initialization problem?");
 836     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 837 
 838     osthread->set_expanding_stack();
 839     os::Linux::manually_expand_stack(jt, addr);
 840     osthread->clear_expanding_stack();
 841   }
 842 
 843   // initialize signal mask for this thread
 844   // and save the caller's signal mask
 845   os::Linux::hotspot_sigmask(thread);
 846 
 847   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 848     os::current_thread_id(), (uintx) pthread_self());
 849 
 850   return true;
 851 }
 852 
 853 void os::pd_start_thread(Thread* thread) {
 854   OSThread * osthread = thread->osthread();
 855   assert(osthread->get_state() != INITIALIZED, "just checking");
 856   Monitor* sync_with_child = osthread->startThread_lock();
 857   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 858   sync_with_child->notify();
 859 }
 860 
 861 // Free Linux resources related to the OSThread
 862 void os::free_thread(OSThread* osthread) {
 863   assert(osthread != NULL, "osthread not set");
 864 
 865   // We are told to free resources of the argument thread,
 866   // but we can only really operate on the current thread.
 867   assert(Thread::current()->osthread() == osthread,
 868          "os::free_thread but not current thread");
 869 
 870 #ifdef ASSERT
 871   sigset_t current;
 872   sigemptyset(&current);
 873   pthread_sigmask(SIG_SETMASK, NULL, &current);
 874   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 875 #endif
 876 
 877   // Restore caller's signal mask
 878   sigset_t sigmask = osthread->caller_sigmask();
 879   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 880 
 881   delete osthread;
 882 }
 883 
 884 //////////////////////////////////////////////////////////////////////////////
 885 // initial thread
 886 
 887 // Check if current thread is the initial thread, similar to Solaris thr_main.
 888 bool os::Linux::is_initial_thread(void) {
 889   char dummy;
 890   // If called before init complete, thread stack bottom will be null.
 891   // Can be called if fatal error occurs before initialization.
 892   if (initial_thread_stack_bottom() == NULL) return false;
 893   assert(initial_thread_stack_bottom() != NULL &&
 894          initial_thread_stack_size()   != 0,
 895          "os::init did not locate initial thread's stack region");
 896   if ((address)&dummy >= initial_thread_stack_bottom() &&
 897       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 898     return true;
 899   } else {
 900     return false;
 901   }
 902 }
 903 
 904 // Find the virtual memory area that contains addr
 905 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 906   FILE *fp = fopen("/proc/self/maps", "r");
 907   if (fp) {
 908     address low, high;
 909     while (!feof(fp)) {
 910       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 911         if (low <= addr && addr < high) {
 912           if (vma_low)  *vma_low  = low;
 913           if (vma_high) *vma_high = high;
 914           fclose(fp);
 915           return true;
 916         }
 917       }
 918       for (;;) {
 919         int ch = fgetc(fp);
 920         if (ch == EOF || ch == (int)'\n') break;
 921       }
 922     }
 923     fclose(fp);
 924   }
 925   return false;
 926 }
 927 
 928 // Locate initial thread stack. This special handling of initial thread stack
 929 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 930 // bogus value for the primordial process thread. While the launcher has created
 931 // the VM in a new thread since JDK 6, we still have to allow for the use of the
 932 // JNI invocation API from a primordial thread.
 933 void os::Linux::capture_initial_stack(size_t max_size) {
 934 
 935   // max_size is either 0 (which means accept OS default for thread stacks) or
 936   // a user-specified value known to be at least the minimum needed. If we
 937   // are actually on the primordial thread we can make it appear that we have a
 938   // smaller max_size stack by inserting the guard pages at that location. But we
 939   // cannot do anything to emulate a larger stack than what has been provided by
 940   // the OS or threading library. In fact if we try to use a stack greater than
 941   // what is set by rlimit then we will crash the hosting process.
 942 
 943   // Maximum stack size is the easy part, get it from RLIMIT_STACK.
 944   // If this is "unlimited" then it will be a huge value.
 945   struct rlimit rlim;
 946   getrlimit(RLIMIT_STACK, &rlim);
 947   size_t stack_size = rlim.rlim_cur;
 948 
 949   // 6308388: a bug in ld.so will relocate its own .data section to the
 950   //   lower end of primordial stack; reduce ulimit -s value a little bit
 951   //   so we won't install guard page on ld.so's data section.
 952   stack_size -= 2 * page_size();
 953 
 954   // Try to figure out where the stack base (top) is. This is harder.
 955   //
 956   // When an application is started, glibc saves the initial stack pointer in
 957   // a global variable "__libc_stack_end", which is then used by system
 958   // libraries. __libc_stack_end should be pretty close to stack top. The
 959   // variable is available since the very early days. However, because it is
 960   // a private interface, it could disappear in the future.
 961   //
 962   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 963   // to __libc_stack_end, it is very close to stack top, but isn't the real
 964   // stack top. Note that /proc may not exist if VM is running as a chroot
 965   // program, so reading /proc/<pid>/stat could fail. Also the contents of
 966   // /proc/<pid>/stat could change in the future (though unlikely).
 967   //
 968   // We try __libc_stack_end first. If that doesn't work, look for
 969   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
 970   // as a hint, which should work well in most cases.
 971 
 972   uintptr_t stack_start;
 973 
 974   // try __libc_stack_end first
 975   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
 976   if (p && *p) {
 977     stack_start = *p;
 978   } else {
 979     // see if we can get the start_stack field from /proc/self/stat
 980     FILE *fp;
 981     int pid;
 982     char state;
 983     int ppid;
 984     int pgrp;
 985     int session;
 986     int nr;
 987     int tpgrp;
 988     unsigned long flags;
 989     unsigned long minflt;
 990     unsigned long cminflt;
 991     unsigned long majflt;
 992     unsigned long cmajflt;
 993     unsigned long utime;
 994     unsigned long stime;
 995     long cutime;
 996     long cstime;
 997     long prio;
 998     long nice;
 999     long junk;
1000     long it_real;
1001     uintptr_t start;
1002     uintptr_t vsize;
1003     intptr_t rss;
1004     uintptr_t rsslim;
1005     uintptr_t scodes;
1006     uintptr_t ecode;
1007     int i;
1008 
1009     // Figure what the primordial thread stack base is. Code is inspired
1010     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1011     // followed by command name surrounded by parentheses, state, etc.
1012     char stat[2048];
1013     int statlen;
1014 
1015     fp = fopen("/proc/self/stat", "r");
1016     if (fp) {
1017       statlen = fread(stat, 1, 2047, fp);
1018       stat[statlen] = '\0';
1019       fclose(fp);
1020 
1021       // Skip pid and the command string. Note that we could be dealing with
1022       // weird command names, e.g. user could decide to rename java launcher
1023       // to "java 1.4.2 :)", then the stat file would look like
1024       //                1234 (java 1.4.2 :)) R ... ...
1025       // We don't really need to know the command string, just find the last
1026       // occurrence of ")" and then start parsing from there. See bug 4726580.
1027       char * s = strrchr(stat, ')');
1028 
1029       i = 0;
1030       if (s) {
1031         // Skip blank chars
1032         do { s++; } while (s && isspace(*s));
1033 
1034 #define _UFM UINTX_FORMAT
1035 #define _DFM INTX_FORMAT
1036 
1037         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1038         //              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
1039         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,
1040                    &state,          // 3  %c
1041                    &ppid,           // 4  %d
1042                    &pgrp,           // 5  %d
1043                    &session,        // 6  %d
1044                    &nr,             // 7  %d
1045                    &tpgrp,          // 8  %d
1046                    &flags,          // 9  %lu
1047                    &minflt,         // 10 %lu
1048                    &cminflt,        // 11 %lu
1049                    &majflt,         // 12 %lu
1050                    &cmajflt,        // 13 %lu
1051                    &utime,          // 14 %lu
1052                    &stime,          // 15 %lu
1053                    &cutime,         // 16 %ld
1054                    &cstime,         // 17 %ld
1055                    &prio,           // 18 %ld
1056                    &nice,           // 19 %ld
1057                    &junk,           // 20 %ld
1058                    &it_real,        // 21 %ld
1059                    &start,          // 22 UINTX_FORMAT
1060                    &vsize,          // 23 UINTX_FORMAT
1061                    &rss,            // 24 INTX_FORMAT
1062                    &rsslim,         // 25 UINTX_FORMAT
1063                    &scodes,         // 26 UINTX_FORMAT
1064                    &ecode,          // 27 UINTX_FORMAT
1065                    &stack_start);   // 28 UINTX_FORMAT
1066       }
1067 
1068 #undef _UFM
1069 #undef _DFM
1070 
1071       if (i != 28 - 2) {
1072         assert(false, "Bad conversion from /proc/self/stat");
1073         // product mode - assume we are the initial thread, good luck in the
1074         // embedded case.
1075         warning("Can't detect initial thread stack location - bad conversion");
1076         stack_start = (uintptr_t) &rlim;
1077       }
1078     } else {
1079       // For some reason we can't open /proc/self/stat (for example, running on
1080       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1081       // most cases, so don't abort:
1082       warning("Can't detect initial thread stack location - no /proc/self/stat");
1083       stack_start = (uintptr_t) &rlim;
1084     }
1085   }
1086 
1087   // Now we have a pointer (stack_start) very close to the stack top, the
1088   // next thing to do is to figure out the exact location of stack top. We
1089   // can find out the virtual memory area that contains stack_start by
1090   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1091   // and its upper limit is the real stack top. (again, this would fail if
1092   // running inside chroot, because /proc may not exist.)
1093 
1094   uintptr_t stack_top;
1095   address low, high;
1096   if (find_vma((address)stack_start, &low, &high)) {
1097     // success, "high" is the true stack top. (ignore "low", because initial
1098     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1099     stack_top = (uintptr_t)high;
1100   } else {
1101     // failed, likely because /proc/self/maps does not exist
1102     warning("Can't detect initial thread stack location - find_vma failed");
1103     // best effort: stack_start is normally within a few pages below the real
1104     // stack top, use it as stack top, and reduce stack size so we won't put
1105     // guard page outside stack.
1106     stack_top = stack_start;
1107     stack_size -= 16 * page_size();
1108   }
1109 
1110   // stack_top could be partially down the page so align it
1111   stack_top = align_size_up(stack_top, page_size());
1112 
1113   // Allowed stack value is minimum of max_size and what we derived from rlimit
1114   if (max_size > 0) {
1115     _initial_thread_stack_size = MIN2(max_size, stack_size);
1116   } else {
1117     // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1118     // clamp it at 8MB as we do on Solaris
1119     _initial_thread_stack_size = MIN2(stack_size, 8*M);
1120   }
1121   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1122   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1123 
1124   assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1125 
1126   if (log_is_enabled(Info, os, thread)) {
1127     // See if we seem to be on primordial process thread
1128     bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1129                       uintptr_t(&rlim) < stack_top;
1130 
1131     log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1132                          SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1133                          primordial ? "primordial" : "user", max_size / K,  _initial_thread_stack_size / K,
1134                          stack_top, intptr_t(_initial_thread_stack_bottom));
1135   }
1136 }
1137 
1138 ////////////////////////////////////////////////////////////////////////////////
1139 // time support
1140 
1141 // Time since start-up in seconds to a fine granularity.
1142 // Used by VMSelfDestructTimer and the MemProfiler.
1143 double os::elapsedTime() {
1144 
1145   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1146 }
1147 
1148 jlong os::elapsed_counter() {
1149   return javaTimeNanos() - initial_time_count;
1150 }
1151 
1152 jlong os::elapsed_frequency() {
1153   return NANOSECS_PER_SEC; // nanosecond resolution
1154 }
1155 
1156 bool os::supports_vtime() { return true; }
1157 bool os::enable_vtime()   { return false; }
1158 bool os::vtime_enabled()  { return false; }
1159 
1160 double os::elapsedVTime() {
1161   struct rusage usage;
1162   int retval = getrusage(RUSAGE_THREAD, &usage);
1163   if (retval == 0) {
1164     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);
1165   } else {
1166     // better than nothing, but not much
1167     return elapsedTime();
1168   }
1169 }
1170 
1171 jlong os::javaTimeMillis() {
1172   timeval time;
1173   int status = gettimeofday(&time, NULL);
1174   assert(status != -1, "linux error");
1175   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1176 }
1177 
1178 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1179   timeval time;
1180   int status = gettimeofday(&time, NULL);
1181   assert(status != -1, "linux error");
1182   seconds = jlong(time.tv_sec);
1183   nanos = jlong(time.tv_usec) * 1000;
1184 }
1185 
1186 
1187 #ifndef CLOCK_MONOTONIC
1188   #define CLOCK_MONOTONIC (1)
1189 #endif
1190 
1191 void os::Linux::clock_init() {
1192   // we do dlopen's in this particular order due to bug in linux
1193   // dynamical loader (see 6348968) leading to crash on exit
1194   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1195   if (handle == NULL) {
1196     handle = dlopen("librt.so", RTLD_LAZY);
1197   }
1198 
1199   if (handle) {
1200     int (*clock_getres_func)(clockid_t, struct timespec*) =
1201            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1202     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1203            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1204     if (clock_getres_func && clock_gettime_func) {
1205       // See if monotonic clock is supported by the kernel. Note that some
1206       // early implementations simply return kernel jiffies (updated every
1207       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1208       // for nano time (though the monotonic property is still nice to have).
1209       // It's fixed in newer kernels, however clock_getres() still returns
1210       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1211       // resolution for now. Hopefully as people move to new kernels, this
1212       // won't be a problem.
1213       struct timespec res;
1214       struct timespec tp;
1215       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1216           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1217         // yes, monotonic clock is supported
1218         _clock_gettime = clock_gettime_func;
1219         return;
1220       } else {
1221         // close librt if there is no monotonic clock
1222         dlclose(handle);
1223       }
1224     }
1225   }
1226   warning("No monotonic clock was available - timed services may " \
1227           "be adversely affected if the time-of-day clock changes");
1228 }
1229 
1230 #ifndef SYS_clock_getres
1231   #if defined(X86) || defined(PPC64) || defined(S390)
1232     #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1233     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1234   #else
1235     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1236     #define sys_clock_getres(x,y)  -1
1237   #endif
1238 #else
1239   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1240 #endif
1241 
1242 void os::Linux::fast_thread_clock_init() {
1243   if (!UseLinuxPosixThreadCPUClocks) {
1244     return;
1245   }
1246   clockid_t clockid;
1247   struct timespec tp;
1248   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1249       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1250 
1251   // Switch to using fast clocks for thread cpu time if
1252   // the sys_clock_getres() returns 0 error code.
1253   // Note, that some kernels may support the current thread
1254   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1255   // returned by the pthread_getcpuclockid().
1256   // If the fast Posix clocks are supported then the sys_clock_getres()
1257   // must return at least tp.tv_sec == 0 which means a resolution
1258   // better than 1 sec. This is extra check for reliability.
1259 
1260   if (pthread_getcpuclockid_func &&
1261       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1262       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1263     _supports_fast_thread_cpu_time = true;
1264     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1265   }
1266 }
1267 
1268 jlong os::javaTimeNanos() {
1269   if (os::supports_monotonic_clock()) {
1270     struct timespec tp;
1271     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1272     assert(status == 0, "gettime error");
1273     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1274     return result;
1275   } else {
1276     timeval time;
1277     int status = gettimeofday(&time, NULL);
1278     assert(status != -1, "linux error");
1279     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1280     return 1000 * usecs;
1281   }
1282 }
1283 
1284 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1285   if (os::supports_monotonic_clock()) {
1286     info_ptr->max_value = ALL_64_BITS;
1287 
1288     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1289     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1290     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1291   } else {
1292     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1293     info_ptr->max_value = ALL_64_BITS;
1294 
1295     // gettimeofday is a real time clock so it skips
1296     info_ptr->may_skip_backward = true;
1297     info_ptr->may_skip_forward = true;
1298   }
1299 
1300   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1301 }
1302 
1303 // Return the real, user, and system times in seconds from an
1304 // arbitrary fixed point in the past.
1305 bool os::getTimesSecs(double* process_real_time,
1306                       double* process_user_time,
1307                       double* process_system_time) {
1308   struct tms ticks;
1309   clock_t real_ticks = times(&ticks);
1310 
1311   if (real_ticks == (clock_t) (-1)) {
1312     return false;
1313   } else {
1314     double ticks_per_second = (double) clock_tics_per_sec;
1315     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1316     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1317     *process_real_time = ((double) real_ticks) / ticks_per_second;
1318 
1319     return true;
1320   }
1321 }
1322 
1323 
1324 char * os::local_time_string(char *buf, size_t buflen) {
1325   struct tm t;
1326   time_t long_time;
1327   time(&long_time);
1328   localtime_r(&long_time, &t);
1329   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1330                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1331                t.tm_hour, t.tm_min, t.tm_sec);
1332   return buf;
1333 }
1334 
1335 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1336   return localtime_r(clock, res);
1337 }
1338 
1339 ////////////////////////////////////////////////////////////////////////////////
1340 // runtime exit support
1341 
1342 // Note: os::shutdown() might be called very early during initialization, or
1343 // called from signal handler. Before adding something to os::shutdown(), make
1344 // sure it is async-safe and can handle partially initialized VM.
1345 void os::shutdown() {
1346 
1347   // allow PerfMemory to attempt cleanup of any persistent resources
1348   perfMemory_exit();
1349 
1350   // needs to remove object in file system
1351   AttachListener::abort();
1352 
1353   // flush buffered output, finish log files
1354   ostream_abort();
1355 
1356   // Check for abort hook
1357   abort_hook_t abort_hook = Arguments::abort_hook();
1358   if (abort_hook != NULL) {
1359     abort_hook();
1360   }
1361 
1362 }
1363 
1364 // Note: os::abort() might be called very early during initialization, or
1365 // called from signal handler. Before adding something to os::abort(), make
1366 // sure it is async-safe and can handle partially initialized VM.
1367 void os::abort(bool dump_core, void* siginfo, const void* context) {
1368   os::shutdown();
1369   if (dump_core) {
1370 #ifndef PRODUCT
1371     fdStream out(defaultStream::output_fd());
1372     out.print_raw("Current thread is ");
1373     char buf[16];
1374     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1375     out.print_raw_cr(buf);
1376     out.print_raw_cr("Dumping core ...");
1377 #endif
1378     ::abort(); // dump core
1379   }
1380 
1381   ::exit(1);
1382 }
1383 
1384 // Die immediately, no exit hook, no abort hook, no cleanup.
1385 void os::die() {
1386   ::abort();
1387 }
1388 
1389 
1390 // This method is a copy of JDK's sysGetLastErrorString
1391 // from src/solaris/hpi/src/system_md.c
1392 
1393 size_t os::lasterror(char *buf, size_t len) {
1394   if (errno == 0)  return 0;
1395 
1396   const char *s = os::strerror(errno);
1397   size_t n = ::strlen(s);
1398   if (n >= len) {
1399     n = len - 1;
1400   }
1401   ::strncpy(buf, s, n);
1402   buf[n] = '\0';
1403   return n;
1404 }
1405 
1406 // thread_id is kernel thread id (similar to Solaris LWP id)
1407 intx os::current_thread_id() { return os::Linux::gettid(); }
1408 int os::current_process_id() {
1409   return ::getpid();
1410 }
1411 
1412 // DLL functions
1413 
1414 const char* os::dll_file_extension() { return ".so"; }
1415 
1416 // This must be hard coded because it's the system's temporary
1417 // directory not the java application's temp directory, ala java.io.tmpdir.
1418 const char* os::get_temp_directory() { return "/tmp"; }
1419 
1420 static bool file_exists(const char* filename) {
1421   struct stat statbuf;
1422   if (filename == NULL || strlen(filename) == 0) {
1423     return false;
1424   }
1425   return os::stat(filename, &statbuf) == 0;
1426 }
1427 
1428 bool os::dll_build_name(char* buffer, size_t buflen,
1429                         const char* pname, const char* fname) {
1430   bool retval = false;
1431   // Copied from libhpi
1432   const size_t pnamelen = pname ? strlen(pname) : 0;
1433 
1434   // Return error on buffer overflow.
1435   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1436     return retval;
1437   }
1438 
1439   if (pnamelen == 0) {
1440     snprintf(buffer, buflen, "lib%s.so", fname);
1441     retval = true;
1442   } else if (strchr(pname, *os::path_separator()) != NULL) {
1443     int n;
1444     char** pelements = split_path(pname, &n);
1445     if (pelements == NULL) {
1446       return false;
1447     }
1448     for (int i = 0; i < n; i++) {
1449       // Really shouldn't be NULL, but check can't hurt
1450       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1451         continue; // skip the empty path values
1452       }
1453       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1454       if (file_exists(buffer)) {
1455         retval = true;
1456         break;
1457       }
1458     }
1459     // release the storage
1460     for (int i = 0; i < n; i++) {
1461       if (pelements[i] != NULL) {
1462         FREE_C_HEAP_ARRAY(char, pelements[i]);
1463       }
1464     }
1465     if (pelements != NULL) {
1466       FREE_C_HEAP_ARRAY(char*, pelements);
1467     }
1468   } else {
1469     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1470     retval = true;
1471   }
1472   return retval;
1473 }
1474 
1475 // check if addr is inside libjvm.so
1476 bool os::address_is_in_vm(address addr) {
1477   static address libjvm_base_addr;
1478   Dl_info dlinfo;
1479 
1480   if (libjvm_base_addr == NULL) {
1481     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1482       libjvm_base_addr = (address)dlinfo.dli_fbase;
1483     }
1484     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1485   }
1486 
1487   if (dladdr((void *)addr, &dlinfo) != 0) {
1488     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1489   }
1490 
1491   return false;
1492 }
1493 
1494 bool os::dll_address_to_function_name(address addr, char *buf,
1495                                       int buflen, int *offset,
1496                                       bool demangle) {
1497   // buf is not optional, but offset is optional
1498   assert(buf != NULL, "sanity check");
1499 
1500   Dl_info dlinfo;
1501 
1502   if (dladdr((void*)addr, &dlinfo) != 0) {
1503     // see if we have a matching symbol
1504     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1505       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1506         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1507       }
1508       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1509       return true;
1510     }
1511     // no matching symbol so try for just file info
1512     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1513       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1514                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1515         return true;
1516       }
1517     }
1518   }
1519 
1520   buf[0] = '\0';
1521   if (offset != NULL) *offset = -1;
1522   return false;
1523 }
1524 
1525 struct _address_to_library_name {
1526   address addr;          // input : memory address
1527   size_t  buflen;        //         size of fname
1528   char*   fname;         // output: library name
1529   address base;          //         library base addr
1530 };
1531 
1532 static int address_to_library_name_callback(struct dl_phdr_info *info,
1533                                             size_t size, void *data) {
1534   int i;
1535   bool found = false;
1536   address libbase = NULL;
1537   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1538 
1539   // iterate through all loadable segments
1540   for (i = 0; i < info->dlpi_phnum; i++) {
1541     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1542     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1543       // base address of a library is the lowest address of its loaded
1544       // segments.
1545       if (libbase == NULL || libbase > segbase) {
1546         libbase = segbase;
1547       }
1548       // see if 'addr' is within current segment
1549       if (segbase <= d->addr &&
1550           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1551         found = true;
1552       }
1553     }
1554   }
1555 
1556   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1557   // so dll_address_to_library_name() can fall through to use dladdr() which
1558   // can figure out executable name from argv[0].
1559   if (found && info->dlpi_name && info->dlpi_name[0]) {
1560     d->base = libbase;
1561     if (d->fname) {
1562       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1563     }
1564     return 1;
1565   }
1566   return 0;
1567 }
1568 
1569 bool os::dll_address_to_library_name(address addr, char* buf,
1570                                      int buflen, int* offset) {
1571   // buf is not optional, but offset is optional
1572   assert(buf != NULL, "sanity check");
1573 
1574   Dl_info dlinfo;
1575   struct _address_to_library_name data;
1576 
1577   // There is a bug in old glibc dladdr() implementation that it could resolve
1578   // to wrong library name if the .so file has a base address != NULL. Here
1579   // we iterate through the program headers of all loaded libraries to find
1580   // out which library 'addr' really belongs to. This workaround can be
1581   // removed once the minimum requirement for glibc is moved to 2.3.x.
1582   data.addr = addr;
1583   data.fname = buf;
1584   data.buflen = buflen;
1585   data.base = NULL;
1586   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1587 
1588   if (rslt) {
1589     // buf already contains library name
1590     if (offset) *offset = addr - data.base;
1591     return true;
1592   }
1593   if (dladdr((void*)addr, &dlinfo) != 0) {
1594     if (dlinfo.dli_fname != NULL) {
1595       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1596     }
1597     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1598       *offset = addr - (address)dlinfo.dli_fbase;
1599     }
1600     return true;
1601   }
1602 
1603   buf[0] = '\0';
1604   if (offset) *offset = -1;
1605   return false;
1606 }
1607 
1608 // Loads .dll/.so and
1609 // in case of error it checks if .dll/.so was built for the
1610 // same architecture as Hotspot is running on
1611 
1612 
1613 // Remember the stack's state. The Linux dynamic linker will change
1614 // the stack to 'executable' at most once, so we must safepoint only once.
1615 bool os::Linux::_stack_is_executable = false;
1616 
1617 // VM operation that loads a library.  This is necessary if stack protection
1618 // of the Java stacks can be lost during loading the library.  If we
1619 // do not stop the Java threads, they can stack overflow before the stacks
1620 // are protected again.
1621 class VM_LinuxDllLoad: public VM_Operation {
1622  private:
1623   const char *_filename;
1624   char *_ebuf;
1625   int _ebuflen;
1626   void *_lib;
1627  public:
1628   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1629     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1630   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1631   void doit() {
1632     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1633     os::Linux::_stack_is_executable = true;
1634   }
1635   void* loaded_library() { return _lib; }
1636 };
1637 
1638 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1639   void * result = NULL;
1640   bool load_attempted = false;
1641 
1642   // Check whether the library to load might change execution rights
1643   // of the stack. If they are changed, the protection of the stack
1644   // guard pages will be lost. We need a safepoint to fix this.
1645   //
1646   // See Linux man page execstack(8) for more info.
1647   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1648     ElfFile ef(filename);
1649     if (!ef.specifies_noexecstack()) {
1650       if (!is_init_completed()) {
1651         os::Linux::_stack_is_executable = true;
1652         // This is OK - No Java threads have been created yet, and hence no
1653         // stack guard pages to fix.
1654         //
1655         // This should happen only when you are building JDK7 using a very
1656         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1657         //
1658         // Dynamic loader will make all stacks executable after
1659         // this function returns, and will not do that again.
1660         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1661       } else {
1662         warning("You have loaded library %s which might have disabled stack guard. "
1663                 "The VM will try to fix the stack guard now.\n"
1664                 "It's highly recommended that you fix the library with "
1665                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1666                 filename);
1667 
1668         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1669         JavaThread *jt = JavaThread::current();
1670         if (jt->thread_state() != _thread_in_native) {
1671           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1672           // that requires ExecStack. Cannot enter safe point. Let's give up.
1673           warning("Unable to fix stack guard. Giving up.");
1674         } else {
1675           if (!LoadExecStackDllInVMThread) {
1676             // This is for the case where the DLL has an static
1677             // constructor function that executes JNI code. We cannot
1678             // load such DLLs in the VMThread.
1679             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1680           }
1681 
1682           ThreadInVMfromNative tiv(jt);
1683           debug_only(VMNativeEntryWrapper vew;)
1684 
1685           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1686           VMThread::execute(&op);
1687           if (LoadExecStackDllInVMThread) {
1688             result = op.loaded_library();
1689           }
1690           load_attempted = true;
1691         }
1692       }
1693     }
1694   }
1695 
1696   if (!load_attempted) {
1697     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1698   }
1699 
1700   if (result != NULL) {
1701     // Successful loading
1702     return result;
1703   }
1704 
1705   Elf32_Ehdr elf_head;
1706   int diag_msg_max_length=ebuflen-strlen(ebuf);
1707   char* diag_msg_buf=ebuf+strlen(ebuf);
1708 
1709   if (diag_msg_max_length==0) {
1710     // No more space in ebuf for additional diagnostics message
1711     return NULL;
1712   }
1713 
1714 
1715   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1716 
1717   if (file_descriptor < 0) {
1718     // Can't open library, report dlerror() message
1719     return NULL;
1720   }
1721 
1722   bool failed_to_read_elf_head=
1723     (sizeof(elf_head)!=
1724      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1725 
1726   ::close(file_descriptor);
1727   if (failed_to_read_elf_head) {
1728     // file i/o error - report dlerror() msg
1729     return NULL;
1730   }
1731 
1732   typedef struct {
1733     Elf32_Half    code;         // Actual value as defined in elf.h
1734     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1735     unsigned char elf_class;    // 32 or 64 bit
1736     unsigned char endianess;    // MSB or LSB
1737     char*         name;         // String representation
1738   } arch_t;
1739 
1740 #ifndef EM_486
1741   #define EM_486          6               /* Intel 80486 */
1742 #endif
1743 #ifndef EM_AARCH64
1744   #define EM_AARCH64    183               /* ARM AARCH64 */
1745 #endif
1746 
1747   static const arch_t arch_array[]={
1748     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1749     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1750     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1751     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1752     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1753     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1754     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1755     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1756 #if defined(VM_LITTLE_ENDIAN)
1757     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1758 #else
1759     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1760 #endif
1761     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1762     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1763     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1764     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1765     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1766     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1767     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1768     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1769   };
1770 
1771 #if  (defined IA32)
1772   static  Elf32_Half running_arch_code=EM_386;
1773 #elif   (defined AMD64)
1774   static  Elf32_Half running_arch_code=EM_X86_64;
1775 #elif  (defined IA64)
1776   static  Elf32_Half running_arch_code=EM_IA_64;
1777 #elif  (defined __sparc) && (defined _LP64)
1778   static  Elf32_Half running_arch_code=EM_SPARCV9;
1779 #elif  (defined __sparc) && (!defined _LP64)
1780   static  Elf32_Half running_arch_code=EM_SPARC;
1781 #elif  (defined __powerpc64__)
1782   static  Elf32_Half running_arch_code=EM_PPC64;
1783 #elif  (defined __powerpc__)
1784   static  Elf32_Half running_arch_code=EM_PPC;
1785 #elif  (defined AARCH64)
1786   static  Elf32_Half running_arch_code=EM_AARCH64;
1787 #elif  (defined ARM)
1788   static  Elf32_Half running_arch_code=EM_ARM;
1789 #elif  (defined S390)
1790   static  Elf32_Half running_arch_code=EM_S390;
1791 #elif  (defined ALPHA)
1792   static  Elf32_Half running_arch_code=EM_ALPHA;
1793 #elif  (defined MIPSEL)
1794   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1795 #elif  (defined PARISC)
1796   static  Elf32_Half running_arch_code=EM_PARISC;
1797 #elif  (defined MIPS)
1798   static  Elf32_Half running_arch_code=EM_MIPS;
1799 #elif  (defined M68K)
1800   static  Elf32_Half running_arch_code=EM_68K;
1801 #else
1802     #error Method os::dll_load requires that one of following is defined:\
1803         AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, __sparc
1804 #endif
1805 
1806   // Identify compatability class for VM's architecture and library's architecture
1807   // Obtain string descriptions for architectures
1808 
1809   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1810   int running_arch_index=-1;
1811 
1812   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1813     if (running_arch_code == arch_array[i].code) {
1814       running_arch_index    = i;
1815     }
1816     if (lib_arch.code == arch_array[i].code) {
1817       lib_arch.compat_class = arch_array[i].compat_class;
1818       lib_arch.name         = arch_array[i].name;
1819     }
1820   }
1821 
1822   assert(running_arch_index != -1,
1823          "Didn't find running architecture code (running_arch_code) in arch_array");
1824   if (running_arch_index == -1) {
1825     // Even though running architecture detection failed
1826     // we may still continue with reporting dlerror() message
1827     return NULL;
1828   }
1829 
1830   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1831     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1832     return NULL;
1833   }
1834 
1835 #ifndef S390
1836   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1837     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1838     return NULL;
1839   }
1840 #endif // !S390
1841 
1842   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1843     if (lib_arch.name!=NULL) {
1844       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1845                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1846                  lib_arch.name, arch_array[running_arch_index].name);
1847     } else {
1848       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1849                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1850                  lib_arch.code,
1851                  arch_array[running_arch_index].name);
1852     }
1853   }
1854 
1855   return NULL;
1856 }
1857 
1858 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1859                                 int ebuflen) {
1860   void * result = ::dlopen(filename, RTLD_LAZY);
1861   if (result == NULL) {
1862     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1863     ebuf[ebuflen-1] = '\0';
1864   }
1865   return result;
1866 }
1867 
1868 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1869                                        int ebuflen) {
1870   void * result = NULL;
1871   if (LoadExecStackDllInVMThread) {
1872     result = dlopen_helper(filename, ebuf, ebuflen);
1873   }
1874 
1875   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1876   // library that requires an executable stack, or which does not have this
1877   // stack attribute set, dlopen changes the stack attribute to executable. The
1878   // read protection of the guard pages gets lost.
1879   //
1880   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1881   // may have been queued at the same time.
1882 
1883   if (!_stack_is_executable) {
1884     JavaThread *jt = Threads::first();
1885 
1886     while (jt) {
1887       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1888           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1889         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1890           warning("Attempt to reguard stack yellow zone failed.");
1891         }
1892       }
1893       jt = jt->next();
1894     }
1895   }
1896 
1897   return result;
1898 }
1899 
1900 void* os::dll_lookup(void* handle, const char* name) {
1901   void* res = dlsym(handle, name);
1902   return res;
1903 }
1904 
1905 void* os::get_default_process_handle() {
1906   return (void*)::dlopen(NULL, RTLD_LAZY);
1907 }
1908 
1909 static bool _print_ascii_file(const char* filename, outputStream* st) {
1910   int fd = ::open(filename, O_RDONLY);
1911   if (fd == -1) {
1912     return false;
1913   }
1914 
1915   char buf[33];
1916   int bytes;
1917   buf[32] = '\0';
1918   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1919     st->print_raw(buf, bytes);
1920   }
1921 
1922   ::close(fd);
1923 
1924   return true;
1925 }
1926 
1927 void os::print_dll_info(outputStream *st) {
1928   st->print_cr("Dynamic libraries:");
1929 
1930   char fname[32];
1931   pid_t pid = os::Linux::gettid();
1932 
1933   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1934 
1935   if (!_print_ascii_file(fname, st)) {
1936     st->print("Can not get library information for pid = %d\n", pid);
1937   }
1938 }
1939 
1940 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1941   FILE *procmapsFile = NULL;
1942 
1943   // Open the procfs maps file for the current process
1944   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1945     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1946     char line[PATH_MAX + 100];
1947 
1948     // Read line by line from 'file'
1949     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1950       u8 base, top, offset, inode;
1951       char permissions[5];
1952       char device[6];
1953       char name[PATH_MAX + 1];
1954 
1955       // Parse fields from line
1956       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1957              &base, &top, permissions, &offset, device, &inode, name);
1958 
1959       // Filter by device id '00:00' so that we only get file system mapped files.
1960       if (strcmp(device, "00:00") != 0) {
1961 
1962         // Call callback with the fields of interest
1963         if(callback(name, (address)base, (address)top, param)) {
1964           // Oops abort, callback aborted
1965           fclose(procmapsFile);
1966           return 1;
1967         }
1968       }
1969     }
1970     fclose(procmapsFile);
1971   }
1972   return 0;
1973 }
1974 
1975 void os::print_os_info_brief(outputStream* st) {
1976   os::Linux::print_distro_info(st);
1977 
1978   os::Posix::print_uname_info(st);
1979 
1980   os::Linux::print_libversion_info(st);
1981 
1982 }
1983 
1984 void os::print_os_info(outputStream* st) {
1985   st->print("OS:");
1986 
1987   os::Linux::print_distro_info(st);
1988 
1989   os::Posix::print_uname_info(st);
1990 
1991   // Print warning if unsafe chroot environment detected
1992   if (unsafe_chroot_detected) {
1993     st->print("WARNING!! ");
1994     st->print_cr("%s", unstable_chroot_error);
1995   }
1996 
1997   os::Linux::print_libversion_info(st);
1998 
1999   os::Posix::print_rlimit_info(st);
2000 
2001   os::Posix::print_load_average(st);
2002 
2003   os::Linux::print_full_memory_info(st);
2004 }
2005 
2006 // Try to identify popular distros.
2007 // Most Linux distributions have a /etc/XXX-release file, which contains
2008 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2009 // file that also contains the OS version string. Some have more than one
2010 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2011 // /etc/redhat-release.), so the order is important.
2012 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2013 // their own specific XXX-release file as well as a redhat-release file.
2014 // Because of this the XXX-release file needs to be searched for before the
2015 // redhat-release file.
2016 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2017 // search for redhat-release / SuSE-release needs to be before lsb-release.
2018 // Since the lsb-release file is the new standard it needs to be searched
2019 // before the older style release files.
2020 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2021 // next to last resort.  The os-release file is a new standard that contains
2022 // distribution information and the system-release file seems to be an old
2023 // standard that has been replaced by the lsb-release and os-release files.
2024 // Searching for the debian_version file is the last resort.  It contains
2025 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2026 // "Debian " is printed before the contents of the debian_version file.
2027 
2028 const char* distro_files[] = {
2029   "/etc/oracle-release",
2030   "/etc/mandriva-release",
2031   "/etc/mandrake-release",
2032   "/etc/sun-release",
2033   "/etc/redhat-release",
2034   "/etc/SuSE-release",
2035   "/etc/lsb-release",
2036   "/etc/turbolinux-release",
2037   "/etc/gentoo-release",
2038   "/etc/ltib-release",
2039   "/etc/angstrom-version",
2040   "/etc/system-release",
2041   "/etc/os-release",
2042   NULL };
2043 
2044 void os::Linux::print_distro_info(outputStream* st) {
2045   for (int i = 0;; i++) {
2046     const char* file = distro_files[i];
2047     if (file == NULL) {
2048       break;  // done
2049     }
2050     // If file prints, we found it.
2051     if (_print_ascii_file(file, st)) {
2052       return;
2053     }
2054   }
2055 
2056   if (file_exists("/etc/debian_version")) {
2057     st->print("Debian ");
2058     _print_ascii_file("/etc/debian_version", st);
2059   } else {
2060     st->print("Linux");
2061   }
2062   st->cr();
2063 }
2064 
2065 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2066   char buf[256];
2067   while (fgets(buf, sizeof(buf), fp)) {
2068     // Edit out extra stuff in expected format
2069     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2070       char* ptr = strstr(buf, "\"");  // the name is in quotes
2071       if (ptr != NULL) {
2072         ptr++; // go beyond first quote
2073         char* nl = strchr(ptr, '\"');
2074         if (nl != NULL) *nl = '\0';
2075         strncpy(distro, ptr, length);
2076       } else {
2077         ptr = strstr(buf, "=");
2078         ptr++; // go beyond equals then
2079         char* nl = strchr(ptr, '\n');
2080         if (nl != NULL) *nl = '\0';
2081         strncpy(distro, ptr, length);
2082       }
2083       return;
2084     } else if (get_first_line) {
2085       char* nl = strchr(buf, '\n');
2086       if (nl != NULL) *nl = '\0';
2087       strncpy(distro, buf, length);
2088       return;
2089     }
2090   }
2091   // print last line and close
2092   char* nl = strchr(buf, '\n');
2093   if (nl != NULL) *nl = '\0';
2094   strncpy(distro, buf, length);
2095 }
2096 
2097 static void parse_os_info(char* distro, size_t length, const char* file) {
2098   FILE* fp = fopen(file, "r");
2099   if (fp != NULL) {
2100     // if suse format, print out first line
2101     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2102     parse_os_info_helper(fp, distro, length, get_first_line);
2103     fclose(fp);
2104   }
2105 }
2106 
2107 void os::get_summary_os_info(char* buf, size_t buflen) {
2108   for (int i = 0;; i++) {
2109     const char* file = distro_files[i];
2110     if (file == NULL) {
2111       break; // ran out of distro_files
2112     }
2113     if (file_exists(file)) {
2114       parse_os_info(buf, buflen, file);
2115       return;
2116     }
2117   }
2118   // special case for debian
2119   if (file_exists("/etc/debian_version")) {
2120     strncpy(buf, "Debian ", buflen);
2121     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2122   } else {
2123     strncpy(buf, "Linux", buflen);
2124   }
2125 }
2126 
2127 void os::Linux::print_libversion_info(outputStream* st) {
2128   // libc, pthread
2129   st->print("libc:");
2130   st->print("%s ", os::Linux::glibc_version());
2131   st->print("%s ", os::Linux::libpthread_version());
2132   st->cr();
2133 }
2134 
2135 void os::Linux::print_full_memory_info(outputStream* st) {
2136   st->print("\n/proc/meminfo:\n");
2137   _print_ascii_file("/proc/meminfo", st);
2138   st->cr();
2139 }
2140 
2141 void os::print_memory_info(outputStream* st) {
2142 
2143   st->print("Memory:");
2144   st->print(" %dk page", os::vm_page_size()>>10);
2145 
2146   // values in struct sysinfo are "unsigned long"
2147   struct sysinfo si;
2148   sysinfo(&si);
2149 
2150   st->print(", physical " UINT64_FORMAT "k",
2151             os::physical_memory() >> 10);
2152   st->print("(" UINT64_FORMAT "k free)",
2153             os::available_memory() >> 10);
2154   st->print(", swap " UINT64_FORMAT "k",
2155             ((jlong)si.totalswap * si.mem_unit) >> 10);
2156   st->print("(" UINT64_FORMAT "k free)",
2157             ((jlong)si.freeswap * si.mem_unit) >> 10);
2158   st->cr();
2159 }
2160 
2161 // Print the first "model name" line and the first "flags" line
2162 // that we find and nothing more. We assume "model name" comes
2163 // before "flags" so if we find a second "model name", then the
2164 // "flags" field is considered missing.
2165 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2166 #if defined(IA32) || defined(AMD64)
2167   // Other platforms have less repetitive cpuinfo files
2168   FILE *fp = fopen("/proc/cpuinfo", "r");
2169   if (fp) {
2170     while (!feof(fp)) {
2171       if (fgets(buf, buflen, fp)) {
2172         // Assume model name comes before flags
2173         bool model_name_printed = false;
2174         if (strstr(buf, "model name") != NULL) {
2175           if (!model_name_printed) {
2176             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2177             st->print_raw(buf);
2178             model_name_printed = true;
2179           } else {
2180             // model name printed but not flags?  Odd, just return
2181             fclose(fp);
2182             return true;
2183           }
2184         }
2185         // print the flags line too
2186         if (strstr(buf, "flags") != NULL) {
2187           st->print_raw(buf);
2188           fclose(fp);
2189           return true;
2190         }
2191       }
2192     }
2193     fclose(fp);
2194   }
2195 #endif // x86 platforms
2196   return false;
2197 }
2198 
2199 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2200   // Only print the model name if the platform provides this as a summary
2201   if (!print_model_name_and_flags(st, buf, buflen)) {
2202     st->print("\n/proc/cpuinfo:\n");
2203     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2204       st->print_cr("  <Not Available>");
2205     }
2206   }
2207 }
2208 
2209 #if defined(AMD64) || defined(IA32) || defined(X32)
2210 const char* search_string = "model name";
2211 #elif defined(PPC64)
2212 const char* search_string = "cpu";
2213 #elif defined(S390)
2214 const char* search_string = "processor";
2215 #elif defined(SPARC)
2216 const char* search_string = "cpu";
2217 #else
2218 const char* search_string = "Processor";
2219 #endif
2220 
2221 // Parses the cpuinfo file for string representing the model name.
2222 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2223   FILE* fp = fopen("/proc/cpuinfo", "r");
2224   if (fp != NULL) {
2225     while (!feof(fp)) {
2226       char buf[256];
2227       if (fgets(buf, sizeof(buf), fp)) {
2228         char* start = strstr(buf, search_string);
2229         if (start != NULL) {
2230           char *ptr = start + strlen(search_string);
2231           char *end = buf + strlen(buf);
2232           while (ptr != end) {
2233              // skip whitespace and colon for the rest of the name.
2234              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2235                break;
2236              }
2237              ptr++;
2238           }
2239           if (ptr != end) {
2240             // reasonable string, get rid of newline and keep the rest
2241             char* nl = strchr(buf, '\n');
2242             if (nl != NULL) *nl = '\0';
2243             strncpy(cpuinfo, ptr, length);
2244             fclose(fp);
2245             return;
2246           }
2247         }
2248       }
2249     }
2250     fclose(fp);
2251   }
2252   // cpuinfo not found or parsing failed, just print generic string.  The entire
2253   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2254 #if   defined(AARCH64)
2255   strncpy(cpuinfo, "AArch64", length);
2256 #elif defined(AMD64)
2257   strncpy(cpuinfo, "x86_64", length);
2258 #elif defined(ARM)  // Order wrt. AARCH64 is relevant!
2259   strncpy(cpuinfo, "ARM", length);
2260 #elif defined(IA32)
2261   strncpy(cpuinfo, "x86_32", length);
2262 #elif defined(IA64)
2263   strncpy(cpuinfo, "IA64", length);
2264 #elif defined(PPC)
2265   strncpy(cpuinfo, "PPC64", length);
2266 #elif defined(S390)
2267   strncpy(cpuinfo, "S390", length);
2268 #elif defined(SPARC)
2269   strncpy(cpuinfo, "sparcv9", length);
2270 #elif defined(ZERO_LIBARCH)
2271   strncpy(cpuinfo, ZERO_LIBARCH, length);
2272 #else
2273   strncpy(cpuinfo, "unknown", length);
2274 #endif
2275 }
2276 
2277 static void print_signal_handler(outputStream* st, int sig,
2278                                  char* buf, size_t buflen);
2279 
2280 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2281   st->print_cr("Signal Handlers:");
2282   print_signal_handler(st, SIGSEGV, buf, buflen);
2283   print_signal_handler(st, SIGBUS , buf, buflen);
2284   print_signal_handler(st, SIGFPE , buf, buflen);
2285   print_signal_handler(st, SIGPIPE, buf, buflen);
2286   print_signal_handler(st, SIGXFSZ, buf, buflen);
2287   print_signal_handler(st, SIGILL , buf, buflen);
2288   print_signal_handler(st, SR_signum, buf, buflen);
2289   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2290   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2291   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2292   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2293 #if defined(PPC64)
2294   print_signal_handler(st, SIGTRAP, buf, buflen);
2295 #endif
2296 }
2297 
2298 static char saved_jvm_path[MAXPATHLEN] = {0};
2299 
2300 // Find the full path to the current module, libjvm.so
2301 void os::jvm_path(char *buf, jint buflen) {
2302   // Error checking.
2303   if (buflen < MAXPATHLEN) {
2304     assert(false, "must use a large-enough buffer");
2305     buf[0] = '\0';
2306     return;
2307   }
2308   // Lazy resolve the path to current module.
2309   if (saved_jvm_path[0] != 0) {
2310     strcpy(buf, saved_jvm_path);
2311     return;
2312   }
2313 
2314   char dli_fname[MAXPATHLEN];
2315   bool ret = dll_address_to_library_name(
2316                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2317                                          dli_fname, sizeof(dli_fname), NULL);
2318   assert(ret, "cannot locate libjvm");
2319   char *rp = NULL;
2320   if (ret && dli_fname[0] != '\0') {
2321     rp = realpath(dli_fname, buf);
2322   }
2323   if (rp == NULL) {
2324     return;
2325   }
2326 
2327   if (Arguments::sun_java_launcher_is_altjvm()) {
2328     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2329     // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2330     // If "/jre/lib/" appears at the right place in the string, then
2331     // assume we are installed in a JDK and we're done. Otherwise, check
2332     // for a JAVA_HOME environment variable and fix up the path so it
2333     // looks like libjvm.so is installed there (append a fake suffix
2334     // hotspot/libjvm.so).
2335     const char *p = buf + strlen(buf) - 1;
2336     for (int count = 0; p > buf && count < 5; ++count) {
2337       for (--p; p > buf && *p != '/'; --p)
2338         /* empty */ ;
2339     }
2340 
2341     if (strncmp(p, "/jre/lib/", 9) != 0) {
2342       // Look for JAVA_HOME in the environment.
2343       char* java_home_var = ::getenv("JAVA_HOME");
2344       if (java_home_var != NULL && java_home_var[0] != 0) {
2345         char* jrelib_p;
2346         int len;
2347 
2348         // Check the current module name "libjvm.so".
2349         p = strrchr(buf, '/');
2350         if (p == NULL) {
2351           return;
2352         }
2353         assert(strstr(p, "/libjvm") == p, "invalid library name");
2354 
2355         rp = realpath(java_home_var, buf);
2356         if (rp == NULL) {
2357           return;
2358         }
2359 
2360         // determine if this is a legacy image or modules image
2361         // modules image doesn't have "jre" subdirectory
2362         len = strlen(buf);
2363         assert(len < buflen, "Ran out of buffer room");
2364         jrelib_p = buf + len;
2365         snprintf(jrelib_p, buflen-len, "/jre/lib");
2366         if (0 != access(buf, F_OK)) {
2367           snprintf(jrelib_p, buflen-len, "/lib");
2368         }
2369 
2370         if (0 == access(buf, F_OK)) {
2371           // Use current module name "libjvm.so"
2372           len = strlen(buf);
2373           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2374         } else {
2375           // Go back to path of .so
2376           rp = realpath(dli_fname, buf);
2377           if (rp == NULL) {
2378             return;
2379           }
2380         }
2381       }
2382     }
2383   }
2384 
2385   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2386   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2387 }
2388 
2389 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2390   // no prefix required, not even "_"
2391 }
2392 
2393 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2394   // no suffix required
2395 }
2396 
2397 ////////////////////////////////////////////////////////////////////////////////
2398 // sun.misc.Signal support
2399 
2400 static volatile jint sigint_count = 0;
2401 
2402 static void UserHandler(int sig, void *siginfo, void *context) {
2403   // 4511530 - sem_post is serialized and handled by the manager thread. When
2404   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2405   // don't want to flood the manager thread with sem_post requests.
2406   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2407     return;
2408   }
2409 
2410   // Ctrl-C is pressed during error reporting, likely because the error
2411   // handler fails to abort. Let VM die immediately.
2412   if (sig == SIGINT && is_error_reported()) {
2413     os::die();
2414   }
2415 
2416   os::signal_notify(sig);
2417 }
2418 
2419 void* os::user_handler() {
2420   return CAST_FROM_FN_PTR(void*, UserHandler);
2421 }
2422 
2423 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2424   struct timespec ts;
2425   // Semaphore's are always associated with CLOCK_REALTIME
2426   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2427   // see unpackTime for discussion on overflow checking
2428   if (sec >= MAX_SECS) {
2429     ts.tv_sec += MAX_SECS;
2430     ts.tv_nsec = 0;
2431   } else {
2432     ts.tv_sec += sec;
2433     ts.tv_nsec += nsec;
2434     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2435       ts.tv_nsec -= NANOSECS_PER_SEC;
2436       ++ts.tv_sec; // note: this must be <= max_secs
2437     }
2438   }
2439 
2440   return ts;
2441 }
2442 
2443 extern "C" {
2444   typedef void (*sa_handler_t)(int);
2445   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2446 }
2447 
2448 void* os::signal(int signal_number, void* handler) {
2449   struct sigaction sigAct, oldSigAct;
2450 
2451   sigfillset(&(sigAct.sa_mask));
2452   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2453   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2454 
2455   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2456     // -1 means registration failed
2457     return (void *)-1;
2458   }
2459 
2460   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2461 }
2462 
2463 void os::signal_raise(int signal_number) {
2464   ::raise(signal_number);
2465 }
2466 
2467 // The following code is moved from os.cpp for making this
2468 // code platform specific, which it is by its very nature.
2469 
2470 // Will be modified when max signal is changed to be dynamic
2471 int os::sigexitnum_pd() {
2472   return NSIG;
2473 }
2474 
2475 // a counter for each possible signal value
2476 static volatile jint pending_signals[NSIG+1] = { 0 };
2477 
2478 // Linux(POSIX) specific hand shaking semaphore.
2479 static sem_t sig_sem;
2480 static PosixSemaphore sr_semaphore;
2481 
2482 void os::signal_init_pd() {
2483   // Initialize signal structures
2484   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2485 
2486   // Initialize signal semaphore
2487   ::sem_init(&sig_sem, 0, 0);
2488 }
2489 
2490 void os::signal_notify(int sig) {
2491   Atomic::inc(&pending_signals[sig]);
2492   ::sem_post(&sig_sem);
2493 }
2494 
2495 static int check_pending_signals(bool wait) {
2496   Atomic::store(0, &sigint_count);
2497   for (;;) {
2498     for (int i = 0; i < NSIG + 1; i++) {
2499       jint n = pending_signals[i];
2500       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2501         return i;
2502       }
2503     }
2504     if (!wait) {
2505       return -1;
2506     }
2507     JavaThread *thread = JavaThread::current();
2508     ThreadBlockInVM tbivm(thread);
2509 
2510     bool threadIsSuspended;
2511     do {
2512       thread->set_suspend_equivalent();
2513       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2514       ::sem_wait(&sig_sem);
2515 
2516       // were we externally suspended while we were waiting?
2517       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2518       if (threadIsSuspended) {
2519         // The semaphore has been incremented, but while we were waiting
2520         // another thread suspended us. We don't want to continue running
2521         // while suspended because that would surprise the thread that
2522         // suspended us.
2523         ::sem_post(&sig_sem);
2524 
2525         thread->java_suspend_self();
2526       }
2527     } while (threadIsSuspended);
2528   }
2529 }
2530 
2531 int os::signal_lookup() {
2532   return check_pending_signals(false);
2533 }
2534 
2535 int os::signal_wait() {
2536   return check_pending_signals(true);
2537 }
2538 
2539 ////////////////////////////////////////////////////////////////////////////////
2540 // Virtual Memory
2541 
2542 int os::vm_page_size() {
2543   // Seems redundant as all get out
2544   assert(os::Linux::page_size() != -1, "must call os::init");
2545   return os::Linux::page_size();
2546 }
2547 
2548 // Solaris allocates memory by pages.
2549 int os::vm_allocation_granularity() {
2550   assert(os::Linux::page_size() != -1, "must call os::init");
2551   return os::Linux::page_size();
2552 }
2553 
2554 // Rationale behind this function:
2555 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2556 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2557 //  samples for JITted code. Here we create private executable mapping over the code cache
2558 //  and then we can use standard (well, almost, as mapping can change) way to provide
2559 //  info for the reporting script by storing timestamp and location of symbol
2560 void linux_wrap_code(char* base, size_t size) {
2561   static volatile jint cnt = 0;
2562 
2563   if (!UseOprofile) {
2564     return;
2565   }
2566 
2567   char buf[PATH_MAX+1];
2568   int num = Atomic::add(1, &cnt);
2569 
2570   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2571            os::get_temp_directory(), os::current_process_id(), num);
2572   unlink(buf);
2573 
2574   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2575 
2576   if (fd != -1) {
2577     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2578     if (rv != (off_t)-1) {
2579       if (::write(fd, "", 1) == 1) {
2580         mmap(base, size,
2581              PROT_READ|PROT_WRITE|PROT_EXEC,
2582              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2583       }
2584     }
2585     ::close(fd);
2586     unlink(buf);
2587   }
2588 }
2589 
2590 static bool recoverable_mmap_error(int err) {
2591   // See if the error is one we can let the caller handle. This
2592   // list of errno values comes from JBS-6843484. I can't find a
2593   // Linux man page that documents this specific set of errno
2594   // values so while this list currently matches Solaris, it may
2595   // change as we gain experience with this failure mode.
2596   switch (err) {
2597   case EBADF:
2598   case EINVAL:
2599   case ENOTSUP:
2600     // let the caller deal with these errors
2601     return true;
2602 
2603   default:
2604     // Any remaining errors on this OS can cause our reserved mapping
2605     // to be lost. That can cause confusion where different data
2606     // structures think they have the same memory mapped. The worst
2607     // scenario is if both the VM and a library think they have the
2608     // same memory mapped.
2609     return false;
2610   }
2611 }
2612 
2613 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2614                                     int err) {
2615   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2616           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2617           os::strerror(err), err);
2618 }
2619 
2620 static void warn_fail_commit_memory(char* addr, size_t size,
2621                                     size_t alignment_hint, bool exec,
2622                                     int err) {
2623   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2624           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2625           alignment_hint, exec, os::strerror(err), err);
2626 }
2627 
2628 // NOTE: Linux kernel does not really reserve the pages for us.
2629 //       All it does is to check if there are enough free pages
2630 //       left at the time of mmap(). This could be a potential
2631 //       problem.
2632 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2633   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2634   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2635                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2636   if (res != (uintptr_t) MAP_FAILED) {
2637     if (UseNUMAInterleaving) {
2638       numa_make_global(addr, size);
2639     }
2640     return 0;
2641   }
2642 
2643   int err = errno;  // save errno from mmap() call above
2644 
2645   if (!recoverable_mmap_error(err)) {
2646     warn_fail_commit_memory(addr, size, exec, err);
2647     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2648   }
2649 
2650   return err;
2651 }
2652 
2653 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2654   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2655 }
2656 
2657 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2658                                   const char* mesg) {
2659   assert(mesg != NULL, "mesg must be specified");
2660   int err = os::Linux::commit_memory_impl(addr, size, exec);
2661   if (err != 0) {
2662     // the caller wants all commit errors to exit with the specified mesg:
2663     warn_fail_commit_memory(addr, size, exec, err);
2664     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2665   }
2666 }
2667 
2668 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2669 #ifndef MAP_HUGETLB
2670   #define MAP_HUGETLB 0x40000
2671 #endif
2672 
2673 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2674 #ifndef MADV_HUGEPAGE
2675   #define MADV_HUGEPAGE 14
2676 #endif
2677 
2678 int os::Linux::commit_memory_impl(char* addr, size_t size,
2679                                   size_t alignment_hint, bool exec) {
2680   int err = os::Linux::commit_memory_impl(addr, size, exec);
2681   if (err == 0) {
2682     realign_memory(addr, size, alignment_hint);
2683   }
2684   return err;
2685 }
2686 
2687 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2688                           bool exec) {
2689   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2690 }
2691 
2692 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2693                                   size_t alignment_hint, bool exec,
2694                                   const char* mesg) {
2695   assert(mesg != NULL, "mesg must be specified");
2696   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2697   if (err != 0) {
2698     // the caller wants all commit errors to exit with the specified mesg:
2699     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2700     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2701   }
2702 }
2703 
2704 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2705   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2706     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2707     // be supported or the memory may already be backed by huge pages.
2708     ::madvise(addr, bytes, MADV_HUGEPAGE);
2709   }
2710 }
2711 
2712 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2713   // This method works by doing an mmap over an existing mmaping and effectively discarding
2714   // the existing pages. However it won't work for SHM-based large pages that cannot be
2715   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2716   // small pages on top of the SHM segment. This method always works for small pages, so we
2717   // allow that in any case.
2718   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2719     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2720   }
2721 }
2722 
2723 void os::numa_make_global(char *addr, size_t bytes) {
2724   Linux::numa_interleave_memory(addr, bytes);
2725 }
2726 
2727 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2728 // bind policy to MPOL_PREFERRED for the current thread.
2729 #define USE_MPOL_PREFERRED 0
2730 
2731 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2732   // To make NUMA and large pages more robust when both enabled, we need to ease
2733   // the requirements on where the memory should be allocated. MPOL_BIND is the
2734   // default policy and it will force memory to be allocated on the specified
2735   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2736   // the specified node, but will not force it. Using this policy will prevent
2737   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2738   // free large pages.
2739   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2740   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2741 }
2742 
2743 bool os::numa_topology_changed() { return false; }
2744 
2745 size_t os::numa_get_groups_num() {
2746   int max_node = Linux::numa_max_node();
2747   return max_node > 0 ? max_node + 1 : 1;
2748 }
2749 
2750 int os::numa_get_group_id() {
2751   int cpu_id = Linux::sched_getcpu();
2752   if (cpu_id != -1) {
2753     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2754     if (lgrp_id != -1) {
2755       return lgrp_id;
2756     }
2757   }
2758   return 0;
2759 }
2760 
2761 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2762   for (size_t i = 0; i < size; i++) {
2763     ids[i] = i;
2764   }
2765   return size;
2766 }
2767 
2768 bool os::get_page_info(char *start, page_info* info) {
2769   return false;
2770 }
2771 
2772 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2773                      page_info* page_found) {
2774   return end;
2775 }
2776 
2777 
2778 int os::Linux::sched_getcpu_syscall(void) {
2779   unsigned int cpu = 0;
2780   int retval = -1;
2781 
2782 #if defined(IA32)
2783   #ifndef SYS_getcpu
2784     #define SYS_getcpu 318
2785   #endif
2786   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2787 #elif defined(AMD64)
2788 // Unfortunately we have to bring all these macros here from vsyscall.h
2789 // to be able to compile on old linuxes.
2790   #define __NR_vgetcpu 2
2791   #define VSYSCALL_START (-10UL << 20)
2792   #define VSYSCALL_SIZE 1024
2793   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2794   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2795   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2796   retval = vgetcpu(&cpu, NULL, NULL);
2797 #endif
2798 
2799   return (retval == -1) ? retval : cpu;
2800 }
2801 
2802 // Something to do with the numa-aware allocator needs these symbols
2803 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2804 extern "C" JNIEXPORT void numa_error(char *where) { }
2805 
2806 
2807 // If we are running with libnuma version > 2, then we should
2808 // be trying to use symbols with versions 1.1
2809 // If we are running with earlier version, which did not have symbol versions,
2810 // we should use the base version.
2811 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2812   void *f = dlvsym(handle, name, "libnuma_1.1");
2813   if (f == NULL) {
2814     f = dlsym(handle, name);
2815   }
2816   return f;
2817 }
2818 
2819 bool os::Linux::libnuma_init() {
2820   // sched_getcpu() should be in libc.
2821   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2822                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2823 
2824   // If it's not, try a direct syscall.
2825   if (sched_getcpu() == -1) {
2826     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2827                                     (void*)&sched_getcpu_syscall));
2828   }
2829 
2830   if (sched_getcpu() != -1) { // Does it work?
2831     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2832     if (handle != NULL) {
2833       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2834                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2835       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2836                                        libnuma_dlsym(handle, "numa_max_node")));
2837       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2838                                         libnuma_dlsym(handle, "numa_available")));
2839       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2840                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2841       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2842                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2843       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2844                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2845 
2846 
2847       if (numa_available() != -1) {
2848         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2849         // Create a cpu -> node mapping
2850         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2851         rebuild_cpu_to_node_map();
2852         return true;
2853       }
2854     }
2855   }
2856   return false;
2857 }
2858 
2859 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2860   // Creating guard page is very expensive. Java thread has HotSpot
2861   // guard pages, only enable glibc guard page for non-Java threads.
2862   // (Remember: compiler thread is a Java thread, too!)
2863   return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2864 }
2865 
2866 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2867 // The table is later used in get_node_by_cpu().
2868 void os::Linux::rebuild_cpu_to_node_map() {
2869   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2870                               // in libnuma (possible values are starting from 16,
2871                               // and continuing up with every other power of 2, but less
2872                               // than the maximum number of CPUs supported by kernel), and
2873                               // is a subject to change (in libnuma version 2 the requirements
2874                               // are more reasonable) we'll just hardcode the number they use
2875                               // in the library.
2876   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2877 
2878   size_t cpu_num = processor_count();
2879   size_t cpu_map_size = NCPUS / BitsPerCLong;
2880   size_t cpu_map_valid_size =
2881     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2882 
2883   cpu_to_node()->clear();
2884   cpu_to_node()->at_grow(cpu_num - 1);
2885   size_t node_num = numa_get_groups_num();
2886 
2887   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2888   for (size_t i = 0; i < node_num; i++) {
2889     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2890       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2891         if (cpu_map[j] != 0) {
2892           for (size_t k = 0; k < BitsPerCLong; k++) {
2893             if (cpu_map[j] & (1UL << k)) {
2894               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2895             }
2896           }
2897         }
2898       }
2899     }
2900   }
2901   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2902 }
2903 
2904 int os::Linux::get_node_by_cpu(int cpu_id) {
2905   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2906     return cpu_to_node()->at(cpu_id);
2907   }
2908   return -1;
2909 }
2910 
2911 GrowableArray<int>* os::Linux::_cpu_to_node;
2912 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2913 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2914 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2915 os::Linux::numa_available_func_t os::Linux::_numa_available;
2916 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2917 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2918 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2919 unsigned long* os::Linux::_numa_all_nodes;
2920 
2921 bool os::pd_uncommit_memory(char* addr, size_t size) {
2922   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2923                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2924   return res  != (uintptr_t) MAP_FAILED;
2925 }
2926 
2927 static address get_stack_commited_bottom(address bottom, size_t size) {
2928   address nbot = bottom;
2929   address ntop = bottom + size;
2930 
2931   size_t page_sz = os::vm_page_size();
2932   unsigned pages = size / page_sz;
2933 
2934   unsigned char vec[1];
2935   unsigned imin = 1, imax = pages + 1, imid;
2936   int mincore_return_value = 0;
2937 
2938   assert(imin <= imax, "Unexpected page size");
2939 
2940   while (imin < imax) {
2941     imid = (imax + imin) / 2;
2942     nbot = ntop - (imid * page_sz);
2943 
2944     // Use a trick with mincore to check whether the page is mapped or not.
2945     // mincore sets vec to 1 if page resides in memory and to 0 if page
2946     // is swapped output but if page we are asking for is unmapped
2947     // it returns -1,ENOMEM
2948     mincore_return_value = mincore(nbot, page_sz, vec);
2949 
2950     if (mincore_return_value == -1) {
2951       // Page is not mapped go up
2952       // to find first mapped page
2953       if (errno != EAGAIN) {
2954         assert(errno == ENOMEM, "Unexpected mincore errno");
2955         imax = imid;
2956       }
2957     } else {
2958       // Page is mapped go down
2959       // to find first not mapped page
2960       imin = imid + 1;
2961     }
2962   }
2963 
2964   nbot = nbot + page_sz;
2965 
2966   // Adjust stack bottom one page up if last checked page is not mapped
2967   if (mincore_return_value == -1) {
2968     nbot = nbot + page_sz;
2969   }
2970 
2971   return nbot;
2972 }
2973 
2974 
2975 // Linux uses a growable mapping for the stack, and if the mapping for
2976 // the stack guard pages is not removed when we detach a thread the
2977 // stack cannot grow beyond the pages where the stack guard was
2978 // mapped.  If at some point later in the process the stack expands to
2979 // that point, the Linux kernel cannot expand the stack any further
2980 // because the guard pages are in the way, and a segfault occurs.
2981 //
2982 // However, it's essential not to split the stack region by unmapping
2983 // a region (leaving a hole) that's already part of the stack mapping,
2984 // so if the stack mapping has already grown beyond the guard pages at
2985 // the time we create them, we have to truncate the stack mapping.
2986 // So, we need to know the extent of the stack mapping when
2987 // create_stack_guard_pages() is called.
2988 
2989 // We only need this for stacks that are growable: at the time of
2990 // writing thread stacks don't use growable mappings (i.e. those
2991 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2992 // only applies to the main thread.
2993 
2994 // If the (growable) stack mapping already extends beyond the point
2995 // where we're going to put our guard pages, truncate the mapping at
2996 // that point by munmap()ping it.  This ensures that when we later
2997 // munmap() the guard pages we don't leave a hole in the stack
2998 // mapping. This only affects the main/initial thread
2999 
3000 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3001   if (os::Linux::is_initial_thread()) {
3002     // As we manually grow stack up to bottom inside create_attached_thread(),
3003     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3004     // we don't need to do anything special.
3005     // Check it first, before calling heavy function.
3006     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3007     unsigned char vec[1];
3008 
3009     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3010       // Fallback to slow path on all errors, including EAGAIN
3011       stack_extent = (uintptr_t) get_stack_commited_bottom(
3012                                                            os::Linux::initial_thread_stack_bottom(),
3013                                                            (size_t)addr - stack_extent);
3014     }
3015 
3016     if (stack_extent < (uintptr_t)addr) {
3017       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3018     }
3019   }
3020 
3021   return os::commit_memory(addr, size, !ExecMem);
3022 }
3023 
3024 // If this is a growable mapping, remove the guard pages entirely by
3025 // munmap()ping them.  If not, just call uncommit_memory(). This only
3026 // affects the main/initial thread, but guard against future OS changes
3027 // It's safe to always unmap guard pages for initial thread because we
3028 // always place it right after end of the mapped region
3029 
3030 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3031   uintptr_t stack_extent, stack_base;
3032 
3033   if (os::Linux::is_initial_thread()) {
3034     return ::munmap(addr, size) == 0;
3035   }
3036 
3037   return os::uncommit_memory(addr, size);
3038 }
3039 
3040 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3041 // at 'requested_addr'. If there are existing memory mappings at the same
3042 // location, however, they will be overwritten. If 'fixed' is false,
3043 // 'requested_addr' is only treated as a hint, the return value may or
3044 // may not start from the requested address. Unlike Linux mmap(), this
3045 // function returns NULL to indicate failure.
3046 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3047   char * addr;
3048   int flags;
3049 
3050   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3051   if (fixed) {
3052     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3053     flags |= MAP_FIXED;
3054   }
3055 
3056   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3057   // touch an uncommitted page. Otherwise, the read/write might
3058   // succeed if we have enough swap space to back the physical page.
3059   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3060                        flags, -1, 0);
3061 
3062   return addr == MAP_FAILED ? NULL : addr;
3063 }
3064 
3065 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3066 //   (req_addr != NULL) or with a given alignment.
3067 //  - bytes shall be a multiple of alignment.
3068 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3069 //  - alignment sets the alignment at which memory shall be allocated.
3070 //     It must be a multiple of allocation granularity.
3071 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3072 //  req_addr or NULL.
3073 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3074 
3075   size_t extra_size = bytes;
3076   if (req_addr == NULL && alignment > 0) {
3077     extra_size += alignment;
3078   }
3079 
3080   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3081     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3082     -1, 0);
3083   if (start == MAP_FAILED) {
3084     start = NULL;
3085   } else {
3086     if (req_addr != NULL) {
3087       if (start != req_addr) {
3088         ::munmap(start, extra_size);
3089         start = NULL;
3090       }
3091     } else {
3092       char* const start_aligned = (char*) align_ptr_up(start, alignment);
3093       char* const end_aligned = start_aligned + bytes;
3094       char* const end = start + extra_size;
3095       if (start_aligned > start) {
3096         ::munmap(start, start_aligned - start);
3097       }
3098       if (end_aligned < end) {
3099         ::munmap(end_aligned, end - end_aligned);
3100       }
3101       start = start_aligned;
3102     }
3103   }
3104   return start;
3105 }
3106 
3107 static int anon_munmap(char * addr, size_t size) {
3108   return ::munmap(addr, size) == 0;
3109 }
3110 
3111 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3112                             size_t alignment_hint) {
3113   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3114 }
3115 
3116 bool os::pd_release_memory(char* addr, size_t size) {
3117   return anon_munmap(addr, size);
3118 }
3119 
3120 static bool linux_mprotect(char* addr, size_t size, int prot) {
3121   // Linux wants the mprotect address argument to be page aligned.
3122   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3123 
3124   // According to SUSv3, mprotect() should only be used with mappings
3125   // established by mmap(), and mmap() always maps whole pages. Unaligned
3126   // 'addr' likely indicates problem in the VM (e.g. trying to change
3127   // protection of malloc'ed or statically allocated memory). Check the
3128   // caller if you hit this assert.
3129   assert(addr == bottom, "sanity check");
3130 
3131   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3132   return ::mprotect(bottom, size, prot) == 0;
3133 }
3134 
3135 // Set protections specified
3136 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3137                         bool is_committed) {
3138   unsigned int p = 0;
3139   switch (prot) {
3140   case MEM_PROT_NONE: p = PROT_NONE; break;
3141   case MEM_PROT_READ: p = PROT_READ; break;
3142   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3143   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3144   default:
3145     ShouldNotReachHere();
3146   }
3147   // is_committed is unused.
3148   return linux_mprotect(addr, bytes, p);
3149 }
3150 
3151 bool os::guard_memory(char* addr, size_t size) {
3152   return linux_mprotect(addr, size, PROT_NONE);
3153 }
3154 
3155 bool os::unguard_memory(char* addr, size_t size) {
3156   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3157 }
3158 
3159 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3160                                                     size_t page_size) {
3161   bool result = false;
3162   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3163                  MAP_ANONYMOUS|MAP_PRIVATE,
3164                  -1, 0);
3165   if (p != MAP_FAILED) {
3166     void *aligned_p = align_ptr_up(p, page_size);
3167 
3168     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3169 
3170     munmap(p, page_size * 2);
3171   }
3172 
3173   if (warn && !result) {
3174     warning("TransparentHugePages is not supported by the operating system.");
3175   }
3176 
3177   return result;
3178 }
3179 
3180 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3181   bool result = false;
3182   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3183                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3184                  -1, 0);
3185 
3186   if (p != MAP_FAILED) {
3187     // We don't know if this really is a huge page or not.
3188     FILE *fp = fopen("/proc/self/maps", "r");
3189     if (fp) {
3190       while (!feof(fp)) {
3191         char chars[257];
3192         long x = 0;
3193         if (fgets(chars, sizeof(chars), fp)) {
3194           if (sscanf(chars, "%lx-%*x", &x) == 1
3195               && x == (long)p) {
3196             if (strstr (chars, "hugepage")) {
3197               result = true;
3198               break;
3199             }
3200           }
3201         }
3202       }
3203       fclose(fp);
3204     }
3205     munmap(p, page_size);
3206   }
3207 
3208   if (warn && !result) {
3209     warning("HugeTLBFS is not supported by the operating system.");
3210   }
3211 
3212   return result;
3213 }
3214 
3215 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3216 //
3217 // From the coredump_filter documentation:
3218 //
3219 // - (bit 0) anonymous private memory
3220 // - (bit 1) anonymous shared memory
3221 // - (bit 2) file-backed private memory
3222 // - (bit 3) file-backed shared memory
3223 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3224 //           effective only if the bit 2 is cleared)
3225 // - (bit 5) hugetlb private memory
3226 // - (bit 6) hugetlb shared memory
3227 //
3228 static void set_coredump_filter(void) {
3229   FILE *f;
3230   long cdm;
3231 
3232   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3233     return;
3234   }
3235 
3236   if (fscanf(f, "%lx", &cdm) != 1) {
3237     fclose(f);
3238     return;
3239   }
3240 
3241   rewind(f);
3242 
3243   if ((cdm & LARGEPAGES_BIT) == 0) {
3244     cdm |= LARGEPAGES_BIT;
3245     fprintf(f, "%#lx", cdm);
3246   }
3247 
3248   fclose(f);
3249 }
3250 
3251 // Large page support
3252 
3253 static size_t _large_page_size = 0;
3254 
3255 size_t os::Linux::find_large_page_size() {
3256   size_t large_page_size = 0;
3257 
3258   // large_page_size on Linux is used to round up heap size. x86 uses either
3259   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3260   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3261   // page as large as 256M.
3262   //
3263   // Here we try to figure out page size by parsing /proc/meminfo and looking
3264   // for a line with the following format:
3265   //    Hugepagesize:     2048 kB
3266   //
3267   // If we can't determine the value (e.g. /proc is not mounted, or the text
3268   // format has been changed), we'll use the largest page size supported by
3269   // the processor.
3270 
3271 #ifndef ZERO
3272   large_page_size =
3273     AARCH64_ONLY(2 * M)
3274     AMD64_ONLY(2 * M)
3275     ARM32_ONLY(2 * M)
3276     IA32_ONLY(4 * M)
3277     IA64_ONLY(256 * M)
3278     PPC_ONLY(4 * M)
3279     S390_ONLY(1 * M)
3280     SPARC_ONLY(4 * M);
3281 #endif // ZERO
3282 
3283   FILE *fp = fopen("/proc/meminfo", "r");
3284   if (fp) {
3285     while (!feof(fp)) {
3286       int x = 0;
3287       char buf[16];
3288       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3289         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3290           large_page_size = x * K;
3291           break;
3292         }
3293       } else {
3294         // skip to next line
3295         for (;;) {
3296           int ch = fgetc(fp);
3297           if (ch == EOF || ch == (int)'\n') break;
3298         }
3299       }
3300     }
3301     fclose(fp);
3302   }
3303 
3304   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3305     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3306             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3307             proper_unit_for_byte_size(large_page_size));
3308   }
3309 
3310   return large_page_size;
3311 }
3312 
3313 size_t os::Linux::setup_large_page_size() {
3314   _large_page_size = Linux::find_large_page_size();
3315   const size_t default_page_size = (size_t)Linux::page_size();
3316   if (_large_page_size > default_page_size) {
3317     _page_sizes[0] = _large_page_size;
3318     _page_sizes[1] = default_page_size;
3319     _page_sizes[2] = 0;
3320   }
3321 
3322   return _large_page_size;
3323 }
3324 
3325 bool os::Linux::setup_large_page_type(size_t page_size) {
3326   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3327       FLAG_IS_DEFAULT(UseSHM) &&
3328       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3329 
3330     // The type of large pages has not been specified by the user.
3331 
3332     // Try UseHugeTLBFS and then UseSHM.
3333     UseHugeTLBFS = UseSHM = true;
3334 
3335     // Don't try UseTransparentHugePages since there are known
3336     // performance issues with it turned on. This might change in the future.
3337     UseTransparentHugePages = false;
3338   }
3339 
3340   if (UseTransparentHugePages) {
3341     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3342     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3343       UseHugeTLBFS = false;
3344       UseSHM = false;
3345       return true;
3346     }
3347     UseTransparentHugePages = false;
3348   }
3349 
3350   if (UseHugeTLBFS) {
3351     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3352     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3353       UseSHM = false;
3354       return true;
3355     }
3356     UseHugeTLBFS = false;
3357   }
3358 
3359   return UseSHM;
3360 }
3361 
3362 void os::large_page_init() {
3363   if (!UseLargePages &&
3364       !UseTransparentHugePages &&
3365       !UseHugeTLBFS &&
3366       !UseSHM) {
3367     // Not using large pages.
3368     return;
3369   }
3370 
3371   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3372     // The user explicitly turned off large pages.
3373     // Ignore the rest of the large pages flags.
3374     UseTransparentHugePages = false;
3375     UseHugeTLBFS = false;
3376     UseSHM = false;
3377     return;
3378   }
3379 
3380   size_t large_page_size = Linux::setup_large_page_size();
3381   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3382 
3383   set_coredump_filter();
3384 }
3385 
3386 #ifndef SHM_HUGETLB
3387   #define SHM_HUGETLB 04000
3388 #endif
3389 
3390 #define shm_warning_format(format, ...)              \
3391   do {                                               \
3392     if (UseLargePages &&                             \
3393         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3394          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3395          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3396       warning(format, __VA_ARGS__);                  \
3397     }                                                \
3398   } while (0)
3399 
3400 #define shm_warning(str) shm_warning_format("%s", str)
3401 
3402 #define shm_warning_with_errno(str)                \
3403   do {                                             \
3404     int err = errno;                               \
3405     shm_warning_format(str " (error = %d)", err);  \
3406   } while (0)
3407 
3408 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3409   assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3410 
3411   if (!is_size_aligned(alignment, SHMLBA)) {
3412     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3413     return NULL;
3414   }
3415 
3416   // To ensure that we get 'alignment' aligned memory from shmat,
3417   // we pre-reserve aligned virtual memory and then attach to that.
3418 
3419   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3420   if (pre_reserved_addr == NULL) {
3421     // Couldn't pre-reserve aligned memory.
3422     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3423     return NULL;
3424   }
3425 
3426   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3427   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3428 
3429   if ((intptr_t)addr == -1) {
3430     int err = errno;
3431     shm_warning_with_errno("Failed to attach shared memory.");
3432 
3433     assert(err != EACCES, "Unexpected error");
3434     assert(err != EIDRM,  "Unexpected error");
3435     assert(err != EINVAL, "Unexpected error");
3436 
3437     // Since we don't know if the kernel unmapped the pre-reserved memory area
3438     // we can't unmap it, since that would potentially unmap memory that was
3439     // mapped from other threads.
3440     return NULL;
3441   }
3442 
3443   return addr;
3444 }
3445 
3446 static char* shmat_at_address(int shmid, char* req_addr) {
3447   if (!is_ptr_aligned(req_addr, SHMLBA)) {
3448     assert(false, "Requested address needs to be SHMLBA aligned");
3449     return NULL;
3450   }
3451 
3452   char* addr = (char*)shmat(shmid, req_addr, 0);
3453 
3454   if ((intptr_t)addr == -1) {
3455     shm_warning_with_errno("Failed to attach shared memory.");
3456     return NULL;
3457   }
3458 
3459   return addr;
3460 }
3461 
3462 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3463   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3464   if (req_addr != NULL) {
3465     assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3466     assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3467     return shmat_at_address(shmid, req_addr);
3468   }
3469 
3470   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3471   // return large page size aligned memory addresses when req_addr == NULL.
3472   // However, if the alignment is larger than the large page size, we have
3473   // to manually ensure that the memory returned is 'alignment' aligned.
3474   if (alignment > os::large_page_size()) {
3475     assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3476     return shmat_with_alignment(shmid, bytes, alignment);
3477   } else {
3478     return shmat_at_address(shmid, NULL);
3479   }
3480 }
3481 
3482 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3483                                             char* req_addr, bool exec) {
3484   // "exec" is passed in but not used.  Creating the shared image for
3485   // the code cache doesn't have an SHM_X executable permission to check.
3486   assert(UseLargePages && UseSHM, "only for SHM large pages");
3487   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3488   assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3489 
3490   if (!is_size_aligned(bytes, os::large_page_size())) {
3491     return NULL; // Fallback to small pages.
3492   }
3493 
3494   // Create a large shared memory region to attach to based on size.
3495   // Currently, size is the total size of the heap.
3496   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3497   if (shmid == -1) {
3498     // Possible reasons for shmget failure:
3499     // 1. shmmax is too small for Java heap.
3500     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3501     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3502     // 2. not enough large page memory.
3503     //    > check available large pages: cat /proc/meminfo
3504     //    > increase amount of large pages:
3505     //          echo new_value > /proc/sys/vm/nr_hugepages
3506     //      Note 1: different Linux may use different name for this property,
3507     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3508     //      Note 2: it's possible there's enough physical memory available but
3509     //            they are so fragmented after a long run that they can't
3510     //            coalesce into large pages. Try to reserve large pages when
3511     //            the system is still "fresh".
3512     shm_warning_with_errno("Failed to reserve shared memory.");
3513     return NULL;
3514   }
3515 
3516   // Attach to the region.
3517   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3518 
3519   // Remove shmid. If shmat() is successful, the actual shared memory segment
3520   // will be deleted when it's detached by shmdt() or when the process
3521   // terminates. If shmat() is not successful this will remove the shared
3522   // segment immediately.
3523   shmctl(shmid, IPC_RMID, NULL);
3524 
3525   return addr;
3526 }
3527 
3528 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3529                                         int error) {
3530   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3531 
3532   bool warn_on_failure = UseLargePages &&
3533       (!FLAG_IS_DEFAULT(UseLargePages) ||
3534        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3535        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3536 
3537   if (warn_on_failure) {
3538     char msg[128];
3539     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3540                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3541     warning("%s", msg);
3542   }
3543 }
3544 
3545 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3546                                                         char* req_addr,
3547                                                         bool exec) {
3548   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3549   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3550   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3551 
3552   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3553   char* addr = (char*)::mmap(req_addr, bytes, prot,
3554                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3555                              -1, 0);
3556 
3557   if (addr == MAP_FAILED) {
3558     warn_on_large_pages_failure(req_addr, bytes, errno);
3559     return NULL;
3560   }
3561 
3562   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3563 
3564   return addr;
3565 }
3566 
3567 // Reserve memory using mmap(MAP_HUGETLB).
3568 //  - bytes shall be a multiple of alignment.
3569 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3570 //  - alignment sets the alignment at which memory shall be allocated.
3571 //     It must be a multiple of allocation granularity.
3572 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3573 //  req_addr or NULL.
3574 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3575                                                          size_t alignment,
3576                                                          char* req_addr,
3577                                                          bool exec) {
3578   size_t large_page_size = os::large_page_size();
3579   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3580 
3581   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3582   assert(is_size_aligned(bytes, alignment), "Must be");
3583 
3584   // First reserve - but not commit - the address range in small pages.
3585   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3586 
3587   if (start == NULL) {
3588     return NULL;
3589   }
3590 
3591   assert(is_ptr_aligned(start, alignment), "Must be");
3592 
3593   char* end = start + bytes;
3594 
3595   // Find the regions of the allocated chunk that can be promoted to large pages.
3596   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3597   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3598 
3599   size_t lp_bytes = lp_end - lp_start;
3600 
3601   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3602 
3603   if (lp_bytes == 0) {
3604     // The mapped region doesn't even span the start and the end of a large page.
3605     // Fall back to allocate a non-special area.
3606     ::munmap(start, end - start);
3607     return NULL;
3608   }
3609 
3610   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3611 
3612   void* result;
3613 
3614   // Commit small-paged leading area.
3615   if (start != lp_start) {
3616     result = ::mmap(start, lp_start - start, prot,
3617                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3618                     -1, 0);
3619     if (result == MAP_FAILED) {
3620       ::munmap(lp_start, end - lp_start);
3621       return NULL;
3622     }
3623   }
3624 
3625   // Commit large-paged area.
3626   result = ::mmap(lp_start, lp_bytes, prot,
3627                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3628                   -1, 0);
3629   if (result == MAP_FAILED) {
3630     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3631     // If the mmap above fails, the large pages region will be unmapped and we
3632     // have regions before and after with small pages. Release these regions.
3633     //
3634     // |  mapped  |  unmapped  |  mapped  |
3635     // ^          ^            ^          ^
3636     // start      lp_start     lp_end     end
3637     //
3638     ::munmap(start, lp_start - start);
3639     ::munmap(lp_end, end - lp_end);
3640     return NULL;
3641   }
3642 
3643   // Commit small-paged trailing area.
3644   if (lp_end != end) {
3645     result = ::mmap(lp_end, end - lp_end, prot,
3646                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3647                     -1, 0);
3648     if (result == MAP_FAILED) {
3649       ::munmap(start, lp_end - start);
3650       return NULL;
3651     }
3652   }
3653 
3654   return start;
3655 }
3656 
3657 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3658                                                    size_t alignment,
3659                                                    char* req_addr,
3660                                                    bool exec) {
3661   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3662   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3663   assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3664   assert(is_power_of_2(os::large_page_size()), "Must be");
3665   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3666 
3667   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3668     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3669   } else {
3670     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3671   }
3672 }
3673 
3674 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3675                                  char* req_addr, bool exec) {
3676   assert(UseLargePages, "only for large pages");
3677 
3678   char* addr;
3679   if (UseSHM) {
3680     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3681   } else {
3682     assert(UseHugeTLBFS, "must be");
3683     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3684   }
3685 
3686   if (addr != NULL) {
3687     if (UseNUMAInterleaving) {
3688       numa_make_global(addr, bytes);
3689     }
3690 
3691     // The memory is committed
3692     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3693   }
3694 
3695   return addr;
3696 }
3697 
3698 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3699   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3700   return shmdt(base) == 0;
3701 }
3702 
3703 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3704   return pd_release_memory(base, bytes);
3705 }
3706 
3707 bool os::release_memory_special(char* base, size_t bytes) {
3708   bool res;
3709   if (MemTracker::tracking_level() > NMT_minimal) {
3710     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3711     res = os::Linux::release_memory_special_impl(base, bytes);
3712     if (res) {
3713       tkr.record((address)base, bytes);
3714     }
3715 
3716   } else {
3717     res = os::Linux::release_memory_special_impl(base, bytes);
3718   }
3719   return res;
3720 }
3721 
3722 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3723   assert(UseLargePages, "only for large pages");
3724   bool res;
3725 
3726   if (UseSHM) {
3727     res = os::Linux::release_memory_special_shm(base, bytes);
3728   } else {
3729     assert(UseHugeTLBFS, "must be");
3730     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3731   }
3732   return res;
3733 }
3734 
3735 size_t os::large_page_size() {
3736   return _large_page_size;
3737 }
3738 
3739 // With SysV SHM the entire memory region must be allocated as shared
3740 // memory.
3741 // HugeTLBFS allows application to commit large page memory on demand.
3742 // However, when committing memory with HugeTLBFS fails, the region
3743 // that was supposed to be committed will lose the old reservation
3744 // and allow other threads to steal that memory region. Because of this
3745 // behavior we can't commit HugeTLBFS memory.
3746 bool os::can_commit_large_page_memory() {
3747   return UseTransparentHugePages;
3748 }
3749 
3750 bool os::can_execute_large_page_memory() {
3751   return UseTransparentHugePages || UseHugeTLBFS;
3752 }
3753 
3754 // Reserve memory at an arbitrary address, only if that area is
3755 // available (and not reserved for something else).
3756 
3757 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3758   const int max_tries = 10;
3759   char* base[max_tries];
3760   size_t size[max_tries];
3761   const size_t gap = 0x000000;
3762 
3763   // Assert only that the size is a multiple of the page size, since
3764   // that's all that mmap requires, and since that's all we really know
3765   // about at this low abstraction level.  If we need higher alignment,
3766   // we can either pass an alignment to this method or verify alignment
3767   // in one of the methods further up the call chain.  See bug 5044738.
3768   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3769 
3770   // Repeatedly allocate blocks until the block is allocated at the
3771   // right spot.
3772 
3773   // Linux mmap allows caller to pass an address as hint; give it a try first,
3774   // if kernel honors the hint then we can return immediately.
3775   char * addr = anon_mmap(requested_addr, bytes, false);
3776   if (addr == requested_addr) {
3777     return requested_addr;
3778   }
3779 
3780   if (addr != NULL) {
3781     // mmap() is successful but it fails to reserve at the requested address
3782     anon_munmap(addr, bytes);
3783   }
3784 
3785   int i;
3786   for (i = 0; i < max_tries; ++i) {
3787     base[i] = reserve_memory(bytes);
3788 
3789     if (base[i] != NULL) {
3790       // Is this the block we wanted?
3791       if (base[i] == requested_addr) {
3792         size[i] = bytes;
3793         break;
3794       }
3795 
3796       // Does this overlap the block we wanted? Give back the overlapped
3797       // parts and try again.
3798 
3799       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3800       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3801         unmap_memory(base[i], top_overlap);
3802         base[i] += top_overlap;
3803         size[i] = bytes - top_overlap;
3804       } else {
3805         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3806         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3807           unmap_memory(requested_addr, bottom_overlap);
3808           size[i] = bytes - bottom_overlap;
3809         } else {
3810           size[i] = bytes;
3811         }
3812       }
3813     }
3814   }
3815 
3816   // Give back the unused reserved pieces.
3817 
3818   for (int j = 0; j < i; ++j) {
3819     if (base[j] != NULL) {
3820       unmap_memory(base[j], size[j]);
3821     }
3822   }
3823 
3824   if (i < max_tries) {
3825     return requested_addr;
3826   } else {
3827     return NULL;
3828   }
3829 }
3830 
3831 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3832   return ::read(fd, buf, nBytes);
3833 }
3834 
3835 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3836   return ::pread(fd, buf, nBytes, offset);
3837 }
3838 
3839 // Short sleep, direct OS call.
3840 //
3841 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3842 // sched_yield(2) will actually give up the CPU:
3843 //
3844 //   * Alone on this pariticular CPU, keeps running.
3845 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3846 //     (pre 2.6.39).
3847 //
3848 // So calling this with 0 is an alternative.
3849 //
3850 void os::naked_short_sleep(jlong ms) {
3851   struct timespec req;
3852 
3853   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3854   req.tv_sec = 0;
3855   if (ms > 0) {
3856     req.tv_nsec = (ms % 1000) * 1000000;
3857   } else {
3858     req.tv_nsec = 1;
3859   }
3860 
3861   nanosleep(&req, NULL);
3862 
3863   return;
3864 }
3865 
3866 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3867 void os::infinite_sleep() {
3868   while (true) {    // sleep forever ...
3869     ::sleep(100);   // ... 100 seconds at a time
3870   }
3871 }
3872 
3873 // Used to convert frequent JVM_Yield() to nops
3874 bool os::dont_yield() {
3875   return DontYieldALot;
3876 }
3877 
3878 void os::naked_yield() {
3879   sched_yield();
3880 }
3881 
3882 ////////////////////////////////////////////////////////////////////////////////
3883 // thread priority support
3884 
3885 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3886 // only supports dynamic priority, static priority must be zero. For real-time
3887 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3888 // However, for large multi-threaded applications, SCHED_RR is not only slower
3889 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3890 // of 5 runs - Sep 2005).
3891 //
3892 // The following code actually changes the niceness of kernel-thread/LWP. It
3893 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3894 // not the entire user process, and user level threads are 1:1 mapped to kernel
3895 // threads. It has always been the case, but could change in the future. For
3896 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3897 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3898 
3899 int os::java_to_os_priority[CriticalPriority + 1] = {
3900   19,              // 0 Entry should never be used
3901 
3902    4,              // 1 MinPriority
3903    3,              // 2
3904    2,              // 3
3905 
3906    1,              // 4
3907    0,              // 5 NormPriority
3908   -1,              // 6
3909 
3910   -2,              // 7
3911   -3,              // 8
3912   -4,              // 9 NearMaxPriority
3913 
3914   -5,              // 10 MaxPriority
3915 
3916   -5               // 11 CriticalPriority
3917 };
3918 
3919 static int prio_init() {
3920   if (ThreadPriorityPolicy == 1) {
3921     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3922     // if effective uid is not root. Perhaps, a more elegant way of doing
3923     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3924     if (geteuid() != 0) {
3925       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3926         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3927       }
3928       ThreadPriorityPolicy = 0;
3929     }
3930   }
3931   if (UseCriticalJavaThreadPriority) {
3932     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3933   }
3934   return 0;
3935 }
3936 
3937 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3938   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3939 
3940   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3941   return (ret == 0) ? OS_OK : OS_ERR;
3942 }
3943 
3944 OSReturn os::get_native_priority(const Thread* const thread,
3945                                  int *priority_ptr) {
3946   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3947     *priority_ptr = java_to_os_priority[NormPriority];
3948     return OS_OK;
3949   }
3950 
3951   errno = 0;
3952   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3953   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3954 }
3955 
3956 // Hint to the underlying OS that a task switch would not be good.
3957 // Void return because it's a hint and can fail.
3958 void os::hint_no_preempt() {}
3959 
3960 ////////////////////////////////////////////////////////////////////////////////
3961 // suspend/resume support
3962 
3963 //  the low-level signal-based suspend/resume support is a remnant from the
3964 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3965 //  within hotspot. Now there is a single use-case for this:
3966 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3967 //      that runs in the watcher thread.
3968 //  The remaining code is greatly simplified from the more general suspension
3969 //  code that used to be used.
3970 //
3971 //  The protocol is quite simple:
3972 //  - suspend:
3973 //      - sends a signal to the target thread
3974 //      - polls the suspend state of the osthread using a yield loop
3975 //      - target thread signal handler (SR_handler) sets suspend state
3976 //        and blocks in sigsuspend until continued
3977 //  - resume:
3978 //      - sets target osthread state to continue
3979 //      - sends signal to end the sigsuspend loop in the SR_handler
3980 //
3981 //  Note that the SR_lock plays no role in this suspend/resume protocol,
3982 //  but is checked for NULL in SR_handler as a thread termination indicator.
3983 
3984 static void resume_clear_context(OSThread *osthread) {
3985   osthread->set_ucontext(NULL);
3986   osthread->set_siginfo(NULL);
3987 }
3988 
3989 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3990                                  ucontext_t* context) {
3991   osthread->set_ucontext(context);
3992   osthread->set_siginfo(siginfo);
3993 }
3994 
3995 // Handler function invoked when a thread's execution is suspended or
3996 // resumed. We have to be careful that only async-safe functions are
3997 // called here (Note: most pthread functions are not async safe and
3998 // should be avoided.)
3999 //
4000 // Note: sigwait() is a more natural fit than sigsuspend() from an
4001 // interface point of view, but sigwait() prevents the signal hander
4002 // from being run. libpthread would get very confused by not having
4003 // its signal handlers run and prevents sigwait()'s use with the
4004 // mutex granting granting signal.
4005 //
4006 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4007 //
4008 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4009   // Save and restore errno to avoid confusing native code with EINTR
4010   // after sigsuspend.
4011   int old_errno = errno;
4012 
4013   Thread* thread = Thread::current_or_null_safe();
4014   assert(thread != NULL, "Missing current thread in SR_handler");
4015 
4016   // On some systems we have seen signal delivery get "stuck" until the signal
4017   // mask is changed as part of thread termination. Check that the current thread
4018   // has not already terminated (via SR_lock()) - else the following assertion
4019   // will fail because the thread is no longer a JavaThread as the ~JavaThread
4020   // destructor has completed.
4021 
4022   if (thread->SR_lock() == NULL) {
4023     return;
4024   }
4025 
4026   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4027 
4028   OSThread* osthread = thread->osthread();
4029 
4030   os::SuspendResume::State current = osthread->sr.state();
4031   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4032     suspend_save_context(osthread, siginfo, context);
4033 
4034     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4035     os::SuspendResume::State state = osthread->sr.suspended();
4036     if (state == os::SuspendResume::SR_SUSPENDED) {
4037       sigset_t suspend_set;  // signals for sigsuspend()
4038       sigemptyset(&suspend_set);
4039       // get current set of blocked signals and unblock resume signal
4040       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4041       sigdelset(&suspend_set, SR_signum);
4042 
4043       sr_semaphore.signal();
4044       // wait here until we are resumed
4045       while (1) {
4046         sigsuspend(&suspend_set);
4047 
4048         os::SuspendResume::State result = osthread->sr.running();
4049         if (result == os::SuspendResume::SR_RUNNING) {
4050           sr_semaphore.signal();
4051           break;
4052         }
4053       }
4054 
4055     } else if (state == os::SuspendResume::SR_RUNNING) {
4056       // request was cancelled, continue
4057     } else {
4058       ShouldNotReachHere();
4059     }
4060 
4061     resume_clear_context(osthread);
4062   } else if (current == os::SuspendResume::SR_RUNNING) {
4063     // request was cancelled, continue
4064   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4065     // ignore
4066   } else {
4067     // ignore
4068   }
4069 
4070   errno = old_errno;
4071 }
4072 
4073 static int SR_initialize() {
4074   struct sigaction act;
4075   char *s;
4076 
4077   // Get signal number to use for suspend/resume
4078   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4079     int sig = ::strtol(s, 0, 10);
4080     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4081         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4082       SR_signum = sig;
4083     } else {
4084       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4085               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4086     }
4087   }
4088 
4089   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4090          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4091 
4092   sigemptyset(&SR_sigset);
4093   sigaddset(&SR_sigset, SR_signum);
4094 
4095   // Set up signal handler for suspend/resume
4096   act.sa_flags = SA_RESTART|SA_SIGINFO;
4097   act.sa_handler = (void (*)(int)) SR_handler;
4098 
4099   // SR_signum is blocked by default.
4100   // 4528190 - We also need to block pthread restart signal (32 on all
4101   // supported Linux platforms). Note that LinuxThreads need to block
4102   // this signal for all threads to work properly. So we don't have
4103   // to use hard-coded signal number when setting up the mask.
4104   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4105 
4106   if (sigaction(SR_signum, &act, 0) == -1) {
4107     return -1;
4108   }
4109 
4110   // Save signal flag
4111   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4112   return 0;
4113 }
4114 
4115 static int sr_notify(OSThread* osthread) {
4116   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4117   assert_status(status == 0, status, "pthread_kill");
4118   return status;
4119 }
4120 
4121 // "Randomly" selected value for how long we want to spin
4122 // before bailing out on suspending a thread, also how often
4123 // we send a signal to a thread we want to resume
4124 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4125 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4126 
4127 // returns true on success and false on error - really an error is fatal
4128 // but this seems the normal response to library errors
4129 static bool do_suspend(OSThread* osthread) {
4130   assert(osthread->sr.is_running(), "thread should be running");
4131   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4132 
4133   // mark as suspended and send signal
4134   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4135     // failed to switch, state wasn't running?
4136     ShouldNotReachHere();
4137     return false;
4138   }
4139 
4140   if (sr_notify(osthread) != 0) {
4141     ShouldNotReachHere();
4142   }
4143 
4144   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4145   while (true) {
4146     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4147       break;
4148     } else {
4149       // timeout
4150       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4151       if (cancelled == os::SuspendResume::SR_RUNNING) {
4152         return false;
4153       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4154         // make sure that we consume the signal on the semaphore as well
4155         sr_semaphore.wait();
4156         break;
4157       } else {
4158         ShouldNotReachHere();
4159         return false;
4160       }
4161     }
4162   }
4163 
4164   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4165   return true;
4166 }
4167 
4168 static void do_resume(OSThread* osthread) {
4169   assert(osthread->sr.is_suspended(), "thread should be suspended");
4170   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4171 
4172   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4173     // failed to switch to WAKEUP_REQUEST
4174     ShouldNotReachHere();
4175     return;
4176   }
4177 
4178   while (true) {
4179     if (sr_notify(osthread) == 0) {
4180       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4181         if (osthread->sr.is_running()) {
4182           return;
4183         }
4184       }
4185     } else {
4186       ShouldNotReachHere();
4187     }
4188   }
4189 
4190   guarantee(osthread->sr.is_running(), "Must be running!");
4191 }
4192 
4193 ///////////////////////////////////////////////////////////////////////////////////
4194 // signal handling (except suspend/resume)
4195 
4196 // This routine may be used by user applications as a "hook" to catch signals.
4197 // The user-defined signal handler must pass unrecognized signals to this
4198 // routine, and if it returns true (non-zero), then the signal handler must
4199 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4200 // routine will never retun false (zero), but instead will execute a VM panic
4201 // routine kill the process.
4202 //
4203 // If this routine returns false, it is OK to call it again.  This allows
4204 // the user-defined signal handler to perform checks either before or after
4205 // the VM performs its own checks.  Naturally, the user code would be making
4206 // a serious error if it tried to handle an exception (such as a null check
4207 // or breakpoint) that the VM was generating for its own correct operation.
4208 //
4209 // This routine may recognize any of the following kinds of signals:
4210 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4211 // It should be consulted by handlers for any of those signals.
4212 //
4213 // The caller of this routine must pass in the three arguments supplied
4214 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4215 // field of the structure passed to sigaction().  This routine assumes that
4216 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4217 //
4218 // Note that the VM will print warnings if it detects conflicting signal
4219 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4220 //
4221 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4222                                                  siginfo_t* siginfo,
4223                                                  void* ucontext,
4224                                                  int abort_if_unrecognized);
4225 
4226 void signalHandler(int sig, siginfo_t* info, void* uc) {
4227   assert(info != NULL && uc != NULL, "it must be old kernel");
4228   int orig_errno = errno;  // Preserve errno value over signal handler.
4229   JVM_handle_linux_signal(sig, info, uc, true);
4230   errno = orig_errno;
4231 }
4232 
4233 
4234 // This boolean allows users to forward their own non-matching signals
4235 // to JVM_handle_linux_signal, harmlessly.
4236 bool os::Linux::signal_handlers_are_installed = false;
4237 
4238 // For signal-chaining
4239 struct sigaction sigact[NSIG];
4240 uint64_t sigs = 0;
4241 #if (64 < NSIG-1)
4242 #error "Not all signals can be encoded in sigs. Adapt its type!"
4243 #endif
4244 bool os::Linux::libjsig_is_loaded = false;
4245 typedef struct sigaction *(*get_signal_t)(int);
4246 get_signal_t os::Linux::get_signal_action = NULL;
4247 
4248 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4249   struct sigaction *actp = NULL;
4250 
4251   if (libjsig_is_loaded) {
4252     // Retrieve the old signal handler from libjsig
4253     actp = (*get_signal_action)(sig);
4254   }
4255   if (actp == NULL) {
4256     // Retrieve the preinstalled signal handler from jvm
4257     actp = get_preinstalled_handler(sig);
4258   }
4259 
4260   return actp;
4261 }
4262 
4263 static bool call_chained_handler(struct sigaction *actp, int sig,
4264                                  siginfo_t *siginfo, void *context) {
4265   // Call the old signal handler
4266   if (actp->sa_handler == SIG_DFL) {
4267     // It's more reasonable to let jvm treat it as an unexpected exception
4268     // instead of taking the default action.
4269     return false;
4270   } else if (actp->sa_handler != SIG_IGN) {
4271     if ((actp->sa_flags & SA_NODEFER) == 0) {
4272       // automaticlly block the signal
4273       sigaddset(&(actp->sa_mask), sig);
4274     }
4275 
4276     sa_handler_t hand = NULL;
4277     sa_sigaction_t sa = NULL;
4278     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4279     // retrieve the chained handler
4280     if (siginfo_flag_set) {
4281       sa = actp->sa_sigaction;
4282     } else {
4283       hand = actp->sa_handler;
4284     }
4285 
4286     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4287       actp->sa_handler = SIG_DFL;
4288     }
4289 
4290     // try to honor the signal mask
4291     sigset_t oset;
4292     sigemptyset(&oset);
4293     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4294 
4295     // call into the chained handler
4296     if (siginfo_flag_set) {
4297       (*sa)(sig, siginfo, context);
4298     } else {
4299       (*hand)(sig);
4300     }
4301 
4302     // restore the signal mask
4303     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4304   }
4305   // Tell jvm's signal handler the signal is taken care of.
4306   return true;
4307 }
4308 
4309 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4310   bool chained = false;
4311   // signal-chaining
4312   if (UseSignalChaining) {
4313     struct sigaction *actp = get_chained_signal_action(sig);
4314     if (actp != NULL) {
4315       chained = call_chained_handler(actp, sig, siginfo, context);
4316     }
4317   }
4318   return chained;
4319 }
4320 
4321 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4322   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4323     return &sigact[sig];
4324   }
4325   return NULL;
4326 }
4327 
4328 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4329   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4330   sigact[sig] = oldAct;
4331   sigs |= (uint64_t)1 << (sig-1);
4332 }
4333 
4334 // for diagnostic
4335 int sigflags[NSIG];
4336 
4337 int os::Linux::get_our_sigflags(int sig) {
4338   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4339   return sigflags[sig];
4340 }
4341 
4342 void os::Linux::set_our_sigflags(int sig, int flags) {
4343   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4344   if (sig > 0 && sig < NSIG) {
4345     sigflags[sig] = flags;
4346   }
4347 }
4348 
4349 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4350   // Check for overwrite.
4351   struct sigaction oldAct;
4352   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4353 
4354   void* oldhand = oldAct.sa_sigaction
4355                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4356                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4357   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4358       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4359       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4360     if (AllowUserSignalHandlers || !set_installed) {
4361       // Do not overwrite; user takes responsibility to forward to us.
4362       return;
4363     } else if (UseSignalChaining) {
4364       // save the old handler in jvm
4365       save_preinstalled_handler(sig, oldAct);
4366       // libjsig also interposes the sigaction() call below and saves the
4367       // old sigaction on it own.
4368     } else {
4369       fatal("Encountered unexpected pre-existing sigaction handler "
4370             "%#lx for signal %d.", (long)oldhand, sig);
4371     }
4372   }
4373 
4374   struct sigaction sigAct;
4375   sigfillset(&(sigAct.sa_mask));
4376   sigAct.sa_handler = SIG_DFL;
4377   if (!set_installed) {
4378     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4379   } else {
4380     sigAct.sa_sigaction = signalHandler;
4381     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4382   }
4383   // Save flags, which are set by ours
4384   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4385   sigflags[sig] = sigAct.sa_flags;
4386 
4387   int ret = sigaction(sig, &sigAct, &oldAct);
4388   assert(ret == 0, "check");
4389 
4390   void* oldhand2  = oldAct.sa_sigaction
4391                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4392                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4393   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4394 }
4395 
4396 // install signal handlers for signals that HotSpot needs to
4397 // handle in order to support Java-level exception handling.
4398 
4399 void os::Linux::install_signal_handlers() {
4400   if (!signal_handlers_are_installed) {
4401     signal_handlers_are_installed = true;
4402 
4403     // signal-chaining
4404     typedef void (*signal_setting_t)();
4405     signal_setting_t begin_signal_setting = NULL;
4406     signal_setting_t end_signal_setting = NULL;
4407     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4408                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4409     if (begin_signal_setting != NULL) {
4410       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4411                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4412       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4413                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4414       libjsig_is_loaded = true;
4415       assert(UseSignalChaining, "should enable signal-chaining");
4416     }
4417     if (libjsig_is_loaded) {
4418       // Tell libjsig jvm is setting signal handlers
4419       (*begin_signal_setting)();
4420     }
4421 
4422     set_signal_handler(SIGSEGV, true);
4423     set_signal_handler(SIGPIPE, true);
4424     set_signal_handler(SIGBUS, true);
4425     set_signal_handler(SIGILL, true);
4426     set_signal_handler(SIGFPE, true);
4427 #if defined(PPC64)
4428     set_signal_handler(SIGTRAP, true);
4429 #endif
4430     set_signal_handler(SIGXFSZ, true);
4431 
4432     if (libjsig_is_loaded) {
4433       // Tell libjsig jvm finishes setting signal handlers
4434       (*end_signal_setting)();
4435     }
4436 
4437     // We don't activate signal checker if libjsig is in place, we trust ourselves
4438     // and if UserSignalHandler is installed all bets are off.
4439     // Log that signal checking is off only if -verbose:jni is specified.
4440     if (CheckJNICalls) {
4441       if (libjsig_is_loaded) {
4442         if (PrintJNIResolving) {
4443           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4444         }
4445         check_signals = false;
4446       }
4447       if (AllowUserSignalHandlers) {
4448         if (PrintJNIResolving) {
4449           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4450         }
4451         check_signals = false;
4452       }
4453     }
4454   }
4455 }
4456 
4457 // This is the fastest way to get thread cpu time on Linux.
4458 // Returns cpu time (user+sys) for any thread, not only for current.
4459 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4460 // It might work on 2.6.10+ with a special kernel/glibc patch.
4461 // For reference, please, see IEEE Std 1003.1-2004:
4462 //   http://www.unix.org/single_unix_specification
4463 
4464 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4465   struct timespec tp;
4466   int rc = os::Linux::clock_gettime(clockid, &tp);
4467   assert(rc == 0, "clock_gettime is expected to return 0 code");
4468 
4469   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4470 }
4471 
4472 void os::Linux::initialize_os_info() {
4473   assert(_os_version == 0, "OS info already initialized");
4474 
4475   struct utsname _uname;
4476 
4477   uint32_t major;
4478   uint32_t minor;
4479   uint32_t fix;
4480 
4481   int rc;
4482 
4483   // Kernel version is unknown if
4484   // verification below fails.
4485   _os_version = 0x01000000;
4486 
4487   rc = uname(&_uname);
4488   if (rc != -1) {
4489 
4490     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4491     if (rc == 3) {
4492 
4493       if (major < 256 && minor < 256 && fix < 256) {
4494         // Kernel version format is as expected,
4495         // set it overriding unknown state.
4496         _os_version = (major << 16) |
4497                       (minor << 8 ) |
4498                       (fix   << 0 ) ;
4499       }
4500     }
4501   }
4502 }
4503 
4504 uint32_t os::Linux::os_version() {
4505   assert(_os_version != 0, "not initialized");
4506   return _os_version & 0x00FFFFFF;
4507 }
4508 
4509 bool os::Linux::os_version_is_known() {
4510   assert(_os_version != 0, "not initialized");
4511   return _os_version & 0x01000000 ? false : true;
4512 }
4513 
4514 /////
4515 // glibc on Linux platform uses non-documented flag
4516 // to indicate, that some special sort of signal
4517 // trampoline is used.
4518 // We will never set this flag, and we should
4519 // ignore this flag in our diagnostic
4520 #ifdef SIGNIFICANT_SIGNAL_MASK
4521   #undef SIGNIFICANT_SIGNAL_MASK
4522 #endif
4523 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4524 
4525 static const char* get_signal_handler_name(address handler,
4526                                            char* buf, int buflen) {
4527   int offset = 0;
4528   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4529   if (found) {
4530     // skip directory names
4531     const char *p1, *p2;
4532     p1 = buf;
4533     size_t len = strlen(os::file_separator());
4534     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4535     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4536   } else {
4537     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4538   }
4539   return buf;
4540 }
4541 
4542 static void print_signal_handler(outputStream* st, int sig,
4543                                  char* buf, size_t buflen) {
4544   struct sigaction sa;
4545 
4546   sigaction(sig, NULL, &sa);
4547 
4548   // See comment for SIGNIFICANT_SIGNAL_MASK define
4549   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4550 
4551   st->print("%s: ", os::exception_name(sig, buf, buflen));
4552 
4553   address handler = (sa.sa_flags & SA_SIGINFO)
4554     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4555     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4556 
4557   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4558     st->print("SIG_DFL");
4559   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4560     st->print("SIG_IGN");
4561   } else {
4562     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4563   }
4564 
4565   st->print(", sa_mask[0]=");
4566   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4567 
4568   address rh = VMError::get_resetted_sighandler(sig);
4569   // May be, handler was resetted by VMError?
4570   if (rh != NULL) {
4571     handler = rh;
4572     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4573   }
4574 
4575   st->print(", sa_flags=");
4576   os::Posix::print_sa_flags(st, sa.sa_flags);
4577 
4578   // Check: is it our handler?
4579   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4580       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4581     // It is our signal handler
4582     // check for flags, reset system-used one!
4583     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4584       st->print(
4585                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4586                 os::Linux::get_our_sigflags(sig));
4587     }
4588   }
4589   st->cr();
4590 }
4591 
4592 
4593 #define DO_SIGNAL_CHECK(sig)                      \
4594   do {                                            \
4595     if (!sigismember(&check_signal_done, sig)) {  \
4596       os::Linux::check_signal_handler(sig);       \
4597     }                                             \
4598   } while (0)
4599 
4600 // This method is a periodic task to check for misbehaving JNI applications
4601 // under CheckJNI, we can add any periodic checks here
4602 
4603 void os::run_periodic_checks() {
4604   if (check_signals == false) return;
4605 
4606   // SEGV and BUS if overridden could potentially prevent
4607   // generation of hs*.log in the event of a crash, debugging
4608   // such a case can be very challenging, so we absolutely
4609   // check the following for a good measure:
4610   DO_SIGNAL_CHECK(SIGSEGV);
4611   DO_SIGNAL_CHECK(SIGILL);
4612   DO_SIGNAL_CHECK(SIGFPE);
4613   DO_SIGNAL_CHECK(SIGBUS);
4614   DO_SIGNAL_CHECK(SIGPIPE);
4615   DO_SIGNAL_CHECK(SIGXFSZ);
4616 #if defined(PPC64)
4617   DO_SIGNAL_CHECK(SIGTRAP);
4618 #endif
4619 
4620   // ReduceSignalUsage allows the user to override these handlers
4621   // see comments at the very top and jvm_solaris.h
4622   if (!ReduceSignalUsage) {
4623     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4624     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4625     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4626     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4627   }
4628 
4629   DO_SIGNAL_CHECK(SR_signum);
4630 }
4631 
4632 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4633 
4634 static os_sigaction_t os_sigaction = NULL;
4635 
4636 void os::Linux::check_signal_handler(int sig) {
4637   char buf[O_BUFLEN];
4638   address jvmHandler = NULL;
4639 
4640 
4641   struct sigaction act;
4642   if (os_sigaction == NULL) {
4643     // only trust the default sigaction, in case it has been interposed
4644     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4645     if (os_sigaction == NULL) return;
4646   }
4647 
4648   os_sigaction(sig, (struct sigaction*)NULL, &act);
4649 
4650 
4651   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4652 
4653   address thisHandler = (act.sa_flags & SA_SIGINFO)
4654     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4655     : CAST_FROM_FN_PTR(address, act.sa_handler);
4656 
4657 
4658   switch (sig) {
4659   case SIGSEGV:
4660   case SIGBUS:
4661   case SIGFPE:
4662   case SIGPIPE:
4663   case SIGILL:
4664   case SIGXFSZ:
4665     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4666     break;
4667 
4668   case SHUTDOWN1_SIGNAL:
4669   case SHUTDOWN2_SIGNAL:
4670   case SHUTDOWN3_SIGNAL:
4671   case BREAK_SIGNAL:
4672     jvmHandler = (address)user_handler();
4673     break;
4674 
4675   default:
4676     if (sig == SR_signum) {
4677       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4678     } else {
4679       return;
4680     }
4681     break;
4682   }
4683 
4684   if (thisHandler != jvmHandler) {
4685     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4686     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4687     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4688     // No need to check this sig any longer
4689     sigaddset(&check_signal_done, sig);
4690     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4691     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4692       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4693                     exception_name(sig, buf, O_BUFLEN));
4694     }
4695   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4696     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4697     tty->print("expected:");
4698     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4699     tty->cr();
4700     tty->print("  found:");
4701     os::Posix::print_sa_flags(tty, act.sa_flags);
4702     tty->cr();
4703     // No need to check this sig any longer
4704     sigaddset(&check_signal_done, sig);
4705   }
4706 
4707   // Dump all the signal
4708   if (sigismember(&check_signal_done, sig)) {
4709     print_signal_handlers(tty, buf, O_BUFLEN);
4710   }
4711 }
4712 
4713 extern void report_error(char* file_name, int line_no, char* title,
4714                          char* format, ...);
4715 
4716 // this is called _before_ the most of global arguments have been parsed
4717 void os::init(void) {
4718   char dummy;   // used to get a guess on initial stack address
4719 //  first_hrtime = gethrtime();
4720 
4721   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4722 
4723   init_random(1234567);
4724 
4725   ThreadCritical::initialize();
4726 
4727   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4728   if (Linux::page_size() == -1) {
4729     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4730           os::strerror(errno));
4731   }
4732   init_page_sizes((size_t) Linux::page_size());
4733 
4734   Linux::initialize_system_info();
4735 
4736   Linux::initialize_os_info();
4737 
4738   // main_thread points to the aboriginal thread
4739   Linux::_main_thread = pthread_self();
4740 
4741   Linux::clock_init();
4742   initial_time_count = javaTimeNanos();
4743 
4744   // pthread_condattr initialization for monotonic clock
4745   int status;
4746   pthread_condattr_t* _condattr = os::Linux::condAttr();
4747   if ((status = pthread_condattr_init(_condattr)) != 0) {
4748     fatal("pthread_condattr_init: %s", os::strerror(status));
4749   }
4750   // Only set the clock if CLOCK_MONOTONIC is available
4751   if (os::supports_monotonic_clock()) {
4752     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4753       if (status == EINVAL) {
4754         warning("Unable to use monotonic clock with relative timed-waits" \
4755                 " - changes to the time-of-day clock may have adverse affects");
4756       } else {
4757         fatal("pthread_condattr_setclock: %s", os::strerror(status));
4758       }
4759     }
4760   }
4761   // else it defaults to CLOCK_REALTIME
4762 
4763   // retrieve entry point for pthread_setname_np
4764   Linux::_pthread_setname_np =
4765     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4766 
4767 }
4768 
4769 // To install functions for atexit system call
4770 extern "C" {
4771   static void perfMemory_exit_helper() {
4772     perfMemory_exit();
4773   }
4774 }
4775 
4776 // this is called _after_ the global arguments have been parsed
4777 jint os::init_2(void) {
4778   Linux::fast_thread_clock_init();
4779 
4780   // Allocate a single page and mark it as readable for safepoint polling
4781   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4782   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4783 
4784   os::set_polling_page(polling_page);
4785   log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4786 
4787   if (!UseMembar) {
4788     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4789     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4790     os::set_memory_serialize_page(mem_serialize_page);
4791     log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4792   }
4793 
4794   // initialize suspend/resume support - must do this before signal_sets_init()
4795   if (SR_initialize() != 0) {
4796     perror("SR_initialize failed");
4797     return JNI_ERR;
4798   }
4799 
4800   Linux::signal_sets_init();
4801   Linux::install_signal_handlers();
4802 
4803   // Check and sets minimum stack sizes against command line options
4804   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4805     return JNI_ERR;
4806   }
4807   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4808 
4809 #if defined(IA32)
4810   workaround_expand_exec_shield_cs_limit();
4811 #endif
4812 
4813   Linux::libpthread_init();
4814   log_info(os)("HotSpot is running with %s, %s",
4815                Linux::glibc_version(), Linux::libpthread_version());
4816 
4817   if (UseNUMA) {
4818     if (!Linux::libnuma_init()) {
4819       UseNUMA = false;
4820     } else {
4821       if ((Linux::numa_max_node() < 1)) {
4822         // There's only one node(they start from 0), disable NUMA.
4823         UseNUMA = false;
4824       }
4825     }
4826     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4827     // we can make the adaptive lgrp chunk resizing work. If the user specified
4828     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4829     // disable adaptive resizing.
4830     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4831       if (FLAG_IS_DEFAULT(UseNUMA)) {
4832         UseNUMA = false;
4833       } else {
4834         if (FLAG_IS_DEFAULT(UseLargePages) &&
4835             FLAG_IS_DEFAULT(UseSHM) &&
4836             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4837           UseLargePages = false;
4838         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4839           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4840           UseAdaptiveSizePolicy = false;
4841           UseAdaptiveNUMAChunkSizing = false;
4842         }
4843       }
4844     }
4845     if (!UseNUMA && ForceNUMA) {
4846       UseNUMA = true;
4847     }
4848   }
4849 
4850   if (MaxFDLimit) {
4851     // set the number of file descriptors to max. print out error
4852     // if getrlimit/setrlimit fails but continue regardless.
4853     struct rlimit nbr_files;
4854     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4855     if (status != 0) {
4856       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4857     } else {
4858       nbr_files.rlim_cur = nbr_files.rlim_max;
4859       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4860       if (status != 0) {
4861         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4862       }
4863     }
4864   }
4865 
4866   // Initialize lock used to serialize thread creation (see os::create_thread)
4867   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4868 
4869   // at-exit methods are called in the reverse order of their registration.
4870   // atexit functions are called on return from main or as a result of a
4871   // call to exit(3C). There can be only 32 of these functions registered
4872   // and atexit() does not set errno.
4873 
4874   if (PerfAllowAtExitRegistration) {
4875     // only register atexit functions if PerfAllowAtExitRegistration is set.
4876     // atexit functions can be delayed until process exit time, which
4877     // can be problematic for embedded VM situations. Embedded VMs should
4878     // call DestroyJavaVM() to assure that VM resources are released.
4879 
4880     // note: perfMemory_exit_helper atexit function may be removed in
4881     // the future if the appropriate cleanup code can be added to the
4882     // VM_Exit VMOperation's doit method.
4883     if (atexit(perfMemory_exit_helper) != 0) {
4884       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4885     }
4886   }
4887 
4888   // initialize thread priority policy
4889   prio_init();
4890 
4891   return JNI_OK;
4892 }
4893 
4894 // Mark the polling page as unreadable
4895 void os::make_polling_page_unreadable(void) {
4896   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4897     fatal("Could not disable polling page");
4898   }
4899 }
4900 
4901 // Mark the polling page as readable
4902 void os::make_polling_page_readable(void) {
4903   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4904     fatal("Could not enable polling page");
4905   }
4906 }
4907 
4908 // older glibc versions don't have this macro (which expands to
4909 // an optimized bit-counting function) so we have to roll our own
4910 #ifndef CPU_COUNT
4911 
4912 static int _cpu_count(const cpu_set_t* cpus) {
4913   int count = 0;
4914   // only look up to the number of configured processors
4915   for (int i = 0; i < os::processor_count(); i++) {
4916     if (CPU_ISSET(i, cpus)) {
4917       count++;
4918     }
4919   }
4920   return count;
4921 }
4922 
4923 #define CPU_COUNT(cpus) _cpu_count(cpus)
4924 
4925 #endif // CPU_COUNT
4926 
4927 // Get the current number of available processors for this process.
4928 // This value can change at any time during a process's lifetime.
4929 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4930 // If it appears there may be more than 1024 processors then we do a
4931 // dynamic check - see 6515172 for details.
4932 // If anything goes wrong we fallback to returning the number of online
4933 // processors - which can be greater than the number available to the process.
4934 int os::active_processor_count() {
4935   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
4936   cpu_set_t* cpus_p = &cpus;
4937   int cpus_size = sizeof(cpu_set_t);
4938 
4939   int configured_cpus = processor_count();  // upper bound on available cpus
4940   int cpu_count = 0;
4941 
4942 // old build platforms may not support dynamic cpu sets
4943 #ifdef CPU_ALLOC
4944 
4945   // To enable easy testing of the dynamic path on different platforms we
4946   // introduce a diagnostic flag: UseCpuAllocPath
4947   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4948     // kernel may use a mask bigger than cpu_set_t
4949     log_trace(os)("active_processor_count: using dynamic path %s"
4950                   "- configured processors: %d",
4951                   UseCpuAllocPath ? "(forced) " : "",
4952                   configured_cpus);
4953     cpus_p = CPU_ALLOC(configured_cpus);
4954     if (cpus_p != NULL) {
4955       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4956       // zero it just to be safe
4957       CPU_ZERO_S(cpus_size, cpus_p);
4958     }
4959     else {
4960        // failed to allocate so fallback to online cpus
4961        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4962        log_trace(os)("active_processor_count: "
4963                      "CPU_ALLOC failed (%s) - using "
4964                      "online processor count: %d",
4965                      os::strerror(errno), online_cpus);
4966        return online_cpus;
4967     }
4968   }
4969   else {
4970     log_trace(os)("active_processor_count: using static path - configured processors: %d",
4971                   configured_cpus);
4972   }
4973 #else // CPU_ALLOC
4974 // these stubs won't be executed
4975 #define CPU_COUNT_S(size, cpus) -1
4976 #define CPU_FREE(cpus)
4977 
4978   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
4979                 configured_cpus);
4980 #endif // CPU_ALLOC
4981 
4982   // pid 0 means the current thread - which we have to assume represents the process
4983   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
4984     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4985       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
4986     }
4987     else {
4988       cpu_count = CPU_COUNT(cpus_p);
4989     }
4990     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
4991   }
4992   else {
4993     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
4994     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
4995             "which may exceed available processors", os::strerror(errno), cpu_count);
4996   }
4997 
4998   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
4999     CPU_FREE(cpus_p);
5000   }
5001 
5002   assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
5003   return cpu_count;
5004 }
5005 
5006 void os::set_native_thread_name(const char *name) {
5007   if (Linux::_pthread_setname_np) {
5008     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5009     snprintf(buf, sizeof(buf), "%s", name);
5010     buf[sizeof(buf) - 1] = '\0';
5011     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5012     // ERANGE should not happen; all other errors should just be ignored.
5013     assert(rc != ERANGE, "pthread_setname_np failed");
5014   }
5015 }
5016 
5017 bool os::distribute_processes(uint length, uint* distribution) {
5018   // Not yet implemented.
5019   return false;
5020 }
5021 
5022 bool os::bind_to_processor(uint processor_id) {
5023   // Not yet implemented.
5024   return false;
5025 }
5026 
5027 ///
5028 
5029 void os::SuspendedThreadTask::internal_do_task() {
5030   if (do_suspend(_thread->osthread())) {
5031     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5032     do_task(context);
5033     do_resume(_thread->osthread());
5034   }
5035 }
5036 
5037 class PcFetcher : public os::SuspendedThreadTask {
5038  public:
5039   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5040   ExtendedPC result();
5041  protected:
5042   void do_task(const os::SuspendedThreadTaskContext& context);
5043  private:
5044   ExtendedPC _epc;
5045 };
5046 
5047 ExtendedPC PcFetcher::result() {
5048   guarantee(is_done(), "task is not done yet.");
5049   return _epc;
5050 }
5051 
5052 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5053   Thread* thread = context.thread();
5054   OSThread* osthread = thread->osthread();
5055   if (osthread->ucontext() != NULL) {
5056     _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
5057   } else {
5058     // NULL context is unexpected, double-check this is the VMThread
5059     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5060   }
5061 }
5062 
5063 // Suspends the target using the signal mechanism and then grabs the PC before
5064 // resuming the target. Used by the flat-profiler only
5065 ExtendedPC os::get_thread_pc(Thread* thread) {
5066   // Make sure that it is called by the watcher for the VMThread
5067   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5068   assert(thread->is_VM_thread(), "Can only be called for VMThread");
5069 
5070   PcFetcher fetcher(thread);
5071   fetcher.run();
5072   return fetcher.result();
5073 }
5074 
5075 ////////////////////////////////////////////////////////////////////////////////
5076 // debug support
5077 
5078 bool os::find(address addr, outputStream* st) {
5079   Dl_info dlinfo;
5080   memset(&dlinfo, 0, sizeof(dlinfo));
5081   if (dladdr(addr, &dlinfo) != 0) {
5082     st->print(PTR_FORMAT ": ", p2i(addr));
5083     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5084       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5085                 p2i(addr) - p2i(dlinfo.dli_saddr));
5086     } else if (dlinfo.dli_fbase != NULL) {
5087       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5088     } else {
5089       st->print("<absolute address>");
5090     }
5091     if (dlinfo.dli_fname != NULL) {
5092       st->print(" in %s", dlinfo.dli_fname);
5093     }
5094     if (dlinfo.dli_fbase != NULL) {
5095       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5096     }
5097     st->cr();
5098 
5099     if (Verbose) {
5100       // decode some bytes around the PC
5101       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5102       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5103       address       lowest = (address) dlinfo.dli_sname;
5104       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5105       if (begin < lowest)  begin = lowest;
5106       Dl_info dlinfo2;
5107       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5108           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5109         end = (address) dlinfo2.dli_saddr;
5110       }
5111       Disassembler::decode(begin, end, st);
5112     }
5113     return true;
5114   }
5115   return false;
5116 }
5117 
5118 ////////////////////////////////////////////////////////////////////////////////
5119 // misc
5120 
5121 // This does not do anything on Linux. This is basically a hook for being
5122 // able to use structured exception handling (thread-local exception filters)
5123 // on, e.g., Win32.
5124 void
5125 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5126                          JavaCallArguments* args, Thread* thread) {
5127   f(value, method, args, thread);
5128 }
5129 
5130 void os::print_statistics() {
5131 }
5132 
5133 bool os::message_box(const char* title, const char* message) {
5134   int i;
5135   fdStream err(defaultStream::error_fd());
5136   for (i = 0; i < 78; i++) err.print_raw("=");
5137   err.cr();
5138   err.print_raw_cr(title);
5139   for (i = 0; i < 78; i++) err.print_raw("-");
5140   err.cr();
5141   err.print_raw_cr(message);
5142   for (i = 0; i < 78; i++) err.print_raw("=");
5143   err.cr();
5144 
5145   char buf[16];
5146   // Prevent process from exiting upon "read error" without consuming all CPU
5147   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5148 
5149   return buf[0] == 'y' || buf[0] == 'Y';
5150 }
5151 
5152 int os::stat(const char *path, struct stat *sbuf) {
5153   char pathbuf[MAX_PATH];
5154   if (strlen(path) > MAX_PATH - 1) {
5155     errno = ENAMETOOLONG;
5156     return -1;
5157   }
5158   os::native_path(strcpy(pathbuf, path));
5159   return ::stat(pathbuf, sbuf);
5160 }
5161 
5162 // Is a (classpath) directory empty?
5163 bool os::dir_is_empty(const char* path) {
5164   DIR *dir = NULL;
5165   struct dirent *ptr;
5166 
5167   dir = opendir(path);
5168   if (dir == NULL) return true;
5169 
5170   // Scan the directory
5171   bool result = true;
5172   char buf[sizeof(struct dirent) + MAX_PATH];
5173   while (result && (ptr = ::readdir(dir)) != NULL) {
5174     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5175       result = false;
5176     }
5177   }
5178   closedir(dir);
5179   return result;
5180 }
5181 
5182 // This code originates from JDK's sysOpen and open64_w
5183 // from src/solaris/hpi/src/system_md.c
5184 
5185 int os::open(const char *path, int oflag, int mode) {
5186   if (strlen(path) > MAX_PATH - 1) {
5187     errno = ENAMETOOLONG;
5188     return -1;
5189   }
5190 
5191   // All file descriptors that are opened in the Java process and not
5192   // specifically destined for a subprocess should have the close-on-exec
5193   // flag set.  If we don't set it, then careless 3rd party native code
5194   // might fork and exec without closing all appropriate file descriptors
5195   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5196   // turn might:
5197   //
5198   // - cause end-of-file to fail to be detected on some file
5199   //   descriptors, resulting in mysterious hangs, or
5200   //
5201   // - might cause an fopen in the subprocess to fail on a system
5202   //   suffering from bug 1085341.
5203   //
5204   // (Yes, the default setting of the close-on-exec flag is a Unix
5205   // design flaw)
5206   //
5207   // See:
5208   // 1085341: 32-bit stdio routines should support file descriptors >255
5209   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5210   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5211   //
5212   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5213   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5214   // because it saves a system call and removes a small window where the flag
5215   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5216   // and we fall back to using FD_CLOEXEC (see below).
5217 #ifdef O_CLOEXEC
5218   oflag |= O_CLOEXEC;
5219 #endif
5220 
5221   int fd = ::open64(path, oflag, mode);
5222   if (fd == -1) return -1;
5223 
5224   //If the open succeeded, the file might still be a directory
5225   {
5226     struct stat64 buf64;
5227     int ret = ::fstat64(fd, &buf64);
5228     int st_mode = buf64.st_mode;
5229 
5230     if (ret != -1) {
5231       if ((st_mode & S_IFMT) == S_IFDIR) {
5232         errno = EISDIR;
5233         ::close(fd);
5234         return -1;
5235       }
5236     } else {
5237       ::close(fd);
5238       return -1;
5239     }
5240   }
5241 
5242 #ifdef FD_CLOEXEC
5243   // Validate that the use of the O_CLOEXEC flag on open above worked.
5244   // With recent kernels, we will perform this check exactly once.
5245   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5246   if (!O_CLOEXEC_is_known_to_work) {
5247     int flags = ::fcntl(fd, F_GETFD);
5248     if (flags != -1) {
5249       if ((flags & FD_CLOEXEC) != 0)
5250         O_CLOEXEC_is_known_to_work = 1;
5251       else
5252         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5253     }
5254   }
5255 #endif
5256 
5257   return fd;
5258 }
5259 
5260 
5261 // create binary file, rewriting existing file if required
5262 int os::create_binary_file(const char* path, bool rewrite_existing) {
5263   int oflags = O_WRONLY | O_CREAT;
5264   if (!rewrite_existing) {
5265     oflags |= O_EXCL;
5266   }
5267   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5268 }
5269 
5270 // return current position of file pointer
5271 jlong os::current_file_offset(int fd) {
5272   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5273 }
5274 
5275 // move file pointer to the specified offset
5276 jlong os::seek_to_file_offset(int fd, jlong offset) {
5277   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5278 }
5279 
5280 // This code originates from JDK's sysAvailable
5281 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5282 
5283 int os::available(int fd, jlong *bytes) {
5284   jlong cur, end;
5285   int mode;
5286   struct stat64 buf64;
5287 
5288   if (::fstat64(fd, &buf64) >= 0) {
5289     mode = buf64.st_mode;
5290     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5291       int n;
5292       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5293         *bytes = n;
5294         return 1;
5295       }
5296     }
5297   }
5298   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5299     return 0;
5300   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5301     return 0;
5302   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5303     return 0;
5304   }
5305   *bytes = end - cur;
5306   return 1;
5307 }
5308 
5309 // Map a block of memory.
5310 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5311                         char *addr, size_t bytes, bool read_only,
5312                         bool allow_exec) {
5313   int prot;
5314   int flags = MAP_PRIVATE;
5315 
5316   if (read_only) {
5317     prot = PROT_READ;
5318   } else {
5319     prot = PROT_READ | PROT_WRITE;
5320   }
5321 
5322   if (allow_exec) {
5323     prot |= PROT_EXEC;
5324   }
5325 
5326   if (addr != NULL) {
5327     flags |= MAP_FIXED;
5328   }
5329 
5330   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5331                                      fd, file_offset);
5332   if (mapped_address == MAP_FAILED) {
5333     return NULL;
5334   }
5335   return mapped_address;
5336 }
5337 
5338 
5339 // Remap a block of memory.
5340 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5341                           char *addr, size_t bytes, bool read_only,
5342                           bool allow_exec) {
5343   // same as map_memory() on this OS
5344   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5345                         allow_exec);
5346 }
5347 
5348 
5349 // Unmap a block of memory.
5350 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5351   return munmap(addr, bytes) == 0;
5352 }
5353 
5354 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5355 
5356 static clockid_t thread_cpu_clockid(Thread* thread) {
5357   pthread_t tid = thread->osthread()->pthread_id();
5358   clockid_t clockid;
5359 
5360   // Get thread clockid
5361   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5362   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5363   return clockid;
5364 }
5365 
5366 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5367 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5368 // of a thread.
5369 //
5370 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5371 // the fast estimate available on the platform.
5372 
5373 jlong os::current_thread_cpu_time() {
5374   if (os::Linux::supports_fast_thread_cpu_time()) {
5375     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5376   } else {
5377     // return user + sys since the cost is the same
5378     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5379   }
5380 }
5381 
5382 jlong os::thread_cpu_time(Thread* thread) {
5383   // consistent with what current_thread_cpu_time() returns
5384   if (os::Linux::supports_fast_thread_cpu_time()) {
5385     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5386   } else {
5387     return slow_thread_cpu_time(thread, true /* user + sys */);
5388   }
5389 }
5390 
5391 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5392   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5393     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5394   } else {
5395     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5396   }
5397 }
5398 
5399 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5400   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5401     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5402   } else {
5403     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5404   }
5405 }
5406 
5407 //  -1 on error.
5408 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5409   pid_t  tid = thread->osthread()->thread_id();
5410   char *s;
5411   char stat[2048];
5412   int statlen;
5413   char proc_name[64];
5414   int count;
5415   long sys_time, user_time;
5416   char cdummy;
5417   int idummy;
5418   long ldummy;
5419   FILE *fp;
5420 
5421   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5422   fp = fopen(proc_name, "r");
5423   if (fp == NULL) return -1;
5424   statlen = fread(stat, 1, 2047, fp);
5425   stat[statlen] = '\0';
5426   fclose(fp);
5427 
5428   // Skip pid and the command string. Note that we could be dealing with
5429   // weird command names, e.g. user could decide to rename java launcher
5430   // to "java 1.4.2 :)", then the stat file would look like
5431   //                1234 (java 1.4.2 :)) R ... ...
5432   // We don't really need to know the command string, just find the last
5433   // occurrence of ")" and then start parsing from there. See bug 4726580.
5434   s = strrchr(stat, ')');
5435   if (s == NULL) return -1;
5436 
5437   // Skip blank chars
5438   do { s++; } while (s && isspace(*s));
5439 
5440   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5441                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5442                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5443                  &user_time, &sys_time);
5444   if (count != 13) return -1;
5445   if (user_sys_cpu_time) {
5446     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5447   } else {
5448     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5449   }
5450 }
5451 
5452 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5453   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5454   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5455   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5456   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5457 }
5458 
5459 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5460   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5461   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5462   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5463   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5464 }
5465 
5466 bool os::is_thread_cpu_time_supported() {
5467   return true;
5468 }
5469 
5470 // System loadavg support.  Returns -1 if load average cannot be obtained.
5471 // Linux doesn't yet have a (official) notion of processor sets,
5472 // so just return the system wide load average.
5473 int os::loadavg(double loadavg[], int nelem) {
5474   return ::getloadavg(loadavg, nelem);
5475 }
5476 
5477 void os::pause() {
5478   char filename[MAX_PATH];
5479   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5480     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5481   } else {
5482     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5483   }
5484 
5485   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5486   if (fd != -1) {
5487     struct stat buf;
5488     ::close(fd);
5489     while (::stat(filename, &buf) == 0) {
5490       (void)::poll(NULL, 0, 100);
5491     }
5492   } else {
5493     jio_fprintf(stderr,
5494                 "Could not open pause file '%s', continuing immediately.\n", filename);
5495   }
5496 }
5497 
5498 
5499 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5500 // comment paragraphs are worth repeating here:
5501 //
5502 // Assumption:
5503 //    Only one parker can exist on an event, which is why we allocate
5504 //    them per-thread. Multiple unparkers can coexist.
5505 //
5506 // _Event serves as a restricted-range semaphore.
5507 //   -1 : thread is blocked, i.e. there is a waiter
5508 //    0 : neutral: thread is running or ready,
5509 //        could have been signaled after a wait started
5510 //    1 : signaled - thread is running or ready
5511 //
5512 
5513 // utility to compute the abstime argument to timedwait:
5514 // millis is the relative timeout time
5515 // abstime will be the absolute timeout time
5516 // TODO: replace compute_abstime() with unpackTime()
5517 
5518 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5519   if (millis < 0)  millis = 0;
5520 
5521   jlong seconds = millis / 1000;
5522   millis %= 1000;
5523   if (seconds > 50000000) { // see man cond_timedwait(3T)
5524     seconds = 50000000;
5525   }
5526 
5527   if (os::supports_monotonic_clock()) {
5528     struct timespec now;
5529     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5530     assert_status(status == 0, status, "clock_gettime");
5531     abstime->tv_sec = now.tv_sec  + seconds;
5532     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5533     if (nanos >= NANOSECS_PER_SEC) {
5534       abstime->tv_sec += 1;
5535       nanos -= NANOSECS_PER_SEC;
5536     }
5537     abstime->tv_nsec = nanos;
5538   } else {
5539     struct timeval now;
5540     int status = gettimeofday(&now, NULL);
5541     assert(status == 0, "gettimeofday");
5542     abstime->tv_sec = now.tv_sec  + seconds;
5543     long usec = now.tv_usec + millis * 1000;
5544     if (usec >= 1000000) {
5545       abstime->tv_sec += 1;
5546       usec -= 1000000;
5547     }
5548     abstime->tv_nsec = usec * 1000;
5549   }
5550   return abstime;
5551 }
5552 
5553 void os::PlatformEvent::park() {       // AKA "down()"
5554   // Transitions for _Event:
5555   //   -1 => -1 : illegal
5556   //    1 =>  0 : pass - return immediately
5557   //    0 => -1 : block; then set _Event to 0 before returning
5558 
5559   // Invariant: Only the thread associated with the Event/PlatformEvent
5560   // may call park().
5561   // TODO: assert that _Assoc != NULL or _Assoc == Self
5562   assert(_nParked == 0, "invariant");
5563 
5564   int v;
5565   for (;;) {
5566     v = _Event;
5567     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5568   }
5569   guarantee(v >= 0, "invariant");
5570   if (v == 0) {
5571     // Do this the hard way by blocking ...
5572     int status = pthread_mutex_lock(_mutex);
5573     assert_status(status == 0, status, "mutex_lock");
5574     guarantee(_nParked == 0, "invariant");
5575     ++_nParked;
5576     while (_Event < 0) {
5577       status = pthread_cond_wait(_cond, _mutex);
5578       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5579       // Treat this the same as if the wait was interrupted
5580       if (status == ETIME) { status = EINTR; }
5581       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5582     }
5583     --_nParked;
5584 
5585     _Event = 0;
5586     status = pthread_mutex_unlock(_mutex);
5587     assert_status(status == 0, status, "mutex_unlock");
5588     // Paranoia to ensure our locked and lock-free paths interact
5589     // correctly with each other.
5590     OrderAccess::fence();
5591   }
5592   guarantee(_Event >= 0, "invariant");
5593 }
5594 
5595 int os::PlatformEvent::park(jlong millis) {
5596   // Transitions for _Event:
5597   //   -1 => -1 : illegal
5598   //    1 =>  0 : pass - return immediately
5599   //    0 => -1 : block; then set _Event to 0 before returning
5600 
5601   guarantee(_nParked == 0, "invariant");
5602 
5603   int v;
5604   for (;;) {
5605     v = _Event;
5606     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5607   }
5608   guarantee(v >= 0, "invariant");
5609   if (v != 0) return OS_OK;
5610 
5611   // We do this the hard way, by blocking the thread.
5612   // Consider enforcing a minimum timeout value.
5613   struct timespec abst;
5614   compute_abstime(&abst, millis);
5615 
5616   int ret = OS_TIMEOUT;
5617   int status = pthread_mutex_lock(_mutex);
5618   assert_status(status == 0, status, "mutex_lock");
5619   guarantee(_nParked == 0, "invariant");
5620   ++_nParked;
5621 
5622   // Object.wait(timo) will return because of
5623   // (a) notification
5624   // (b) timeout
5625   // (c) thread.interrupt
5626   //
5627   // Thread.interrupt and object.notify{All} both call Event::set.
5628   // That is, we treat thread.interrupt as a special case of notification.
5629   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5630   // We assume all ETIME returns are valid.
5631   //
5632   // TODO: properly differentiate simultaneous notify+interrupt.
5633   // In that case, we should propagate the notify to another waiter.
5634 
5635   while (_Event < 0) {
5636     status = pthread_cond_timedwait(_cond, _mutex, &abst);
5637     assert_status(status == 0 || status == EINTR ||
5638                   status == ETIME || status == ETIMEDOUT,
5639                   status, "cond_timedwait");
5640     if (!FilterSpuriousWakeups) break;                 // previous semantics
5641     if (status == ETIME || status == ETIMEDOUT) break;
5642     // We consume and ignore EINTR and spurious wakeups.
5643   }
5644   --_nParked;
5645   if (_Event >= 0) {
5646     ret = OS_OK;
5647   }
5648   _Event = 0;
5649   status = pthread_mutex_unlock(_mutex);
5650   assert_status(status == 0, status, "mutex_unlock");
5651   assert(_nParked == 0, "invariant");
5652   // Paranoia to ensure our locked and lock-free paths interact
5653   // correctly with each other.
5654   OrderAccess::fence();
5655   return ret;
5656 }
5657 
5658 void os::PlatformEvent::unpark() {
5659   // Transitions for _Event:
5660   //    0 => 1 : just return
5661   //    1 => 1 : just return
5662   //   -1 => either 0 or 1; must signal target thread
5663   //         That is, we can safely transition _Event from -1 to either
5664   //         0 or 1.
5665   // See also: "Semaphores in Plan 9" by Mullender & Cox
5666   //
5667   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5668   // that it will take two back-to-back park() calls for the owning
5669   // thread to block. This has the benefit of forcing a spurious return
5670   // from the first park() call after an unpark() call which will help
5671   // shake out uses of park() and unpark() without condition variables.
5672 
5673   if (Atomic::xchg(1, &_Event) >= 0) return;
5674 
5675   // Wait for the thread associated with the event to vacate
5676   int status = pthread_mutex_lock(_mutex);
5677   assert_status(status == 0, status, "mutex_lock");
5678   int AnyWaiters = _nParked;
5679   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5680   status = pthread_mutex_unlock(_mutex);
5681   assert_status(status == 0, status, "mutex_unlock");
5682   if (AnyWaiters != 0) {
5683     // Note that we signal() *after* dropping the lock for "immortal" Events.
5684     // This is safe and avoids a common class of  futile wakeups.  In rare
5685     // circumstances this can cause a thread to return prematurely from
5686     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5687     // will simply re-test the condition and re-park itself.
5688     // This provides particular benefit if the underlying platform does not
5689     // provide wait morphing.
5690     status = pthread_cond_signal(_cond);
5691     assert_status(status == 0, status, "cond_signal");
5692   }
5693 }
5694 
5695 
5696 // JSR166
5697 // -------------------------------------------------------
5698 
5699 // The solaris and linux implementations of park/unpark are fairly
5700 // conservative for now, but can be improved. They currently use a
5701 // mutex/condvar pair, plus a a count.
5702 // Park decrements count if > 0, else does a condvar wait.  Unpark
5703 // sets count to 1 and signals condvar.  Only one thread ever waits
5704 // on the condvar. Contention seen when trying to park implies that someone
5705 // is unparking you, so don't wait. And spurious returns are fine, so there
5706 // is no need to track notifications.
5707 
5708 // This code is common to linux and solaris and will be moved to a
5709 // common place in dolphin.
5710 //
5711 // The passed in time value is either a relative time in nanoseconds
5712 // or an absolute time in milliseconds. Either way it has to be unpacked
5713 // into suitable seconds and nanoseconds components and stored in the
5714 // given timespec structure.
5715 // Given time is a 64-bit value and the time_t used in the timespec is only
5716 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5717 // overflow if times way in the future are given. Further on Solaris versions
5718 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5719 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5720 // As it will be 28 years before "now + 100000000" will overflow we can
5721 // ignore overflow and just impose a hard-limit on seconds using the value
5722 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5723 // years from "now".
5724 
5725 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5726   assert(time > 0, "convertTime");
5727   time_t max_secs = 0;
5728 
5729   if (!os::supports_monotonic_clock() || isAbsolute) {
5730     struct timeval now;
5731     int status = gettimeofday(&now, NULL);
5732     assert(status == 0, "gettimeofday");
5733 
5734     max_secs = now.tv_sec + MAX_SECS;
5735 
5736     if (isAbsolute) {
5737       jlong secs = time / 1000;
5738       if (secs > max_secs) {
5739         absTime->tv_sec = max_secs;
5740       } else {
5741         absTime->tv_sec = secs;
5742       }
5743       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5744     } else {
5745       jlong secs = time / NANOSECS_PER_SEC;
5746       if (secs >= MAX_SECS) {
5747         absTime->tv_sec = max_secs;
5748         absTime->tv_nsec = 0;
5749       } else {
5750         absTime->tv_sec = now.tv_sec + secs;
5751         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5752         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5753           absTime->tv_nsec -= NANOSECS_PER_SEC;
5754           ++absTime->tv_sec; // note: this must be <= max_secs
5755         }
5756       }
5757     }
5758   } else {
5759     // must be relative using monotonic clock
5760     struct timespec now;
5761     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5762     assert_status(status == 0, status, "clock_gettime");
5763     max_secs = now.tv_sec + MAX_SECS;
5764     jlong secs = time / NANOSECS_PER_SEC;
5765     if (secs >= MAX_SECS) {
5766       absTime->tv_sec = max_secs;
5767       absTime->tv_nsec = 0;
5768     } else {
5769       absTime->tv_sec = now.tv_sec + secs;
5770       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5771       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5772         absTime->tv_nsec -= NANOSECS_PER_SEC;
5773         ++absTime->tv_sec; // note: this must be <= max_secs
5774       }
5775     }
5776   }
5777   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5778   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5779   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5780   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5781 }
5782 
5783 void Parker::park(bool isAbsolute, jlong time) {
5784   // Ideally we'd do something useful while spinning, such
5785   // as calling unpackTime().
5786 
5787   // Optional fast-path check:
5788   // Return immediately if a permit is available.
5789   // We depend on Atomic::xchg() having full barrier semantics
5790   // since we are doing a lock-free update to _counter.
5791   if (Atomic::xchg(0, &_counter) > 0) return;
5792 
5793   Thread* thread = Thread::current();
5794   assert(thread->is_Java_thread(), "Must be JavaThread");
5795   JavaThread *jt = (JavaThread *)thread;
5796 
5797   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5798   // Check interrupt before trying to wait
5799   if (Thread::is_interrupted(thread, false)) {
5800     return;
5801   }
5802 
5803   // Next, demultiplex/decode time arguments
5804   timespec absTime;
5805   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5806     return;
5807   }
5808   if (time > 0) {
5809     unpackTime(&absTime, isAbsolute, time);
5810   }
5811 
5812 
5813   // Enter safepoint region
5814   // Beware of deadlocks such as 6317397.
5815   // The per-thread Parker:: mutex is a classic leaf-lock.
5816   // In particular a thread must never block on the Threads_lock while
5817   // holding the Parker:: mutex.  If safepoints are pending both the
5818   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5819   ThreadBlockInVM tbivm(jt);
5820 
5821   // Don't wait if cannot get lock since interference arises from
5822   // unblocking.  Also. check interrupt before trying wait
5823   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5824     return;
5825   }
5826 
5827   int status;
5828   if (_counter > 0)  { // no wait needed
5829     _counter = 0;
5830     status = pthread_mutex_unlock(_mutex);
5831     assert_status(status == 0, status, "invariant");
5832     // Paranoia to ensure our locked and lock-free paths interact
5833     // correctly with each other and Java-level accesses.
5834     OrderAccess::fence();
5835     return;
5836   }
5837 
5838 #ifdef ASSERT
5839   // Don't catch signals while blocked; let the running threads have the signals.
5840   // (This allows a debugger to break into the running thread.)
5841   sigset_t oldsigs;
5842   sigemptyset(&oldsigs);
5843   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5844   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5845 #endif
5846 
5847   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5848   jt->set_suspend_equivalent();
5849   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5850 
5851   assert(_cur_index == -1, "invariant");
5852   if (time == 0) {
5853     _cur_index = REL_INDEX; // arbitrary choice when not timed
5854     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5855   } else {
5856     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5857     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5858   }
5859   _cur_index = -1;
5860   assert_status(status == 0 || status == EINTR ||
5861                 status == ETIME || status == ETIMEDOUT,
5862                 status, "cond_timedwait");
5863 
5864 #ifdef ASSERT
5865   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5866 #endif
5867 
5868   _counter = 0;
5869   status = pthread_mutex_unlock(_mutex);
5870   assert_status(status == 0, status, "invariant");
5871   // Paranoia to ensure our locked and lock-free paths interact
5872   // correctly with each other and Java-level accesses.
5873   OrderAccess::fence();
5874 
5875   // If externally suspended while waiting, re-suspend
5876   if (jt->handle_special_suspend_equivalent_condition()) {
5877     jt->java_suspend_self();
5878   }
5879 }
5880 
5881 void Parker::unpark() {
5882   int status = pthread_mutex_lock(_mutex);
5883   assert_status(status == 0, status, "invariant");
5884   const int s = _counter;
5885   _counter = 1;
5886   // must capture correct index before unlocking
5887   int index = _cur_index;
5888   status = pthread_mutex_unlock(_mutex);
5889   assert_status(status == 0, status, "invariant");
5890   if (s < 1 && index != -1) {
5891     // thread is definitely parked
5892     status = pthread_cond_signal(&_cond[index]);
5893     assert_status(status == 0, status, "invariant");
5894   }
5895 }
5896 
5897 
5898 extern char** environ;
5899 
5900 // Run the specified command in a separate process. Return its exit value,
5901 // or -1 on failure (e.g. can't fork a new process).
5902 // Unlike system(), this function can be called from signal handler. It
5903 // doesn't block SIGINT et al.
5904 int os::fork_and_exec(char* cmd) {
5905   const char * argv[4] = {"sh", "-c", cmd, NULL};
5906 
5907   pid_t pid = fork();
5908 
5909   if (pid < 0) {
5910     // fork failed
5911     return -1;
5912 
5913   } else if (pid == 0) {
5914     // child process
5915 
5916     execve("/bin/sh", (char* const*)argv, environ);
5917 
5918     // execve failed
5919     _exit(-1);
5920 
5921   } else  {
5922     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5923     // care about the actual exit code, for now.
5924 
5925     int status;
5926 
5927     // Wait for the child process to exit.  This returns immediately if
5928     // the child has already exited. */
5929     while (waitpid(pid, &status, 0) < 0) {
5930       switch (errno) {
5931       case ECHILD: return 0;
5932       case EINTR: break;
5933       default: return -1;
5934       }
5935     }
5936 
5937     if (WIFEXITED(status)) {
5938       // The child exited normally; get its exit code.
5939       return WEXITSTATUS(status);
5940     } else if (WIFSIGNALED(status)) {
5941       // The child exited because of a signal
5942       // The best value to return is 0x80 + signal number,
5943       // because that is what all Unix shells do, and because
5944       // it allows callers to distinguish between process exit and
5945       // process death by signal.
5946       return 0x80 + WTERMSIG(status);
5947     } else {
5948       // Unknown exit code; pass it through
5949       return status;
5950     }
5951   }
5952 }
5953 
5954 // is_headless_jre()
5955 //
5956 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5957 // in order to report if we are running in a headless jre
5958 //
5959 // Since JDK8 xawt/libmawt.so was moved into the same directory
5960 // as libawt.so, and renamed libawt_xawt.so
5961 //
5962 bool os::is_headless_jre() {
5963   struct stat statbuf;
5964   char buf[MAXPATHLEN];
5965   char libmawtpath[MAXPATHLEN];
5966   const char *xawtstr  = "/xawt/libmawt.so";
5967   const char *new_xawtstr = "/libawt_xawt.so";
5968   char *p;
5969 
5970   // Get path to libjvm.so
5971   os::jvm_path(buf, sizeof(buf));
5972 
5973   // Get rid of libjvm.so
5974   p = strrchr(buf, '/');
5975   if (p == NULL) {
5976     return false;
5977   } else {
5978     *p = '\0';
5979   }
5980 
5981   // Get rid of client or server
5982   p = strrchr(buf, '/');
5983   if (p == NULL) {
5984     return false;
5985   } else {
5986     *p = '\0';
5987   }
5988 
5989   // check xawt/libmawt.so
5990   strcpy(libmawtpath, buf);
5991   strcat(libmawtpath, xawtstr);
5992   if (::stat(libmawtpath, &statbuf) == 0) return false;
5993 
5994   // check libawt_xawt.so
5995   strcpy(libmawtpath, buf);
5996   strcat(libmawtpath, new_xawtstr);
5997   if (::stat(libmawtpath, &statbuf) == 0) return false;
5998 
5999   return true;
6000 }
6001 
6002 // Get the default path to the core file
6003 // Returns the length of the string
6004 int os::get_core_path(char* buffer, size_t bufferSize) {
6005   /*
6006    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
6007    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
6008    */
6009   const int core_pattern_len = 129;
6010   char core_pattern[core_pattern_len] = {0};
6011 
6012   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
6013   if (core_pattern_file == -1) {
6014     return -1;
6015   }
6016 
6017   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6018   ::close(core_pattern_file);
6019   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
6020     return -1;
6021   }
6022   if (core_pattern[ret-1] == '\n') {
6023     core_pattern[ret-1] = '\0';
6024   } else {
6025     core_pattern[ret] = '\0';
6026   }
6027 
6028   char *pid_pos = strstr(core_pattern, "%p");
6029   int written;
6030 
6031   if (core_pattern[0] == '/') {
6032     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6033   } else {
6034     char cwd[PATH_MAX];
6035 
6036     const char* p = get_current_directory(cwd, PATH_MAX);
6037     if (p == NULL) {
6038       return -1;
6039     }
6040 
6041     if (core_pattern[0] == '|') {
6042       written = jio_snprintf(buffer, bufferSize,
6043                              "\"%s\" (or dumping to %s/core.%d)",
6044                              &core_pattern[1], p, current_process_id());
6045     } else {
6046       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6047     }
6048   }
6049 
6050   if (written < 0) {
6051     return -1;
6052   }
6053 
6054   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6055     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6056 
6057     if (core_uses_pid_file != -1) {
6058       char core_uses_pid = 0;
6059       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6060       ::close(core_uses_pid_file);
6061 
6062       if (core_uses_pid == '1') {
6063         jio_snprintf(buffer + written, bufferSize - written,
6064                                           ".%d", current_process_id());
6065       }
6066     }
6067   }
6068 
6069   return strlen(buffer);
6070 }
6071 
6072 bool os::start_debugging(char *buf, int buflen) {
6073   int len = (int)strlen(buf);
6074   char *p = &buf[len];
6075 
6076   jio_snprintf(p, buflen-len,
6077                "\n\n"
6078                "Do you want to debug the problem?\n\n"
6079                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6080                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6081                "Otherwise, press RETURN to abort...",
6082                os::current_process_id(), os::current_process_id(),
6083                os::current_thread_id(), os::current_thread_id());
6084 
6085   bool yes = os::message_box("Unexpected Error", buf);
6086 
6087   if (yes) {
6088     // yes, user asked VM to launch debugger
6089     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6090                  os::current_process_id(), os::current_process_id());
6091 
6092     os::fork_and_exec(buf);
6093     yes = false;
6094   }
6095   return yes;
6096 }
6097 
6098 
6099 // Java/Compiler thread:
6100 //
6101 //   Low memory addresses
6102 // P0 +------------------------+
6103 //    |                        |\  Java thread created by VM does not have glibc
6104 //    |    glibc guard page    | - guard page, attached Java thread usually has
6105 //    |                        |/  1 glibc guard page.
6106 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6107 //    |                        |\
6108 //    |  HotSpot Guard Pages   | - red, yellow and reserved pages
6109 //    |                        |/
6110 //    +------------------------+ JavaThread::stack_reserved_zone_base()
6111 //    |                        |\
6112 //    |      Normal Stack      | -
6113 //    |                        |/
6114 // P2 +------------------------+ Thread::stack_base()
6115 //
6116 // Non-Java thread:
6117 //
6118 //   Low memory addresses
6119 // P0 +------------------------+
6120 //    |                        |\
6121 //    |  glibc guard page      | - usually 1 page
6122 //    |                        |/
6123 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6124 //    |                        |\
6125 //    |      Normal Stack      | -
6126 //    |                        |/
6127 // P2 +------------------------+ Thread::stack_base()
6128 //
6129 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6130 //    returned from pthread_attr_getstack().
6131 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6132 //    of the stack size given in pthread_attr. We work around this for
6133 //    threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6134 //
6135 #ifndef ZERO
6136 static void current_stack_region(address * bottom, size_t * size) {
6137   if (os::Linux::is_initial_thread()) {
6138     // initial thread needs special handling because pthread_getattr_np()
6139     // may return bogus value.
6140     *bottom = os::Linux::initial_thread_stack_bottom();
6141     *size   = os::Linux::initial_thread_stack_size();
6142   } else {
6143     pthread_attr_t attr;
6144 
6145     int rslt = pthread_getattr_np(pthread_self(), &attr);
6146 
6147     // JVM needs to know exact stack location, abort if it fails
6148     if (rslt != 0) {
6149       if (rslt == ENOMEM) {
6150         vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6151       } else {
6152         fatal("pthread_getattr_np failed with error = %d", rslt);
6153       }
6154     }
6155 
6156     if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6157       fatal("Cannot locate current stack attributes!");
6158     }
6159 
6160     // Work around NPTL stack guard error.
6161     size_t guard_size = 0;
6162     rslt = pthread_attr_getguardsize(&attr, &guard_size);
6163     if (rslt != 0) {
6164       fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6165     }
6166     *bottom += guard_size;
6167     *size   -= guard_size;
6168 
6169     pthread_attr_destroy(&attr);
6170 
6171   }
6172   assert(os::current_stack_pointer() >= *bottom &&
6173          os::current_stack_pointer() < *bottom + *size, "just checking");
6174 }
6175 
6176 address os::current_stack_base() {
6177   address bottom;
6178   size_t size;
6179   current_stack_region(&bottom, &size);
6180   return (bottom + size);
6181 }
6182 
6183 size_t os::current_stack_size() {
6184   // This stack size includes the usable stack and HotSpot guard pages
6185   // (for the threads that have Hotspot guard pages).
6186   address bottom;
6187   size_t size;
6188   current_stack_region(&bottom, &size);
6189   return size;
6190 }
6191 #endif
6192 
6193 static inline struct timespec get_mtime(const char* filename) {
6194   struct stat st;
6195   int ret = os::stat(filename, &st);
6196   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
6197   return st.st_mtim;
6198 }
6199 
6200 int os::compare_file_modified_times(const char* file1, const char* file2) {
6201   struct timespec filetime1 = get_mtime(file1);
6202   struct timespec filetime2 = get_mtime(file2);
6203   int diff = filetime1.tv_sec - filetime2.tv_sec;
6204   if (diff == 0) {
6205     return filetime1.tv_nsec - filetime2.tv_nsec;
6206   }
6207   return diff;
6208 }
6209 
6210 /////////////// Unit tests ///////////////
6211 
6212 #ifndef PRODUCT
6213 
6214 #define test_log(...)              \
6215   do {                             \
6216     if (VerboseInternalVMTests) {  \
6217       tty->print_cr(__VA_ARGS__);  \
6218       tty->flush();                \
6219     }                              \
6220   } while (false)
6221 
6222 class TestReserveMemorySpecial : AllStatic {
6223  public:
6224   static void small_page_write(void* addr, size_t size) {
6225     size_t page_size = os::vm_page_size();
6226 
6227     char* end = (char*)addr + size;
6228     for (char* p = (char*)addr; p < end; p += page_size) {
6229       *p = 1;
6230     }
6231   }
6232 
6233   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6234     if (!UseHugeTLBFS) {
6235       return;
6236     }
6237 
6238     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6239 
6240     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6241 
6242     if (addr != NULL) {
6243       small_page_write(addr, size);
6244 
6245       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6246     }
6247   }
6248 
6249   static void test_reserve_memory_special_huge_tlbfs_only() {
6250     if (!UseHugeTLBFS) {
6251       return;
6252     }
6253 
6254     size_t lp = os::large_page_size();
6255 
6256     for (size_t size = lp; size <= lp * 10; size += lp) {
6257       test_reserve_memory_special_huge_tlbfs_only(size);
6258     }
6259   }
6260 
6261   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6262     size_t lp = os::large_page_size();
6263     size_t ag = os::vm_allocation_granularity();
6264 
6265     // sizes to test
6266     const size_t sizes[] = {
6267       lp, lp + ag, lp + lp / 2, lp * 2,
6268       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6269       lp * 10, lp * 10 + lp / 2
6270     };
6271     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6272 
6273     // For each size/alignment combination, we test three scenarios:
6274     // 1) with req_addr == NULL
6275     // 2) with a non-null req_addr at which we expect to successfully allocate
6276     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6277     //    expect the allocation to either fail or to ignore req_addr
6278 
6279     // Pre-allocate two areas; they shall be as large as the largest allocation
6280     //  and aligned to the largest alignment we will be testing.
6281     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6282     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6283       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6284       -1, 0);
6285     assert(mapping1 != MAP_FAILED, "should work");
6286 
6287     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6288       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6289       -1, 0);
6290     assert(mapping2 != MAP_FAILED, "should work");
6291 
6292     // Unmap the first mapping, but leave the second mapping intact: the first
6293     // mapping will serve as a value for a "good" req_addr (case 2). The second
6294     // mapping, still intact, as "bad" req_addr (case 3).
6295     ::munmap(mapping1, mapping_size);
6296 
6297     // Case 1
6298     test_log("%s, req_addr NULL:", __FUNCTION__);
6299     test_log("size            align           result");
6300 
6301     for (int i = 0; i < num_sizes; i++) {
6302       const size_t size = sizes[i];
6303       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6304         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6305         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6306                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6307         if (p != NULL) {
6308           assert(is_ptr_aligned(p, alignment), "must be");
6309           small_page_write(p, size);
6310           os::Linux::release_memory_special_huge_tlbfs(p, size);
6311         }
6312       }
6313     }
6314 
6315     // Case 2
6316     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6317     test_log("size            align           req_addr         result");
6318 
6319     for (int i = 0; i < num_sizes; i++) {
6320       const size_t size = sizes[i];
6321       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6322         char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6323         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6324         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6325                  size, alignment, p2i(req_addr), p2i(p),
6326                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6327         if (p != NULL) {
6328           assert(p == req_addr, "must be");
6329           small_page_write(p, size);
6330           os::Linux::release_memory_special_huge_tlbfs(p, size);
6331         }
6332       }
6333     }
6334 
6335     // Case 3
6336     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6337     test_log("size            align           req_addr         result");
6338 
6339     for (int i = 0; i < num_sizes; i++) {
6340       const size_t size = sizes[i];
6341       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6342         char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6343         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6344         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6345                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6346         // as the area around req_addr contains already existing mappings, the API should always
6347         // return NULL (as per contract, it cannot return another address)
6348         assert(p == NULL, "must be");
6349       }
6350     }
6351 
6352     ::munmap(mapping2, mapping_size);
6353 
6354   }
6355 
6356   static void test_reserve_memory_special_huge_tlbfs() {
6357     if (!UseHugeTLBFS) {
6358       return;
6359     }
6360 
6361     test_reserve_memory_special_huge_tlbfs_only();
6362     test_reserve_memory_special_huge_tlbfs_mixed();
6363   }
6364 
6365   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6366     if (!UseSHM) {
6367       return;
6368     }
6369 
6370     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6371 
6372     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6373 
6374     if (addr != NULL) {
6375       assert(is_ptr_aligned(addr, alignment), "Check");
6376       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6377 
6378       small_page_write(addr, size);
6379 
6380       os::Linux::release_memory_special_shm(addr, size);
6381     }
6382   }
6383 
6384   static void test_reserve_memory_special_shm() {
6385     size_t lp = os::large_page_size();
6386     size_t ag = os::vm_allocation_granularity();
6387 
6388     for (size_t size = ag; size < lp * 3; size += ag) {
6389       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6390         test_reserve_memory_special_shm(size, alignment);
6391       }
6392     }
6393   }
6394 
6395   static void test() {
6396     test_reserve_memory_special_huge_tlbfs();
6397     test_reserve_memory_special_shm();
6398   }
6399 };
6400 
6401 void TestReserveMemorySpecial_test() {
6402   TestReserveMemorySpecial::test();
6403 }
6404 
6405 #endif