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 // Cpu architecture string
 222 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
 223 
 224 
 225 // pid_t gettid()
 226 //
 227 // Returns the kernel thread id of the currently running thread. Kernel
 228 // thread id is used to access /proc.
 229 pid_t os::Linux::gettid() {
 230   int rslt = syscall(SYS_gettid);
 231   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 232   return (pid_t)rslt;
 233 }
 234 
 235 // Most versions of linux have a bug where the number of processors are
 236 // determined by looking at the /proc file system.  In a chroot environment,
 237 // the system call returns 1.  This causes the VM to act as if it is
 238 // a single processor and elide locking (see is_MP() call).
 239 static bool unsafe_chroot_detected = false;
 240 static const char *unstable_chroot_error = "/proc file system not found.\n"
 241                      "Java may be unstable running multithreaded in a chroot "
 242                      "environment on Linux when /proc filesystem is not mounted.";
 243 
 244 void os::Linux::initialize_system_info() {
 245   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 246   if (processor_count() == 1) {
 247     pid_t pid = os::Linux::gettid();
 248     char fname[32];
 249     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 250     FILE *fp = fopen(fname, "r");
 251     if (fp == NULL) {
 252       unsafe_chroot_detected = true;
 253     } else {
 254       fclose(fp);
 255     }
 256   }
 257   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 258   assert(processor_count() > 0, "linux error");
 259 }
 260 
 261 void os::init_system_properties_values() {
 262   // The next steps are taken in the product version:
 263   //
 264   // Obtain the JAVA_HOME value from the location of libjvm.so.
 265   // This library should be located at:
 266   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 267   //
 268   // If "/jre/lib/" appears at the right place in the path, then we
 269   // assume libjvm.so is installed in a JDK and we use this path.
 270   //
 271   // Otherwise exit with message: "Could not create the Java virtual machine."
 272   //
 273   // The following extra steps are taken in the debugging version:
 274   //
 275   // If "/jre/lib/" does NOT appear at the right place in the path
 276   // instead of exit check for $JAVA_HOME environment variable.
 277   //
 278   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 279   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 280   // it looks like libjvm.so is installed there
 281   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 282   //
 283   // Otherwise exit.
 284   //
 285   // Important note: if the location of libjvm.so changes this
 286   // code needs to be changed accordingly.
 287 
 288   // See ld(1):
 289   //      The linker uses the following search paths to locate required
 290   //      shared libraries:
 291   //        1: ...
 292   //        ...
 293   //        7: The default directories, normally /lib and /usr/lib.
 294 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 295   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 296 #else
 297   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 298 #endif
 299 
 300 // Base path of extensions installed on the system.
 301 #define SYS_EXT_DIR     "/usr/java/packages"
 302 #define EXTENSIONS_DIR  "/lib/ext"
 303 
 304   // Buffer that fits several sprintfs.
 305   // Note that the space for the colon and the trailing null are provided
 306   // by the nulls included by the sizeof operator.
 307   const size_t bufsize =
 308     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 309          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 310   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 311 
 312   // sysclasspath, java_home, dll_dir
 313   {
 314     char *pslash;
 315     os::jvm_path(buf, bufsize);
 316 
 317     // Found the full path to libjvm.so.
 318     // Now cut the path to <java_home>/jre if we can.
 319     pslash = strrchr(buf, '/');
 320     if (pslash != NULL) {
 321       *pslash = '\0';            // Get rid of /libjvm.so.
 322     }
 323     pslash = strrchr(buf, '/');
 324     if (pslash != NULL) {
 325       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 326     }
 327     Arguments::set_dll_dir(buf);
 328 
 329     if (pslash != NULL) {
 330       pslash = strrchr(buf, '/');
 331       if (pslash != NULL) {
 332         *pslash = '\0';          // Get rid of /<arch>.
 333         pslash = strrchr(buf, '/');
 334         if (pslash != NULL) {
 335           *pslash = '\0';        // Get rid of /lib.
 336         }
 337       }
 338     }
 339     Arguments::set_java_home(buf);
 340     set_boot_path('/', ':');
 341   }
 342 
 343   // Where to look for native libraries.
 344   //
 345   // Note: Due to a legacy implementation, most of the library path
 346   // is set in the launcher. This was to accomodate linking restrictions
 347   // on legacy Linux implementations (which are no longer supported).
 348   // Eventually, all the library path setting will be done here.
 349   //
 350   // However, to prevent the proliferation of improperly built native
 351   // libraries, the new path component /usr/java/packages is added here.
 352   // Eventually, all the library path setting will be done here.
 353   {
 354     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 355     // should always exist (until the legacy problem cited above is
 356     // addressed).
 357     const char *v = ::getenv("LD_LIBRARY_PATH");
 358     const char *v_colon = ":";
 359     if (v == NULL) { v = ""; v_colon = ""; }
 360     // That's +1 for the colon and +1 for the trailing '\0'.
 361     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 362                                                      strlen(v) + 1 +
 363                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 364                                                      mtInternal);
 365     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 366     Arguments::set_library_path(ld_library_path);
 367     FREE_C_HEAP_ARRAY(char, ld_library_path);
 368   }
 369 
 370   // Extensions directories.
 371   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 372   Arguments::set_ext_dirs(buf);
 373 
 374   FREE_C_HEAP_ARRAY(char, buf);
 375 
 376 #undef DEFAULT_LIBPATH
 377 #undef SYS_EXT_DIR
 378 #undef EXTENSIONS_DIR
 379 }
 380 
 381 ////////////////////////////////////////////////////////////////////////////////
 382 // breakpoint support
 383 
 384 void os::breakpoint() {
 385   BREAKPOINT;
 386 }
 387 
 388 extern "C" void breakpoint() {
 389   // use debugger to set breakpoint here
 390 }
 391 
 392 ////////////////////////////////////////////////////////////////////////////////
 393 // signal support
 394 
 395 debug_only(static bool signal_sets_initialized = false);
 396 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 397 
 398 bool os::Linux::is_sig_ignored(int sig) {
 399   struct sigaction oact;
 400   sigaction(sig, (struct sigaction*)NULL, &oact);
 401   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 402                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 403   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 404     return true;
 405   } else {
 406     return false;
 407   }
 408 }
 409 
 410 void os::Linux::signal_sets_init() {
 411   // Should also have an assertion stating we are still single-threaded.
 412   assert(!signal_sets_initialized, "Already initialized");
 413   // Fill in signals that are necessarily unblocked for all threads in
 414   // the VM. Currently, we unblock the following signals:
 415   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 416   //                         by -Xrs (=ReduceSignalUsage));
 417   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 418   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 419   // the dispositions or masks wrt these signals.
 420   // Programs embedding the VM that want to use the above signals for their
 421   // own purposes must, at this time, use the "-Xrs" option to prevent
 422   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 423   // (See bug 4345157, and other related bugs).
 424   // In reality, though, unblocking these signals is really a nop, since
 425   // these signals are not blocked by default.
 426   sigemptyset(&unblocked_sigs);
 427   sigemptyset(&allowdebug_blocked_sigs);
 428   sigaddset(&unblocked_sigs, SIGILL);
 429   sigaddset(&unblocked_sigs, SIGSEGV);
 430   sigaddset(&unblocked_sigs, SIGBUS);
 431   sigaddset(&unblocked_sigs, SIGFPE);
 432 #if defined(PPC64)
 433   sigaddset(&unblocked_sigs, SIGTRAP);
 434 #endif
 435   sigaddset(&unblocked_sigs, SR_signum);
 436 
 437   if (!ReduceSignalUsage) {
 438     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 439       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 440       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 441     }
 442     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 443       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 444       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 445     }
 446     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 447       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 448       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 449     }
 450   }
 451   // Fill in signals that are blocked by all but the VM thread.
 452   sigemptyset(&vm_sigs);
 453   if (!ReduceSignalUsage) {
 454     sigaddset(&vm_sigs, BREAK_SIGNAL);
 455   }
 456   debug_only(signal_sets_initialized = true);
 457 
 458 }
 459 
 460 // These are signals that are unblocked while a thread is running Java.
 461 // (For some reason, they get blocked by default.)
 462 sigset_t* os::Linux::unblocked_signals() {
 463   assert(signal_sets_initialized, "Not initialized");
 464   return &unblocked_sigs;
 465 }
 466 
 467 // These are the signals that are blocked while a (non-VM) thread is
 468 // running Java. Only the VM thread handles these signals.
 469 sigset_t* os::Linux::vm_signals() {
 470   assert(signal_sets_initialized, "Not initialized");
 471   return &vm_sigs;
 472 }
 473 
 474 // These are signals that are blocked during cond_wait to allow debugger in
 475 sigset_t* os::Linux::allowdebug_blocked_signals() {
 476   assert(signal_sets_initialized, "Not initialized");
 477   return &allowdebug_blocked_sigs;
 478 }
 479 
 480 void os::Linux::hotspot_sigmask(Thread* thread) {
 481 
 482   //Save caller's signal mask before setting VM signal mask
 483   sigset_t caller_sigmask;
 484   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 485 
 486   OSThread* osthread = thread->osthread();
 487   osthread->set_caller_sigmask(caller_sigmask);
 488 
 489   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 490 
 491   if (!ReduceSignalUsage) {
 492     if (thread->is_VM_thread()) {
 493       // Only the VM thread handles BREAK_SIGNAL ...
 494       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 495     } else {
 496       // ... all other threads block BREAK_SIGNAL
 497       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 498     }
 499   }
 500 }
 501 
 502 //////////////////////////////////////////////////////////////////////////////
 503 // detecting pthread library
 504 
 505 void os::Linux::libpthread_init() {
 506   // Save glibc and pthread version strings.
 507 #if !defined(_CS_GNU_LIBC_VERSION) || \
 508     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 509   #error "glibc too old (< 2.3.2)"
 510 #endif
 511 
 512   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 513   assert(n > 0, "cannot retrieve glibc version");
 514   char *str = (char *)malloc(n, mtInternal);
 515   confstr(_CS_GNU_LIBC_VERSION, str, n);
 516   os::Linux::set_glibc_version(str);
 517 
 518   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 519   assert(n > 0, "cannot retrieve pthread version");
 520   str = (char *)malloc(n, mtInternal);
 521   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 522   os::Linux::set_libpthread_version(str);
 523 }
 524 
 525 /////////////////////////////////////////////////////////////////////////////
 526 // thread stack expansion
 527 
 528 // os::Linux::manually_expand_stack() takes care of expanding the thread
 529 // stack. Note that this is normally not needed: pthread stacks allocate
 530 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 531 // committed. Therefore it is not necessary to expand the stack manually.
 532 //
 533 // Manually expanding the stack was historically needed on LinuxThreads
 534 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 535 // it is kept to deal with very rare corner cases:
 536 //
 537 // For one, user may run the VM on an own implementation of threads
 538 // whose stacks are - like the old LinuxThreads - implemented using
 539 // mmap(MAP_GROWSDOWN).
 540 //
 541 // Also, this coding may be needed if the VM is running on the primordial
 542 // thread. Normally we avoid running on the primordial thread; however,
 543 // user may still invoke the VM on the primordial thread.
 544 //
 545 // The following historical comment describes the details about running
 546 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 547 
 548 
 549 // Force Linux kernel to expand current thread stack. If "bottom" is close
 550 // to the stack guard, caller should block all signals.
 551 //
 552 // MAP_GROWSDOWN:
 553 //   A special mmap() flag that is used to implement thread stacks. It tells
 554 //   kernel that the memory region should extend downwards when needed. This
 555 //   allows early versions of LinuxThreads to only mmap the first few pages
 556 //   when creating a new thread. Linux kernel will automatically expand thread
 557 //   stack as needed (on page faults).
 558 //
 559 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 560 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 561 //   region, it's hard to tell if the fault is due to a legitimate stack
 562 //   access or because of reading/writing non-exist memory (e.g. buffer
 563 //   overrun). As a rule, if the fault happens below current stack pointer,
 564 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 565 //   application (see Linux kernel fault.c).
 566 //
 567 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 568 //   stack overflow detection.
 569 //
 570 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 571 //   not use MAP_GROWSDOWN.
 572 //
 573 // To get around the problem and allow stack banging on Linux, we need to
 574 // manually expand thread stack after receiving the SIGSEGV.
 575 //
 576 // There are two ways to expand thread stack to address "bottom", we used
 577 // both of them in JVM before 1.5:
 578 //   1. adjust stack pointer first so that it is below "bottom", and then
 579 //      touch "bottom"
 580 //   2. mmap() the page in question
 581 //
 582 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 583 // if current sp is already near the lower end of page 101, and we need to
 584 // call mmap() to map page 100, it is possible that part of the mmap() frame
 585 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 586 // That will destroy the mmap() frame and cause VM to crash.
 587 //
 588 // The following code works by adjusting sp first, then accessing the "bottom"
 589 // page to force a page fault. Linux kernel will then automatically expand the
 590 // stack mapping.
 591 //
 592 // _expand_stack_to() assumes its frame size is less than page size, which
 593 // should always be true if the function is not inlined.
 594 
 595 static void NOINLINE _expand_stack_to(address bottom) {
 596   address sp;
 597   size_t size;
 598   volatile char *p;
 599 
 600   // Adjust bottom to point to the largest address within the same page, it
 601   // gives us a one-page buffer if alloca() allocates slightly more memory.
 602   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 603   bottom += os::Linux::page_size() - 1;
 604 
 605   // sp might be slightly above current stack pointer; if that's the case, we
 606   // will alloca() a little more space than necessary, which is OK. Don't use
 607   // os::current_stack_pointer(), as its result can be slightly below current
 608   // stack pointer, causing us to not alloca enough to reach "bottom".
 609   sp = (address)&sp;
 610 
 611   if (sp > bottom) {
 612     size = sp - bottom;
 613     p = (volatile char *)alloca(size);
 614     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 615     p[0] = '\0';
 616   }
 617 }
 618 
 619 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 620   assert(t!=NULL, "just checking");
 621   assert(t->osthread()->expanding_stack(), "expand should be set");
 622   assert(t->stack_base() != NULL, "stack_base was not initialized");
 623 
 624   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 625     sigset_t mask_all, old_sigset;
 626     sigfillset(&mask_all);
 627     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 628     _expand_stack_to(addr);
 629     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 630     return true;
 631   }
 632   return false;
 633 }
 634 
 635 //////////////////////////////////////////////////////////////////////////////
 636 // create new thread
 637 
 638 // Thread start routine for all newly created threads
 639 static void *thread_native_entry(Thread *thread) {
 640   // Try to randomize the cache line index of hot stack frames.
 641   // This helps when threads of the same stack traces evict each other's
 642   // cache lines. The threads can be either from the same JVM instance, or
 643   // from different JVM instances. The benefit is especially true for
 644   // processors with hyperthreading technology.
 645   static int counter = 0;
 646   int pid = os::current_process_id();
 647   alloca(((pid ^ counter++) & 7) * 128);
 648 
 649   thread->initialize_thread_current();
 650 
 651   OSThread* osthread = thread->osthread();
 652   Monitor* sync = osthread->startThread_lock();
 653 
 654   osthread->set_thread_id(os::current_thread_id());
 655 
 656   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 657     os::current_thread_id(), (uintx) pthread_self());
 658 
 659   if (UseNUMA) {
 660     int lgrp_id = os::numa_get_group_id();
 661     if (lgrp_id != -1) {
 662       thread->set_lgrp_id(lgrp_id);
 663     }
 664   }
 665   // initialize signal mask for this thread
 666   os::Linux::hotspot_sigmask(thread);
 667 
 668   // initialize floating point control register
 669   os::Linux::init_thread_fpu_state();
 670 
 671   // handshaking with parent thread
 672   {
 673     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 674 
 675     // notify parent thread
 676     osthread->set_state(INITIALIZED);
 677     sync->notify_all();
 678 
 679     // wait until os::start_thread()
 680     while (osthread->get_state() == INITIALIZED) {
 681       sync->wait(Mutex::_no_safepoint_check_flag);
 682     }
 683   }
 684 
 685   // call one more level start routine
 686   thread->run();
 687 
 688   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 689     os::current_thread_id(), (uintx) pthread_self());
 690 
 691   // If a thread has not deleted itself ("delete this") as part of its
 692   // termination sequence, we have to ensure thread-local-storage is
 693   // cleared before we actually terminate. No threads should ever be
 694   // deleted asynchronously with respect to their termination.
 695   if (Thread::current_or_null_safe() != NULL) {
 696     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 697     thread->clear_thread_current();
 698   }
 699 
 700   return 0;
 701 }
 702 
 703 bool os::create_thread(Thread* thread, ThreadType thr_type,
 704                        size_t stack_size) {
 705   assert(thread->osthread() == NULL, "caller responsible");
 706 
 707   // Allocate the OSThread object
 708   OSThread* osthread = new OSThread(NULL, NULL);
 709   if (osthread == NULL) {
 710     return false;
 711   }
 712 
 713   // set the correct thread state
 714   osthread->set_thread_type(thr_type);
 715 
 716   // Initial state is ALLOCATED but not INITIALIZED
 717   osthread->set_state(ALLOCATED);
 718 
 719   thread->set_osthread(osthread);
 720 
 721   // init thread attributes
 722   pthread_attr_t attr;
 723   pthread_attr_init(&attr);
 724   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 725 
 726   // stack size
 727   // calculate stack size if it's not specified by caller
 728   if (stack_size == 0) {
 729     stack_size = os::Linux::default_stack_size(thr_type);
 730 
 731     switch (thr_type) {
 732     case os::java_thread:
 733       // Java threads use ThreadStackSize which default value can be
 734       // changed with the flag -Xss
 735       assert(JavaThread::stack_size_at_create() > 0, "this should be set");
 736       stack_size = JavaThread::stack_size_at_create();
 737       break;
 738     case os::compiler_thread:
 739       if (CompilerThreadStackSize > 0) {
 740         stack_size = (size_t)(CompilerThreadStackSize * K);
 741         break;
 742       } // else fall through:
 743         // use VMThreadStackSize if CompilerThreadStackSize is not defined
 744     case os::vm_thread:
 745     case os::pgc_thread:
 746     case os::cgc_thread:
 747     case os::watcher_thread:
 748       if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 749       break;
 750     }
 751   }
 752 
 753   stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 754   pthread_attr_setstacksize(&attr, stack_size);
 755 
 756   // glibc guard page
 757   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 758 
 759   ThreadState state;
 760 
 761   {
 762     pthread_t tid;
 763     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 764 
 765     char buf[64];
 766     if (ret == 0) {
 767       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 768         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 769     } else {
 770       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 771         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 772     }
 773 
 774     pthread_attr_destroy(&attr);
 775 
 776     if (ret != 0) {
 777       // Need to clean up stuff we've allocated so far
 778       thread->set_osthread(NULL);
 779       delete osthread;
 780       return false;
 781     }
 782 
 783     // Store pthread info into the OSThread
 784     osthread->set_pthread_id(tid);
 785 
 786     // Wait until child thread is either initialized or aborted
 787     {
 788       Monitor* sync_with_child = osthread->startThread_lock();
 789       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 790       while ((state = osthread->get_state()) == ALLOCATED) {
 791         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 792       }
 793     }
 794   }
 795 
 796   // Aborted due to thread limit being reached
 797   if (state == ZOMBIE) {
 798     thread->set_osthread(NULL);
 799     delete osthread;
 800     return false;
 801   }
 802 
 803   // The thread is returned suspended (in state INITIALIZED),
 804   // and is started higher up in the call chain
 805   assert(state == INITIALIZED, "race condition");
 806   return true;
 807 }
 808 
 809 /////////////////////////////////////////////////////////////////////////////
 810 // attach existing thread
 811 
 812 // bootstrap the main thread
 813 bool os::create_main_thread(JavaThread* thread) {
 814   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 815   return create_attached_thread(thread);
 816 }
 817 
 818 bool os::create_attached_thread(JavaThread* thread) {
 819 #ifdef ASSERT
 820   thread->verify_not_published();
 821 #endif
 822 
 823   // Allocate the OSThread object
 824   OSThread* osthread = new OSThread(NULL, NULL);
 825 
 826   if (osthread == NULL) {
 827     return false;
 828   }
 829 
 830   // Store pthread info into the OSThread
 831   osthread->set_thread_id(os::Linux::gettid());
 832   osthread->set_pthread_id(::pthread_self());
 833 
 834   // initialize floating point control register
 835   os::Linux::init_thread_fpu_state();
 836 
 837   // Initial thread state is RUNNABLE
 838   osthread->set_state(RUNNABLE);
 839 
 840   thread->set_osthread(osthread);
 841 
 842   if (UseNUMA) {
 843     int lgrp_id = os::numa_get_group_id();
 844     if (lgrp_id != -1) {
 845       thread->set_lgrp_id(lgrp_id);
 846     }
 847   }
 848 
 849   if (os::Linux::is_initial_thread()) {
 850     // If current thread is initial thread, its stack is mapped on demand,
 851     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 852     // the entire stack region to avoid SEGV in stack banging.
 853     // It is also useful to get around the heap-stack-gap problem on SuSE
 854     // kernel (see 4821821 for details). We first expand stack to the top
 855     // of yellow zone, then enable stack yellow zone (order is significant,
 856     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 857     // is no gap between the last two virtual memory regions.
 858 
 859     JavaThread *jt = (JavaThread *)thread;
 860     address addr = jt->stack_reserved_zone_base();
 861     assert(addr != NULL, "initialization problem?");
 862     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 863 
 864     osthread->set_expanding_stack();
 865     os::Linux::manually_expand_stack(jt, addr);
 866     osthread->clear_expanding_stack();
 867   }
 868 
 869   // initialize signal mask for this thread
 870   // and save the caller's signal mask
 871   os::Linux::hotspot_sigmask(thread);
 872 
 873   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 874     os::current_thread_id(), (uintx) pthread_self());
 875 
 876   return true;
 877 }
 878 
 879 void os::pd_start_thread(Thread* thread) {
 880   OSThread * osthread = thread->osthread();
 881   assert(osthread->get_state() != INITIALIZED, "just checking");
 882   Monitor* sync_with_child = osthread->startThread_lock();
 883   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 884   sync_with_child->notify();
 885 }
 886 
 887 // Free Linux resources related to the OSThread
 888 void os::free_thread(OSThread* osthread) {
 889   assert(osthread != NULL, "osthread not set");
 890 
 891   // We are told to free resources of the argument thread,
 892   // but we can only really operate on the current thread.
 893   assert(Thread::current()->osthread() == osthread,
 894          "os::free_thread but not current thread");
 895 
 896 #ifdef ASSERT
 897   sigset_t current;
 898   sigemptyset(&current);
 899   pthread_sigmask(SIG_SETMASK, NULL, &current);
 900   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 901 #endif
 902 
 903   // Restore caller's signal mask
 904   sigset_t sigmask = osthread->caller_sigmask();
 905   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 906 
 907   delete osthread;
 908 }
 909 
 910 //////////////////////////////////////////////////////////////////////////////
 911 // initial thread
 912 
 913 // Check if current thread is the initial thread, similar to Solaris thr_main.
 914 bool os::Linux::is_initial_thread(void) {
 915   char dummy;
 916   // If called before init complete, thread stack bottom will be null.
 917   // Can be called if fatal error occurs before initialization.
 918   if (initial_thread_stack_bottom() == NULL) return false;
 919   assert(initial_thread_stack_bottom() != NULL &&
 920          initial_thread_stack_size()   != 0,
 921          "os::init did not locate initial thread's stack region");
 922   if ((address)&dummy >= initial_thread_stack_bottom() &&
 923       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 924     return true;
 925   } else {
 926     return false;
 927   }
 928 }
 929 
 930 // Find the virtual memory area that contains addr
 931 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 932   FILE *fp = fopen("/proc/self/maps", "r");
 933   if (fp) {
 934     address low, high;
 935     while (!feof(fp)) {
 936       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 937         if (low <= addr && addr < high) {
 938           if (vma_low)  *vma_low  = low;
 939           if (vma_high) *vma_high = high;
 940           fclose(fp);
 941           return true;
 942         }
 943       }
 944       for (;;) {
 945         int ch = fgetc(fp);
 946         if (ch == EOF || ch == (int)'\n') break;
 947       }
 948     }
 949     fclose(fp);
 950   }
 951   return false;
 952 }
 953 
 954 // Locate initial thread stack. This special handling of initial thread stack
 955 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 956 // bogus value for initial thread.
 957 void os::Linux::capture_initial_stack(size_t max_size) {
 958   // stack size is the easy part, get it from RLIMIT_STACK
 959   size_t stack_size;
 960   struct rlimit rlim;
 961   getrlimit(RLIMIT_STACK, &rlim);
 962   stack_size = rlim.rlim_cur;
 963 
 964   // 6308388: a bug in ld.so will relocate its own .data section to the
 965   //   lower end of primordial stack; reduce ulimit -s value a little bit
 966   //   so we won't install guard page on ld.so's data section.
 967   stack_size -= 2 * page_size();
 968 
 969   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
 970   //   7.1, in both cases we will get 2G in return value.
 971   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
 972   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
 973   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
 974   //   in case other parts in glibc still assumes 2M max stack size.
 975   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
 976   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
 977   if (stack_size > 2 * K * K IA64_ONLY(*2)) {
 978     stack_size = 2 * K * K IA64_ONLY(*2);
 979   }
 980   // Try to figure out where the stack base (top) is. This is harder.
 981   //
 982   // When an application is started, glibc saves the initial stack pointer in
 983   // a global variable "__libc_stack_end", which is then used by system
 984   // libraries. __libc_stack_end should be pretty close to stack top. The
 985   // variable is available since the very early days. However, because it is
 986   // a private interface, it could disappear in the future.
 987   //
 988   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 989   // to __libc_stack_end, it is very close to stack top, but isn't the real
 990   // stack top. Note that /proc may not exist if VM is running as a chroot
 991   // program, so reading /proc/<pid>/stat could fail. Also the contents of
 992   // /proc/<pid>/stat could change in the future (though unlikely).
 993   //
 994   // We try __libc_stack_end first. If that doesn't work, look for
 995   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
 996   // as a hint, which should work well in most cases.
 997 
 998   uintptr_t stack_start;
 999 
1000   // try __libc_stack_end first
1001   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1002   if (p && *p) {
1003     stack_start = *p;
1004   } else {
1005     // see if we can get the start_stack field from /proc/self/stat
1006     FILE *fp;
1007     int pid;
1008     char state;
1009     int ppid;
1010     int pgrp;
1011     int session;
1012     int nr;
1013     int tpgrp;
1014     unsigned long flags;
1015     unsigned long minflt;
1016     unsigned long cminflt;
1017     unsigned long majflt;
1018     unsigned long cmajflt;
1019     unsigned long utime;
1020     unsigned long stime;
1021     long cutime;
1022     long cstime;
1023     long prio;
1024     long nice;
1025     long junk;
1026     long it_real;
1027     uintptr_t start;
1028     uintptr_t vsize;
1029     intptr_t rss;
1030     uintptr_t rsslim;
1031     uintptr_t scodes;
1032     uintptr_t ecode;
1033     int i;
1034 
1035     // Figure what the primordial thread stack base is. Code is inspired
1036     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1037     // followed by command name surrounded by parentheses, state, etc.
1038     char stat[2048];
1039     int statlen;
1040 
1041     fp = fopen("/proc/self/stat", "r");
1042     if (fp) {
1043       statlen = fread(stat, 1, 2047, fp);
1044       stat[statlen] = '\0';
1045       fclose(fp);
1046 
1047       // Skip pid and the command string. Note that we could be dealing with
1048       // weird command names, e.g. user could decide to rename java launcher
1049       // to "java 1.4.2 :)", then the stat file would look like
1050       //                1234 (java 1.4.2 :)) R ... ...
1051       // We don't really need to know the command string, just find the last
1052       // occurrence of ")" and then start parsing from there. See bug 4726580.
1053       char * s = strrchr(stat, ')');
1054 
1055       i = 0;
1056       if (s) {
1057         // Skip blank chars
1058         do { s++; } while (s && isspace(*s));
1059 
1060 #define _UFM UINTX_FORMAT
1061 #define _DFM INTX_FORMAT
1062 
1063         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1064         //              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
1065         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,
1066                    &state,          // 3  %c
1067                    &ppid,           // 4  %d
1068                    &pgrp,           // 5  %d
1069                    &session,        // 6  %d
1070                    &nr,             // 7  %d
1071                    &tpgrp,          // 8  %d
1072                    &flags,          // 9  %lu
1073                    &minflt,         // 10 %lu
1074                    &cminflt,        // 11 %lu
1075                    &majflt,         // 12 %lu
1076                    &cmajflt,        // 13 %lu
1077                    &utime,          // 14 %lu
1078                    &stime,          // 15 %lu
1079                    &cutime,         // 16 %ld
1080                    &cstime,         // 17 %ld
1081                    &prio,           // 18 %ld
1082                    &nice,           // 19 %ld
1083                    &junk,           // 20 %ld
1084                    &it_real,        // 21 %ld
1085                    &start,          // 22 UINTX_FORMAT
1086                    &vsize,          // 23 UINTX_FORMAT
1087                    &rss,            // 24 INTX_FORMAT
1088                    &rsslim,         // 25 UINTX_FORMAT
1089                    &scodes,         // 26 UINTX_FORMAT
1090                    &ecode,          // 27 UINTX_FORMAT
1091                    &stack_start);   // 28 UINTX_FORMAT
1092       }
1093 
1094 #undef _UFM
1095 #undef _DFM
1096 
1097       if (i != 28 - 2) {
1098         assert(false, "Bad conversion from /proc/self/stat");
1099         // product mode - assume we are the initial thread, good luck in the
1100         // embedded case.
1101         warning("Can't detect initial thread stack location - bad conversion");
1102         stack_start = (uintptr_t) &rlim;
1103       }
1104     } else {
1105       // For some reason we can't open /proc/self/stat (for example, running on
1106       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1107       // most cases, so don't abort:
1108       warning("Can't detect initial thread stack location - no /proc/self/stat");
1109       stack_start = (uintptr_t) &rlim;
1110     }
1111   }
1112 
1113   // Now we have a pointer (stack_start) very close to the stack top, the
1114   // next thing to do is to figure out the exact location of stack top. We
1115   // can find out the virtual memory area that contains stack_start by
1116   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1117   // and its upper limit is the real stack top. (again, this would fail if
1118   // running inside chroot, because /proc may not exist.)
1119 
1120   uintptr_t stack_top;
1121   address low, high;
1122   if (find_vma((address)stack_start, &low, &high)) {
1123     // success, "high" is the true stack top. (ignore "low", because initial
1124     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1125     stack_top = (uintptr_t)high;
1126   } else {
1127     // failed, likely because /proc/self/maps does not exist
1128     warning("Can't detect initial thread stack location - find_vma failed");
1129     // best effort: stack_start is normally within a few pages below the real
1130     // stack top, use it as stack top, and reduce stack size so we won't put
1131     // guard page outside stack.
1132     stack_top = stack_start;
1133     stack_size -= 16 * page_size();
1134   }
1135 
1136   // stack_top could be partially down the page so align it
1137   stack_top = align_size_up(stack_top, page_size());
1138 
1139   if (max_size && stack_size > max_size) {
1140     _initial_thread_stack_size = max_size;
1141   } else {
1142     _initial_thread_stack_size = stack_size;
1143   }
1144 
1145   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1146   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1147 }
1148 
1149 ////////////////////////////////////////////////////////////////////////////////
1150 // time support
1151 
1152 // Time since start-up in seconds to a fine granularity.
1153 // Used by VMSelfDestructTimer and the MemProfiler.
1154 double os::elapsedTime() {
1155 
1156   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1157 }
1158 
1159 jlong os::elapsed_counter() {
1160   return javaTimeNanos() - initial_time_count;
1161 }
1162 
1163 jlong os::elapsed_frequency() {
1164   return NANOSECS_PER_SEC; // nanosecond resolution
1165 }
1166 
1167 bool os::supports_vtime() { return true; }
1168 bool os::enable_vtime()   { return false; }
1169 bool os::vtime_enabled()  { return false; }
1170 
1171 double os::elapsedVTime() {
1172   struct rusage usage;
1173   int retval = getrusage(RUSAGE_THREAD, &usage);
1174   if (retval == 0) {
1175     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);
1176   } else {
1177     // better than nothing, but not much
1178     return elapsedTime();
1179   }
1180 }
1181 
1182 jlong os::javaTimeMillis() {
1183   timeval time;
1184   int status = gettimeofday(&time, NULL);
1185   assert(status != -1, "linux error");
1186   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1187 }
1188 
1189 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1190   timeval time;
1191   int status = gettimeofday(&time, NULL);
1192   assert(status != -1, "linux error");
1193   seconds = jlong(time.tv_sec);
1194   nanos = jlong(time.tv_usec) * 1000;
1195 }
1196 
1197 
1198 #ifndef CLOCK_MONOTONIC
1199   #define CLOCK_MONOTONIC (1)
1200 #endif
1201 
1202 void os::Linux::clock_init() {
1203   // we do dlopen's in this particular order due to bug in linux
1204   // dynamical loader (see 6348968) leading to crash on exit
1205   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1206   if (handle == NULL) {
1207     handle = dlopen("librt.so", RTLD_LAZY);
1208   }
1209 
1210   if (handle) {
1211     int (*clock_getres_func)(clockid_t, struct timespec*) =
1212            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1213     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1214            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1215     if (clock_getres_func && clock_gettime_func) {
1216       // See if monotonic clock is supported by the kernel. Note that some
1217       // early implementations simply return kernel jiffies (updated every
1218       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1219       // for nano time (though the monotonic property is still nice to have).
1220       // It's fixed in newer kernels, however clock_getres() still returns
1221       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1222       // resolution for now. Hopefully as people move to new kernels, this
1223       // won't be a problem.
1224       struct timespec res;
1225       struct timespec tp;
1226       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1227           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1228         // yes, monotonic clock is supported
1229         _clock_gettime = clock_gettime_func;
1230         return;
1231       } else {
1232         // close librt if there is no monotonic clock
1233         dlclose(handle);
1234       }
1235     }
1236   }
1237   warning("No monotonic clock was available - timed services may " \
1238           "be adversely affected if the time-of-day clock changes");
1239 }
1240 
1241 #ifndef SYS_clock_getres
1242   #if defined(IA32) || defined(AMD64)
1243     #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1244     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1245   #else
1246     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1247     #define sys_clock_getres(x,y)  -1
1248   #endif
1249 #else
1250   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1251 #endif
1252 
1253 void os::Linux::fast_thread_clock_init() {
1254   if (!UseLinuxPosixThreadCPUClocks) {
1255     return;
1256   }
1257   clockid_t clockid;
1258   struct timespec tp;
1259   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1260       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1261 
1262   // Switch to using fast clocks for thread cpu time if
1263   // the sys_clock_getres() returns 0 error code.
1264   // Note, that some kernels may support the current thread
1265   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1266   // returned by the pthread_getcpuclockid().
1267   // If the fast Posix clocks are supported then the sys_clock_getres()
1268   // must return at least tp.tv_sec == 0 which means a resolution
1269   // better than 1 sec. This is extra check for reliability.
1270 
1271   if (pthread_getcpuclockid_func &&
1272       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1273       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1274     _supports_fast_thread_cpu_time = true;
1275     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1276   }
1277 }
1278 
1279 jlong os::javaTimeNanos() {
1280   if (os::supports_monotonic_clock()) {
1281     struct timespec tp;
1282     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1283     assert(status == 0, "gettime error");
1284     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1285     return result;
1286   } else {
1287     timeval time;
1288     int status = gettimeofday(&time, NULL);
1289     assert(status != -1, "linux error");
1290     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1291     return 1000 * usecs;
1292   }
1293 }
1294 
1295 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1296   if (os::supports_monotonic_clock()) {
1297     info_ptr->max_value = ALL_64_BITS;
1298 
1299     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1300     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1301     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1302   } else {
1303     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1304     info_ptr->max_value = ALL_64_BITS;
1305 
1306     // gettimeofday is a real time clock so it skips
1307     info_ptr->may_skip_backward = true;
1308     info_ptr->may_skip_forward = true;
1309   }
1310 
1311   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1312 }
1313 
1314 // Return the real, user, and system times in seconds from an
1315 // arbitrary fixed point in the past.
1316 bool os::getTimesSecs(double* process_real_time,
1317                       double* process_user_time,
1318                       double* process_system_time) {
1319   struct tms ticks;
1320   clock_t real_ticks = times(&ticks);
1321 
1322   if (real_ticks == (clock_t) (-1)) {
1323     return false;
1324   } else {
1325     double ticks_per_second = (double) clock_tics_per_sec;
1326     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1327     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1328     *process_real_time = ((double) real_ticks) / ticks_per_second;
1329 
1330     return true;
1331   }
1332 }
1333 
1334 
1335 char * os::local_time_string(char *buf, size_t buflen) {
1336   struct tm t;
1337   time_t long_time;
1338   time(&long_time);
1339   localtime_r(&long_time, &t);
1340   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1341                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1342                t.tm_hour, t.tm_min, t.tm_sec);
1343   return buf;
1344 }
1345 
1346 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1347   return localtime_r(clock, res);
1348 }
1349 
1350 ////////////////////////////////////////////////////////////////////////////////
1351 // runtime exit support
1352 
1353 // Note: os::shutdown() might be called very early during initialization, or
1354 // called from signal handler. Before adding something to os::shutdown(), make
1355 // sure it is async-safe and can handle partially initialized VM.
1356 void os::shutdown() {
1357 
1358   // allow PerfMemory to attempt cleanup of any persistent resources
1359   perfMemory_exit();
1360 
1361   // needs to remove object in file system
1362   AttachListener::abort();
1363 
1364   // flush buffered output, finish log files
1365   ostream_abort();
1366 
1367   // Check for abort hook
1368   abort_hook_t abort_hook = Arguments::abort_hook();
1369   if (abort_hook != NULL) {
1370     abort_hook();
1371   }
1372 
1373 }
1374 
1375 // Note: os::abort() might be called very early during initialization, or
1376 // called from signal handler. Before adding something to os::abort(), make
1377 // sure it is async-safe and can handle partially initialized VM.
1378 void os::abort(bool dump_core, void* siginfo, const void* context) {
1379   os::shutdown();
1380   if (dump_core) {
1381 #ifndef PRODUCT
1382     fdStream out(defaultStream::output_fd());
1383     out.print_raw("Current thread is ");
1384     char buf[16];
1385     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1386     out.print_raw_cr(buf);
1387     out.print_raw_cr("Dumping core ...");
1388 #endif
1389     ::abort(); // dump core
1390   }
1391 
1392   ::exit(1);
1393 }
1394 
1395 // Die immediately, no exit hook, no abort hook, no cleanup.
1396 void os::die() {
1397   ::abort();
1398 }
1399 
1400 
1401 // This method is a copy of JDK's sysGetLastErrorString
1402 // from src/solaris/hpi/src/system_md.c
1403 
1404 size_t os::lasterror(char *buf, size_t len) {
1405   if (errno == 0)  return 0;
1406 
1407   const char *s = os::strerror(errno);
1408   size_t n = ::strlen(s);
1409   if (n >= len) {
1410     n = len - 1;
1411   }
1412   ::strncpy(buf, s, n);
1413   buf[n] = '\0';
1414   return n;
1415 }
1416 
1417 // thread_id is kernel thread id (similar to Solaris LWP id)
1418 intx os::current_thread_id() { return os::Linux::gettid(); }
1419 int os::current_process_id() {
1420   return ::getpid();
1421 }
1422 
1423 // DLL functions
1424 
1425 const char* os::dll_file_extension() { return ".so"; }
1426 
1427 // This must be hard coded because it's the system's temporary
1428 // directory not the java application's temp directory, ala java.io.tmpdir.
1429 const char* os::get_temp_directory() { return "/tmp"; }
1430 
1431 static bool file_exists(const char* filename) {
1432   struct stat statbuf;
1433   if (filename == NULL || strlen(filename) == 0) {
1434     return false;
1435   }
1436   return os::stat(filename, &statbuf) == 0;
1437 }
1438 
1439 bool os::dll_build_name(char* buffer, size_t buflen,
1440                         const char* pname, const char* fname) {
1441   bool retval = false;
1442   // Copied from libhpi
1443   const size_t pnamelen = pname ? strlen(pname) : 0;
1444 
1445   // Return error on buffer overflow.
1446   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1447     return retval;
1448   }
1449 
1450   if (pnamelen == 0) {
1451     snprintf(buffer, buflen, "lib%s.so", fname);
1452     retval = true;
1453   } else if (strchr(pname, *os::path_separator()) != NULL) {
1454     int n;
1455     char** pelements = split_path(pname, &n);
1456     if (pelements == NULL) {
1457       return false;
1458     }
1459     for (int i = 0; i < n; i++) {
1460       // Really shouldn't be NULL, but check can't hurt
1461       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1462         continue; // skip the empty path values
1463       }
1464       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1465       if (file_exists(buffer)) {
1466         retval = true;
1467         break;
1468       }
1469     }
1470     // release the storage
1471     for (int i = 0; i < n; i++) {
1472       if (pelements[i] != NULL) {
1473         FREE_C_HEAP_ARRAY(char, pelements[i]);
1474       }
1475     }
1476     if (pelements != NULL) {
1477       FREE_C_HEAP_ARRAY(char*, pelements);
1478     }
1479   } else {
1480     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1481     retval = true;
1482   }
1483   return retval;
1484 }
1485 
1486 // check if addr is inside libjvm.so
1487 bool os::address_is_in_vm(address addr) {
1488   static address libjvm_base_addr;
1489   Dl_info dlinfo;
1490 
1491   if (libjvm_base_addr == NULL) {
1492     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1493       libjvm_base_addr = (address)dlinfo.dli_fbase;
1494     }
1495     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1496   }
1497 
1498   if (dladdr((void *)addr, &dlinfo) != 0) {
1499     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1500   }
1501 
1502   return false;
1503 }
1504 
1505 bool os::dll_address_to_function_name(address addr, char *buf,
1506                                       int buflen, int *offset,
1507                                       bool demangle) {
1508   // buf is not optional, but offset is optional
1509   assert(buf != NULL, "sanity check");
1510 
1511   Dl_info dlinfo;
1512 
1513   if (dladdr((void*)addr, &dlinfo) != 0) {
1514     // see if we have a matching symbol
1515     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1516       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1517         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1518       }
1519       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1520       return true;
1521     }
1522     // no matching symbol so try for just file info
1523     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1524       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1525                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1526         return true;
1527       }
1528     }
1529   }
1530 
1531   buf[0] = '\0';
1532   if (offset != NULL) *offset = -1;
1533   return false;
1534 }
1535 
1536 struct _address_to_library_name {
1537   address addr;          // input : memory address
1538   size_t  buflen;        //         size of fname
1539   char*   fname;         // output: library name
1540   address base;          //         library base addr
1541 };
1542 
1543 static int address_to_library_name_callback(struct dl_phdr_info *info,
1544                                             size_t size, void *data) {
1545   int i;
1546   bool found = false;
1547   address libbase = NULL;
1548   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1549 
1550   // iterate through all loadable segments
1551   for (i = 0; i < info->dlpi_phnum; i++) {
1552     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1553     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1554       // base address of a library is the lowest address of its loaded
1555       // segments.
1556       if (libbase == NULL || libbase > segbase) {
1557         libbase = segbase;
1558       }
1559       // see if 'addr' is within current segment
1560       if (segbase <= d->addr &&
1561           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1562         found = true;
1563       }
1564     }
1565   }
1566 
1567   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1568   // so dll_address_to_library_name() can fall through to use dladdr() which
1569   // can figure out executable name from argv[0].
1570   if (found && info->dlpi_name && info->dlpi_name[0]) {
1571     d->base = libbase;
1572     if (d->fname) {
1573       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1574     }
1575     return 1;
1576   }
1577   return 0;
1578 }
1579 
1580 bool os::dll_address_to_library_name(address addr, char* buf,
1581                                      int buflen, int* offset) {
1582   // buf is not optional, but offset is optional
1583   assert(buf != NULL, "sanity check");
1584 
1585   Dl_info dlinfo;
1586   struct _address_to_library_name data;
1587 
1588   // There is a bug in old glibc dladdr() implementation that it could resolve
1589   // to wrong library name if the .so file has a base address != NULL. Here
1590   // we iterate through the program headers of all loaded libraries to find
1591   // out which library 'addr' really belongs to. This workaround can be
1592   // removed once the minimum requirement for glibc is moved to 2.3.x.
1593   data.addr = addr;
1594   data.fname = buf;
1595   data.buflen = buflen;
1596   data.base = NULL;
1597   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1598 
1599   if (rslt) {
1600     // buf already contains library name
1601     if (offset) *offset = addr - data.base;
1602     return true;
1603   }
1604   if (dladdr((void*)addr, &dlinfo) != 0) {
1605     if (dlinfo.dli_fname != NULL) {
1606       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1607     }
1608     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1609       *offset = addr - (address)dlinfo.dli_fbase;
1610     }
1611     return true;
1612   }
1613 
1614   buf[0] = '\0';
1615   if (offset) *offset = -1;
1616   return false;
1617 }
1618 
1619 // Loads .dll/.so and
1620 // in case of error it checks if .dll/.so was built for the
1621 // same architecture as Hotspot is running on
1622 
1623 
1624 // Remember the stack's state. The Linux dynamic linker will change
1625 // the stack to 'executable' at most once, so we must safepoint only once.
1626 bool os::Linux::_stack_is_executable = false;
1627 
1628 // VM operation that loads a library.  This is necessary if stack protection
1629 // of the Java stacks can be lost during loading the library.  If we
1630 // do not stop the Java threads, they can stack overflow before the stacks
1631 // are protected again.
1632 class VM_LinuxDllLoad: public VM_Operation {
1633  private:
1634   const char *_filename;
1635   char *_ebuf;
1636   int _ebuflen;
1637   void *_lib;
1638  public:
1639   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1640     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1641   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1642   void doit() {
1643     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1644     os::Linux::_stack_is_executable = true;
1645   }
1646   void* loaded_library() { return _lib; }
1647 };
1648 
1649 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1650   void * result = NULL;
1651   bool load_attempted = false;
1652 
1653   // Check whether the library to load might change execution rights
1654   // of the stack. If they are changed, the protection of the stack
1655   // guard pages will be lost. We need a safepoint to fix this.
1656   //
1657   // See Linux man page execstack(8) for more info.
1658   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1659     ElfFile ef(filename);
1660     if (!ef.specifies_noexecstack()) {
1661       if (!is_init_completed()) {
1662         os::Linux::_stack_is_executable = true;
1663         // This is OK - No Java threads have been created yet, and hence no
1664         // stack guard pages to fix.
1665         //
1666         // This should happen only when you are building JDK7 using a very
1667         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1668         //
1669         // Dynamic loader will make all stacks executable after
1670         // this function returns, and will not do that again.
1671         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1672       } else {
1673         warning("You have loaded library %s which might have disabled stack guard. "
1674                 "The VM will try to fix the stack guard now.\n"
1675                 "It's highly recommended that you fix the library with "
1676                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1677                 filename);
1678 
1679         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1680         JavaThread *jt = JavaThread::current();
1681         if (jt->thread_state() != _thread_in_native) {
1682           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1683           // that requires ExecStack. Cannot enter safe point. Let's give up.
1684           warning("Unable to fix stack guard. Giving up.");
1685         } else {
1686           if (!LoadExecStackDllInVMThread) {
1687             // This is for the case where the DLL has an static
1688             // constructor function that executes JNI code. We cannot
1689             // load such DLLs in the VMThread.
1690             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1691           }
1692 
1693           ThreadInVMfromNative tiv(jt);
1694           debug_only(VMNativeEntryWrapper vew;)
1695 
1696           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1697           VMThread::execute(&op);
1698           if (LoadExecStackDllInVMThread) {
1699             result = op.loaded_library();
1700           }
1701           load_attempted = true;
1702         }
1703       }
1704     }
1705   }
1706 
1707   if (!load_attempted) {
1708     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1709   }
1710 
1711   if (result != NULL) {
1712     // Successful loading
1713     return result;
1714   }
1715 
1716   Elf32_Ehdr elf_head;
1717   int diag_msg_max_length=ebuflen-strlen(ebuf);
1718   char* diag_msg_buf=ebuf+strlen(ebuf);
1719 
1720   if (diag_msg_max_length==0) {
1721     // No more space in ebuf for additional diagnostics message
1722     return NULL;
1723   }
1724 
1725 
1726   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1727 
1728   if (file_descriptor < 0) {
1729     // Can't open library, report dlerror() message
1730     return NULL;
1731   }
1732 
1733   bool failed_to_read_elf_head=
1734     (sizeof(elf_head)!=
1735      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1736 
1737   ::close(file_descriptor);
1738   if (failed_to_read_elf_head) {
1739     // file i/o error - report dlerror() msg
1740     return NULL;
1741   }
1742 
1743   typedef struct {
1744     Elf32_Half    code;         // Actual value as defined in elf.h
1745     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1746     unsigned char elf_class;    // 32 or 64 bit
1747     unsigned char endianess;    // MSB or LSB
1748     char*         name;         // String representation
1749   } arch_t;
1750 
1751 #ifndef EM_486
1752   #define EM_486          6               /* Intel 80486 */
1753 #endif
1754 #ifndef EM_AARCH64
1755   #define EM_AARCH64    183               /* ARM AARCH64 */
1756 #endif
1757 
1758   static const arch_t arch_array[]={
1759     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1760     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1761     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1762     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1763     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1764     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1765     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1766     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1767 #if defined(VM_LITTLE_ENDIAN)
1768     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1769 #else
1770     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1771 #endif
1772     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1773     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1774     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1775     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1776     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1777     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1778     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1779     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1780   };
1781 
1782 #if  (defined IA32)
1783   static  Elf32_Half running_arch_code=EM_386;
1784 #elif   (defined AMD64)
1785   static  Elf32_Half running_arch_code=EM_X86_64;
1786 #elif  (defined IA64)
1787   static  Elf32_Half running_arch_code=EM_IA_64;
1788 #elif  (defined __sparc) && (defined _LP64)
1789   static  Elf32_Half running_arch_code=EM_SPARCV9;
1790 #elif  (defined __sparc) && (!defined _LP64)
1791   static  Elf32_Half running_arch_code=EM_SPARC;
1792 #elif  (defined __powerpc64__)
1793   static  Elf32_Half running_arch_code=EM_PPC64;
1794 #elif  (defined __powerpc__)
1795   static  Elf32_Half running_arch_code=EM_PPC;
1796 #elif  (defined ARM)
1797   static  Elf32_Half running_arch_code=EM_ARM;
1798 #elif  (defined S390)
1799   static  Elf32_Half running_arch_code=EM_S390;
1800 #elif  (defined ALPHA)
1801   static  Elf32_Half running_arch_code=EM_ALPHA;
1802 #elif  (defined MIPSEL)
1803   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1804 #elif  (defined PARISC)
1805   static  Elf32_Half running_arch_code=EM_PARISC;
1806 #elif  (defined MIPS)
1807   static  Elf32_Half running_arch_code=EM_MIPS;
1808 #elif  (defined M68K)
1809   static  Elf32_Half running_arch_code=EM_68K;
1810 #elif  (defined AARCH64)
1811   static  Elf32_Half running_arch_code=EM_AARCH64;
1812 #else
1813     #error Method os::dll_load requires that one of following is defined:\
1814          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1815 #endif
1816 
1817   // Identify compatability class for VM's architecture and library's architecture
1818   // Obtain string descriptions for architectures
1819 
1820   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1821   int running_arch_index=-1;
1822 
1823   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1824     if (running_arch_code == arch_array[i].code) {
1825       running_arch_index    = i;
1826     }
1827     if (lib_arch.code == arch_array[i].code) {
1828       lib_arch.compat_class = arch_array[i].compat_class;
1829       lib_arch.name         = arch_array[i].name;
1830     }
1831   }
1832 
1833   assert(running_arch_index != -1,
1834          "Didn't find running architecture code (running_arch_code) in arch_array");
1835   if (running_arch_index == -1) {
1836     // Even though running architecture detection failed
1837     // we may still continue with reporting dlerror() message
1838     return NULL;
1839   }
1840 
1841   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1842     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1843     return NULL;
1844   }
1845 
1846 #ifndef S390
1847   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1848     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1849     return NULL;
1850   }
1851 #endif // !S390
1852 
1853   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1854     if (lib_arch.name!=NULL) {
1855       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1856                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1857                  lib_arch.name, arch_array[running_arch_index].name);
1858     } else {
1859       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1860                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1861                  lib_arch.code,
1862                  arch_array[running_arch_index].name);
1863     }
1864   }
1865 
1866   return NULL;
1867 }
1868 
1869 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1870                                 int ebuflen) {
1871   void * result = ::dlopen(filename, RTLD_LAZY);
1872   if (result == NULL) {
1873     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1874     ebuf[ebuflen-1] = '\0';
1875   }
1876   return result;
1877 }
1878 
1879 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1880                                        int ebuflen) {
1881   void * result = NULL;
1882   if (LoadExecStackDllInVMThread) {
1883     result = dlopen_helper(filename, ebuf, ebuflen);
1884   }
1885 
1886   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1887   // library that requires an executable stack, or which does not have this
1888   // stack attribute set, dlopen changes the stack attribute to executable. The
1889   // read protection of the guard pages gets lost.
1890   //
1891   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1892   // may have been queued at the same time.
1893 
1894   if (!_stack_is_executable) {
1895     JavaThread *jt = Threads::first();
1896 
1897     while (jt) {
1898       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1899           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1900         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1901           warning("Attempt to reguard stack yellow zone failed.");
1902         }
1903       }
1904       jt = jt->next();
1905     }
1906   }
1907 
1908   return result;
1909 }
1910 
1911 void* os::dll_lookup(void* handle, const char* name) {
1912   void* res = dlsym(handle, name);
1913   return res;
1914 }
1915 
1916 void* os::get_default_process_handle() {
1917   return (void*)::dlopen(NULL, RTLD_LAZY);
1918 }
1919 
1920 static bool _print_ascii_file(const char* filename, outputStream* st) {
1921   int fd = ::open(filename, O_RDONLY);
1922   if (fd == -1) {
1923     return false;
1924   }
1925 
1926   char buf[33];
1927   int bytes;
1928   buf[32] = '\0';
1929   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1930     st->print_raw(buf, bytes);
1931   }
1932 
1933   ::close(fd);
1934 
1935   return true;
1936 }
1937 
1938 void os::print_dll_info(outputStream *st) {
1939   st->print_cr("Dynamic libraries:");
1940 
1941   char fname[32];
1942   pid_t pid = os::Linux::gettid();
1943 
1944   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1945 
1946   if (!_print_ascii_file(fname, st)) {
1947     st->print("Can not get library information for pid = %d\n", pid);
1948   }
1949 }
1950 
1951 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1952   FILE *procmapsFile = NULL;
1953 
1954   // Open the procfs maps file for the current process
1955   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1956     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1957     char line[PATH_MAX + 100];
1958 
1959     // Read line by line from 'file'
1960     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1961       u8 base, top, offset, inode;
1962       char permissions[5];
1963       char device[6];
1964       char name[PATH_MAX + 1];
1965 
1966       // Parse fields from line
1967       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1968              &base, &top, permissions, &offset, device, &inode, name);
1969 
1970       // Filter by device id '00:00' so that we only get file system mapped files.
1971       if (strcmp(device, "00:00") != 0) {
1972 
1973         // Call callback with the fields of interest
1974         if(callback(name, (address)base, (address)top, param)) {
1975           // Oops abort, callback aborted
1976           fclose(procmapsFile);
1977           return 1;
1978         }
1979       }
1980     }
1981     fclose(procmapsFile);
1982   }
1983   return 0;
1984 }
1985 
1986 void os::print_os_info_brief(outputStream* st) {
1987   os::Linux::print_distro_info(st);
1988 
1989   os::Posix::print_uname_info(st);
1990 
1991   os::Linux::print_libversion_info(st);
1992 
1993 }
1994 
1995 void os::print_os_info(outputStream* st) {
1996   st->print("OS:");
1997 
1998   os::Linux::print_distro_info(st);
1999 
2000   os::Posix::print_uname_info(st);
2001 
2002   // Print warning if unsafe chroot environment detected
2003   if (unsafe_chroot_detected) {
2004     st->print("WARNING!! ");
2005     st->print_cr("%s", unstable_chroot_error);
2006   }
2007 
2008   os::Linux::print_libversion_info(st);
2009 
2010   os::Posix::print_rlimit_info(st);
2011 
2012   os::Posix::print_load_average(st);
2013 
2014   os::Linux::print_full_memory_info(st);
2015 }
2016 
2017 // Try to identify popular distros.
2018 // Most Linux distributions have a /etc/XXX-release file, which contains
2019 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2020 // file that also contains the OS version string. Some have more than one
2021 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2022 // /etc/redhat-release.), so the order is important.
2023 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2024 // their own specific XXX-release file as well as a redhat-release file.
2025 // Because of this the XXX-release file needs to be searched for before the
2026 // redhat-release file.
2027 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2028 // search for redhat-release / SuSE-release needs to be before lsb-release.
2029 // Since the lsb-release file is the new standard it needs to be searched
2030 // before the older style release files.
2031 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2032 // next to last resort.  The os-release file is a new standard that contains
2033 // distribution information and the system-release file seems to be an old
2034 // standard that has been replaced by the lsb-release and os-release files.
2035 // Searching for the debian_version file is the last resort.  It contains
2036 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2037 // "Debian " is printed before the contents of the debian_version file.
2038 
2039 const char* distro_files[] = {
2040   "/etc/oracle-release",
2041   "/etc/mandriva-release",
2042   "/etc/mandrake-release",
2043   "/etc/sun-release",
2044   "/etc/redhat-release",
2045   "/etc/SuSE-release",
2046   "/etc/lsb-release",
2047   "/etc/turbolinux-release",
2048   "/etc/gentoo-release",
2049   "/etc/ltib-release",
2050   "/etc/angstrom-version",
2051   "/etc/system-release",
2052   "/etc/os-release",
2053   NULL };
2054 
2055 void os::Linux::print_distro_info(outputStream* st) {
2056   for (int i = 0;; i++) {
2057     const char* file = distro_files[i];
2058     if (file == NULL) {
2059       break;  // done
2060     }
2061     // If file prints, we found it.
2062     if (_print_ascii_file(file, st)) {
2063       return;
2064     }
2065   }
2066 
2067   if (file_exists("/etc/debian_version")) {
2068     st->print("Debian ");
2069     _print_ascii_file("/etc/debian_version", st);
2070   } else {
2071     st->print("Linux");
2072   }
2073   st->cr();
2074 }
2075 
2076 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2077   char buf[256];
2078   while (fgets(buf, sizeof(buf), fp)) {
2079     // Edit out extra stuff in expected format
2080     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2081       char* ptr = strstr(buf, "\"");  // the name is in quotes
2082       if (ptr != NULL) {
2083         ptr++; // go beyond first quote
2084         char* nl = strchr(ptr, '\"');
2085         if (nl != NULL) *nl = '\0';
2086         strncpy(distro, ptr, length);
2087       } else {
2088         ptr = strstr(buf, "=");
2089         ptr++; // go beyond equals then
2090         char* nl = strchr(ptr, '\n');
2091         if (nl != NULL) *nl = '\0';
2092         strncpy(distro, ptr, length);
2093       }
2094       return;
2095     } else if (get_first_line) {
2096       char* nl = strchr(buf, '\n');
2097       if (nl != NULL) *nl = '\0';
2098       strncpy(distro, buf, length);
2099       return;
2100     }
2101   }
2102   // print last line and close
2103   char* nl = strchr(buf, '\n');
2104   if (nl != NULL) *nl = '\0';
2105   strncpy(distro, buf, length);
2106 }
2107 
2108 static void parse_os_info(char* distro, size_t length, const char* file) {
2109   FILE* fp = fopen(file, "r");
2110   if (fp != NULL) {
2111     // if suse format, print out first line
2112     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2113     parse_os_info_helper(fp, distro, length, get_first_line);
2114     fclose(fp);
2115   }
2116 }
2117 
2118 void os::get_summary_os_info(char* buf, size_t buflen) {
2119   for (int i = 0;; i++) {
2120     const char* file = distro_files[i];
2121     if (file == NULL) {
2122       break; // ran out of distro_files
2123     }
2124     if (file_exists(file)) {
2125       parse_os_info(buf, buflen, file);
2126       return;
2127     }
2128   }
2129   // special case for debian
2130   if (file_exists("/etc/debian_version")) {
2131     strncpy(buf, "Debian ", buflen);
2132     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2133   } else {
2134     strncpy(buf, "Linux", buflen);
2135   }
2136 }
2137 
2138 void os::Linux::print_libversion_info(outputStream* st) {
2139   // libc, pthread
2140   st->print("libc:");
2141   st->print("%s ", os::Linux::glibc_version());
2142   st->print("%s ", os::Linux::libpthread_version());
2143   st->cr();
2144 }
2145 
2146 void os::Linux::print_full_memory_info(outputStream* st) {
2147   st->print("\n/proc/meminfo:\n");
2148   _print_ascii_file("/proc/meminfo", st);
2149   st->cr();
2150 }
2151 
2152 void os::print_memory_info(outputStream* st) {
2153 
2154   st->print("Memory:");
2155   st->print(" %dk page", os::vm_page_size()>>10);
2156 
2157   // values in struct sysinfo are "unsigned long"
2158   struct sysinfo si;
2159   sysinfo(&si);
2160 
2161   st->print(", physical " UINT64_FORMAT "k",
2162             os::physical_memory() >> 10);
2163   st->print("(" UINT64_FORMAT "k free)",
2164             os::available_memory() >> 10);
2165   st->print(", swap " UINT64_FORMAT "k",
2166             ((jlong)si.totalswap * si.mem_unit) >> 10);
2167   st->print("(" UINT64_FORMAT "k free)",
2168             ((jlong)si.freeswap * si.mem_unit) >> 10);
2169   st->cr();
2170 }
2171 
2172 // Print the first "model name" line and the first "flags" line
2173 // that we find and nothing more. We assume "model name" comes
2174 // before "flags" so if we find a second "model name", then the
2175 // "flags" field is considered missing.
2176 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2177 #if defined(IA32) || defined(AMD64)
2178   // Other platforms have less repetitive cpuinfo files
2179   FILE *fp = fopen("/proc/cpuinfo", "r");
2180   if (fp) {
2181     while (!feof(fp)) {
2182       if (fgets(buf, buflen, fp)) {
2183         // Assume model name comes before flags
2184         bool model_name_printed = false;
2185         if (strstr(buf, "model name") != NULL) {
2186           if (!model_name_printed) {
2187             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2188             st->print_raw(buf);
2189             model_name_printed = true;
2190           } else {
2191             // model name printed but not flags?  Odd, just return
2192             fclose(fp);
2193             return true;
2194           }
2195         }
2196         // print the flags line too
2197         if (strstr(buf, "flags") != NULL) {
2198           st->print_raw(buf);
2199           fclose(fp);
2200           return true;
2201         }
2202       }
2203     }
2204     fclose(fp);
2205   }
2206 #endif // x86 platforms
2207   return false;
2208 }
2209 
2210 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2211   // Only print the model name if the platform provides this as a summary
2212   if (!print_model_name_and_flags(st, buf, buflen)) {
2213     st->print("\n/proc/cpuinfo:\n");
2214     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2215       st->print_cr("  <Not Available>");
2216     }
2217   }
2218 }
2219 
2220 #if defined(AMD64) || defined(IA32) || defined(X32)
2221 const char* search_string = "model name";
2222 #elif defined(SPARC)
2223 const char* search_string = "cpu";
2224 #elif defined(PPC64)
2225 const char* search_string = "cpu";
2226 #else
2227 const char* search_string = "Processor";
2228 #endif
2229 
2230 // Parses the cpuinfo file for string representing the model name.
2231 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2232   FILE* fp = fopen("/proc/cpuinfo", "r");
2233   if (fp != NULL) {
2234     while (!feof(fp)) {
2235       char buf[256];
2236       if (fgets(buf, sizeof(buf), fp)) {
2237         char* start = strstr(buf, search_string);
2238         if (start != NULL) {
2239           char *ptr = start + strlen(search_string);
2240           char *end = buf + strlen(buf);
2241           while (ptr != end) {
2242              // skip whitespace and colon for the rest of the name.
2243              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2244                break;
2245              }
2246              ptr++;
2247           }
2248           if (ptr != end) {
2249             // reasonable string, get rid of newline and keep the rest
2250             char* nl = strchr(buf, '\n');
2251             if (nl != NULL) *nl = '\0';
2252             strncpy(cpuinfo, ptr, length);
2253             fclose(fp);
2254             return;
2255           }
2256         }
2257       }
2258     }
2259     fclose(fp);
2260   }
2261   // cpuinfo not found or parsing failed, just print generic string.  The entire
2262   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2263 #if defined(AMD64)
2264   strncpy(cpuinfo, "x86_64", length);
2265 #elif defined(IA32)
2266   strncpy(cpuinfo, "x86_32", length);
2267 #elif defined(IA64)
2268   strncpy(cpuinfo, "IA64", length);
2269 #elif defined(SPARC)
2270   strncpy(cpuinfo, "sparcv9", length);
2271 #elif defined(AARCH64)
2272   strncpy(cpuinfo, "AArch64", length);
2273 #elif defined(ARM)
2274   strncpy(cpuinfo, "ARM", length);
2275 #elif defined(PPC)
2276   strncpy(cpuinfo, "PPC64", length);
2277 #elif defined(ZERO_LIBARCH)
2278   strncpy(cpuinfo, ZERO_LIBARCH, length);
2279 #else
2280   strncpy(cpuinfo, "unknown", length);
2281 #endif
2282 }
2283 
2284 static void print_signal_handler(outputStream* st, int sig,
2285                                  char* buf, size_t buflen);
2286 
2287 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2288   st->print_cr("Signal Handlers:");
2289   print_signal_handler(st, SIGSEGV, buf, buflen);
2290   print_signal_handler(st, SIGBUS , buf, buflen);
2291   print_signal_handler(st, SIGFPE , buf, buflen);
2292   print_signal_handler(st, SIGPIPE, buf, buflen);
2293   print_signal_handler(st, SIGXFSZ, buf, buflen);
2294   print_signal_handler(st, SIGILL , buf, buflen);
2295   print_signal_handler(st, SR_signum, buf, buflen);
2296   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2297   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2298   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2299   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2300 #if defined(PPC64)
2301   print_signal_handler(st, SIGTRAP, buf, buflen);
2302 #endif
2303 }
2304 
2305 static char saved_jvm_path[MAXPATHLEN] = {0};
2306 
2307 // Find the full path to the current module, libjvm.so
2308 void os::jvm_path(char *buf, jint buflen) {
2309   // Error checking.
2310   if (buflen < MAXPATHLEN) {
2311     assert(false, "must use a large-enough buffer");
2312     buf[0] = '\0';
2313     return;
2314   }
2315   // Lazy resolve the path to current module.
2316   if (saved_jvm_path[0] != 0) {
2317     strcpy(buf, saved_jvm_path);
2318     return;
2319   }
2320 
2321   char dli_fname[MAXPATHLEN];
2322   bool ret = dll_address_to_library_name(
2323                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2324                                          dli_fname, sizeof(dli_fname), NULL);
2325   assert(ret, "cannot locate libjvm");
2326   char *rp = NULL;
2327   if (ret && dli_fname[0] != '\0') {
2328     rp = realpath(dli_fname, buf);
2329   }
2330   if (rp == NULL) {
2331     return;
2332   }
2333 
2334   if (Arguments::sun_java_launcher_is_altjvm()) {
2335     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2336     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2337     // If "/jre/lib/" appears at the right place in the string, then
2338     // assume we are installed in a JDK and we're done. Otherwise, check
2339     // for a JAVA_HOME environment variable and fix up the path so it
2340     // looks like libjvm.so is installed there (append a fake suffix
2341     // hotspot/libjvm.so).
2342     const char *p = buf + strlen(buf) - 1;
2343     for (int count = 0; p > buf && count < 5; ++count) {
2344       for (--p; p > buf && *p != '/'; --p)
2345         /* empty */ ;
2346     }
2347 
2348     if (strncmp(p, "/jre/lib/", 9) != 0) {
2349       // Look for JAVA_HOME in the environment.
2350       char* java_home_var = ::getenv("JAVA_HOME");
2351       if (java_home_var != NULL && java_home_var[0] != 0) {
2352         char* jrelib_p;
2353         int len;
2354 
2355         // Check the current module name "libjvm.so".
2356         p = strrchr(buf, '/');
2357         if (p == NULL) {
2358           return;
2359         }
2360         assert(strstr(p, "/libjvm") == p, "invalid library name");
2361 
2362         rp = realpath(java_home_var, buf);
2363         if (rp == NULL) {
2364           return;
2365         }
2366 
2367         // determine if this is a legacy image or modules image
2368         // modules image doesn't have "jre" subdirectory
2369         len = strlen(buf);
2370         assert(len < buflen, "Ran out of buffer room");
2371         jrelib_p = buf + len;
2372         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2373         if (0 != access(buf, F_OK)) {
2374           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2375         }
2376 
2377         if (0 == access(buf, F_OK)) {
2378           // Use current module name "libjvm.so"
2379           len = strlen(buf);
2380           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2381         } else {
2382           // Go back to path of .so
2383           rp = realpath(dli_fname, buf);
2384           if (rp == NULL) {
2385             return;
2386           }
2387         }
2388       }
2389     }
2390   }
2391 
2392   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2393   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2394 }
2395 
2396 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2397   // no prefix required, not even "_"
2398 }
2399 
2400 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2401   // no suffix required
2402 }
2403 
2404 ////////////////////////////////////////////////////////////////////////////////
2405 // sun.misc.Signal support
2406 
2407 static volatile jint sigint_count = 0;
2408 
2409 static void UserHandler(int sig, void *siginfo, void *context) {
2410   // 4511530 - sem_post is serialized and handled by the manager thread. When
2411   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2412   // don't want to flood the manager thread with sem_post requests.
2413   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2414     return;
2415   }
2416 
2417   // Ctrl-C is pressed during error reporting, likely because the error
2418   // handler fails to abort. Let VM die immediately.
2419   if (sig == SIGINT && is_error_reported()) {
2420     os::die();
2421   }
2422 
2423   os::signal_notify(sig);
2424 }
2425 
2426 void* os::user_handler() {
2427   return CAST_FROM_FN_PTR(void*, UserHandler);
2428 }
2429 
2430 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2431   struct timespec ts;
2432   // Semaphore's are always associated with CLOCK_REALTIME
2433   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2434   // see unpackTime for discussion on overflow checking
2435   if (sec >= MAX_SECS) {
2436     ts.tv_sec += MAX_SECS;
2437     ts.tv_nsec = 0;
2438   } else {
2439     ts.tv_sec += sec;
2440     ts.tv_nsec += nsec;
2441     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2442       ts.tv_nsec -= NANOSECS_PER_SEC;
2443       ++ts.tv_sec; // note: this must be <= max_secs
2444     }
2445   }
2446 
2447   return ts;
2448 }
2449 
2450 extern "C" {
2451   typedef void (*sa_handler_t)(int);
2452   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2453 }
2454 
2455 void* os::signal(int signal_number, void* handler) {
2456   struct sigaction sigAct, oldSigAct;
2457 
2458   sigfillset(&(sigAct.sa_mask));
2459   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2460   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2461 
2462   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2463     // -1 means registration failed
2464     return (void *)-1;
2465   }
2466 
2467   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2468 }
2469 
2470 void os::signal_raise(int signal_number) {
2471   ::raise(signal_number);
2472 }
2473 
2474 // The following code is moved from os.cpp for making this
2475 // code platform specific, which it is by its very nature.
2476 
2477 // Will be modified when max signal is changed to be dynamic
2478 int os::sigexitnum_pd() {
2479   return NSIG;
2480 }
2481 
2482 // a counter for each possible signal value
2483 static volatile jint pending_signals[NSIG+1] = { 0 };
2484 
2485 // Linux(POSIX) specific hand shaking semaphore.
2486 static sem_t sig_sem;
2487 static PosixSemaphore sr_semaphore;
2488 
2489 void os::signal_init_pd() {
2490   // Initialize signal structures
2491   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2492 
2493   // Initialize signal semaphore
2494   ::sem_init(&sig_sem, 0, 0);
2495 }
2496 
2497 void os::signal_notify(int sig) {
2498   Atomic::inc(&pending_signals[sig]);
2499   ::sem_post(&sig_sem);
2500 }
2501 
2502 static int check_pending_signals(bool wait) {
2503   Atomic::store(0, &sigint_count);
2504   for (;;) {
2505     for (int i = 0; i < NSIG + 1; i++) {
2506       jint n = pending_signals[i];
2507       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2508         return i;
2509       }
2510     }
2511     if (!wait) {
2512       return -1;
2513     }
2514     JavaThread *thread = JavaThread::current();
2515     ThreadBlockInVM tbivm(thread);
2516 
2517     bool threadIsSuspended;
2518     do {
2519       thread->set_suspend_equivalent();
2520       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2521       ::sem_wait(&sig_sem);
2522 
2523       // were we externally suspended while we were waiting?
2524       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2525       if (threadIsSuspended) {
2526         // The semaphore has been incremented, but while we were waiting
2527         // another thread suspended us. We don't want to continue running
2528         // while suspended because that would surprise the thread that
2529         // suspended us.
2530         ::sem_post(&sig_sem);
2531 
2532         thread->java_suspend_self();
2533       }
2534     } while (threadIsSuspended);
2535   }
2536 }
2537 
2538 int os::signal_lookup() {
2539   return check_pending_signals(false);
2540 }
2541 
2542 int os::signal_wait() {
2543   return check_pending_signals(true);
2544 }
2545 
2546 ////////////////////////////////////////////////////////////////////////////////
2547 // Virtual Memory
2548 
2549 int os::vm_page_size() {
2550   // Seems redundant as all get out
2551   assert(os::Linux::page_size() != -1, "must call os::init");
2552   return os::Linux::page_size();
2553 }
2554 
2555 // Solaris allocates memory by pages.
2556 int os::vm_allocation_granularity() {
2557   assert(os::Linux::page_size() != -1, "must call os::init");
2558   return os::Linux::page_size();
2559 }
2560 
2561 // Rationale behind this function:
2562 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2563 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2564 //  samples for JITted code. Here we create private executable mapping over the code cache
2565 //  and then we can use standard (well, almost, as mapping can change) way to provide
2566 //  info for the reporting script by storing timestamp and location of symbol
2567 void linux_wrap_code(char* base, size_t size) {
2568   static volatile jint cnt = 0;
2569 
2570   if (!UseOprofile) {
2571     return;
2572   }
2573 
2574   char buf[PATH_MAX+1];
2575   int num = Atomic::add(1, &cnt);
2576 
2577   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2578            os::get_temp_directory(), os::current_process_id(), num);
2579   unlink(buf);
2580 
2581   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2582 
2583   if (fd != -1) {
2584     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2585     if (rv != (off_t)-1) {
2586       if (::write(fd, "", 1) == 1) {
2587         mmap(base, size,
2588              PROT_READ|PROT_WRITE|PROT_EXEC,
2589              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2590       }
2591     }
2592     ::close(fd);
2593     unlink(buf);
2594   }
2595 }
2596 
2597 static bool recoverable_mmap_error(int err) {
2598   // See if the error is one we can let the caller handle. This
2599   // list of errno values comes from JBS-6843484. I can't find a
2600   // Linux man page that documents this specific set of errno
2601   // values so while this list currently matches Solaris, it may
2602   // change as we gain experience with this failure mode.
2603   switch (err) {
2604   case EBADF:
2605   case EINVAL:
2606   case ENOTSUP:
2607     // let the caller deal with these errors
2608     return true;
2609 
2610   default:
2611     // Any remaining errors on this OS can cause our reserved mapping
2612     // to be lost. That can cause confusion where different data
2613     // structures think they have the same memory mapped. The worst
2614     // scenario is if both the VM and a library think they have the
2615     // same memory mapped.
2616     return false;
2617   }
2618 }
2619 
2620 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2621                                     int err) {
2622   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2623           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2624           os::strerror(err), err);
2625 }
2626 
2627 static void warn_fail_commit_memory(char* addr, size_t size,
2628                                     size_t alignment_hint, bool exec,
2629                                     int err) {
2630   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2631           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2632           alignment_hint, exec, os::strerror(err), err);
2633 }
2634 
2635 // NOTE: Linux kernel does not really reserve the pages for us.
2636 //       All it does is to check if there are enough free pages
2637 //       left at the time of mmap(). This could be a potential
2638 //       problem.
2639 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2640   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2641   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2642                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2643   if (res != (uintptr_t) MAP_FAILED) {
2644     if (UseNUMAInterleaving) {
2645       numa_make_global(addr, size);
2646     }
2647     return 0;
2648   }
2649 
2650   int err = errno;  // save errno from mmap() call above
2651 
2652   if (!recoverable_mmap_error(err)) {
2653     warn_fail_commit_memory(addr, size, exec, err);
2654     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2655   }
2656 
2657   return err;
2658 }
2659 
2660 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2661   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2662 }
2663 
2664 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2665                                   const char* mesg) {
2666   assert(mesg != NULL, "mesg must be specified");
2667   int err = os::Linux::commit_memory_impl(addr, size, exec);
2668   if (err != 0) {
2669     // the caller wants all commit errors to exit with the specified mesg:
2670     warn_fail_commit_memory(addr, size, exec, err);
2671     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2672   }
2673 }
2674 
2675 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2676 #ifndef MAP_HUGETLB
2677   #define MAP_HUGETLB 0x40000
2678 #endif
2679 
2680 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2681 #ifndef MADV_HUGEPAGE
2682   #define MADV_HUGEPAGE 14
2683 #endif
2684 
2685 int os::Linux::commit_memory_impl(char* addr, size_t size,
2686                                   size_t alignment_hint, bool exec) {
2687   int err = os::Linux::commit_memory_impl(addr, size, exec);
2688   if (err == 0) {
2689     realign_memory(addr, size, alignment_hint);
2690   }
2691   return err;
2692 }
2693 
2694 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2695                           bool exec) {
2696   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2697 }
2698 
2699 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2700                                   size_t alignment_hint, bool exec,
2701                                   const char* mesg) {
2702   assert(mesg != NULL, "mesg must be specified");
2703   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2704   if (err != 0) {
2705     // the caller wants all commit errors to exit with the specified mesg:
2706     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2707     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2708   }
2709 }
2710 
2711 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2712   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2713     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2714     // be supported or the memory may already be backed by huge pages.
2715     ::madvise(addr, bytes, MADV_HUGEPAGE);
2716   }
2717 }
2718 
2719 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2720   // This method works by doing an mmap over an existing mmaping and effectively discarding
2721   // the existing pages. However it won't work for SHM-based large pages that cannot be
2722   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2723   // small pages on top of the SHM segment. This method always works for small pages, so we
2724   // allow that in any case.
2725   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2726     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2727   }
2728 }
2729 
2730 void os::numa_make_global(char *addr, size_t bytes) {
2731   Linux::numa_interleave_memory(addr, bytes);
2732 }
2733 
2734 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2735 // bind policy to MPOL_PREFERRED for the current thread.
2736 #define USE_MPOL_PREFERRED 0
2737 
2738 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2739   // To make NUMA and large pages more robust when both enabled, we need to ease
2740   // the requirements on where the memory should be allocated. MPOL_BIND is the
2741   // default policy and it will force memory to be allocated on the specified
2742   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2743   // the specified node, but will not force it. Using this policy will prevent
2744   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2745   // free large pages.
2746   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2747   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2748 }
2749 
2750 bool os::numa_topology_changed() { return false; }
2751 
2752 size_t os::numa_get_groups_num() {
2753   int max_node = Linux::numa_max_node();
2754   return max_node > 0 ? max_node + 1 : 1;
2755 }
2756 
2757 int os::numa_get_group_id() {
2758   int cpu_id = Linux::sched_getcpu();
2759   if (cpu_id != -1) {
2760     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2761     if (lgrp_id != -1) {
2762       return lgrp_id;
2763     }
2764   }
2765   return 0;
2766 }
2767 
2768 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2769   for (size_t i = 0; i < size; i++) {
2770     ids[i] = i;
2771   }
2772   return size;
2773 }
2774 
2775 bool os::get_page_info(char *start, page_info* info) {
2776   return false;
2777 }
2778 
2779 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2780                      page_info* page_found) {
2781   return end;
2782 }
2783 
2784 
2785 int os::Linux::sched_getcpu_syscall(void) {
2786   unsigned int cpu = 0;
2787   int retval = -1;
2788 
2789 #if defined(IA32)
2790   #ifndef SYS_getcpu
2791     #define SYS_getcpu 318
2792   #endif
2793   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2794 #elif defined(AMD64)
2795 // Unfortunately we have to bring all these macros here from vsyscall.h
2796 // to be able to compile on old linuxes.
2797   #define __NR_vgetcpu 2
2798   #define VSYSCALL_START (-10UL << 20)
2799   #define VSYSCALL_SIZE 1024
2800   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2801   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2802   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2803   retval = vgetcpu(&cpu, NULL, NULL);
2804 #endif
2805 
2806   return (retval == -1) ? retval : cpu;
2807 }
2808 
2809 // Something to do with the numa-aware allocator needs these symbols
2810 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2811 extern "C" JNIEXPORT void numa_error(char *where) { }
2812 
2813 
2814 // If we are running with libnuma version > 2, then we should
2815 // be trying to use symbols with versions 1.1
2816 // If we are running with earlier version, which did not have symbol versions,
2817 // we should use the base version.
2818 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2819   void *f = dlvsym(handle, name, "libnuma_1.1");
2820   if (f == NULL) {
2821     f = dlsym(handle, name);
2822   }
2823   return f;
2824 }
2825 
2826 bool os::Linux::libnuma_init() {
2827   // sched_getcpu() should be in libc.
2828   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2829                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2830 
2831   // If it's not, try a direct syscall.
2832   if (sched_getcpu() == -1) {
2833     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2834                                     (void*)&sched_getcpu_syscall));
2835   }
2836 
2837   if (sched_getcpu() != -1) { // Does it work?
2838     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2839     if (handle != NULL) {
2840       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2841                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2842       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2843                                        libnuma_dlsym(handle, "numa_max_node")));
2844       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2845                                         libnuma_dlsym(handle, "numa_available")));
2846       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2847                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2848       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2849                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2850       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2851                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2852 
2853 
2854       if (numa_available() != -1) {
2855         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2856         // Create a cpu -> node mapping
2857         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2858         rebuild_cpu_to_node_map();
2859         return true;
2860       }
2861     }
2862   }
2863   return false;
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 = os::active_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 = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3273                      ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3274 #endif // ZERO
3275 
3276   FILE *fp = fopen("/proc/meminfo", "r");
3277   if (fp) {
3278     while (!feof(fp)) {
3279       int x = 0;
3280       char buf[16];
3281       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3282         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3283           large_page_size = x * K;
3284           break;
3285         }
3286       } else {
3287         // skip to next line
3288         for (;;) {
3289           int ch = fgetc(fp);
3290           if (ch == EOF || ch == (int)'\n') break;
3291         }
3292       }
3293     }
3294     fclose(fp);
3295   }
3296 
3297   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3298     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3299             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3300             proper_unit_for_byte_size(large_page_size));
3301   }
3302 
3303   return large_page_size;
3304 }
3305 
3306 size_t os::Linux::setup_large_page_size() {
3307   _large_page_size = Linux::find_large_page_size();
3308   const size_t default_page_size = (size_t)Linux::page_size();
3309   if (_large_page_size > default_page_size) {
3310     _page_sizes[0] = _large_page_size;
3311     _page_sizes[1] = default_page_size;
3312     _page_sizes[2] = 0;
3313   }
3314 
3315   return _large_page_size;
3316 }
3317 
3318 bool os::Linux::setup_large_page_type(size_t page_size) {
3319   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3320       FLAG_IS_DEFAULT(UseSHM) &&
3321       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3322 
3323     // The type of large pages has not been specified by the user.
3324 
3325     // Try UseHugeTLBFS and then UseSHM.
3326     UseHugeTLBFS = UseSHM = true;
3327 
3328     // Don't try UseTransparentHugePages since there are known
3329     // performance issues with it turned on. This might change in the future.
3330     UseTransparentHugePages = false;
3331   }
3332 
3333   if (UseTransparentHugePages) {
3334     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3335     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3336       UseHugeTLBFS = false;
3337       UseSHM = false;
3338       return true;
3339     }
3340     UseTransparentHugePages = false;
3341   }
3342 
3343   if (UseHugeTLBFS) {
3344     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3345     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3346       UseSHM = false;
3347       return true;
3348     }
3349     UseHugeTLBFS = false;
3350   }
3351 
3352   return UseSHM;
3353 }
3354 
3355 void os::large_page_init() {
3356   if (!UseLargePages &&
3357       !UseTransparentHugePages &&
3358       !UseHugeTLBFS &&
3359       !UseSHM) {
3360     // Not using large pages.
3361     return;
3362   }
3363 
3364   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3365     // The user explicitly turned off large pages.
3366     // Ignore the rest of the large pages flags.
3367     UseTransparentHugePages = false;
3368     UseHugeTLBFS = false;
3369     UseSHM = false;
3370     return;
3371   }
3372 
3373   size_t large_page_size = Linux::setup_large_page_size();
3374   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3375 
3376   set_coredump_filter();
3377 }
3378 
3379 #ifndef SHM_HUGETLB
3380   #define SHM_HUGETLB 04000
3381 #endif
3382 
3383 #define shm_warning_format(format, ...)              \
3384   do {                                               \
3385     if (UseLargePages &&                             \
3386         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3387          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3388          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3389       warning(format, __VA_ARGS__);                  \
3390     }                                                \
3391   } while (0)
3392 
3393 #define shm_warning(str) shm_warning_format("%s", str)
3394 
3395 #define shm_warning_with_errno(str)                \
3396   do {                                             \
3397     int err = errno;                               \
3398     shm_warning_format(str " (error = %d)", err);  \
3399   } while (0)
3400 
3401 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3402   assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3403 
3404   if (!is_size_aligned(alignment, SHMLBA)) {
3405     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3406     return NULL;
3407   }
3408 
3409   // To ensure that we get 'alignment' aligned memory from shmat,
3410   // we pre-reserve aligned virtual memory and then attach to that.
3411 
3412   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3413   if (pre_reserved_addr == NULL) {
3414     // Couldn't pre-reserve aligned memory.
3415     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3416     return NULL;
3417   }
3418 
3419   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3420   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3421 
3422   if ((intptr_t)addr == -1) {
3423     int err = errno;
3424     shm_warning_with_errno("Failed to attach shared memory.");
3425 
3426     assert(err != EACCES, "Unexpected error");
3427     assert(err != EIDRM,  "Unexpected error");
3428     assert(err != EINVAL, "Unexpected error");
3429 
3430     // Since we don't know if the kernel unmapped the pre-reserved memory area
3431     // we can't unmap it, since that would potentially unmap memory that was
3432     // mapped from other threads.
3433     return NULL;
3434   }
3435 
3436   return addr;
3437 }
3438 
3439 static char* shmat_at_address(int shmid, char* req_addr) {
3440   if (!is_ptr_aligned(req_addr, SHMLBA)) {
3441     assert(false, "Requested address needs to be SHMLBA aligned");
3442     return NULL;
3443   }
3444 
3445   char* addr = (char*)shmat(shmid, req_addr, 0);
3446 
3447   if ((intptr_t)addr == -1) {
3448     shm_warning_with_errno("Failed to attach shared memory.");
3449     return NULL;
3450   }
3451 
3452   return addr;
3453 }
3454 
3455 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3456   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3457   if (req_addr != NULL) {
3458     assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3459     assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3460     return shmat_at_address(shmid, req_addr);
3461   }
3462 
3463   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3464   // return large page size aligned memory addresses when req_addr == NULL.
3465   // However, if the alignment is larger than the large page size, we have
3466   // to manually ensure that the memory returned is 'alignment' aligned.
3467   if (alignment > os::large_page_size()) {
3468     assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3469     return shmat_with_alignment(shmid, bytes, alignment);
3470   } else {
3471     return shmat_at_address(shmid, NULL);
3472   }
3473 }
3474 
3475 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3476                                             char* req_addr, bool exec) {
3477   // "exec" is passed in but not used.  Creating the shared image for
3478   // the code cache doesn't have an SHM_X executable permission to check.
3479   assert(UseLargePages && UseSHM, "only for SHM large pages");
3480   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3481   assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3482 
3483   if (!is_size_aligned(bytes, os::large_page_size())) {
3484     return NULL; // Fallback to small pages.
3485   }
3486 
3487   // Create a large shared memory region to attach to based on size.
3488   // Currently, size is the total size of the heap.
3489   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3490   if (shmid == -1) {
3491     // Possible reasons for shmget failure:
3492     // 1. shmmax is too small for Java heap.
3493     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3494     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3495     // 2. not enough large page memory.
3496     //    > check available large pages: cat /proc/meminfo
3497     //    > increase amount of large pages:
3498     //          echo new_value > /proc/sys/vm/nr_hugepages
3499     //      Note 1: different Linux may use different name for this property,
3500     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3501     //      Note 2: it's possible there's enough physical memory available but
3502     //            they are so fragmented after a long run that they can't
3503     //            coalesce into large pages. Try to reserve large pages when
3504     //            the system is still "fresh".
3505     shm_warning_with_errno("Failed to reserve shared memory.");
3506     return NULL;
3507   }
3508 
3509   // Attach to the region.
3510   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3511 
3512   // Remove shmid. If shmat() is successful, the actual shared memory segment
3513   // will be deleted when it's detached by shmdt() or when the process
3514   // terminates. If shmat() is not successful this will remove the shared
3515   // segment immediately.
3516   shmctl(shmid, IPC_RMID, NULL);
3517 
3518   return addr;
3519 }
3520 
3521 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3522                                         int error) {
3523   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3524 
3525   bool warn_on_failure = UseLargePages &&
3526       (!FLAG_IS_DEFAULT(UseLargePages) ||
3527        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3528        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3529 
3530   if (warn_on_failure) {
3531     char msg[128];
3532     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3533                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3534     warning("%s", msg);
3535   }
3536 }
3537 
3538 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3539                                                         char* req_addr,
3540                                                         bool exec) {
3541   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3542   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3543   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3544 
3545   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3546   char* addr = (char*)::mmap(req_addr, bytes, prot,
3547                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3548                              -1, 0);
3549 
3550   if (addr == MAP_FAILED) {
3551     warn_on_large_pages_failure(req_addr, bytes, errno);
3552     return NULL;
3553   }
3554 
3555   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3556 
3557   return addr;
3558 }
3559 
3560 // Reserve memory using mmap(MAP_HUGETLB).
3561 //  - bytes shall be a multiple of alignment.
3562 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3563 //  - alignment sets the alignment at which memory shall be allocated.
3564 //     It must be a multiple of allocation granularity.
3565 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3566 //  req_addr or NULL.
3567 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3568                                                          size_t alignment,
3569                                                          char* req_addr,
3570                                                          bool exec) {
3571   size_t large_page_size = os::large_page_size();
3572   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3573 
3574   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3575   assert(is_size_aligned(bytes, alignment), "Must be");
3576 
3577   // First reserve - but not commit - the address range in small pages.
3578   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3579 
3580   if (start == NULL) {
3581     return NULL;
3582   }
3583 
3584   assert(is_ptr_aligned(start, alignment), "Must be");
3585 
3586   char* end = start + bytes;
3587 
3588   // Find the regions of the allocated chunk that can be promoted to large pages.
3589   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3590   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3591 
3592   size_t lp_bytes = lp_end - lp_start;
3593 
3594   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3595 
3596   if (lp_bytes == 0) {
3597     // The mapped region doesn't even span the start and the end of a large page.
3598     // Fall back to allocate a non-special area.
3599     ::munmap(start, end - start);
3600     return NULL;
3601   }
3602 
3603   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3604 
3605   void* result;
3606 
3607   // Commit small-paged leading area.
3608   if (start != lp_start) {
3609     result = ::mmap(start, lp_start - start, prot,
3610                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3611                     -1, 0);
3612     if (result == MAP_FAILED) {
3613       ::munmap(lp_start, end - lp_start);
3614       return NULL;
3615     }
3616   }
3617 
3618   // Commit large-paged area.
3619   result = ::mmap(lp_start, lp_bytes, prot,
3620                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3621                   -1, 0);
3622   if (result == MAP_FAILED) {
3623     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3624     // If the mmap above fails, the large pages region will be unmapped and we
3625     // have regions before and after with small pages. Release these regions.
3626     //
3627     // |  mapped  |  unmapped  |  mapped  |
3628     // ^          ^            ^          ^
3629     // start      lp_start     lp_end     end
3630     //
3631     ::munmap(start, lp_start - start);
3632     ::munmap(lp_end, end - lp_end);
3633     return NULL;
3634   }
3635 
3636   // Commit small-paged trailing area.
3637   if (lp_end != end) {
3638     result = ::mmap(lp_end, end - lp_end, prot,
3639                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3640                     -1, 0);
3641     if (result == MAP_FAILED) {
3642       ::munmap(start, lp_end - start);
3643       return NULL;
3644     }
3645   }
3646 
3647   return start;
3648 }
3649 
3650 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3651                                                    size_t alignment,
3652                                                    char* req_addr,
3653                                                    bool exec) {
3654   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3655   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3656   assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3657   assert(is_power_of_2(os::large_page_size()), "Must be");
3658   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3659 
3660   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3661     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3662   } else {
3663     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3664   }
3665 }
3666 
3667 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3668                                  char* req_addr, bool exec) {
3669   assert(UseLargePages, "only for large pages");
3670 
3671   char* addr;
3672   if (UseSHM) {
3673     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3674   } else {
3675     assert(UseHugeTLBFS, "must be");
3676     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3677   }
3678 
3679   if (addr != NULL) {
3680     if (UseNUMAInterleaving) {
3681       numa_make_global(addr, bytes);
3682     }
3683 
3684     // The memory is committed
3685     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3686   }
3687 
3688   return addr;
3689 }
3690 
3691 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3692   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3693   return shmdt(base) == 0;
3694 }
3695 
3696 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3697   return pd_release_memory(base, bytes);
3698 }
3699 
3700 bool os::release_memory_special(char* base, size_t bytes) {
3701   bool res;
3702   if (MemTracker::tracking_level() > NMT_minimal) {
3703     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3704     res = os::Linux::release_memory_special_impl(base, bytes);
3705     if (res) {
3706       tkr.record((address)base, bytes);
3707     }
3708 
3709   } else {
3710     res = os::Linux::release_memory_special_impl(base, bytes);
3711   }
3712   return res;
3713 }
3714 
3715 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3716   assert(UseLargePages, "only for large pages");
3717   bool res;
3718 
3719   if (UseSHM) {
3720     res = os::Linux::release_memory_special_shm(base, bytes);
3721   } else {
3722     assert(UseHugeTLBFS, "must be");
3723     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3724   }
3725   return res;
3726 }
3727 
3728 size_t os::large_page_size() {
3729   return _large_page_size;
3730 }
3731 
3732 // With SysV SHM the entire memory region must be allocated as shared
3733 // memory.
3734 // HugeTLBFS allows application to commit large page memory on demand.
3735 // However, when committing memory with HugeTLBFS fails, the region
3736 // that was supposed to be committed will lose the old reservation
3737 // and allow other threads to steal that memory region. Because of this
3738 // behavior we can't commit HugeTLBFS memory.
3739 bool os::can_commit_large_page_memory() {
3740   return UseTransparentHugePages;
3741 }
3742 
3743 bool os::can_execute_large_page_memory() {
3744   return UseTransparentHugePages || UseHugeTLBFS;
3745 }
3746 
3747 // Reserve memory at an arbitrary address, only if that area is
3748 // available (and not reserved for something else).
3749 
3750 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3751   const int max_tries = 10;
3752   char* base[max_tries];
3753   size_t size[max_tries];
3754   const size_t gap = 0x000000;
3755 
3756   // Assert only that the size is a multiple of the page size, since
3757   // that's all that mmap requires, and since that's all we really know
3758   // about at this low abstraction level.  If we need higher alignment,
3759   // we can either pass an alignment to this method or verify alignment
3760   // in one of the methods further up the call chain.  See bug 5044738.
3761   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3762 
3763   // Repeatedly allocate blocks until the block is allocated at the
3764   // right spot.
3765 
3766   // Linux mmap allows caller to pass an address as hint; give it a try first,
3767   // if kernel honors the hint then we can return immediately.
3768   char * addr = anon_mmap(requested_addr, bytes, false);
3769   if (addr == requested_addr) {
3770     return requested_addr;
3771   }
3772 
3773   if (addr != NULL) {
3774     // mmap() is successful but it fails to reserve at the requested address
3775     anon_munmap(addr, bytes);
3776   }
3777 
3778   int i;
3779   for (i = 0; i < max_tries; ++i) {
3780     base[i] = reserve_memory(bytes);
3781 
3782     if (base[i] != NULL) {
3783       // Is this the block we wanted?
3784       if (base[i] == requested_addr) {
3785         size[i] = bytes;
3786         break;
3787       }
3788 
3789       // Does this overlap the block we wanted? Give back the overlapped
3790       // parts and try again.
3791 
3792       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3793       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3794         unmap_memory(base[i], top_overlap);
3795         base[i] += top_overlap;
3796         size[i] = bytes - top_overlap;
3797       } else {
3798         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3799         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3800           unmap_memory(requested_addr, bottom_overlap);
3801           size[i] = bytes - bottom_overlap;
3802         } else {
3803           size[i] = bytes;
3804         }
3805       }
3806     }
3807   }
3808 
3809   // Give back the unused reserved pieces.
3810 
3811   for (int j = 0; j < i; ++j) {
3812     if (base[j] != NULL) {
3813       unmap_memory(base[j], size[j]);
3814     }
3815   }
3816 
3817   if (i < max_tries) {
3818     return requested_addr;
3819   } else {
3820     return NULL;
3821   }
3822 }
3823 
3824 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3825   return ::read(fd, buf, nBytes);
3826 }
3827 
3828 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3829   return ::pread(fd, buf, nBytes, offset);
3830 }
3831 
3832 // Short sleep, direct OS call.
3833 //
3834 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3835 // sched_yield(2) will actually give up the CPU:
3836 //
3837 //   * Alone on this pariticular CPU, keeps running.
3838 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3839 //     (pre 2.6.39).
3840 //
3841 // So calling this with 0 is an alternative.
3842 //
3843 void os::naked_short_sleep(jlong ms) {
3844   struct timespec req;
3845 
3846   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3847   req.tv_sec = 0;
3848   if (ms > 0) {
3849     req.tv_nsec = (ms % 1000) * 1000000;
3850   } else {
3851     req.tv_nsec = 1;
3852   }
3853 
3854   nanosleep(&req, NULL);
3855 
3856   return;
3857 }
3858 
3859 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3860 void os::infinite_sleep() {
3861   while (true) {    // sleep forever ...
3862     ::sleep(100);   // ... 100 seconds at a time
3863   }
3864 }
3865 
3866 // Used to convert frequent JVM_Yield() to nops
3867 bool os::dont_yield() {
3868   return DontYieldALot;
3869 }
3870 
3871 void os::naked_yield() {
3872   sched_yield();
3873 }
3874 
3875 ////////////////////////////////////////////////////////////////////////////////
3876 // thread priority support
3877 
3878 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3879 // only supports dynamic priority, static priority must be zero. For real-time
3880 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3881 // However, for large multi-threaded applications, SCHED_RR is not only slower
3882 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3883 // of 5 runs - Sep 2005).
3884 //
3885 // The following code actually changes the niceness of kernel-thread/LWP. It
3886 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3887 // not the entire user process, and user level threads are 1:1 mapped to kernel
3888 // threads. It has always been the case, but could change in the future. For
3889 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3890 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3891 
3892 int os::java_to_os_priority[CriticalPriority + 1] = {
3893   19,              // 0 Entry should never be used
3894 
3895    4,              // 1 MinPriority
3896    3,              // 2
3897    2,              // 3
3898 
3899    1,              // 4
3900    0,              // 5 NormPriority
3901   -1,              // 6
3902 
3903   -2,              // 7
3904   -3,              // 8
3905   -4,              // 9 NearMaxPriority
3906 
3907   -5,              // 10 MaxPriority
3908 
3909   -5               // 11 CriticalPriority
3910 };
3911 
3912 static int prio_init() {
3913   if (ThreadPriorityPolicy == 1) {
3914     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3915     // if effective uid is not root. Perhaps, a more elegant way of doing
3916     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3917     if (geteuid() != 0) {
3918       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3919         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3920       }
3921       ThreadPriorityPolicy = 0;
3922     }
3923   }
3924   if (UseCriticalJavaThreadPriority) {
3925     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3926   }
3927   return 0;
3928 }
3929 
3930 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3931   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3932 
3933   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3934   return (ret == 0) ? OS_OK : OS_ERR;
3935 }
3936 
3937 OSReturn os::get_native_priority(const Thread* const thread,
3938                                  int *priority_ptr) {
3939   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3940     *priority_ptr = java_to_os_priority[NormPriority];
3941     return OS_OK;
3942   }
3943 
3944   errno = 0;
3945   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3946   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3947 }
3948 
3949 // Hint to the underlying OS that a task switch would not be good.
3950 // Void return because it's a hint and can fail.
3951 void os::hint_no_preempt() {}
3952 
3953 ////////////////////////////////////////////////////////////////////////////////
3954 // suspend/resume support
3955 
3956 //  the low-level signal-based suspend/resume support is a remnant from the
3957 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3958 //  within hotspot. Now there is a single use-case for this:
3959 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
3960 //      that runs in the watcher thread.
3961 //  The remaining code is greatly simplified from the more general suspension
3962 //  code that used to be used.
3963 //
3964 //  The protocol is quite simple:
3965 //  - suspend:
3966 //      - sends a signal to the target thread
3967 //      - polls the suspend state of the osthread using a yield loop
3968 //      - target thread signal handler (SR_handler) sets suspend state
3969 //        and blocks in sigsuspend until continued
3970 //  - resume:
3971 //      - sets target osthread state to continue
3972 //      - sends signal to end the sigsuspend loop in the SR_handler
3973 //
3974 //  Note that the SR_lock plays no role in this suspend/resume protocol,
3975 //  but is checked for NULL in SR_handler as a thread termination indicator.
3976 
3977 static void resume_clear_context(OSThread *osthread) {
3978   osthread->set_ucontext(NULL);
3979   osthread->set_siginfo(NULL);
3980 }
3981 
3982 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3983                                  ucontext_t* context) {
3984   osthread->set_ucontext(context);
3985   osthread->set_siginfo(siginfo);
3986 }
3987 
3988 // Handler function invoked when a thread's execution is suspended or
3989 // resumed. We have to be careful that only async-safe functions are
3990 // called here (Note: most pthread functions are not async safe and
3991 // should be avoided.)
3992 //
3993 // Note: sigwait() is a more natural fit than sigsuspend() from an
3994 // interface point of view, but sigwait() prevents the signal hander
3995 // from being run. libpthread would get very confused by not having
3996 // its signal handlers run and prevents sigwait()'s use with the
3997 // mutex granting granting signal.
3998 //
3999 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4000 //
4001 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4002   // Save and restore errno to avoid confusing native code with EINTR
4003   // after sigsuspend.
4004   int old_errno = errno;
4005 
4006   Thread* thread = Thread::current_or_null_safe();
4007   assert(thread != NULL, "Missing current thread in SR_handler");
4008 
4009   // On some systems we have seen signal delivery get "stuck" until the signal
4010   // mask is changed as part of thread termination. Check that the current thread
4011   // has not already terminated (via SR_lock()) - else the following assertion
4012   // will fail because the thread is no longer a JavaThread as the ~JavaThread
4013   // destructor has completed.
4014 
4015   if (thread->SR_lock() == NULL) {
4016     return;
4017   }
4018 
4019   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4020 
4021   OSThread* osthread = thread->osthread();
4022 
4023   os::SuspendResume::State current = osthread->sr.state();
4024   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4025     suspend_save_context(osthread, siginfo, context);
4026 
4027     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4028     os::SuspendResume::State state = osthread->sr.suspended();
4029     if (state == os::SuspendResume::SR_SUSPENDED) {
4030       sigset_t suspend_set;  // signals for sigsuspend()
4031       sigemptyset(&suspend_set);
4032       // get current set of blocked signals and unblock resume signal
4033       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4034       sigdelset(&suspend_set, SR_signum);
4035 
4036       sr_semaphore.signal();
4037       // wait here until we are resumed
4038       while (1) {
4039         sigsuspend(&suspend_set);
4040 
4041         os::SuspendResume::State result = osthread->sr.running();
4042         if (result == os::SuspendResume::SR_RUNNING) {
4043           sr_semaphore.signal();
4044           break;
4045         }
4046       }
4047 
4048     } else if (state == os::SuspendResume::SR_RUNNING) {
4049       // request was cancelled, continue
4050     } else {
4051       ShouldNotReachHere();
4052     }
4053 
4054     resume_clear_context(osthread);
4055   } else if (current == os::SuspendResume::SR_RUNNING) {
4056     // request was cancelled, continue
4057   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4058     // ignore
4059   } else {
4060     // ignore
4061   }
4062 
4063   errno = old_errno;
4064 }
4065 
4066 static int SR_initialize() {
4067   struct sigaction act;
4068   char *s;
4069 
4070   // Get signal number to use for suspend/resume
4071   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4072     int sig = ::strtol(s, 0, 10);
4073     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4074         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4075       SR_signum = sig;
4076     } else {
4077       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4078               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4079     }
4080   }
4081 
4082   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4083          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4084 
4085   sigemptyset(&SR_sigset);
4086   sigaddset(&SR_sigset, SR_signum);
4087 
4088   // Set up signal handler for suspend/resume
4089   act.sa_flags = SA_RESTART|SA_SIGINFO;
4090   act.sa_handler = (void (*)(int)) SR_handler;
4091 
4092   // SR_signum is blocked by default.
4093   // 4528190 - We also need to block pthread restart signal (32 on all
4094   // supported Linux platforms). Note that LinuxThreads need to block
4095   // this signal for all threads to work properly. So we don't have
4096   // to use hard-coded signal number when setting up the mask.
4097   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4098 
4099   if (sigaction(SR_signum, &act, 0) == -1) {
4100     return -1;
4101   }
4102 
4103   // Save signal flag
4104   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4105   return 0;
4106 }
4107 
4108 static int sr_notify(OSThread* osthread) {
4109   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4110   assert_status(status == 0, status, "pthread_kill");
4111   return status;
4112 }
4113 
4114 // "Randomly" selected value for how long we want to spin
4115 // before bailing out on suspending a thread, also how often
4116 // we send a signal to a thread we want to resume
4117 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4118 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4119 
4120 // returns true on success and false on error - really an error is fatal
4121 // but this seems the normal response to library errors
4122 static bool do_suspend(OSThread* osthread) {
4123   assert(osthread->sr.is_running(), "thread should be running");
4124   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4125 
4126   // mark as suspended and send signal
4127   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4128     // failed to switch, state wasn't running?
4129     ShouldNotReachHere();
4130     return false;
4131   }
4132 
4133   if (sr_notify(osthread) != 0) {
4134     ShouldNotReachHere();
4135   }
4136 
4137   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4138   while (true) {
4139     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4140       break;
4141     } else {
4142       // timeout
4143       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4144       if (cancelled == os::SuspendResume::SR_RUNNING) {
4145         return false;
4146       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4147         // make sure that we consume the signal on the semaphore as well
4148         sr_semaphore.wait();
4149         break;
4150       } else {
4151         ShouldNotReachHere();
4152         return false;
4153       }
4154     }
4155   }
4156 
4157   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4158   return true;
4159 }
4160 
4161 static void do_resume(OSThread* osthread) {
4162   assert(osthread->sr.is_suspended(), "thread should be suspended");
4163   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4164 
4165   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4166     // failed to switch to WAKEUP_REQUEST
4167     ShouldNotReachHere();
4168     return;
4169   }
4170 
4171   while (true) {
4172     if (sr_notify(osthread) == 0) {
4173       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4174         if (osthread->sr.is_running()) {
4175           return;
4176         }
4177       }
4178     } else {
4179       ShouldNotReachHere();
4180     }
4181   }
4182 
4183   guarantee(osthread->sr.is_running(), "Must be running!");
4184 }
4185 
4186 ///////////////////////////////////////////////////////////////////////////////////
4187 // signal handling (except suspend/resume)
4188 
4189 // This routine may be used by user applications as a "hook" to catch signals.
4190 // The user-defined signal handler must pass unrecognized signals to this
4191 // routine, and if it returns true (non-zero), then the signal handler must
4192 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4193 // routine will never retun false (zero), but instead will execute a VM panic
4194 // routine kill the process.
4195 //
4196 // If this routine returns false, it is OK to call it again.  This allows
4197 // the user-defined signal handler to perform checks either before or after
4198 // the VM performs its own checks.  Naturally, the user code would be making
4199 // a serious error if it tried to handle an exception (such as a null check
4200 // or breakpoint) that the VM was generating for its own correct operation.
4201 //
4202 // This routine may recognize any of the following kinds of signals:
4203 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4204 // It should be consulted by handlers for any of those signals.
4205 //
4206 // The caller of this routine must pass in the three arguments supplied
4207 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4208 // field of the structure passed to sigaction().  This routine assumes that
4209 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4210 //
4211 // Note that the VM will print warnings if it detects conflicting signal
4212 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4213 //
4214 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4215                                                  siginfo_t* siginfo,
4216                                                  void* ucontext,
4217                                                  int abort_if_unrecognized);
4218 
4219 void signalHandler(int sig, siginfo_t* info, void* uc) {
4220   assert(info != NULL && uc != NULL, "it must be old kernel");
4221   int orig_errno = errno;  // Preserve errno value over signal handler.
4222   JVM_handle_linux_signal(sig, info, uc, true);
4223   errno = orig_errno;
4224 }
4225 
4226 
4227 // This boolean allows users to forward their own non-matching signals
4228 // to JVM_handle_linux_signal, harmlessly.
4229 bool os::Linux::signal_handlers_are_installed = false;
4230 
4231 // For signal-chaining
4232 struct sigaction sigact[NSIG];
4233 uint64_t sigs = 0;
4234 #if (64 < NSIG-1)
4235 #error "Not all signals can be encoded in sigs. Adapt its type!"
4236 #endif
4237 bool os::Linux::libjsig_is_loaded = false;
4238 typedef struct sigaction *(*get_signal_t)(int);
4239 get_signal_t os::Linux::get_signal_action = NULL;
4240 
4241 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4242   struct sigaction *actp = NULL;
4243 
4244   if (libjsig_is_loaded) {
4245     // Retrieve the old signal handler from libjsig
4246     actp = (*get_signal_action)(sig);
4247   }
4248   if (actp == NULL) {
4249     // Retrieve the preinstalled signal handler from jvm
4250     actp = get_preinstalled_handler(sig);
4251   }
4252 
4253   return actp;
4254 }
4255 
4256 static bool call_chained_handler(struct sigaction *actp, int sig,
4257                                  siginfo_t *siginfo, void *context) {
4258   // Call the old signal handler
4259   if (actp->sa_handler == SIG_DFL) {
4260     // It's more reasonable to let jvm treat it as an unexpected exception
4261     // instead of taking the default action.
4262     return false;
4263   } else if (actp->sa_handler != SIG_IGN) {
4264     if ((actp->sa_flags & SA_NODEFER) == 0) {
4265       // automaticlly block the signal
4266       sigaddset(&(actp->sa_mask), sig);
4267     }
4268 
4269     sa_handler_t hand = NULL;
4270     sa_sigaction_t sa = NULL;
4271     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4272     // retrieve the chained handler
4273     if (siginfo_flag_set) {
4274       sa = actp->sa_sigaction;
4275     } else {
4276       hand = actp->sa_handler;
4277     }
4278 
4279     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4280       actp->sa_handler = SIG_DFL;
4281     }
4282 
4283     // try to honor the signal mask
4284     sigset_t oset;
4285     sigemptyset(&oset);
4286     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4287 
4288     // call into the chained handler
4289     if (siginfo_flag_set) {
4290       (*sa)(sig, siginfo, context);
4291     } else {
4292       (*hand)(sig);
4293     }
4294 
4295     // restore the signal mask
4296     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4297   }
4298   // Tell jvm's signal handler the signal is taken care of.
4299   return true;
4300 }
4301 
4302 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4303   bool chained = false;
4304   // signal-chaining
4305   if (UseSignalChaining) {
4306     struct sigaction *actp = get_chained_signal_action(sig);
4307     if (actp != NULL) {
4308       chained = call_chained_handler(actp, sig, siginfo, context);
4309     }
4310   }
4311   return chained;
4312 }
4313 
4314 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4315   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4316     return &sigact[sig];
4317   }
4318   return NULL;
4319 }
4320 
4321 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4322   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4323   sigact[sig] = oldAct;
4324   sigs |= (uint64_t)1 << (sig-1);
4325 }
4326 
4327 // for diagnostic
4328 int sigflags[NSIG];
4329 
4330 int os::Linux::get_our_sigflags(int sig) {
4331   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4332   return sigflags[sig];
4333 }
4334 
4335 void os::Linux::set_our_sigflags(int sig, int flags) {
4336   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4337   if (sig > 0 && sig < NSIG) {
4338     sigflags[sig] = flags;
4339   }
4340 }
4341 
4342 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4343   // Check for overwrite.
4344   struct sigaction oldAct;
4345   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4346 
4347   void* oldhand = oldAct.sa_sigaction
4348                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4349                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4350   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4351       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4352       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4353     if (AllowUserSignalHandlers || !set_installed) {
4354       // Do not overwrite; user takes responsibility to forward to us.
4355       return;
4356     } else if (UseSignalChaining) {
4357       // save the old handler in jvm
4358       save_preinstalled_handler(sig, oldAct);
4359       // libjsig also interposes the sigaction() call below and saves the
4360       // old sigaction on it own.
4361     } else {
4362       fatal("Encountered unexpected pre-existing sigaction handler "
4363             "%#lx for signal %d.", (long)oldhand, sig);
4364     }
4365   }
4366 
4367   struct sigaction sigAct;
4368   sigfillset(&(sigAct.sa_mask));
4369   sigAct.sa_handler = SIG_DFL;
4370   if (!set_installed) {
4371     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4372   } else {
4373     sigAct.sa_sigaction = signalHandler;
4374     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4375   }
4376   // Save flags, which are set by ours
4377   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4378   sigflags[sig] = sigAct.sa_flags;
4379 
4380   int ret = sigaction(sig, &sigAct, &oldAct);
4381   assert(ret == 0, "check");
4382 
4383   void* oldhand2  = oldAct.sa_sigaction
4384                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4385                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4386   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4387 }
4388 
4389 // install signal handlers for signals that HotSpot needs to
4390 // handle in order to support Java-level exception handling.
4391 
4392 void os::Linux::install_signal_handlers() {
4393   if (!signal_handlers_are_installed) {
4394     signal_handlers_are_installed = true;
4395 
4396     // signal-chaining
4397     typedef void (*signal_setting_t)();
4398     signal_setting_t begin_signal_setting = NULL;
4399     signal_setting_t end_signal_setting = NULL;
4400     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4401                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4402     if (begin_signal_setting != NULL) {
4403       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4404                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4405       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4406                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4407       libjsig_is_loaded = true;
4408       assert(UseSignalChaining, "should enable signal-chaining");
4409     }
4410     if (libjsig_is_loaded) {
4411       // Tell libjsig jvm is setting signal handlers
4412       (*begin_signal_setting)();
4413     }
4414 
4415     set_signal_handler(SIGSEGV, true);
4416     set_signal_handler(SIGPIPE, true);
4417     set_signal_handler(SIGBUS, true);
4418     set_signal_handler(SIGILL, true);
4419     set_signal_handler(SIGFPE, true);
4420 #if defined(PPC64)
4421     set_signal_handler(SIGTRAP, true);
4422 #endif
4423     set_signal_handler(SIGXFSZ, true);
4424 
4425     if (libjsig_is_loaded) {
4426       // Tell libjsig jvm finishes setting signal handlers
4427       (*end_signal_setting)();
4428     }
4429 
4430     // We don't activate signal checker if libjsig is in place, we trust ourselves
4431     // and if UserSignalHandler is installed all bets are off.
4432     // Log that signal checking is off only if -verbose:jni is specified.
4433     if (CheckJNICalls) {
4434       if (libjsig_is_loaded) {
4435         if (PrintJNIResolving) {
4436           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4437         }
4438         check_signals = false;
4439       }
4440       if (AllowUserSignalHandlers) {
4441         if (PrintJNIResolving) {
4442           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4443         }
4444         check_signals = false;
4445       }
4446     }
4447   }
4448 }
4449 
4450 // This is the fastest way to get thread cpu time on Linux.
4451 // Returns cpu time (user+sys) for any thread, not only for current.
4452 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4453 // It might work on 2.6.10+ with a special kernel/glibc patch.
4454 // For reference, please, see IEEE Std 1003.1-2004:
4455 //   http://www.unix.org/single_unix_specification
4456 
4457 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4458   struct timespec tp;
4459   int rc = os::Linux::clock_gettime(clockid, &tp);
4460   assert(rc == 0, "clock_gettime is expected to return 0 code");
4461 
4462   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4463 }
4464 
4465 void os::Linux::initialize_os_info() {
4466   assert(_os_version == 0, "OS info already initialized");
4467 
4468   struct utsname _uname;
4469 
4470   uint32_t major;
4471   uint32_t minor;
4472   uint32_t fix;
4473 
4474   int rc;
4475 
4476   // Kernel version is unknown if
4477   // verification below fails.
4478   _os_version = 0x01000000;
4479 
4480   rc = uname(&_uname);
4481   if (rc != -1) {
4482 
4483     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4484     if (rc == 3) {
4485 
4486       if (major < 256 && minor < 256 && fix < 256) {
4487         // Kernel version format is as expected,
4488         // set it overriding unknown state.
4489         _os_version = (major << 16) |
4490                       (minor << 8 ) |
4491                       (fix   << 0 ) ;
4492       }
4493     }
4494   }
4495 }
4496 
4497 uint32_t os::Linux::os_version() {
4498   assert(_os_version != 0, "not initialized");
4499   return _os_version & 0x00FFFFFF;
4500 }
4501 
4502 bool os::Linux::os_version_is_known() {
4503   assert(_os_version != 0, "not initialized");
4504   return _os_version & 0x01000000 ? false : true;
4505 }
4506 
4507 /////
4508 // glibc on Linux platform uses non-documented flag
4509 // to indicate, that some special sort of signal
4510 // trampoline is used.
4511 // We will never set this flag, and we should
4512 // ignore this flag in our diagnostic
4513 #ifdef SIGNIFICANT_SIGNAL_MASK
4514   #undef SIGNIFICANT_SIGNAL_MASK
4515 #endif
4516 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4517 
4518 static const char* get_signal_handler_name(address handler,
4519                                            char* buf, int buflen) {
4520   int offset = 0;
4521   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4522   if (found) {
4523     // skip directory names
4524     const char *p1, *p2;
4525     p1 = buf;
4526     size_t len = strlen(os::file_separator());
4527     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4528     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4529   } else {
4530     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4531   }
4532   return buf;
4533 }
4534 
4535 static void print_signal_handler(outputStream* st, int sig,
4536                                  char* buf, size_t buflen) {
4537   struct sigaction sa;
4538 
4539   sigaction(sig, NULL, &sa);
4540 
4541   // See comment for SIGNIFICANT_SIGNAL_MASK define
4542   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4543 
4544   st->print("%s: ", os::exception_name(sig, buf, buflen));
4545 
4546   address handler = (sa.sa_flags & SA_SIGINFO)
4547     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4548     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4549 
4550   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4551     st->print("SIG_DFL");
4552   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4553     st->print("SIG_IGN");
4554   } else {
4555     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4556   }
4557 
4558   st->print(", sa_mask[0]=");
4559   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4560 
4561   address rh = VMError::get_resetted_sighandler(sig);
4562   // May be, handler was resetted by VMError?
4563   if (rh != NULL) {
4564     handler = rh;
4565     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4566   }
4567 
4568   st->print(", sa_flags=");
4569   os::Posix::print_sa_flags(st, sa.sa_flags);
4570 
4571   // Check: is it our handler?
4572   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4573       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4574     // It is our signal handler
4575     // check for flags, reset system-used one!
4576     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4577       st->print(
4578                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4579                 os::Linux::get_our_sigflags(sig));
4580     }
4581   }
4582   st->cr();
4583 }
4584 
4585 
4586 #define DO_SIGNAL_CHECK(sig)                      \
4587   do {                                            \
4588     if (!sigismember(&check_signal_done, sig)) {  \
4589       os::Linux::check_signal_handler(sig);       \
4590     }                                             \
4591   } while (0)
4592 
4593 // This method is a periodic task to check for misbehaving JNI applications
4594 // under CheckJNI, we can add any periodic checks here
4595 
4596 void os::run_periodic_checks() {
4597   if (check_signals == false) return;
4598 
4599   // SEGV and BUS if overridden could potentially prevent
4600   // generation of hs*.log in the event of a crash, debugging
4601   // such a case can be very challenging, so we absolutely
4602   // check the following for a good measure:
4603   DO_SIGNAL_CHECK(SIGSEGV);
4604   DO_SIGNAL_CHECK(SIGILL);
4605   DO_SIGNAL_CHECK(SIGFPE);
4606   DO_SIGNAL_CHECK(SIGBUS);
4607   DO_SIGNAL_CHECK(SIGPIPE);
4608   DO_SIGNAL_CHECK(SIGXFSZ);
4609 #if defined(PPC64)
4610   DO_SIGNAL_CHECK(SIGTRAP);
4611 #endif
4612 
4613   // ReduceSignalUsage allows the user to override these handlers
4614   // see comments at the very top and jvm_solaris.h
4615   if (!ReduceSignalUsage) {
4616     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4617     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4618     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4619     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4620   }
4621 
4622   DO_SIGNAL_CHECK(SR_signum);
4623 }
4624 
4625 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4626 
4627 static os_sigaction_t os_sigaction = NULL;
4628 
4629 void os::Linux::check_signal_handler(int sig) {
4630   char buf[O_BUFLEN];
4631   address jvmHandler = NULL;
4632 
4633 
4634   struct sigaction act;
4635   if (os_sigaction == NULL) {
4636     // only trust the default sigaction, in case it has been interposed
4637     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4638     if (os_sigaction == NULL) return;
4639   }
4640 
4641   os_sigaction(sig, (struct sigaction*)NULL, &act);
4642 
4643 
4644   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4645 
4646   address thisHandler = (act.sa_flags & SA_SIGINFO)
4647     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4648     : CAST_FROM_FN_PTR(address, act.sa_handler);
4649 
4650 
4651   switch (sig) {
4652   case SIGSEGV:
4653   case SIGBUS:
4654   case SIGFPE:
4655   case SIGPIPE:
4656   case SIGILL:
4657   case SIGXFSZ:
4658     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4659     break;
4660 
4661   case SHUTDOWN1_SIGNAL:
4662   case SHUTDOWN2_SIGNAL:
4663   case SHUTDOWN3_SIGNAL:
4664   case BREAK_SIGNAL:
4665     jvmHandler = (address)user_handler();
4666     break;
4667 
4668   default:
4669     if (sig == SR_signum) {
4670       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4671     } else {
4672       return;
4673     }
4674     break;
4675   }
4676 
4677   if (thisHandler != jvmHandler) {
4678     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4679     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4680     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4681     // No need to check this sig any longer
4682     sigaddset(&check_signal_done, sig);
4683     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4684     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4685       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4686                     exception_name(sig, buf, O_BUFLEN));
4687     }
4688   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4689     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4690     tty->print("expected:");
4691     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4692     tty->cr();
4693     tty->print("  found:");
4694     os::Posix::print_sa_flags(tty, act.sa_flags);
4695     tty->cr();
4696     // No need to check this sig any longer
4697     sigaddset(&check_signal_done, sig);
4698   }
4699 
4700   // Dump all the signal
4701   if (sigismember(&check_signal_done, sig)) {
4702     print_signal_handlers(tty, buf, O_BUFLEN);
4703   }
4704 }
4705 
4706 extern void report_error(char* file_name, int line_no, char* title,
4707                          char* format, ...);
4708 
4709 // this is called _before_ the most of global arguments have been parsed
4710 void os::init(void) {
4711   char dummy;   // used to get a guess on initial stack address
4712 //  first_hrtime = gethrtime();
4713 
4714   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4715 
4716   init_random(1234567);
4717 
4718   ThreadCritical::initialize();
4719 
4720   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4721   if (Linux::page_size() == -1) {
4722     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4723           os::strerror(errno));
4724   }
4725   init_page_sizes((size_t) Linux::page_size());
4726 
4727   Linux::initialize_system_info();
4728 
4729   Linux::initialize_os_info();
4730 
4731   // main_thread points to the aboriginal thread
4732   Linux::_main_thread = pthread_self();
4733 
4734   Linux::clock_init();
4735   initial_time_count = javaTimeNanos();
4736 
4737   // pthread_condattr initialization for monotonic clock
4738   int status;
4739   pthread_condattr_t* _condattr = os::Linux::condAttr();
4740   if ((status = pthread_condattr_init(_condattr)) != 0) {
4741     fatal("pthread_condattr_init: %s", os::strerror(status));
4742   }
4743   // Only set the clock if CLOCK_MONOTONIC is available
4744   if (os::supports_monotonic_clock()) {
4745     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4746       if (status == EINVAL) {
4747         warning("Unable to use monotonic clock with relative timed-waits" \
4748                 " - changes to the time-of-day clock may have adverse affects");
4749       } else {
4750         fatal("pthread_condattr_setclock: %s", os::strerror(status));
4751       }
4752     }
4753   }
4754   // else it defaults to CLOCK_REALTIME
4755 
4756   // retrieve entry point for pthread_setname_np
4757   Linux::_pthread_setname_np =
4758     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4759 
4760 }
4761 
4762 // To install functions for atexit system call
4763 extern "C" {
4764   static void perfMemory_exit_helper() {
4765     perfMemory_exit();
4766   }
4767 }
4768 
4769 // this is called _after_ the global arguments have been parsed
4770 jint os::init_2(void) {
4771   Linux::fast_thread_clock_init();
4772 
4773   // Allocate a single page and mark it as readable for safepoint polling
4774   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4775   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4776 
4777   os::set_polling_page(polling_page);
4778   log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4779 
4780   if (!UseMembar) {
4781     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4782     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4783     os::set_memory_serialize_page(mem_serialize_page);
4784     log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4785   }
4786 
4787   // initialize suspend/resume support - must do this before signal_sets_init()
4788   if (SR_initialize() != 0) {
4789     perror("SR_initialize failed");
4790     return JNI_ERR;
4791   }
4792 
4793   Linux::signal_sets_init();
4794   Linux::install_signal_handlers();
4795 
4796   // Check minimum allowable stack size for thread creation and to initialize
4797   // the java system classes, including StackOverflowError - depends on page
4798   // size.  Add two 4K pages for compiler2 recursion in main thread.
4799   // Add in 4*BytesPerWord 4K pages to account for VM stack during
4800   // class initialization depending on 32 or 64 bit VM.
4801   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4802                                       JavaThread::stack_guard_zone_size() +
4803                                       JavaThread::stack_shadow_zone_size() +
4804                                       (4*BytesPerWord COMPILER2_PRESENT(+2)) * 4 * K);
4805 
4806   os::Linux::min_stack_allowed = align_size_up(os::Linux::min_stack_allowed, os::vm_page_size());
4807 
4808   size_t threadStackSizeInBytes = ThreadStackSize * K;
4809   if (threadStackSizeInBytes != 0 &&
4810       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4811     tty->print_cr("\nThe stack size specified is too small, "
4812                   "Specify at least " SIZE_FORMAT "k",
4813                   os::Linux::min_stack_allowed/ K);
4814     return JNI_ERR;
4815   }
4816 
4817   // Make the stack size a multiple of the page size so that
4818   // the yellow/red zones can be guarded.
4819   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4820                                                 vm_page_size()));
4821 
4822   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4823 
4824 #if defined(IA32)
4825   workaround_expand_exec_shield_cs_limit();
4826 #endif
4827 
4828   Linux::libpthread_init();
4829   log_info(os)("HotSpot is running with %s, %s",
4830                Linux::glibc_version(), Linux::libpthread_version());
4831 
4832   if (UseNUMA) {
4833     if (!Linux::libnuma_init()) {
4834       UseNUMA = false;
4835     } else {
4836       if ((Linux::numa_max_node() < 1)) {
4837         // There's only one node(they start from 0), disable NUMA.
4838         UseNUMA = false;
4839       }
4840     }
4841     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4842     // we can make the adaptive lgrp chunk resizing work. If the user specified
4843     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4844     // disable adaptive resizing.
4845     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4846       if (FLAG_IS_DEFAULT(UseNUMA)) {
4847         UseNUMA = false;
4848       } else {
4849         if (FLAG_IS_DEFAULT(UseLargePages) &&
4850             FLAG_IS_DEFAULT(UseSHM) &&
4851             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4852           UseLargePages = false;
4853         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4854           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4855           UseAdaptiveSizePolicy = false;
4856           UseAdaptiveNUMAChunkSizing = false;
4857         }
4858       }
4859     }
4860     if (!UseNUMA && ForceNUMA) {
4861       UseNUMA = true;
4862     }
4863   }
4864 
4865   if (MaxFDLimit) {
4866     // set the number of file descriptors to max. print out error
4867     // if getrlimit/setrlimit fails but continue regardless.
4868     struct rlimit nbr_files;
4869     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4870     if (status != 0) {
4871       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4872     } else {
4873       nbr_files.rlim_cur = nbr_files.rlim_max;
4874       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4875       if (status != 0) {
4876         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4877       }
4878     }
4879   }
4880 
4881   // Initialize lock used to serialize thread creation (see os::create_thread)
4882   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4883 
4884   // at-exit methods are called in the reverse order of their registration.
4885   // atexit functions are called on return from main or as a result of a
4886   // call to exit(3C). There can be only 32 of these functions registered
4887   // and atexit() does not set errno.
4888 
4889   if (PerfAllowAtExitRegistration) {
4890     // only register atexit functions if PerfAllowAtExitRegistration is set.
4891     // atexit functions can be delayed until process exit time, which
4892     // can be problematic for embedded VM situations. Embedded VMs should
4893     // call DestroyJavaVM() to assure that VM resources are released.
4894 
4895     // note: perfMemory_exit_helper atexit function may be removed in
4896     // the future if the appropriate cleanup code can be added to the
4897     // VM_Exit VMOperation's doit method.
4898     if (atexit(perfMemory_exit_helper) != 0) {
4899       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4900     }
4901   }
4902 
4903   // initialize thread priority policy
4904   prio_init();
4905 
4906   return JNI_OK;
4907 }
4908 
4909 // Mark the polling page as unreadable
4910 void os::make_polling_page_unreadable(void) {
4911   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4912     fatal("Could not disable polling page");
4913   }
4914 }
4915 
4916 // Mark the polling page as readable
4917 void os::make_polling_page_readable(void) {
4918   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4919     fatal("Could not enable polling page");
4920   }
4921 }
4922 
4923 // older glibc versions don't have this macro (which expands to
4924 // an optimized bit-counting function) so we have to roll our own
4925 #ifndef CPU_COUNT
4926 
4927 static int _cpu_count(const cpu_set_t* cpus) {
4928   int count = 0;
4929   // only look up to the number of configured processors
4930   for (int i = 0; i < os::processor_count(); i++) {
4931     if (CPU_ISSET(i, cpus)) {
4932       count++;
4933     }
4934   }
4935   return count;
4936 }
4937 
4938 #define CPU_COUNT(cpus) _cpu_count(cpus)
4939 
4940 #endif // CPU_COUNT
4941 
4942 // Get the current number of available processors for this process.
4943 // This value can change at any time during a process's lifetime.
4944 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
4945 // If it appears there may be more than 1024 processors then we do a
4946 // dynamic check - see 6515172 for details.
4947 // If anything goes wrong we fallback to returning the number of online
4948 // processors - which can be greater than the number available to the process.
4949 int os::active_processor_count() {
4950   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
4951   cpu_set_t* cpus_p = &cpus;
4952   int cpus_size = sizeof(cpu_set_t);
4953 
4954   int configured_cpus = processor_count();  // upper bound on available cpus
4955   int cpu_count = 0;
4956 
4957 // old build platforms may not support dynamic cpu sets
4958 #ifdef CPU_ALLOC
4959 
4960   // To enable easy testing of the dynamic path on different platforms we
4961   // introduce a diagnostic flag: UseCpuAllocPath
4962   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
4963     // kernel may use a mask bigger than cpu_set_t
4964     log_trace(os)("active_processor_count: using dynamic path %s"
4965                   "- configured processors: %d",
4966                   UseCpuAllocPath ? "(forced) " : "",
4967                   configured_cpus);
4968     cpus_p = CPU_ALLOC(configured_cpus);
4969     if (cpus_p != NULL) {
4970       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
4971       // zero it just to be safe
4972       CPU_ZERO_S(cpus_size, cpus_p);
4973     }
4974     else {
4975        // failed to allocate so fallback to online cpus
4976        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4977        log_trace(os)("active_processor_count: "
4978                      "CPU_ALLOC failed (%s) - using "
4979                      "online processor count: %d",
4980                      os::strerror(errno), online_cpus);
4981        return online_cpus;
4982     }
4983   }
4984   else {
4985     log_trace(os)("active_processor_count: using static path - configured processors: %d",
4986                   configured_cpus);
4987   }
4988 #else // CPU_ALLOC
4989 // these stubs won't be executed
4990 #define CPU_COUNT_S(size, cpus) -1
4991 #define CPU_FREE(cpus)
4992 
4993   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
4994                 configured_cpus);
4995 #endif // CPU_ALLOC
4996 
4997   // pid 0 means the current thread - which we have to assume represents the process
4998   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
4999     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5000       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5001     }
5002     else {
5003       cpu_count = CPU_COUNT(cpus_p);
5004     }
5005     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5006   }
5007   else {
5008     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5009     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5010             "which may exceed available processors", os::strerror(errno), cpu_count);
5011   }
5012 
5013   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5014     CPU_FREE(cpus_p);
5015   }
5016 
5017   assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
5018   return cpu_count;
5019 }
5020 
5021 void os::set_native_thread_name(const char *name) {
5022   if (Linux::_pthread_setname_np) {
5023     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5024     snprintf(buf, sizeof(buf), "%s", name);
5025     buf[sizeof(buf) - 1] = '\0';
5026     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5027     // ERANGE should not happen; all other errors should just be ignored.
5028     assert(rc != ERANGE, "pthread_setname_np failed");
5029   }
5030 }
5031 
5032 bool os::distribute_processes(uint length, uint* distribution) {
5033   // Not yet implemented.
5034   return false;
5035 }
5036 
5037 bool os::bind_to_processor(uint processor_id) {
5038   // Not yet implemented.
5039   return false;
5040 }
5041 
5042 ///
5043 
5044 void os::SuspendedThreadTask::internal_do_task() {
5045   if (do_suspend(_thread->osthread())) {
5046     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5047     do_task(context);
5048     do_resume(_thread->osthread());
5049   }
5050 }
5051 
5052 class PcFetcher : public os::SuspendedThreadTask {
5053  public:
5054   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5055   ExtendedPC result();
5056  protected:
5057   void do_task(const os::SuspendedThreadTaskContext& context);
5058  private:
5059   ExtendedPC _epc;
5060 };
5061 
5062 ExtendedPC PcFetcher::result() {
5063   guarantee(is_done(), "task is not done yet.");
5064   return _epc;
5065 }
5066 
5067 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5068   Thread* thread = context.thread();
5069   OSThread* osthread = thread->osthread();
5070   if (osthread->ucontext() != NULL) {
5071     _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
5072   } else {
5073     // NULL context is unexpected, double-check this is the VMThread
5074     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5075   }
5076 }
5077 
5078 // Suspends the target using the signal mechanism and then grabs the PC before
5079 // resuming the target. Used by the flat-profiler only
5080 ExtendedPC os::get_thread_pc(Thread* thread) {
5081   // Make sure that it is called by the watcher for the VMThread
5082   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5083   assert(thread->is_VM_thread(), "Can only be called for VMThread");
5084 
5085   PcFetcher fetcher(thread);
5086   fetcher.run();
5087   return fetcher.result();
5088 }
5089 
5090 ////////////////////////////////////////////////////////////////////////////////
5091 // debug support
5092 
5093 bool os::find(address addr, outputStream* st) {
5094   Dl_info dlinfo;
5095   memset(&dlinfo, 0, sizeof(dlinfo));
5096   if (dladdr(addr, &dlinfo) != 0) {
5097     st->print(PTR_FORMAT ": ", p2i(addr));
5098     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5099       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5100                 p2i(addr) - p2i(dlinfo.dli_saddr));
5101     } else if (dlinfo.dli_fbase != NULL) {
5102       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5103     } else {
5104       st->print("<absolute address>");
5105     }
5106     if (dlinfo.dli_fname != NULL) {
5107       st->print(" in %s", dlinfo.dli_fname);
5108     }
5109     if (dlinfo.dli_fbase != NULL) {
5110       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5111     }
5112     st->cr();
5113 
5114     if (Verbose) {
5115       // decode some bytes around the PC
5116       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5117       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5118       address       lowest = (address) dlinfo.dli_sname;
5119       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5120       if (begin < lowest)  begin = lowest;
5121       Dl_info dlinfo2;
5122       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5123           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5124         end = (address) dlinfo2.dli_saddr;
5125       }
5126       Disassembler::decode(begin, end, st);
5127     }
5128     return true;
5129   }
5130   return false;
5131 }
5132 
5133 ////////////////////////////////////////////////////////////////////////////////
5134 // misc
5135 
5136 // This does not do anything on Linux. This is basically a hook for being
5137 // able to use structured exception handling (thread-local exception filters)
5138 // on, e.g., Win32.
5139 void
5140 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5141                          JavaCallArguments* args, Thread* thread) {
5142   f(value, method, args, thread);
5143 }
5144 
5145 void os::print_statistics() {
5146 }
5147 
5148 bool os::message_box(const char* title, const char* message) {
5149   int i;
5150   fdStream err(defaultStream::error_fd());
5151   for (i = 0; i < 78; i++) err.print_raw("=");
5152   err.cr();
5153   err.print_raw_cr(title);
5154   for (i = 0; i < 78; i++) err.print_raw("-");
5155   err.cr();
5156   err.print_raw_cr(message);
5157   for (i = 0; i < 78; i++) err.print_raw("=");
5158   err.cr();
5159 
5160   char buf[16];
5161   // Prevent process from exiting upon "read error" without consuming all CPU
5162   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5163 
5164   return buf[0] == 'y' || buf[0] == 'Y';
5165 }
5166 
5167 int os::stat(const char *path, struct stat *sbuf) {
5168   char pathbuf[MAX_PATH];
5169   if (strlen(path) > MAX_PATH - 1) {
5170     errno = ENAMETOOLONG;
5171     return -1;
5172   }
5173   os::native_path(strcpy(pathbuf, path));
5174   return ::stat(pathbuf, sbuf);
5175 }
5176 
5177 // Is a (classpath) directory empty?
5178 bool os::dir_is_empty(const char* path) {
5179   DIR *dir = NULL;
5180   struct dirent *ptr;
5181 
5182   dir = opendir(path);
5183   if (dir == NULL) return true;
5184 
5185   // Scan the directory
5186   bool result = true;
5187   char buf[sizeof(struct dirent) + MAX_PATH];
5188   while (result && (ptr = ::readdir(dir)) != NULL) {
5189     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5190       result = false;
5191     }
5192   }
5193   closedir(dir);
5194   return result;
5195 }
5196 
5197 // This code originates from JDK's sysOpen and open64_w
5198 // from src/solaris/hpi/src/system_md.c
5199 
5200 int os::open(const char *path, int oflag, int mode) {
5201   if (strlen(path) > MAX_PATH - 1) {
5202     errno = ENAMETOOLONG;
5203     return -1;
5204   }
5205 
5206   // All file descriptors that are opened in the Java process and not
5207   // specifically destined for a subprocess should have the close-on-exec
5208   // flag set.  If we don't set it, then careless 3rd party native code
5209   // might fork and exec without closing all appropriate file descriptors
5210   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5211   // turn might:
5212   //
5213   // - cause end-of-file to fail to be detected on some file
5214   //   descriptors, resulting in mysterious hangs, or
5215   //
5216   // - might cause an fopen in the subprocess to fail on a system
5217   //   suffering from bug 1085341.
5218   //
5219   // (Yes, the default setting of the close-on-exec flag is a Unix
5220   // design flaw)
5221   //
5222   // See:
5223   // 1085341: 32-bit stdio routines should support file descriptors >255
5224   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5225   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5226   //
5227   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5228   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5229   // because it saves a system call and removes a small window where the flag
5230   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5231   // and we fall back to using FD_CLOEXEC (see below).
5232 #ifdef O_CLOEXEC
5233   oflag |= O_CLOEXEC;
5234 #endif
5235 
5236   int fd = ::open64(path, oflag, mode);
5237   if (fd == -1) return -1;
5238 
5239   //If the open succeeded, the file might still be a directory
5240   {
5241     struct stat64 buf64;
5242     int ret = ::fstat64(fd, &buf64);
5243     int st_mode = buf64.st_mode;
5244 
5245     if (ret != -1) {
5246       if ((st_mode & S_IFMT) == S_IFDIR) {
5247         errno = EISDIR;
5248         ::close(fd);
5249         return -1;
5250       }
5251     } else {
5252       ::close(fd);
5253       return -1;
5254     }
5255   }
5256 
5257 #ifdef FD_CLOEXEC
5258   // Validate that the use of the O_CLOEXEC flag on open above worked.
5259   // With recent kernels, we will perform this check exactly once.
5260   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5261   if (!O_CLOEXEC_is_known_to_work) {
5262     int flags = ::fcntl(fd, F_GETFD);
5263     if (flags != -1) {
5264       if ((flags & FD_CLOEXEC) != 0)
5265         O_CLOEXEC_is_known_to_work = 1;
5266       else
5267         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5268     }
5269   }
5270 #endif
5271 
5272   return fd;
5273 }
5274 
5275 
5276 // create binary file, rewriting existing file if required
5277 int os::create_binary_file(const char* path, bool rewrite_existing) {
5278   int oflags = O_WRONLY | O_CREAT;
5279   if (!rewrite_existing) {
5280     oflags |= O_EXCL;
5281   }
5282   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5283 }
5284 
5285 // return current position of file pointer
5286 jlong os::current_file_offset(int fd) {
5287   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5288 }
5289 
5290 // move file pointer to the specified offset
5291 jlong os::seek_to_file_offset(int fd, jlong offset) {
5292   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5293 }
5294 
5295 // This code originates from JDK's sysAvailable
5296 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5297 
5298 int os::available(int fd, jlong *bytes) {
5299   jlong cur, end;
5300   int mode;
5301   struct stat64 buf64;
5302 
5303   if (::fstat64(fd, &buf64) >= 0) {
5304     mode = buf64.st_mode;
5305     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5306       int n;
5307       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5308         *bytes = n;
5309         return 1;
5310       }
5311     }
5312   }
5313   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5314     return 0;
5315   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5316     return 0;
5317   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5318     return 0;
5319   }
5320   *bytes = end - cur;
5321   return 1;
5322 }
5323 
5324 // Map a block of memory.
5325 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5326                         char *addr, size_t bytes, bool read_only,
5327                         bool allow_exec) {
5328   int prot;
5329   int flags = MAP_PRIVATE;
5330 
5331   if (read_only) {
5332     prot = PROT_READ;
5333   } else {
5334     prot = PROT_READ | PROT_WRITE;
5335   }
5336 
5337   if (allow_exec) {
5338     prot |= PROT_EXEC;
5339   }
5340 
5341   if (addr != NULL) {
5342     flags |= MAP_FIXED;
5343   }
5344 
5345   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5346                                      fd, file_offset);
5347   if (mapped_address == MAP_FAILED) {
5348     return NULL;
5349   }
5350   return mapped_address;
5351 }
5352 
5353 
5354 // Remap a block of memory.
5355 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5356                           char *addr, size_t bytes, bool read_only,
5357                           bool allow_exec) {
5358   // same as map_memory() on this OS
5359   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5360                         allow_exec);
5361 }
5362 
5363 
5364 // Unmap a block of memory.
5365 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5366   return munmap(addr, bytes) == 0;
5367 }
5368 
5369 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5370 
5371 static clockid_t thread_cpu_clockid(Thread* thread) {
5372   pthread_t tid = thread->osthread()->pthread_id();
5373   clockid_t clockid;
5374 
5375   // Get thread clockid
5376   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5377   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5378   return clockid;
5379 }
5380 
5381 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5382 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5383 // of a thread.
5384 //
5385 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5386 // the fast estimate available on the platform.
5387 
5388 jlong os::current_thread_cpu_time() {
5389   if (os::Linux::supports_fast_thread_cpu_time()) {
5390     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5391   } else {
5392     // return user + sys since the cost is the same
5393     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5394   }
5395 }
5396 
5397 jlong os::thread_cpu_time(Thread* thread) {
5398   // consistent with what current_thread_cpu_time() returns
5399   if (os::Linux::supports_fast_thread_cpu_time()) {
5400     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5401   } else {
5402     return slow_thread_cpu_time(thread, true /* user + sys */);
5403   }
5404 }
5405 
5406 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5407   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5408     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5409   } else {
5410     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5411   }
5412 }
5413 
5414 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5415   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5416     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5417   } else {
5418     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5419   }
5420 }
5421 
5422 //  -1 on error.
5423 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5424   pid_t  tid = thread->osthread()->thread_id();
5425   char *s;
5426   char stat[2048];
5427   int statlen;
5428   char proc_name[64];
5429   int count;
5430   long sys_time, user_time;
5431   char cdummy;
5432   int idummy;
5433   long ldummy;
5434   FILE *fp;
5435 
5436   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5437   fp = fopen(proc_name, "r");
5438   if (fp == NULL) return -1;
5439   statlen = fread(stat, 1, 2047, fp);
5440   stat[statlen] = '\0';
5441   fclose(fp);
5442 
5443   // Skip pid and the command string. Note that we could be dealing with
5444   // weird command names, e.g. user could decide to rename java launcher
5445   // to "java 1.4.2 :)", then the stat file would look like
5446   //                1234 (java 1.4.2 :)) R ... ...
5447   // We don't really need to know the command string, just find the last
5448   // occurrence of ")" and then start parsing from there. See bug 4726580.
5449   s = strrchr(stat, ')');
5450   if (s == NULL) return -1;
5451 
5452   // Skip blank chars
5453   do { s++; } while (s && isspace(*s));
5454 
5455   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5456                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5457                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5458                  &user_time, &sys_time);
5459   if (count != 13) return -1;
5460   if (user_sys_cpu_time) {
5461     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5462   } else {
5463     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5464   }
5465 }
5466 
5467 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5468   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5469   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5470   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5471   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5472 }
5473 
5474 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5475   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5476   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5477   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5478   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5479 }
5480 
5481 bool os::is_thread_cpu_time_supported() {
5482   return true;
5483 }
5484 
5485 // System loadavg support.  Returns -1 if load average cannot be obtained.
5486 // Linux doesn't yet have a (official) notion of processor sets,
5487 // so just return the system wide load average.
5488 int os::loadavg(double loadavg[], int nelem) {
5489   return ::getloadavg(loadavg, nelem);
5490 }
5491 
5492 void os::pause() {
5493   char filename[MAX_PATH];
5494   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5495     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5496   } else {
5497     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5498   }
5499 
5500   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5501   if (fd != -1) {
5502     struct stat buf;
5503     ::close(fd);
5504     while (::stat(filename, &buf) == 0) {
5505       (void)::poll(NULL, 0, 100);
5506     }
5507   } else {
5508     jio_fprintf(stderr,
5509                 "Could not open pause file '%s', continuing immediately.\n", filename);
5510   }
5511 }
5512 
5513 
5514 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5515 // comment paragraphs are worth repeating here:
5516 //
5517 // Assumption:
5518 //    Only one parker can exist on an event, which is why we allocate
5519 //    them per-thread. Multiple unparkers can coexist.
5520 //
5521 // _Event serves as a restricted-range semaphore.
5522 //   -1 : thread is blocked, i.e. there is a waiter
5523 //    0 : neutral: thread is running or ready,
5524 //        could have been signaled after a wait started
5525 //    1 : signaled - thread is running or ready
5526 //
5527 
5528 // utility to compute the abstime argument to timedwait:
5529 // millis is the relative timeout time
5530 // abstime will be the absolute timeout time
5531 // TODO: replace compute_abstime() with unpackTime()
5532 
5533 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5534   if (millis < 0)  millis = 0;
5535 
5536   jlong seconds = millis / 1000;
5537   millis %= 1000;
5538   if (seconds > 50000000) { // see man cond_timedwait(3T)
5539     seconds = 50000000;
5540   }
5541 
5542   if (os::supports_monotonic_clock()) {
5543     struct timespec now;
5544     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5545     assert_status(status == 0, status, "clock_gettime");
5546     abstime->tv_sec = now.tv_sec  + seconds;
5547     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5548     if (nanos >= NANOSECS_PER_SEC) {
5549       abstime->tv_sec += 1;
5550       nanos -= NANOSECS_PER_SEC;
5551     }
5552     abstime->tv_nsec = nanos;
5553   } else {
5554     struct timeval now;
5555     int status = gettimeofday(&now, NULL);
5556     assert(status == 0, "gettimeofday");
5557     abstime->tv_sec = now.tv_sec  + seconds;
5558     long usec = now.tv_usec + millis * 1000;
5559     if (usec >= 1000000) {
5560       abstime->tv_sec += 1;
5561       usec -= 1000000;
5562     }
5563     abstime->tv_nsec = usec * 1000;
5564   }
5565   return abstime;
5566 }
5567 
5568 void os::PlatformEvent::park() {       // AKA "down()"
5569   // Transitions for _Event:
5570   //   -1 => -1 : illegal
5571   //    1 =>  0 : pass - return immediately
5572   //    0 => -1 : block; then set _Event to 0 before returning
5573 
5574   // Invariant: Only the thread associated with the Event/PlatformEvent
5575   // may call park().
5576   // TODO: assert that _Assoc != NULL or _Assoc == Self
5577   assert(_nParked == 0, "invariant");
5578 
5579   int v;
5580   for (;;) {
5581     v = _Event;
5582     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5583   }
5584   guarantee(v >= 0, "invariant");
5585   if (v == 0) {
5586     // Do this the hard way by blocking ...
5587     int status = pthread_mutex_lock(_mutex);
5588     assert_status(status == 0, status, "mutex_lock");
5589     guarantee(_nParked == 0, "invariant");
5590     ++_nParked;
5591     while (_Event < 0) {
5592       status = pthread_cond_wait(_cond, _mutex);
5593       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5594       // Treat this the same as if the wait was interrupted
5595       if (status == ETIME) { status = EINTR; }
5596       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5597     }
5598     --_nParked;
5599 
5600     _Event = 0;
5601     status = pthread_mutex_unlock(_mutex);
5602     assert_status(status == 0, status, "mutex_unlock");
5603     // Paranoia to ensure our locked and lock-free paths interact
5604     // correctly with each other.
5605     OrderAccess::fence();
5606   }
5607   guarantee(_Event >= 0, "invariant");
5608 }
5609 
5610 int os::PlatformEvent::park(jlong millis) {
5611   // Transitions for _Event:
5612   //   -1 => -1 : illegal
5613   //    1 =>  0 : pass - return immediately
5614   //    0 => -1 : block; then set _Event to 0 before returning
5615 
5616   guarantee(_nParked == 0, "invariant");
5617 
5618   int v;
5619   for (;;) {
5620     v = _Event;
5621     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5622   }
5623   guarantee(v >= 0, "invariant");
5624   if (v != 0) return OS_OK;
5625 
5626   // We do this the hard way, by blocking the thread.
5627   // Consider enforcing a minimum timeout value.
5628   struct timespec abst;
5629   compute_abstime(&abst, millis);
5630 
5631   int ret = OS_TIMEOUT;
5632   int status = pthread_mutex_lock(_mutex);
5633   assert_status(status == 0, status, "mutex_lock");
5634   guarantee(_nParked == 0, "invariant");
5635   ++_nParked;
5636 
5637   // Object.wait(timo) will return because of
5638   // (a) notification
5639   // (b) timeout
5640   // (c) thread.interrupt
5641   //
5642   // Thread.interrupt and object.notify{All} both call Event::set.
5643   // That is, we treat thread.interrupt as a special case of notification.
5644   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5645   // We assume all ETIME returns are valid.
5646   //
5647   // TODO: properly differentiate simultaneous notify+interrupt.
5648   // In that case, we should propagate the notify to another waiter.
5649 
5650   while (_Event < 0) {
5651     status = pthread_cond_timedwait(_cond, _mutex, &abst);
5652     assert_status(status == 0 || status == EINTR ||
5653                   status == ETIME || status == ETIMEDOUT,
5654                   status, "cond_timedwait");
5655     if (!FilterSpuriousWakeups) break;                 // previous semantics
5656     if (status == ETIME || status == ETIMEDOUT) break;
5657     // We consume and ignore EINTR and spurious wakeups.
5658   }
5659   --_nParked;
5660   if (_Event >= 0) {
5661     ret = OS_OK;
5662   }
5663   _Event = 0;
5664   status = pthread_mutex_unlock(_mutex);
5665   assert_status(status == 0, status, "mutex_unlock");
5666   assert(_nParked == 0, "invariant");
5667   // Paranoia to ensure our locked and lock-free paths interact
5668   // correctly with each other.
5669   OrderAccess::fence();
5670   return ret;
5671 }
5672 
5673 void os::PlatformEvent::unpark() {
5674   // Transitions for _Event:
5675   //    0 => 1 : just return
5676   //    1 => 1 : just return
5677   //   -1 => either 0 or 1; must signal target thread
5678   //         That is, we can safely transition _Event from -1 to either
5679   //         0 or 1.
5680   // See also: "Semaphores in Plan 9" by Mullender & Cox
5681   //
5682   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5683   // that it will take two back-to-back park() calls for the owning
5684   // thread to block. This has the benefit of forcing a spurious return
5685   // from the first park() call after an unpark() call which will help
5686   // shake out uses of park() and unpark() without condition variables.
5687 
5688   if (Atomic::xchg(1, &_Event) >= 0) return;
5689 
5690   // Wait for the thread associated with the event to vacate
5691   int status = pthread_mutex_lock(_mutex);
5692   assert_status(status == 0, status, "mutex_lock");
5693   int AnyWaiters = _nParked;
5694   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5695   status = pthread_mutex_unlock(_mutex);
5696   assert_status(status == 0, status, "mutex_unlock");
5697   if (AnyWaiters != 0) {
5698     // Note that we signal() *after* dropping the lock for "immortal" Events.
5699     // This is safe and avoids a common class of  futile wakeups.  In rare
5700     // circumstances this can cause a thread to return prematurely from
5701     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5702     // will simply re-test the condition and re-park itself.
5703     // This provides particular benefit if the underlying platform does not
5704     // provide wait morphing.
5705     status = pthread_cond_signal(_cond);
5706     assert_status(status == 0, status, "cond_signal");
5707   }
5708 }
5709 
5710 
5711 // JSR166
5712 // -------------------------------------------------------
5713 
5714 // The solaris and linux implementations of park/unpark are fairly
5715 // conservative for now, but can be improved. They currently use a
5716 // mutex/condvar pair, plus a a count.
5717 // Park decrements count if > 0, else does a condvar wait.  Unpark
5718 // sets count to 1 and signals condvar.  Only one thread ever waits
5719 // on the condvar. Contention seen when trying to park implies that someone
5720 // is unparking you, so don't wait. And spurious returns are fine, so there
5721 // is no need to track notifications.
5722 
5723 // This code is common to linux and solaris and will be moved to a
5724 // common place in dolphin.
5725 //
5726 // The passed in time value is either a relative time in nanoseconds
5727 // or an absolute time in milliseconds. Either way it has to be unpacked
5728 // into suitable seconds and nanoseconds components and stored in the
5729 // given timespec structure.
5730 // Given time is a 64-bit value and the time_t used in the timespec is only
5731 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5732 // overflow if times way in the future are given. Further on Solaris versions
5733 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5734 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5735 // As it will be 28 years before "now + 100000000" will overflow we can
5736 // ignore overflow and just impose a hard-limit on seconds using the value
5737 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5738 // years from "now".
5739 
5740 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5741   assert(time > 0, "convertTime");
5742   time_t max_secs = 0;
5743 
5744   if (!os::supports_monotonic_clock() || isAbsolute) {
5745     struct timeval now;
5746     int status = gettimeofday(&now, NULL);
5747     assert(status == 0, "gettimeofday");
5748 
5749     max_secs = now.tv_sec + MAX_SECS;
5750 
5751     if (isAbsolute) {
5752       jlong secs = time / 1000;
5753       if (secs > max_secs) {
5754         absTime->tv_sec = max_secs;
5755       } else {
5756         absTime->tv_sec = secs;
5757       }
5758       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5759     } else {
5760       jlong secs = time / NANOSECS_PER_SEC;
5761       if (secs >= MAX_SECS) {
5762         absTime->tv_sec = max_secs;
5763         absTime->tv_nsec = 0;
5764       } else {
5765         absTime->tv_sec = now.tv_sec + secs;
5766         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5767         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5768           absTime->tv_nsec -= NANOSECS_PER_SEC;
5769           ++absTime->tv_sec; // note: this must be <= max_secs
5770         }
5771       }
5772     }
5773   } else {
5774     // must be relative using monotonic clock
5775     struct timespec now;
5776     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5777     assert_status(status == 0, status, "clock_gettime");
5778     max_secs = now.tv_sec + MAX_SECS;
5779     jlong secs = time / NANOSECS_PER_SEC;
5780     if (secs >= MAX_SECS) {
5781       absTime->tv_sec = max_secs;
5782       absTime->tv_nsec = 0;
5783     } else {
5784       absTime->tv_sec = now.tv_sec + secs;
5785       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5786       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5787         absTime->tv_nsec -= NANOSECS_PER_SEC;
5788         ++absTime->tv_sec; // note: this must be <= max_secs
5789       }
5790     }
5791   }
5792   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5793   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5794   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5795   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5796 }
5797 
5798 void Parker::park(bool isAbsolute, jlong time) {
5799   // Ideally we'd do something useful while spinning, such
5800   // as calling unpackTime().
5801 
5802   // Optional fast-path check:
5803   // Return immediately if a permit is available.
5804   // We depend on Atomic::xchg() having full barrier semantics
5805   // since we are doing a lock-free update to _counter.
5806   if (Atomic::xchg(0, &_counter) > 0) return;
5807 
5808   Thread* thread = Thread::current();
5809   assert(thread->is_Java_thread(), "Must be JavaThread");
5810   JavaThread *jt = (JavaThread *)thread;
5811 
5812   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5813   // Check interrupt before trying to wait
5814   if (Thread::is_interrupted(thread, false)) {
5815     return;
5816   }
5817 
5818   // Next, demultiplex/decode time arguments
5819   timespec absTime;
5820   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5821     return;
5822   }
5823   if (time > 0) {
5824     unpackTime(&absTime, isAbsolute, time);
5825   }
5826 
5827 
5828   // Enter safepoint region
5829   // Beware of deadlocks such as 6317397.
5830   // The per-thread Parker:: mutex is a classic leaf-lock.
5831   // In particular a thread must never block on the Threads_lock while
5832   // holding the Parker:: mutex.  If safepoints are pending both the
5833   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5834   ThreadBlockInVM tbivm(jt);
5835 
5836   // Don't wait if cannot get lock since interference arises from
5837   // unblocking.  Also. check interrupt before trying wait
5838   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5839     return;
5840   }
5841 
5842   int status;
5843   if (_counter > 0)  { // no wait needed
5844     _counter = 0;
5845     status = pthread_mutex_unlock(_mutex);
5846     assert_status(status == 0, status, "invariant");
5847     // Paranoia to ensure our locked and lock-free paths interact
5848     // correctly with each other and Java-level accesses.
5849     OrderAccess::fence();
5850     return;
5851   }
5852 
5853 #ifdef ASSERT
5854   // Don't catch signals while blocked; let the running threads have the signals.
5855   // (This allows a debugger to break into the running thread.)
5856   sigset_t oldsigs;
5857   sigemptyset(&oldsigs);
5858   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5859   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5860 #endif
5861 
5862   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5863   jt->set_suspend_equivalent();
5864   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5865 
5866   assert(_cur_index == -1, "invariant");
5867   if (time == 0) {
5868     _cur_index = REL_INDEX; // arbitrary choice when not timed
5869     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5870   } else {
5871     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5872     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5873   }
5874   _cur_index = -1;
5875   assert_status(status == 0 || status == EINTR ||
5876                 status == ETIME || status == ETIMEDOUT,
5877                 status, "cond_timedwait");
5878 
5879 #ifdef ASSERT
5880   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5881 #endif
5882 
5883   _counter = 0;
5884   status = pthread_mutex_unlock(_mutex);
5885   assert_status(status == 0, status, "invariant");
5886   // Paranoia to ensure our locked and lock-free paths interact
5887   // correctly with each other and Java-level accesses.
5888   OrderAccess::fence();
5889 
5890   // If externally suspended while waiting, re-suspend
5891   if (jt->handle_special_suspend_equivalent_condition()) {
5892     jt->java_suspend_self();
5893   }
5894 }
5895 
5896 void Parker::unpark() {
5897   int status = pthread_mutex_lock(_mutex);
5898   assert_status(status == 0, status, "invariant");
5899   const int s = _counter;
5900   _counter = 1;
5901   // must capture correct index before unlocking
5902   int index = _cur_index;
5903   status = pthread_mutex_unlock(_mutex);
5904   assert_status(status == 0, status, "invariant");
5905   if (s < 1 && index != -1) {
5906     // thread is definitely parked
5907     status = pthread_cond_signal(&_cond[index]);
5908     assert_status(status == 0, status, "invariant");
5909   }
5910 }
5911 
5912 
5913 extern char** environ;
5914 
5915 // Run the specified command in a separate process. Return its exit value,
5916 // or -1 on failure (e.g. can't fork a new process).
5917 // Unlike system(), this function can be called from signal handler. It
5918 // doesn't block SIGINT et al.
5919 int os::fork_and_exec(char* cmd) {
5920   const char * argv[4] = {"sh", "-c", cmd, NULL};
5921 
5922   pid_t pid = fork();
5923 
5924   if (pid < 0) {
5925     // fork failed
5926     return -1;
5927 
5928   } else if (pid == 0) {
5929     // child process
5930 
5931     execve("/bin/sh", (char* const*)argv, environ);
5932 
5933     // execve failed
5934     _exit(-1);
5935 
5936   } else  {
5937     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5938     // care about the actual exit code, for now.
5939 
5940     int status;
5941 
5942     // Wait for the child process to exit.  This returns immediately if
5943     // the child has already exited. */
5944     while (waitpid(pid, &status, 0) < 0) {
5945       switch (errno) {
5946       case ECHILD: return 0;
5947       case EINTR: break;
5948       default: return -1;
5949       }
5950     }
5951 
5952     if (WIFEXITED(status)) {
5953       // The child exited normally; get its exit code.
5954       return WEXITSTATUS(status);
5955     } else if (WIFSIGNALED(status)) {
5956       // The child exited because of a signal
5957       // The best value to return is 0x80 + signal number,
5958       // because that is what all Unix shells do, and because
5959       // it allows callers to distinguish between process exit and
5960       // process death by signal.
5961       return 0x80 + WTERMSIG(status);
5962     } else {
5963       // Unknown exit code; pass it through
5964       return status;
5965     }
5966   }
5967 }
5968 
5969 // is_headless_jre()
5970 //
5971 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5972 // in order to report if we are running in a headless jre
5973 //
5974 // Since JDK8 xawt/libmawt.so was moved into the same directory
5975 // as libawt.so, and renamed libawt_xawt.so
5976 //
5977 bool os::is_headless_jre() {
5978   struct stat statbuf;
5979   char buf[MAXPATHLEN];
5980   char libmawtpath[MAXPATHLEN];
5981   const char *xawtstr  = "/xawt/libmawt.so";
5982   const char *new_xawtstr = "/libawt_xawt.so";
5983   char *p;
5984 
5985   // Get path to libjvm.so
5986   os::jvm_path(buf, sizeof(buf));
5987 
5988   // Get rid of libjvm.so
5989   p = strrchr(buf, '/');
5990   if (p == NULL) {
5991     return false;
5992   } else {
5993     *p = '\0';
5994   }
5995 
5996   // Get rid of client or server
5997   p = strrchr(buf, '/');
5998   if (p == NULL) {
5999     return false;
6000   } else {
6001     *p = '\0';
6002   }
6003 
6004   // check xawt/libmawt.so
6005   strcpy(libmawtpath, buf);
6006   strcat(libmawtpath, xawtstr);
6007   if (::stat(libmawtpath, &statbuf) == 0) return false;
6008 
6009   // check libawt_xawt.so
6010   strcpy(libmawtpath, buf);
6011   strcat(libmawtpath, new_xawtstr);
6012   if (::stat(libmawtpath, &statbuf) == 0) return false;
6013 
6014   return true;
6015 }
6016 
6017 // Get the default path to the core file
6018 // Returns the length of the string
6019 int os::get_core_path(char* buffer, size_t bufferSize) {
6020   /*
6021    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
6022    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
6023    */
6024   const int core_pattern_len = 129;
6025   char core_pattern[core_pattern_len] = {0};
6026 
6027   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
6028   if (core_pattern_file == -1) {
6029     return -1;
6030   }
6031 
6032   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6033   ::close(core_pattern_file);
6034   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
6035     return -1;
6036   }
6037   if (core_pattern[ret-1] == '\n') {
6038     core_pattern[ret-1] = '\0';
6039   } else {
6040     core_pattern[ret] = '\0';
6041   }
6042 
6043   char *pid_pos = strstr(core_pattern, "%p");
6044   int written;
6045 
6046   if (core_pattern[0] == '/') {
6047     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6048   } else {
6049     char cwd[PATH_MAX];
6050 
6051     const char* p = get_current_directory(cwd, PATH_MAX);
6052     if (p == NULL) {
6053       return -1;
6054     }
6055 
6056     if (core_pattern[0] == '|') {
6057       written = jio_snprintf(buffer, bufferSize,
6058                              "\"%s\" (or dumping to %s/core.%d)",
6059                              &core_pattern[1], p, current_process_id());
6060     } else {
6061       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6062     }
6063   }
6064 
6065   if (written < 0) {
6066     return -1;
6067   }
6068 
6069   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6070     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6071 
6072     if (core_uses_pid_file != -1) {
6073       char core_uses_pid = 0;
6074       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6075       ::close(core_uses_pid_file);
6076 
6077       if (core_uses_pid == '1') {
6078         jio_snprintf(buffer + written, bufferSize - written,
6079                                           ".%d", current_process_id());
6080       }
6081     }
6082   }
6083 
6084   return strlen(buffer);
6085 }
6086 
6087 bool os::start_debugging(char *buf, int buflen) {
6088   int len = (int)strlen(buf);
6089   char *p = &buf[len];
6090 
6091   jio_snprintf(p, buflen-len,
6092                "\n\n"
6093                "Do you want to debug the problem?\n\n"
6094                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6095                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6096                "Otherwise, press RETURN to abort...",
6097                os::current_process_id(), os::current_process_id(),
6098                os::current_thread_id(), os::current_thread_id());
6099 
6100   bool yes = os::message_box("Unexpected Error", buf);
6101 
6102   if (yes) {
6103     // yes, user asked VM to launch debugger
6104     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6105                  os::current_process_id(), os::current_process_id());
6106 
6107     os::fork_and_exec(buf);
6108     yes = false;
6109   }
6110   return yes;
6111 }
6112 
6113 static inline struct timespec get_mtime(const char* filename) {
6114   struct stat st;
6115   int ret = os::stat(filename, &st);
6116   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
6117   return st.st_mtim;
6118 }
6119 
6120 int os::compare_file_modified_times(const char* file1, const char* file2) {
6121   struct timespec filetime1 = get_mtime(file1);
6122   struct timespec filetime2 = get_mtime(file2);
6123   int diff = filetime1.tv_sec - filetime2.tv_sec;
6124   if (diff == 0) {
6125     return filetime1.tv_nsec - filetime2.tv_nsec;
6126   }
6127   return diff;
6128 }
6129 
6130 /////////////// Unit tests ///////////////
6131 
6132 #ifndef PRODUCT
6133 
6134 #define test_log(...)              \
6135   do {                             \
6136     if (VerboseInternalVMTests) {  \
6137       tty->print_cr(__VA_ARGS__);  \
6138       tty->flush();                \
6139     }                              \
6140   } while (false)
6141 
6142 class TestReserveMemorySpecial : AllStatic {
6143  public:
6144   static void small_page_write(void* addr, size_t size) {
6145     size_t page_size = os::vm_page_size();
6146 
6147     char* end = (char*)addr + size;
6148     for (char* p = (char*)addr; p < end; p += page_size) {
6149       *p = 1;
6150     }
6151   }
6152 
6153   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6154     if (!UseHugeTLBFS) {
6155       return;
6156     }
6157 
6158     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6159 
6160     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6161 
6162     if (addr != NULL) {
6163       small_page_write(addr, size);
6164 
6165       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6166     }
6167   }
6168 
6169   static void test_reserve_memory_special_huge_tlbfs_only() {
6170     if (!UseHugeTLBFS) {
6171       return;
6172     }
6173 
6174     size_t lp = os::large_page_size();
6175 
6176     for (size_t size = lp; size <= lp * 10; size += lp) {
6177       test_reserve_memory_special_huge_tlbfs_only(size);
6178     }
6179   }
6180 
6181   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6182     size_t lp = os::large_page_size();
6183     size_t ag = os::vm_allocation_granularity();
6184 
6185     // sizes to test
6186     const size_t sizes[] = {
6187       lp, lp + ag, lp + lp / 2, lp * 2,
6188       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6189       lp * 10, lp * 10 + lp / 2
6190     };
6191     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6192 
6193     // For each size/alignment combination, we test three scenarios:
6194     // 1) with req_addr == NULL
6195     // 2) with a non-null req_addr at which we expect to successfully allocate
6196     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6197     //    expect the allocation to either fail or to ignore req_addr
6198 
6199     // Pre-allocate two areas; they shall be as large as the largest allocation
6200     //  and aligned to the largest alignment we will be testing.
6201     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6202     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6203       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6204       -1, 0);
6205     assert(mapping1 != MAP_FAILED, "should work");
6206 
6207     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6208       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6209       -1, 0);
6210     assert(mapping2 != MAP_FAILED, "should work");
6211 
6212     // Unmap the first mapping, but leave the second mapping intact: the first
6213     // mapping will serve as a value for a "good" req_addr (case 2). The second
6214     // mapping, still intact, as "bad" req_addr (case 3).
6215     ::munmap(mapping1, mapping_size);
6216 
6217     // Case 1
6218     test_log("%s, req_addr NULL:", __FUNCTION__);
6219     test_log("size            align           result");
6220 
6221     for (int i = 0; i < num_sizes; i++) {
6222       const size_t size = sizes[i];
6223       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6224         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6225         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6226                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6227         if (p != NULL) {
6228           assert(is_ptr_aligned(p, alignment), "must be");
6229           small_page_write(p, size);
6230           os::Linux::release_memory_special_huge_tlbfs(p, size);
6231         }
6232       }
6233     }
6234 
6235     // Case 2
6236     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6237     test_log("size            align           req_addr         result");
6238 
6239     for (int i = 0; i < num_sizes; i++) {
6240       const size_t size = sizes[i];
6241       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6242         char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6243         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6244         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6245                  size, alignment, p2i(req_addr), p2i(p),
6246                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6247         if (p != NULL) {
6248           assert(p == req_addr, "must be");
6249           small_page_write(p, size);
6250           os::Linux::release_memory_special_huge_tlbfs(p, size);
6251         }
6252       }
6253     }
6254 
6255     // Case 3
6256     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6257     test_log("size            align           req_addr         result");
6258 
6259     for (int i = 0; i < num_sizes; i++) {
6260       const size_t size = sizes[i];
6261       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6262         char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6263         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6264         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6265                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6266         // as the area around req_addr contains already existing mappings, the API should always
6267         // return NULL (as per contract, it cannot return another address)
6268         assert(p == NULL, "must be");
6269       }
6270     }
6271 
6272     ::munmap(mapping2, mapping_size);
6273 
6274   }
6275 
6276   static void test_reserve_memory_special_huge_tlbfs() {
6277     if (!UseHugeTLBFS) {
6278       return;
6279     }
6280 
6281     test_reserve_memory_special_huge_tlbfs_only();
6282     test_reserve_memory_special_huge_tlbfs_mixed();
6283   }
6284 
6285   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6286     if (!UseSHM) {
6287       return;
6288     }
6289 
6290     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6291 
6292     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6293 
6294     if (addr != NULL) {
6295       assert(is_ptr_aligned(addr, alignment), "Check");
6296       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6297 
6298       small_page_write(addr, size);
6299 
6300       os::Linux::release_memory_special_shm(addr, size);
6301     }
6302   }
6303 
6304   static void test_reserve_memory_special_shm() {
6305     size_t lp = os::large_page_size();
6306     size_t ag = os::vm_allocation_granularity();
6307 
6308     for (size_t size = ag; size < lp * 3; size += ag) {
6309       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6310         test_reserve_memory_special_shm(size, alignment);
6311       }
6312     }
6313   }
6314 
6315   static void test() {
6316     test_reserve_memory_special_huge_tlbfs();
6317     test_reserve_memory_special_shm();
6318   }
6319 };
6320 
6321 void TestReserveMemorySpecial_test() {
6322   TestReserveMemorySpecial::test();
6323 }
6324 
6325 #endif