1 /*
   2  * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_linux.h"
  35 #include "logging/log.hpp"
  36 #include "memory/allocation.inline.hpp"
  37 #include "memory/filemap.hpp"
  38 #include "mutex_linux.inline.hpp"
  39 #include "oops/oop.inline.hpp"
  40 #include "os_linux.inline.hpp"
  41 #include "os_share_linux.hpp"
  42 #include "prims/jniFastGetField.hpp"
  43 #include "prims/jvm.h"
  44 #include "prims/jvm_misc.hpp"
  45 #include "runtime/arguments.hpp"
  46 #include "runtime/atomic.inline.hpp"
  47 #include "runtime/extendedPC.hpp"
  48 #include "runtime/globals.hpp"
  49 #include "runtime/interfaceSupport.hpp"
  50 #include "runtime/init.hpp"
  51 #include "runtime/java.hpp"
  52 #include "runtime/javaCalls.hpp"
  53 #include "runtime/mutexLocker.hpp"
  54 #include "runtime/objectMonitor.hpp"
  55 #include "runtime/orderAccess.inline.hpp"
  56 #include "runtime/osThread.hpp"
  57 #include "runtime/perfMemory.hpp"
  58 #include "runtime/sharedRuntime.hpp"
  59 #include "runtime/statSampler.hpp"
  60 #include "runtime/stubRoutines.hpp"
  61 #include "runtime/thread.inline.hpp"
  62 #include "runtime/threadCritical.hpp"
  63 #include "runtime/timer.hpp"
  64 #include "semaphore_posix.hpp"
  65 #include "services/attachListener.hpp"
  66 #include "services/memTracker.hpp"
  67 #include "services/runtimeService.hpp"
  68 #include "utilities/decoder.hpp"
  69 #include "utilities/defaultStream.hpp"
  70 #include "utilities/events.hpp"
  71 #include "utilities/elfFile.hpp"
  72 #include "utilities/growableArray.hpp"
  73 #include "utilities/macros.hpp"
  74 #include "utilities/vmError.hpp"
  75 
  76 // put OS-includes here
  77 # include <sys/types.h>
  78 # include <sys/mman.h>
  79 # include <sys/stat.h>
  80 # include <sys/select.h>
  81 # include <pthread.h>
  82 # include <signal.h>
  83 # include <errno.h>
  84 # include <dlfcn.h>
  85 # include <stdio.h>
  86 # include <unistd.h>
  87 # include <sys/resource.h>
  88 # include <pthread.h>
  89 # include <sys/stat.h>
  90 # include <sys/time.h>
  91 # include <sys/times.h>
  92 # include <sys/utsname.h>
  93 # include <sys/socket.h>
  94 # include <sys/wait.h>
  95 # include <pwd.h>
  96 # include <poll.h>
  97 # include <semaphore.h>
  98 # include <fcntl.h>
  99 # include <string.h>
 100 # include <syscall.h>
 101 # include <sys/sysinfo.h>
 102 # include <gnu/libc-version.h>
 103 # include <sys/ipc.h>
 104 # include <sys/shm.h>
 105 # include <link.h>
 106 # include <stdint.h>
 107 # include <inttypes.h>
 108 # include <sys/ioctl.h>
 109 
 110 #ifndef _GNU_SOURCE
 111   #define _GNU_SOURCE
 112   #include <sched.h>
 113   #undef _GNU_SOURCE
 114 #else
 115   #include <sched.h>
 116 #endif
 117 
 118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 119 // getrusage() is prepared to handle the associated failure.
 120 #ifndef RUSAGE_THREAD
 121   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 122 #endif
 123 
 124 #define MAX_PATH    (2 * K)
 125 
 126 #define MAX_SECS 100000000
 127 
 128 // for timer info max values which include all bits
 129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 130 
 131 #define LARGEPAGES_BIT (1 << 6)
 132 ////////////////////////////////////////////////////////////////////////////////
 133 // global variables
 134 julong os::Linux::_physical_memory = 0;
 135 
 136 address   os::Linux::_initial_thread_stack_bottom = NULL;
 137 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 138 
 139 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 141 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 142 Mutex* os::Linux::_createThread_lock = NULL;
 143 pthread_t os::Linux::_main_thread;
 144 int os::Linux::_page_size = -1;
 145 const int os::Linux::_vm_default_page_size = (8 * K);
 146 bool os::Linux::_supports_fast_thread_cpu_time = false;
 147 uint32_t os::Linux::_os_version = 0;
 148 const char * os::Linux::_glibc_version = NULL;
 149 const char * os::Linux::_libpthread_version = NULL;
 150 pthread_condattr_t os::Linux::_condattr[1];
 151 
 152 static jlong initial_time_count=0;
 153 
 154 static int clock_tics_per_sec = 100;
 155 
 156 // For diagnostics to print a message once. see run_periodic_checks
 157 static sigset_t check_signal_done;
 158 static bool check_signals = true;
 159 
 160 // Signal number used to suspend/resume a thread
 161 
 162 // do not use any signal number less than SIGSEGV, see 4355769
 163 static int SR_signum = SIGUSR2;
 164 sigset_t SR_sigset;
 165 
 166 // Declarations
 167 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 168 
 169 // utility functions
 170 
 171 static int SR_initialize();
 172 
 173 julong os::available_memory() {
 174   return Linux::available_memory();
 175 }
 176 
 177 julong os::Linux::available_memory() {
 178   // values in struct sysinfo are "unsigned long"
 179   struct sysinfo si;
 180   sysinfo(&si);
 181 
 182   return (julong)si.freeram * si.mem_unit;
 183 }
 184 
 185 julong os::physical_memory() {
 186   return Linux::physical_memory();
 187 }
 188 
 189 // Return true if user is running as root.
 190 
 191 bool os::have_special_privileges() {
 192   static bool init = false;
 193   static bool privileges = false;
 194   if (!init) {
 195     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 196     init = true;
 197   }
 198   return privileges;
 199 }
 200 
 201 
 202 #ifndef SYS_gettid
 203 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 204   #ifdef __ia64__
 205     #define SYS_gettid 1105
 206   #else
 207     #ifdef __i386__
 208       #define SYS_gettid 224
 209     #else
 210       #ifdef __amd64__
 211         #define SYS_gettid 186
 212       #else
 213         #ifdef __sparc__
 214           #define SYS_gettid 143
 215         #else
 216           #error define gettid for the arch
 217         #endif
 218       #endif
 219     #endif
 220   #endif
 221 #endif
 222 
 223 // Cpu architecture string
 224 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
 225 
 226 
 227 // pid_t gettid()
 228 //
 229 // Returns the kernel thread id of the currently running thread. Kernel
 230 // thread id is used to access /proc.
 231 pid_t os::Linux::gettid() {
 232   int rslt = syscall(SYS_gettid);
 233   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 234   return (pid_t)rslt;
 235 }
 236 
 237 // Most versions of linux have a bug where the number of processors are
 238 // determined by looking at the /proc file system.  In a chroot environment,
 239 // the system call returns 1.  This causes the VM to act as if it is
 240 // a single processor and elide locking (see is_MP() call).
 241 static bool unsafe_chroot_detected = false;
 242 static const char *unstable_chroot_error = "/proc file system not found.\n"
 243                      "Java may be unstable running multithreaded in a chroot "
 244                      "environment on Linux when /proc filesystem is not mounted.";
 245 
 246 void os::Linux::initialize_system_info() {
 247   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 248   if (processor_count() == 1) {
 249     pid_t pid = os::Linux::gettid();
 250     char fname[32];
 251     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 252     FILE *fp = fopen(fname, "r");
 253     if (fp == NULL) {
 254       unsafe_chroot_detected = true;
 255     } else {
 256       fclose(fp);
 257     }
 258   }
 259   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 260   assert(processor_count() > 0, "linux error");
 261 }
 262 
 263 void os::init_system_properties_values() {
 264   // The next steps are taken in the product version:
 265   //
 266   // Obtain the JAVA_HOME value from the location of libjvm.so.
 267   // This library should be located at:
 268   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 269   //
 270   // If "/jre/lib/" appears at the right place in the path, then we
 271   // assume libjvm.so is installed in a JDK and we use this path.
 272   //
 273   // Otherwise exit with message: "Could not create the Java virtual machine."
 274   //
 275   // The following extra steps are taken in the debugging version:
 276   //
 277   // If "/jre/lib/" does NOT appear at the right place in the path
 278   // instead of exit check for $JAVA_HOME environment variable.
 279   //
 280   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 281   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 282   // it looks like libjvm.so is installed there
 283   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 284   //
 285   // Otherwise exit.
 286   //
 287   // Important note: if the location of libjvm.so changes this
 288   // code needs to be changed accordingly.
 289 
 290   // See ld(1):
 291   //      The linker uses the following search paths to locate required
 292   //      shared libraries:
 293   //        1: ...
 294   //        ...
 295   //        7: The default directories, normally /lib and /usr/lib.
 296 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
 297   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 298 #else
 299   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 300 #endif
 301 
 302 // Base path of extensions installed on the system.
 303 #define SYS_EXT_DIR     "/usr/java/packages"
 304 #define EXTENSIONS_DIR  "/lib/ext"
 305 
 306   // Buffer that fits several sprintfs.
 307   // Note that the space for the colon and the trailing null are provided
 308   // by the nulls included by the sizeof operator.
 309   const size_t bufsize =
 310     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 311          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 312   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 313 
 314   // sysclasspath, java_home, dll_dir
 315   {
 316     char *pslash;
 317     os::jvm_path(buf, bufsize);
 318 
 319     // Found the full path to libjvm.so.
 320     // Now cut the path to <java_home>/jre if we can.
 321     pslash = strrchr(buf, '/');
 322     if (pslash != NULL) {
 323       *pslash = '\0';            // Get rid of /libjvm.so.
 324     }
 325     pslash = strrchr(buf, '/');
 326     if (pslash != NULL) {
 327       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 328     }
 329     Arguments::set_dll_dir(buf);
 330 
 331     if (pslash != NULL) {
 332       pslash = strrchr(buf, '/');
 333       if (pslash != NULL) {
 334         *pslash = '\0';          // Get rid of /<arch>.
 335         pslash = strrchr(buf, '/');
 336         if (pslash != NULL) {
 337           *pslash = '\0';        // Get rid of /lib.
 338         }
 339       }
 340     }
 341     Arguments::set_java_home(buf);
 342     set_boot_path('/', ':');
 343   }
 344 
 345   // Where to look for native libraries.
 346   //
 347   // Note: Due to a legacy implementation, most of the library path
 348   // is set in the launcher. This was to accomodate linking restrictions
 349   // on legacy Linux implementations (which are no longer supported).
 350   // Eventually, all the library path setting will be done here.
 351   //
 352   // However, to prevent the proliferation of improperly built native
 353   // libraries, the new path component /usr/java/packages is added here.
 354   // Eventually, all the library path setting will be done here.
 355   {
 356     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 357     // should always exist (until the legacy problem cited above is
 358     // addressed).
 359     const char *v = ::getenv("LD_LIBRARY_PATH");
 360     const char *v_colon = ":";
 361     if (v == NULL) { v = ""; v_colon = ""; }
 362     // That's +1 for the colon and +1 for the trailing '\0'.
 363     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 364                                                      strlen(v) + 1 +
 365                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
 366                                                      mtInternal);
 367     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
 368     Arguments::set_library_path(ld_library_path);
 369     FREE_C_HEAP_ARRAY(char, ld_library_path);
 370   }
 371 
 372   // Extensions directories.
 373   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 374   Arguments::set_ext_dirs(buf);
 375 
 376   FREE_C_HEAP_ARRAY(char, buf);
 377 
 378 #undef DEFAULT_LIBPATH
 379 #undef SYS_EXT_DIR
 380 #undef EXTENSIONS_DIR
 381 }
 382 
 383 ////////////////////////////////////////////////////////////////////////////////
 384 // breakpoint support
 385 
 386 void os::breakpoint() {
 387   BREAKPOINT;
 388 }
 389 
 390 extern "C" void breakpoint() {
 391   // use debugger to set breakpoint here
 392 }
 393 
 394 ////////////////////////////////////////////////////////////////////////////////
 395 // signal support
 396 
 397 debug_only(static bool signal_sets_initialized = false);
 398 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
 399 
 400 bool os::Linux::is_sig_ignored(int sig) {
 401   struct sigaction oact;
 402   sigaction(sig, (struct sigaction*)NULL, &oact);
 403   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 404                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 405   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 406     return true;
 407   } else {
 408     return false;
 409   }
 410 }
 411 
 412 void os::Linux::signal_sets_init() {
 413   // Should also have an assertion stating we are still single-threaded.
 414   assert(!signal_sets_initialized, "Already initialized");
 415   // Fill in signals that are necessarily unblocked for all threads in
 416   // the VM. Currently, we unblock the following signals:
 417   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 418   //                         by -Xrs (=ReduceSignalUsage));
 419   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 420   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 421   // the dispositions or masks wrt these signals.
 422   // Programs embedding the VM that want to use the above signals for their
 423   // own purposes must, at this time, use the "-Xrs" option to prevent
 424   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 425   // (See bug 4345157, and other related bugs).
 426   // In reality, though, unblocking these signals is really a nop, since
 427   // these signals are not blocked by default.
 428   sigemptyset(&unblocked_sigs);
 429   sigemptyset(&allowdebug_blocked_sigs);
 430   sigaddset(&unblocked_sigs, SIGILL);
 431   sigaddset(&unblocked_sigs, SIGSEGV);
 432   sigaddset(&unblocked_sigs, SIGBUS);
 433   sigaddset(&unblocked_sigs, SIGFPE);
 434 #if defined(PPC64)
 435   sigaddset(&unblocked_sigs, SIGTRAP);
 436 #endif
 437   sigaddset(&unblocked_sigs, SR_signum);
 438 
 439   if (!ReduceSignalUsage) {
 440     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 441       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 442       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
 443     }
 444     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 445       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 446       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
 447     }
 448     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 449       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 450       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
 451     }
 452   }
 453   // Fill in signals that are blocked by all but the VM thread.
 454   sigemptyset(&vm_sigs);
 455   if (!ReduceSignalUsage) {
 456     sigaddset(&vm_sigs, BREAK_SIGNAL);
 457   }
 458   debug_only(signal_sets_initialized = true);
 459 
 460 }
 461 
 462 // These are signals that are unblocked while a thread is running Java.
 463 // (For some reason, they get blocked by default.)
 464 sigset_t* os::Linux::unblocked_signals() {
 465   assert(signal_sets_initialized, "Not initialized");
 466   return &unblocked_sigs;
 467 }
 468 
 469 // These are the signals that are blocked while a (non-VM) thread is
 470 // running Java. Only the VM thread handles these signals.
 471 sigset_t* os::Linux::vm_signals() {
 472   assert(signal_sets_initialized, "Not initialized");
 473   return &vm_sigs;
 474 }
 475 
 476 // These are signals that are blocked during cond_wait to allow debugger in
 477 sigset_t* os::Linux::allowdebug_blocked_signals() {
 478   assert(signal_sets_initialized, "Not initialized");
 479   return &allowdebug_blocked_sigs;
 480 }
 481 
 482 void os::Linux::hotspot_sigmask(Thread* thread) {
 483 
 484   //Save caller's signal mask before setting VM signal mask
 485   sigset_t caller_sigmask;
 486   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 487 
 488   OSThread* osthread = thread->osthread();
 489   osthread->set_caller_sigmask(caller_sigmask);
 490 
 491   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 492 
 493   if (!ReduceSignalUsage) {
 494     if (thread->is_VM_thread()) {
 495       // Only the VM thread handles BREAK_SIGNAL ...
 496       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 497     } else {
 498       // ... all other threads block BREAK_SIGNAL
 499       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 500     }
 501   }
 502 }
 503 
 504 //////////////////////////////////////////////////////////////////////////////
 505 // detecting pthread library
 506 
 507 void os::Linux::libpthread_init() {
 508   // Save glibc and pthread version strings.
 509 #if !defined(_CS_GNU_LIBC_VERSION) || \
 510     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 511   #error "glibc too old (< 2.3.2)"
 512 #endif
 513 
 514   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 515   assert(n > 0, "cannot retrieve glibc version");
 516   char *str = (char *)malloc(n, mtInternal);
 517   confstr(_CS_GNU_LIBC_VERSION, str, n);
 518   os::Linux::set_glibc_version(str);
 519 
 520   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 521   assert(n > 0, "cannot retrieve pthread version");
 522   str = (char *)malloc(n, mtInternal);
 523   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 524   os::Linux::set_libpthread_version(str);
 525 }
 526 
 527 /////////////////////////////////////////////////////////////////////////////
 528 // thread stack expansion
 529 
 530 // os::Linux::manually_expand_stack() takes care of expanding the thread
 531 // stack. Note that this is normally not needed: pthread stacks allocate
 532 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 533 // committed. Therefore it is not necessary to expand the stack manually.
 534 //
 535 // Manually expanding the stack was historically needed on LinuxThreads
 536 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 537 // it is kept to deal with very rare corner cases:
 538 //
 539 // For one, user may run the VM on an own implementation of threads
 540 // whose stacks are - like the old LinuxThreads - implemented using
 541 // mmap(MAP_GROWSDOWN).
 542 //
 543 // Also, this coding may be needed if the VM is running on the primordial
 544 // thread. Normally we avoid running on the primordial thread; however,
 545 // user may still invoke the VM on the primordial thread.
 546 //
 547 // The following historical comment describes the details about running
 548 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 549 
 550 
 551 // Force Linux kernel to expand current thread stack. If "bottom" is close
 552 // to the stack guard, caller should block all signals.
 553 //
 554 // MAP_GROWSDOWN:
 555 //   A special mmap() flag that is used to implement thread stacks. It tells
 556 //   kernel that the memory region should extend downwards when needed. This
 557 //   allows early versions of LinuxThreads to only mmap the first few pages
 558 //   when creating a new thread. Linux kernel will automatically expand thread
 559 //   stack as needed (on page faults).
 560 //
 561 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 562 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 563 //   region, it's hard to tell if the fault is due to a legitimate stack
 564 //   access or because of reading/writing non-exist memory (e.g. buffer
 565 //   overrun). As a rule, if the fault happens below current stack pointer,
 566 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 567 //   application (see Linux kernel fault.c).
 568 //
 569 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 570 //   stack overflow detection.
 571 //
 572 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 573 //   not use MAP_GROWSDOWN.
 574 //
 575 // To get around the problem and allow stack banging on Linux, we need to
 576 // manually expand thread stack after receiving the SIGSEGV.
 577 //
 578 // There are two ways to expand thread stack to address "bottom", we used
 579 // both of them in JVM before 1.5:
 580 //   1. adjust stack pointer first so that it is below "bottom", and then
 581 //      touch "bottom"
 582 //   2. mmap() the page in question
 583 //
 584 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 585 // if current sp is already near the lower end of page 101, and we need to
 586 // call mmap() to map page 100, it is possible that part of the mmap() frame
 587 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 588 // That will destroy the mmap() frame and cause VM to crash.
 589 //
 590 // The following code works by adjusting sp first, then accessing the "bottom"
 591 // page to force a page fault. Linux kernel will then automatically expand the
 592 // stack mapping.
 593 //
 594 // _expand_stack_to() assumes its frame size is less than page size, which
 595 // should always be true if the function is not inlined.
 596 
 597 static void NOINLINE _expand_stack_to(address bottom) {
 598   address sp;
 599   size_t size;
 600   volatile char *p;
 601 
 602   // Adjust bottom to point to the largest address within the same page, it
 603   // gives us a one-page buffer if alloca() allocates slightly more memory.
 604   bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
 605   bottom += os::Linux::page_size() - 1;
 606 
 607   // sp might be slightly above current stack pointer; if that's the case, we
 608   // will alloca() a little more space than necessary, which is OK. Don't use
 609   // os::current_stack_pointer(), as its result can be slightly below current
 610   // stack pointer, causing us to not alloca enough to reach "bottom".
 611   sp = (address)&sp;
 612 
 613   if (sp > bottom) {
 614     size = sp - bottom;
 615     p = (volatile char *)alloca(size);
 616     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 617     p[0] = '\0';
 618   }
 619 }
 620 
 621 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 622   assert(t!=NULL, "just checking");
 623   assert(t->osthread()->expanding_stack(), "expand should be set");
 624   assert(t->stack_base() != NULL, "stack_base was not initialized");
 625 
 626   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 627     sigset_t mask_all, old_sigset;
 628     sigfillset(&mask_all);
 629     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 630     _expand_stack_to(addr);
 631     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 632     return true;
 633   }
 634   return false;
 635 }
 636 
 637 //////////////////////////////////////////////////////////////////////////////
 638 // create new thread
 639 
 640 // Thread start routine for all newly created threads
 641 static void *thread_native_entry(Thread *thread) {
 642   // Try to randomize the cache line index of hot stack frames.
 643   // This helps when threads of the same stack traces evict each other's
 644   // cache lines. The threads can be either from the same JVM instance, or
 645   // from different JVM instances. The benefit is especially true for
 646   // processors with hyperthreading technology.
 647   static int counter = 0;
 648   int pid = os::current_process_id();
 649   alloca(((pid ^ counter++) & 7) * 128);
 650 
 651   thread->initialize_thread_current();
 652 
 653   OSThread* osthread = thread->osthread();
 654   Monitor* sync = osthread->startThread_lock();
 655 
 656   osthread->set_thread_id(os::current_thread_id());
 657 
 658   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 659     os::current_thread_id(), (uintx) pthread_self());
 660 
 661   if (UseNUMA) {
 662     int lgrp_id = os::numa_get_group_id();
 663     if (lgrp_id != -1) {
 664       thread->set_lgrp_id(lgrp_id);
 665     }
 666   }
 667   // initialize signal mask for this thread
 668   os::Linux::hotspot_sigmask(thread);
 669 
 670   // initialize floating point control register
 671   os::Linux::init_thread_fpu_state();
 672 
 673   // handshaking with parent thread
 674   {
 675     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 676 
 677     // notify parent thread
 678     osthread->set_state(INITIALIZED);
 679     sync->notify_all();
 680 
 681     // wait until os::start_thread()
 682     while (osthread->get_state() == INITIALIZED) {
 683       sync->wait(Mutex::_no_safepoint_check_flag);
 684     }
 685   }
 686 
 687   // call one more level start routine
 688   thread->run();
 689 
 690   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 691     os::current_thread_id(), (uintx) pthread_self());
 692 
 693   // If a thread has not deleted itself ("delete this") as part of its
 694   // termination sequence, we have to ensure thread-local-storage is
 695   // cleared before we actually terminate. No threads should ever be
 696   // deleted asynchronously with respect to their termination.
 697   if (Thread::current_or_null_safe() != NULL) {
 698     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 699     thread->clear_thread_current();
 700   }
 701 
 702   return 0;
 703 }
 704 
 705 bool os::create_thread(Thread* thread, ThreadType thr_type,
 706                        size_t stack_size) {
 707   assert(thread->osthread() == NULL, "caller responsible");
 708 
 709   // Allocate the OSThread object
 710   OSThread* osthread = new OSThread(NULL, NULL);
 711   if (osthread == NULL) {
 712     return false;
 713   }
 714 
 715   // set the correct thread state
 716   osthread->set_thread_type(thr_type);
 717 
 718   // Initial state is ALLOCATED but not INITIALIZED
 719   osthread->set_state(ALLOCATED);
 720 
 721   thread->set_osthread(osthread);
 722 
 723   // init thread attributes
 724   pthread_attr_t attr;
 725   pthread_attr_init(&attr);
 726   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 727 
 728   // stack size
 729   // calculate stack size if it's not specified by caller
 730   if (stack_size == 0) {
 731     stack_size = os::Linux::default_stack_size(thr_type);
 732 
 733     switch (thr_type) {
 734     case os::java_thread:
 735       // Java threads use ThreadStackSize which default value can be
 736       // changed with the flag -Xss
 737       assert(JavaThread::stack_size_at_create() > 0, "this should be set");
 738       stack_size = JavaThread::stack_size_at_create();
 739       break;
 740     case os::compiler_thread:
 741       if (CompilerThreadStackSize > 0) {
 742         stack_size = (size_t)(CompilerThreadStackSize * K);
 743         break;
 744       } // else fall through:
 745         // use VMThreadStackSize if CompilerThreadStackSize is not defined
 746     case os::vm_thread:
 747     case os::pgc_thread:
 748     case os::cgc_thread:
 749     case os::watcher_thread:
 750       if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
 751       break;
 752     }
 753   }
 754 
 755   stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
 756   pthread_attr_setstacksize(&attr, stack_size);
 757 
 758   // glibc guard page
 759   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 760 
 761   ThreadState state;
 762 
 763   {
 764     pthread_t tid;
 765     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 766 
 767     char buf[64];
 768     if (ret == 0) {
 769       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 770         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 771     } else {
 772       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 773         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 774     }
 775 
 776     pthread_attr_destroy(&attr);
 777 
 778     if (ret != 0) {
 779       // Need to clean up stuff we've allocated so far
 780       thread->set_osthread(NULL);
 781       delete osthread;
 782       return false;
 783     }
 784 
 785     // Store pthread info into the OSThread
 786     osthread->set_pthread_id(tid);
 787 
 788     // Wait until child thread is either initialized or aborted
 789     {
 790       Monitor* sync_with_child = osthread->startThread_lock();
 791       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 792       while ((state = osthread->get_state()) == ALLOCATED) {
 793         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 794       }
 795     }
 796   }
 797 
 798   // Aborted due to thread limit being reached
 799   if (state == ZOMBIE) {
 800     thread->set_osthread(NULL);
 801     delete osthread;
 802     return false;
 803   }
 804 
 805   // The thread is returned suspended (in state INITIALIZED),
 806   // and is started higher up in the call chain
 807   assert(state == INITIALIZED, "race condition");
 808   return true;
 809 }
 810 
 811 /////////////////////////////////////////////////////////////////////////////
 812 // attach existing thread
 813 
 814 // bootstrap the main thread
 815 bool os::create_main_thread(JavaThread* thread) {
 816   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 817   return create_attached_thread(thread);
 818 }
 819 
 820 bool os::create_attached_thread(JavaThread* thread) {
 821 #ifdef ASSERT
 822   thread->verify_not_published();
 823 #endif
 824 
 825   // Allocate the OSThread object
 826   OSThread* osthread = new OSThread(NULL, NULL);
 827 
 828   if (osthread == NULL) {
 829     return false;
 830   }
 831 
 832   // Store pthread info into the OSThread
 833   osthread->set_thread_id(os::Linux::gettid());
 834   osthread->set_pthread_id(::pthread_self());
 835 
 836   // initialize floating point control register
 837   os::Linux::init_thread_fpu_state();
 838 
 839   // Initial thread state is RUNNABLE
 840   osthread->set_state(RUNNABLE);
 841 
 842   thread->set_osthread(osthread);
 843 
 844   if (UseNUMA) {
 845     int lgrp_id = os::numa_get_group_id();
 846     if (lgrp_id != -1) {
 847       thread->set_lgrp_id(lgrp_id);
 848     }
 849   }
 850 
 851   if (os::Linux::is_initial_thread()) {
 852     // If current thread is initial thread, its stack is mapped on demand,
 853     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 854     // the entire stack region to avoid SEGV in stack banging.
 855     // It is also useful to get around the heap-stack-gap problem on SuSE
 856     // kernel (see 4821821 for details). We first expand stack to the top
 857     // of yellow zone, then enable stack yellow zone (order is significant,
 858     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 859     // is no gap between the last two virtual memory regions.
 860 
 861     JavaThread *jt = (JavaThread *)thread;
 862     address addr = jt->stack_reserved_zone_base();
 863     assert(addr != NULL, "initialization problem?");
 864     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 865 
 866     osthread->set_expanding_stack();
 867     os::Linux::manually_expand_stack(jt, addr);
 868     osthread->clear_expanding_stack();
 869   }
 870 
 871   // initialize signal mask for this thread
 872   // and save the caller's signal mask
 873   os::Linux::hotspot_sigmask(thread);
 874 
 875   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 876     os::current_thread_id(), (uintx) pthread_self());
 877 
 878   return true;
 879 }
 880 
 881 void os::pd_start_thread(Thread* thread) {
 882   OSThread * osthread = thread->osthread();
 883   assert(osthread->get_state() != INITIALIZED, "just checking");
 884   Monitor* sync_with_child = osthread->startThread_lock();
 885   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 886   sync_with_child->notify();
 887 }
 888 
 889 // Free Linux resources related to the OSThread
 890 void os::free_thread(OSThread* osthread) {
 891   assert(osthread != NULL, "osthread not set");
 892 
 893   // We are told to free resources of the argument thread,
 894   // but we can only really operate on the current thread.
 895   assert(Thread::current()->osthread() == osthread,
 896          "os::free_thread but not current thread");
 897 
 898 #ifdef ASSERT
 899   sigset_t current;
 900   sigemptyset(&current);
 901   pthread_sigmask(SIG_SETMASK, NULL, &current);
 902   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 903 #endif
 904 
 905   // Restore caller's signal mask
 906   sigset_t sigmask = osthread->caller_sigmask();
 907   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 908 
 909   delete osthread;
 910 }
 911 
 912 //////////////////////////////////////////////////////////////////////////////
 913 // initial thread
 914 
 915 // Check if current thread is the initial thread, similar to Solaris thr_main.
 916 bool os::Linux::is_initial_thread(void) {
 917   char dummy;
 918   // If called before init complete, thread stack bottom will be null.
 919   // Can be called if fatal error occurs before initialization.
 920   if (initial_thread_stack_bottom() == NULL) return false;
 921   assert(initial_thread_stack_bottom() != NULL &&
 922          initial_thread_stack_size()   != 0,
 923          "os::init did not locate initial thread's stack region");
 924   if ((address)&dummy >= initial_thread_stack_bottom() &&
 925       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 926     return true;
 927   } else {
 928     return false;
 929   }
 930 }
 931 
 932 // Find the virtual memory area that contains addr
 933 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 934   FILE *fp = fopen("/proc/self/maps", "r");
 935   if (fp) {
 936     address low, high;
 937     while (!feof(fp)) {
 938       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 939         if (low <= addr && addr < high) {
 940           if (vma_low)  *vma_low  = low;
 941           if (vma_high) *vma_high = high;
 942           fclose(fp);
 943           return true;
 944         }
 945       }
 946       for (;;) {
 947         int ch = fgetc(fp);
 948         if (ch == EOF || ch == (int)'\n') break;
 949       }
 950     }
 951     fclose(fp);
 952   }
 953   return false;
 954 }
 955 
 956 // Locate initial thread stack. This special handling of initial thread stack
 957 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 958 // bogus value for initial thread.
 959 void os::Linux::capture_initial_stack(size_t max_size) {
 960   // stack size is the easy part, get it from RLIMIT_STACK
 961   size_t stack_size;
 962   struct rlimit rlim;
 963   getrlimit(RLIMIT_STACK, &rlim);
 964   stack_size = rlim.rlim_cur;
 965 
 966   // 6308388: a bug in ld.so will relocate its own .data section to the
 967   //   lower end of primordial stack; reduce ulimit -s value a little bit
 968   //   so we won't install guard page on ld.so's data section.
 969   stack_size -= 2 * page_size();
 970 
 971   // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
 972   //   7.1, in both cases we will get 2G in return value.
 973   // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
 974   //   SuSE 7.2, Debian) can not handle alternate signal stack correctly
 975   //   for initial thread if its stack size exceeds 6M. Cap it at 2M,
 976   //   in case other parts in glibc still assumes 2M max stack size.
 977   // FIXME: alt signal stack is gone, maybe we can relax this constraint?
 978   // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
 979   if (stack_size > 2 * K * K IA64_ONLY(*2)) {
 980     stack_size = 2 * K * K IA64_ONLY(*2);
 981   }
 982   // Try to figure out where the stack base (top) is. This is harder.
 983   //
 984   // When an application is started, glibc saves the initial stack pointer in
 985   // a global variable "__libc_stack_end", which is then used by system
 986   // libraries. __libc_stack_end should be pretty close to stack top. The
 987   // variable is available since the very early days. However, because it is
 988   // a private interface, it could disappear in the future.
 989   //
 990   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 991   // to __libc_stack_end, it is very close to stack top, but isn't the real
 992   // stack top. Note that /proc may not exist if VM is running as a chroot
 993   // program, so reading /proc/<pid>/stat could fail. Also the contents of
 994   // /proc/<pid>/stat could change in the future (though unlikely).
 995   //
 996   // We try __libc_stack_end first. If that doesn't work, look for
 997   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
 998   // as a hint, which should work well in most cases.
 999 
1000   uintptr_t stack_start;
1001 
1002   // try __libc_stack_end first
1003   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1004   if (p && *p) {
1005     stack_start = *p;
1006   } else {
1007     // see if we can get the start_stack field from /proc/self/stat
1008     FILE *fp;
1009     int pid;
1010     char state;
1011     int ppid;
1012     int pgrp;
1013     int session;
1014     int nr;
1015     int tpgrp;
1016     unsigned long flags;
1017     unsigned long minflt;
1018     unsigned long cminflt;
1019     unsigned long majflt;
1020     unsigned long cmajflt;
1021     unsigned long utime;
1022     unsigned long stime;
1023     long cutime;
1024     long cstime;
1025     long prio;
1026     long nice;
1027     long junk;
1028     long it_real;
1029     uintptr_t start;
1030     uintptr_t vsize;
1031     intptr_t rss;
1032     uintptr_t rsslim;
1033     uintptr_t scodes;
1034     uintptr_t ecode;
1035     int i;
1036 
1037     // Figure what the primordial thread stack base is. Code is inspired
1038     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1039     // followed by command name surrounded by parentheses, state, etc.
1040     char stat[2048];
1041     int statlen;
1042 
1043     fp = fopen("/proc/self/stat", "r");
1044     if (fp) {
1045       statlen = fread(stat, 1, 2047, fp);
1046       stat[statlen] = '\0';
1047       fclose(fp);
1048 
1049       // Skip pid and the command string. Note that we could be dealing with
1050       // weird command names, e.g. user could decide to rename java launcher
1051       // to "java 1.4.2 :)", then the stat file would look like
1052       //                1234 (java 1.4.2 :)) R ... ...
1053       // We don't really need to know the command string, just find the last
1054       // occurrence of ")" and then start parsing from there. See bug 4726580.
1055       char * s = strrchr(stat, ')');
1056 
1057       i = 0;
1058       if (s) {
1059         // Skip blank chars
1060         do { s++; } while (s && isspace(*s));
1061 
1062 #define _UFM UINTX_FORMAT
1063 #define _DFM INTX_FORMAT
1064 
1065         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1066         //              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
1067         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,
1068                    &state,          // 3  %c
1069                    &ppid,           // 4  %d
1070                    &pgrp,           // 5  %d
1071                    &session,        // 6  %d
1072                    &nr,             // 7  %d
1073                    &tpgrp,          // 8  %d
1074                    &flags,          // 9  %lu
1075                    &minflt,         // 10 %lu
1076                    &cminflt,        // 11 %lu
1077                    &majflt,         // 12 %lu
1078                    &cmajflt,        // 13 %lu
1079                    &utime,          // 14 %lu
1080                    &stime,          // 15 %lu
1081                    &cutime,         // 16 %ld
1082                    &cstime,         // 17 %ld
1083                    &prio,           // 18 %ld
1084                    &nice,           // 19 %ld
1085                    &junk,           // 20 %ld
1086                    &it_real,        // 21 %ld
1087                    &start,          // 22 UINTX_FORMAT
1088                    &vsize,          // 23 UINTX_FORMAT
1089                    &rss,            // 24 INTX_FORMAT
1090                    &rsslim,         // 25 UINTX_FORMAT
1091                    &scodes,         // 26 UINTX_FORMAT
1092                    &ecode,          // 27 UINTX_FORMAT
1093                    &stack_start);   // 28 UINTX_FORMAT
1094       }
1095 
1096 #undef _UFM
1097 #undef _DFM
1098 
1099       if (i != 28 - 2) {
1100         assert(false, "Bad conversion from /proc/self/stat");
1101         // product mode - assume we are the initial thread, good luck in the
1102         // embedded case.
1103         warning("Can't detect initial thread stack location - bad conversion");
1104         stack_start = (uintptr_t) &rlim;
1105       }
1106     } else {
1107       // For some reason we can't open /proc/self/stat (for example, running on
1108       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1109       // most cases, so don't abort:
1110       warning("Can't detect initial thread stack location - no /proc/self/stat");
1111       stack_start = (uintptr_t) &rlim;
1112     }
1113   }
1114 
1115   // Now we have a pointer (stack_start) very close to the stack top, the
1116   // next thing to do is to figure out the exact location of stack top. We
1117   // can find out the virtual memory area that contains stack_start by
1118   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1119   // and its upper limit is the real stack top. (again, this would fail if
1120   // running inside chroot, because /proc may not exist.)
1121 
1122   uintptr_t stack_top;
1123   address low, high;
1124   if (find_vma((address)stack_start, &low, &high)) {
1125     // success, "high" is the true stack top. (ignore "low", because initial
1126     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1127     stack_top = (uintptr_t)high;
1128   } else {
1129     // failed, likely because /proc/self/maps does not exist
1130     warning("Can't detect initial thread stack location - find_vma failed");
1131     // best effort: stack_start is normally within a few pages below the real
1132     // stack top, use it as stack top, and reduce stack size so we won't put
1133     // guard page outside stack.
1134     stack_top = stack_start;
1135     stack_size -= 16 * page_size();
1136   }
1137 
1138   // stack_top could be partially down the page so align it
1139   stack_top = align_size_up(stack_top, page_size());
1140 
1141   if (max_size && stack_size > max_size) {
1142     _initial_thread_stack_size = max_size;
1143   } else {
1144     _initial_thread_stack_size = stack_size;
1145   }
1146 
1147   _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1148   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1149 }
1150 
1151 ////////////////////////////////////////////////////////////////////////////////
1152 // time support
1153 
1154 // Time since start-up in seconds to a fine granularity.
1155 // Used by VMSelfDestructTimer and the MemProfiler.
1156 double os::elapsedTime() {
1157 
1158   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1159 }
1160 
1161 jlong os::elapsed_counter() {
1162   return javaTimeNanos() - initial_time_count;
1163 }
1164 
1165 jlong os::elapsed_frequency() {
1166   return NANOSECS_PER_SEC; // nanosecond resolution
1167 }
1168 
1169 bool os::supports_vtime() { return true; }
1170 bool os::enable_vtime()   { return false; }
1171 bool os::vtime_enabled()  { return false; }
1172 
1173 double os::elapsedVTime() {
1174   struct rusage usage;
1175   int retval = getrusage(RUSAGE_THREAD, &usage);
1176   if (retval == 0) {
1177     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);
1178   } else {
1179     // better than nothing, but not much
1180     return elapsedTime();
1181   }
1182 }
1183 
1184 jlong os::javaTimeMillis() {
1185   timeval time;
1186   int status = gettimeofday(&time, NULL);
1187   assert(status != -1, "linux error");
1188   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1189 }
1190 
1191 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1192   timeval time;
1193   int status = gettimeofday(&time, NULL);
1194   assert(status != -1, "linux error");
1195   seconds = jlong(time.tv_sec);
1196   nanos = jlong(time.tv_usec) * 1000;
1197 }
1198 
1199 
1200 #ifndef CLOCK_MONOTONIC
1201   #define CLOCK_MONOTONIC (1)
1202 #endif
1203 
1204 void os::Linux::clock_init() {
1205   // we do dlopen's in this particular order due to bug in linux
1206   // dynamical loader (see 6348968) leading to crash on exit
1207   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1208   if (handle == NULL) {
1209     handle = dlopen("librt.so", RTLD_LAZY);
1210   }
1211 
1212   if (handle) {
1213     int (*clock_getres_func)(clockid_t, struct timespec*) =
1214            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1215     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1216            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1217     if (clock_getres_func && clock_gettime_func) {
1218       // See if monotonic clock is supported by the kernel. Note that some
1219       // early implementations simply return kernel jiffies (updated every
1220       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1221       // for nano time (though the monotonic property is still nice to have).
1222       // It's fixed in newer kernels, however clock_getres() still returns
1223       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1224       // resolution for now. Hopefully as people move to new kernels, this
1225       // won't be a problem.
1226       struct timespec res;
1227       struct timespec tp;
1228       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1229           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1230         // yes, monotonic clock is supported
1231         _clock_gettime = clock_gettime_func;
1232         return;
1233       } else {
1234         // close librt if there is no monotonic clock
1235         dlclose(handle);
1236       }
1237     }
1238   }
1239   warning("No monotonic clock was available - timed services may " \
1240           "be adversely affected if the time-of-day clock changes");
1241 }
1242 
1243 #ifndef SYS_clock_getres
1244   #if defined(IA32) || defined(AMD64)
1245     #define SYS_clock_getres IA32_ONLY(266)  AMD64_ONLY(229)
1246     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1247   #else
1248     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1249     #define sys_clock_getres(x,y)  -1
1250   #endif
1251 #else
1252   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1253 #endif
1254 
1255 void os::Linux::fast_thread_clock_init() {
1256   if (!UseLinuxPosixThreadCPUClocks) {
1257     return;
1258   }
1259   clockid_t clockid;
1260   struct timespec tp;
1261   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1262       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1263 
1264   // Switch to using fast clocks for thread cpu time if
1265   // the sys_clock_getres() returns 0 error code.
1266   // Note, that some kernels may support the current thread
1267   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1268   // returned by the pthread_getcpuclockid().
1269   // If the fast Posix clocks are supported then the sys_clock_getres()
1270   // must return at least tp.tv_sec == 0 which means a resolution
1271   // better than 1 sec. This is extra check for reliability.
1272 
1273   if (pthread_getcpuclockid_func &&
1274       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1275       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1276     _supports_fast_thread_cpu_time = true;
1277     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1278   }
1279 }
1280 
1281 jlong os::javaTimeNanos() {
1282   if (os::supports_monotonic_clock()) {
1283     struct timespec tp;
1284     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1285     assert(status == 0, "gettime error");
1286     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1287     return result;
1288   } else {
1289     timeval time;
1290     int status = gettimeofday(&time, NULL);
1291     assert(status != -1, "linux error");
1292     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1293     return 1000 * usecs;
1294   }
1295 }
1296 
1297 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1298   if (os::supports_monotonic_clock()) {
1299     info_ptr->max_value = ALL_64_BITS;
1300 
1301     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1302     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1303     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1304   } else {
1305     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1306     info_ptr->max_value = ALL_64_BITS;
1307 
1308     // gettimeofday is a real time clock so it skips
1309     info_ptr->may_skip_backward = true;
1310     info_ptr->may_skip_forward = true;
1311   }
1312 
1313   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1314 }
1315 
1316 // Return the real, user, and system times in seconds from an
1317 // arbitrary fixed point in the past.
1318 bool os::getTimesSecs(double* process_real_time,
1319                       double* process_user_time,
1320                       double* process_system_time) {
1321   struct tms ticks;
1322   clock_t real_ticks = times(&ticks);
1323 
1324   if (real_ticks == (clock_t) (-1)) {
1325     return false;
1326   } else {
1327     double ticks_per_second = (double) clock_tics_per_sec;
1328     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1329     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1330     *process_real_time = ((double) real_ticks) / ticks_per_second;
1331 
1332     return true;
1333   }
1334 }
1335 
1336 
1337 char * os::local_time_string(char *buf, size_t buflen) {
1338   struct tm t;
1339   time_t long_time;
1340   time(&long_time);
1341   localtime_r(&long_time, &t);
1342   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1343                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1344                t.tm_hour, t.tm_min, t.tm_sec);
1345   return buf;
1346 }
1347 
1348 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1349   return localtime_r(clock, res);
1350 }
1351 
1352 ////////////////////////////////////////////////////////////////////////////////
1353 // runtime exit support
1354 
1355 // Note: os::shutdown() might be called very early during initialization, or
1356 // called from signal handler. Before adding something to os::shutdown(), make
1357 // sure it is async-safe and can handle partially initialized VM.
1358 void os::shutdown() {
1359 
1360   // allow PerfMemory to attempt cleanup of any persistent resources
1361   perfMemory_exit();
1362 
1363   // needs to remove object in file system
1364   AttachListener::abort();
1365 
1366   // flush buffered output, finish log files
1367   ostream_abort();
1368 
1369   // Check for abort hook
1370   abort_hook_t abort_hook = Arguments::abort_hook();
1371   if (abort_hook != NULL) {
1372     abort_hook();
1373   }
1374 
1375 }
1376 
1377 // Note: os::abort() might be called very early during initialization, or
1378 // called from signal handler. Before adding something to os::abort(), make
1379 // sure it is async-safe and can handle partially initialized VM.
1380 void os::abort(bool dump_core, void* siginfo, const void* context) {
1381   os::shutdown();
1382   if (dump_core) {
1383 #ifndef PRODUCT
1384     fdStream out(defaultStream::output_fd());
1385     out.print_raw("Current thread is ");
1386     char buf[16];
1387     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1388     out.print_raw_cr(buf);
1389     out.print_raw_cr("Dumping core ...");
1390 #endif
1391     ::abort(); // dump core
1392   }
1393 
1394   ::exit(1);
1395 }
1396 
1397 // Die immediately, no exit hook, no abort hook, no cleanup.
1398 void os::die() {
1399   ::abort();
1400 }
1401 
1402 
1403 // This method is a copy of JDK's sysGetLastErrorString
1404 // from src/solaris/hpi/src/system_md.c
1405 
1406 size_t os::lasterror(char *buf, size_t len) {
1407   if (errno == 0)  return 0;
1408 
1409   const char *s = os::strerror(errno);
1410   size_t n = ::strlen(s);
1411   if (n >= len) {
1412     n = len - 1;
1413   }
1414   ::strncpy(buf, s, n);
1415   buf[n] = '\0';
1416   return n;
1417 }
1418 
1419 // thread_id is kernel thread id (similar to Solaris LWP id)
1420 intx os::current_thread_id() { return os::Linux::gettid(); }
1421 int os::current_process_id() {
1422   return ::getpid();
1423 }
1424 
1425 // DLL functions
1426 
1427 const char* os::dll_file_extension() { return ".so"; }
1428 
1429 // This must be hard coded because it's the system's temporary
1430 // directory not the java application's temp directory, ala java.io.tmpdir.
1431 const char* os::get_temp_directory() { return "/tmp"; }
1432 
1433 static bool file_exists(const char* filename) {
1434   struct stat statbuf;
1435   if (filename == NULL || strlen(filename) == 0) {
1436     return false;
1437   }
1438   return os::stat(filename, &statbuf) == 0;
1439 }
1440 
1441 bool os::dll_build_name(char* buffer, size_t buflen,
1442                         const char* pname, const char* fname) {
1443   bool retval = false;
1444   // Copied from libhpi
1445   const size_t pnamelen = pname ? strlen(pname) : 0;
1446 
1447   // Return error on buffer overflow.
1448   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1449     return retval;
1450   }
1451 
1452   if (pnamelen == 0) {
1453     snprintf(buffer, buflen, "lib%s.so", fname);
1454     retval = true;
1455   } else if (strchr(pname, *os::path_separator()) != NULL) {
1456     int n;
1457     char** pelements = split_path(pname, &n);
1458     if (pelements == NULL) {
1459       return false;
1460     }
1461     for (int i = 0; i < n; i++) {
1462       // Really shouldn't be NULL, but check can't hurt
1463       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1464         continue; // skip the empty path values
1465       }
1466       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1467       if (file_exists(buffer)) {
1468         retval = true;
1469         break;
1470       }
1471     }
1472     // release the storage
1473     for (int i = 0; i < n; i++) {
1474       if (pelements[i] != NULL) {
1475         FREE_C_HEAP_ARRAY(char, pelements[i]);
1476       }
1477     }
1478     if (pelements != NULL) {
1479       FREE_C_HEAP_ARRAY(char*, pelements);
1480     }
1481   } else {
1482     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1483     retval = true;
1484   }
1485   return retval;
1486 }
1487 
1488 // check if addr is inside libjvm.so
1489 bool os::address_is_in_vm(address addr) {
1490   static address libjvm_base_addr;
1491   Dl_info dlinfo;
1492 
1493   if (libjvm_base_addr == NULL) {
1494     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1495       libjvm_base_addr = (address)dlinfo.dli_fbase;
1496     }
1497     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1498   }
1499 
1500   if (dladdr((void *)addr, &dlinfo) != 0) {
1501     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1502   }
1503 
1504   return false;
1505 }
1506 
1507 bool os::dll_address_to_function_name(address addr, char *buf,
1508                                       int buflen, int *offset,
1509                                       bool demangle) {
1510   // buf is not optional, but offset is optional
1511   assert(buf != NULL, "sanity check");
1512 
1513   Dl_info dlinfo;
1514 
1515   if (dladdr((void*)addr, &dlinfo) != 0) {
1516     // see if we have a matching symbol
1517     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1518       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1519         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1520       }
1521       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1522       return true;
1523     }
1524     // no matching symbol so try for just file info
1525     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1526       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1527                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1528         return true;
1529       }
1530     }
1531   }
1532 
1533   buf[0] = '\0';
1534   if (offset != NULL) *offset = -1;
1535   return false;
1536 }
1537 
1538 struct _address_to_library_name {
1539   address addr;          // input : memory address
1540   size_t  buflen;        //         size of fname
1541   char*   fname;         // output: library name
1542   address base;          //         library base addr
1543 };
1544 
1545 static int address_to_library_name_callback(struct dl_phdr_info *info,
1546                                             size_t size, void *data) {
1547   int i;
1548   bool found = false;
1549   address libbase = NULL;
1550   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1551 
1552   // iterate through all loadable segments
1553   for (i = 0; i < info->dlpi_phnum; i++) {
1554     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1555     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1556       // base address of a library is the lowest address of its loaded
1557       // segments.
1558       if (libbase == NULL || libbase > segbase) {
1559         libbase = segbase;
1560       }
1561       // see if 'addr' is within current segment
1562       if (segbase <= d->addr &&
1563           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1564         found = true;
1565       }
1566     }
1567   }
1568 
1569   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1570   // so dll_address_to_library_name() can fall through to use dladdr() which
1571   // can figure out executable name from argv[0].
1572   if (found && info->dlpi_name && info->dlpi_name[0]) {
1573     d->base = libbase;
1574     if (d->fname) {
1575       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1576     }
1577     return 1;
1578   }
1579   return 0;
1580 }
1581 
1582 bool os::dll_address_to_library_name(address addr, char* buf,
1583                                      int buflen, int* offset) {
1584   // buf is not optional, but offset is optional
1585   assert(buf != NULL, "sanity check");
1586 
1587   Dl_info dlinfo;
1588   struct _address_to_library_name data;
1589 
1590   // There is a bug in old glibc dladdr() implementation that it could resolve
1591   // to wrong library name if the .so file has a base address != NULL. Here
1592   // we iterate through the program headers of all loaded libraries to find
1593   // out which library 'addr' really belongs to. This workaround can be
1594   // removed once the minimum requirement for glibc is moved to 2.3.x.
1595   data.addr = addr;
1596   data.fname = buf;
1597   data.buflen = buflen;
1598   data.base = NULL;
1599   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1600 
1601   if (rslt) {
1602     // buf already contains library name
1603     if (offset) *offset = addr - data.base;
1604     return true;
1605   }
1606   if (dladdr((void*)addr, &dlinfo) != 0) {
1607     if (dlinfo.dli_fname != NULL) {
1608       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1609     }
1610     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1611       *offset = addr - (address)dlinfo.dli_fbase;
1612     }
1613     return true;
1614   }
1615 
1616   buf[0] = '\0';
1617   if (offset) *offset = -1;
1618   return false;
1619 }
1620 
1621 // Loads .dll/.so and
1622 // in case of error it checks if .dll/.so was built for the
1623 // same architecture as Hotspot is running on
1624 
1625 
1626 // Remember the stack's state. The Linux dynamic linker will change
1627 // the stack to 'executable' at most once, so we must safepoint only once.
1628 bool os::Linux::_stack_is_executable = false;
1629 
1630 // VM operation that loads a library.  This is necessary if stack protection
1631 // of the Java stacks can be lost during loading the library.  If we
1632 // do not stop the Java threads, they can stack overflow before the stacks
1633 // are protected again.
1634 class VM_LinuxDllLoad: public VM_Operation {
1635  private:
1636   const char *_filename;
1637   char *_ebuf;
1638   int _ebuflen;
1639   void *_lib;
1640  public:
1641   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1642     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1643   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1644   void doit() {
1645     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1646     os::Linux::_stack_is_executable = true;
1647   }
1648   void* loaded_library() { return _lib; }
1649 };
1650 
1651 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1652   void * result = NULL;
1653   bool load_attempted = false;
1654 
1655   // Check whether the library to load might change execution rights
1656   // of the stack. If they are changed, the protection of the stack
1657   // guard pages will be lost. We need a safepoint to fix this.
1658   //
1659   // See Linux man page execstack(8) for more info.
1660   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1661     ElfFile ef(filename);
1662     if (!ef.specifies_noexecstack()) {
1663       if (!is_init_completed()) {
1664         os::Linux::_stack_is_executable = true;
1665         // This is OK - No Java threads have been created yet, and hence no
1666         // stack guard pages to fix.
1667         //
1668         // This should happen only when you are building JDK7 using a very
1669         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1670         //
1671         // Dynamic loader will make all stacks executable after
1672         // this function returns, and will not do that again.
1673         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1674       } else {
1675         warning("You have loaded library %s which might have disabled stack guard. "
1676                 "The VM will try to fix the stack guard now.\n"
1677                 "It's highly recommended that you fix the library with "
1678                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1679                 filename);
1680 
1681         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1682         JavaThread *jt = JavaThread::current();
1683         if (jt->thread_state() != _thread_in_native) {
1684           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1685           // that requires ExecStack. Cannot enter safe point. Let's give up.
1686           warning("Unable to fix stack guard. Giving up.");
1687         } else {
1688           if (!LoadExecStackDllInVMThread) {
1689             // This is for the case where the DLL has an static
1690             // constructor function that executes JNI code. We cannot
1691             // load such DLLs in the VMThread.
1692             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1693           }
1694 
1695           ThreadInVMfromNative tiv(jt);
1696           debug_only(VMNativeEntryWrapper vew;)
1697 
1698           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1699           VMThread::execute(&op);
1700           if (LoadExecStackDllInVMThread) {
1701             result = op.loaded_library();
1702           }
1703           load_attempted = true;
1704         }
1705       }
1706     }
1707   }
1708 
1709   if (!load_attempted) {
1710     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1711   }
1712 
1713   if (result != NULL) {
1714     // Successful loading
1715     return result;
1716   }
1717 
1718   Elf32_Ehdr elf_head;
1719   int diag_msg_max_length=ebuflen-strlen(ebuf);
1720   char* diag_msg_buf=ebuf+strlen(ebuf);
1721 
1722   if (diag_msg_max_length==0) {
1723     // No more space in ebuf for additional diagnostics message
1724     return NULL;
1725   }
1726 
1727 
1728   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1729 
1730   if (file_descriptor < 0) {
1731     // Can't open library, report dlerror() message
1732     return NULL;
1733   }
1734 
1735   bool failed_to_read_elf_head=
1736     (sizeof(elf_head)!=
1737      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1738 
1739   ::close(file_descriptor);
1740   if (failed_to_read_elf_head) {
1741     // file i/o error - report dlerror() msg
1742     return NULL;
1743   }
1744 
1745   typedef struct {
1746     Elf32_Half  code;         // Actual value as defined in elf.h
1747     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1748     char        elf_class;    // 32 or 64 bit
1749     char        endianess;    // MSB or LSB
1750     char*       name;         // String representation
1751   } arch_t;
1752 
1753 #ifndef EM_486
1754   #define EM_486          6               /* Intel 80486 */
1755 #endif
1756 #ifndef EM_AARCH64
1757   #define EM_AARCH64    183               /* ARM AARCH64 */
1758 #endif
1759 
1760   static const arch_t arch_array[]={
1761     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1762     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1763     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1764     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1765     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1766     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1767     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1768     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1769 #if defined(VM_LITTLE_ENDIAN)
1770     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1771 #else
1772     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64 LE"},
1773 #endif
1774     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1775     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1776     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1777     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1778     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1779     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1780     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1781     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1782   };
1783 
1784 #if  (defined IA32)
1785   static  Elf32_Half running_arch_code=EM_386;
1786 #elif   (defined AMD64)
1787   static  Elf32_Half running_arch_code=EM_X86_64;
1788 #elif  (defined IA64)
1789   static  Elf32_Half running_arch_code=EM_IA_64;
1790 #elif  (defined __sparc) && (defined _LP64)
1791   static  Elf32_Half running_arch_code=EM_SPARCV9;
1792 #elif  (defined __sparc) && (!defined _LP64)
1793   static  Elf32_Half running_arch_code=EM_SPARC;
1794 #elif  (defined __powerpc64__)
1795   static  Elf32_Half running_arch_code=EM_PPC64;
1796 #elif  (defined __powerpc__)
1797   static  Elf32_Half running_arch_code=EM_PPC;
1798 #elif  (defined ARM)
1799   static  Elf32_Half running_arch_code=EM_ARM;
1800 #elif  (defined S390)
1801   static  Elf32_Half running_arch_code=EM_S390;
1802 #elif  (defined ALPHA)
1803   static  Elf32_Half running_arch_code=EM_ALPHA;
1804 #elif  (defined MIPSEL)
1805   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1806 #elif  (defined PARISC)
1807   static  Elf32_Half running_arch_code=EM_PARISC;
1808 #elif  (defined MIPS)
1809   static  Elf32_Half running_arch_code=EM_MIPS;
1810 #elif  (defined M68K)
1811   static  Elf32_Half running_arch_code=EM_68K;
1812 #elif  (defined AARCH64)
1813   static  Elf32_Half running_arch_code=EM_AARCH64;
1814 #else
1815     #error Method os::dll_load requires that one of following is defined:\
1816          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1817 #endif
1818 
1819   // Identify compatability class for VM's architecture and library's architecture
1820   // Obtain string descriptions for architectures
1821 
1822   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1823   int running_arch_index=-1;
1824 
1825   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1826     if (running_arch_code == arch_array[i].code) {
1827       running_arch_index    = i;
1828     }
1829     if (lib_arch.code == arch_array[i].code) {
1830       lib_arch.compat_class = arch_array[i].compat_class;
1831       lib_arch.name         = arch_array[i].name;
1832     }
1833   }
1834 
1835   assert(running_arch_index != -1,
1836          "Didn't find running architecture code (running_arch_code) in arch_array");
1837   if (running_arch_index == -1) {
1838     // Even though running architecture detection failed
1839     // we may still continue with reporting dlerror() message
1840     return NULL;
1841   }
1842 
1843   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1844     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1845     return NULL;
1846   }
1847 
1848 #ifndef S390
1849   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1850     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1851     return NULL;
1852   }
1853 #endif // !S390
1854 
1855   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1856     if (lib_arch.name!=NULL) {
1857       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1858                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1859                  lib_arch.name, arch_array[running_arch_index].name);
1860     } else {
1861       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1862                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1863                  lib_arch.code,
1864                  arch_array[running_arch_index].name);
1865     }
1866   }
1867 
1868   return NULL;
1869 }
1870 
1871 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1872                                 int ebuflen) {
1873   void * result = ::dlopen(filename, RTLD_LAZY);
1874   if (result == NULL) {
1875     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1876     ebuf[ebuflen-1] = '\0';
1877   }
1878   return result;
1879 }
1880 
1881 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1882                                        int ebuflen) {
1883   void * result = NULL;
1884   if (LoadExecStackDllInVMThread) {
1885     result = dlopen_helper(filename, ebuf, ebuflen);
1886   }
1887 
1888   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1889   // library that requires an executable stack, or which does not have this
1890   // stack attribute set, dlopen changes the stack attribute to executable. The
1891   // read protection of the guard pages gets lost.
1892   //
1893   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1894   // may have been queued at the same time.
1895 
1896   if (!_stack_is_executable) {
1897     JavaThread *jt = Threads::first();
1898 
1899     while (jt) {
1900       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1901           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1902         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1903           warning("Attempt to reguard stack yellow zone failed.");
1904         }
1905       }
1906       jt = jt->next();
1907     }
1908   }
1909 
1910   return result;
1911 }
1912 
1913 void* os::dll_lookup(void* handle, const char* name) {
1914   void* res = dlsym(handle, name);
1915   return res;
1916 }
1917 
1918 void* os::get_default_process_handle() {
1919   return (void*)::dlopen(NULL, RTLD_LAZY);
1920 }
1921 
1922 static bool _print_ascii_file(const char* filename, outputStream* st) {
1923   int fd = ::open(filename, O_RDONLY);
1924   if (fd == -1) {
1925     return false;
1926   }
1927 
1928   char buf[33];
1929   int bytes;
1930   buf[32] = '\0';
1931   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1932     st->print_raw(buf, bytes);
1933   }
1934 
1935   ::close(fd);
1936 
1937   return true;
1938 }
1939 
1940 void os::print_dll_info(outputStream *st) {
1941   st->print_cr("Dynamic libraries:");
1942 
1943   char fname[32];
1944   pid_t pid = os::Linux::gettid();
1945 
1946   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1947 
1948   if (!_print_ascii_file(fname, st)) {
1949     st->print("Can not get library information for pid = %d\n", pid);
1950   }
1951 }
1952 
1953 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1954   FILE *procmapsFile = NULL;
1955 
1956   // Open the procfs maps file for the current process
1957   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1958     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1959     char line[PATH_MAX + 100];
1960 
1961     // Read line by line from 'file'
1962     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1963       u8 base, top, offset, inode;
1964       char permissions[5];
1965       char device[6];
1966       char name[PATH_MAX + 1];
1967 
1968       // Parse fields from line
1969       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1970              &base, &top, permissions, &offset, device, &inode, name);
1971 
1972       // Filter by device id '00:00' so that we only get file system mapped files.
1973       if (strcmp(device, "00:00") != 0) {
1974 
1975         // Call callback with the fields of interest
1976         if(callback(name, (address)base, (address)top, param)) {
1977           // Oops abort, callback aborted
1978           fclose(procmapsFile);
1979           return 1;
1980         }
1981       }
1982     }
1983     fclose(procmapsFile);
1984   }
1985   return 0;
1986 }
1987 
1988 void os::print_os_info_brief(outputStream* st) {
1989   os::Linux::print_distro_info(st);
1990 
1991   os::Posix::print_uname_info(st);
1992 
1993   os::Linux::print_libversion_info(st);
1994 
1995 }
1996 
1997 void os::print_os_info(outputStream* st) {
1998   st->print("OS:");
1999 
2000   os::Linux::print_distro_info(st);
2001 
2002   os::Posix::print_uname_info(st);
2003 
2004   // Print warning if unsafe chroot environment detected
2005   if (unsafe_chroot_detected) {
2006     st->print("WARNING!! ");
2007     st->print_cr("%s", unstable_chroot_error);
2008   }
2009 
2010   os::Linux::print_libversion_info(st);
2011 
2012   os::Posix::print_rlimit_info(st);
2013 
2014   os::Posix::print_load_average(st);
2015 
2016   os::Linux::print_full_memory_info(st);
2017 }
2018 
2019 // Try to identify popular distros.
2020 // Most Linux distributions have a /etc/XXX-release file, which contains
2021 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2022 // file that also contains the OS version string. Some have more than one
2023 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2024 // /etc/redhat-release.), so the order is important.
2025 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2026 // their own specific XXX-release file as well as a redhat-release file.
2027 // Because of this the XXX-release file needs to be searched for before the
2028 // redhat-release file.
2029 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2030 // search for redhat-release / SuSE-release needs to be before lsb-release.
2031 // Since the lsb-release file is the new standard it needs to be searched
2032 // before the older style release files.
2033 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2034 // next to last resort.  The os-release file is a new standard that contains
2035 // distribution information and the system-release file seems to be an old
2036 // standard that has been replaced by the lsb-release and os-release files.
2037 // Searching for the debian_version file is the last resort.  It contains
2038 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2039 // "Debian " is printed before the contents of the debian_version file.
2040 
2041 const char* distro_files[] = {
2042   "/etc/oracle-release",
2043   "/etc/mandriva-release",
2044   "/etc/mandrake-release",
2045   "/etc/sun-release",
2046   "/etc/redhat-release",
2047   "/etc/SuSE-release",
2048   "/etc/lsb-release",
2049   "/etc/turbolinux-release",
2050   "/etc/gentoo-release",
2051   "/etc/ltib-release",
2052   "/etc/angstrom-version",
2053   "/etc/system-release",
2054   "/etc/os-release",
2055   NULL };
2056 
2057 void os::Linux::print_distro_info(outputStream* st) {
2058   for (int i = 0;; i++) {
2059     const char* file = distro_files[i];
2060     if (file == NULL) {
2061       break;  // done
2062     }
2063     // If file prints, we found it.
2064     if (_print_ascii_file(file, st)) {
2065       return;
2066     }
2067   }
2068 
2069   if (file_exists("/etc/debian_version")) {
2070     st->print("Debian ");
2071     _print_ascii_file("/etc/debian_version", st);
2072   } else {
2073     st->print("Linux");
2074   }
2075   st->cr();
2076 }
2077 
2078 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2079   char buf[256];
2080   while (fgets(buf, sizeof(buf), fp)) {
2081     // Edit out extra stuff in expected format
2082     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2083       char* ptr = strstr(buf, "\"");  // the name is in quotes
2084       if (ptr != NULL) {
2085         ptr++; // go beyond first quote
2086         char* nl = strchr(ptr, '\"');
2087         if (nl != NULL) *nl = '\0';
2088         strncpy(distro, ptr, length);
2089       } else {
2090         ptr = strstr(buf, "=");
2091         ptr++; // go beyond equals then
2092         char* nl = strchr(ptr, '\n');
2093         if (nl != NULL) *nl = '\0';
2094         strncpy(distro, ptr, length);
2095       }
2096       return;
2097     } else if (get_first_line) {
2098       char* nl = strchr(buf, '\n');
2099       if (nl != NULL) *nl = '\0';
2100       strncpy(distro, buf, length);
2101       return;
2102     }
2103   }
2104   // print last line and close
2105   char* nl = strchr(buf, '\n');
2106   if (nl != NULL) *nl = '\0';
2107   strncpy(distro, buf, length);
2108 }
2109 
2110 static void parse_os_info(char* distro, size_t length, const char* file) {
2111   FILE* fp = fopen(file, "r");
2112   if (fp != NULL) {
2113     // if suse format, print out first line
2114     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2115     parse_os_info_helper(fp, distro, length, get_first_line);
2116     fclose(fp);
2117   }
2118 }
2119 
2120 void os::get_summary_os_info(char* buf, size_t buflen) {
2121   for (int i = 0;; i++) {
2122     const char* file = distro_files[i];
2123     if (file == NULL) {
2124       break; // ran out of distro_files
2125     }
2126     if (file_exists(file)) {
2127       parse_os_info(buf, buflen, file);
2128       return;
2129     }
2130   }
2131   // special case for debian
2132   if (file_exists("/etc/debian_version")) {
2133     strncpy(buf, "Debian ", buflen);
2134     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2135   } else {
2136     strncpy(buf, "Linux", buflen);
2137   }
2138 }
2139 
2140 void os::Linux::print_libversion_info(outputStream* st) {
2141   // libc, pthread
2142   st->print("libc:");
2143   st->print("%s ", os::Linux::glibc_version());
2144   st->print("%s ", os::Linux::libpthread_version());
2145   st->cr();
2146 }
2147 
2148 void os::Linux::print_full_memory_info(outputStream* st) {
2149   st->print("\n/proc/meminfo:\n");
2150   _print_ascii_file("/proc/meminfo", st);
2151   st->cr();
2152 }
2153 
2154 void os::print_memory_info(outputStream* st) {
2155 
2156   st->print("Memory:");
2157   st->print(" %dk page", os::vm_page_size()>>10);
2158 
2159   // values in struct sysinfo are "unsigned long"
2160   struct sysinfo si;
2161   sysinfo(&si);
2162 
2163   st->print(", physical " UINT64_FORMAT "k",
2164             os::physical_memory() >> 10);
2165   st->print("(" UINT64_FORMAT "k free)",
2166             os::available_memory() >> 10);
2167   st->print(", swap " UINT64_FORMAT "k",
2168             ((jlong)si.totalswap * si.mem_unit) >> 10);
2169   st->print("(" UINT64_FORMAT "k free)",
2170             ((jlong)si.freeswap * si.mem_unit) >> 10);
2171   st->cr();
2172 }
2173 
2174 // Print the first "model name" line and the first "flags" line
2175 // that we find and nothing more. We assume "model name" comes
2176 // before "flags" so if we find a second "model name", then the
2177 // "flags" field is considered missing.
2178 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2179 #if defined(IA32) || defined(AMD64)
2180   // Other platforms have less repetitive cpuinfo files
2181   FILE *fp = fopen("/proc/cpuinfo", "r");
2182   if (fp) {
2183     while (!feof(fp)) {
2184       if (fgets(buf, buflen, fp)) {
2185         // Assume model name comes before flags
2186         bool model_name_printed = false;
2187         if (strstr(buf, "model name") != NULL) {
2188           if (!model_name_printed) {
2189             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2190             st->print_raw(buf);
2191             model_name_printed = true;
2192           } else {
2193             // model name printed but not flags?  Odd, just return
2194             fclose(fp);
2195             return true;
2196           }
2197         }
2198         // print the flags line too
2199         if (strstr(buf, "flags") != NULL) {
2200           st->print_raw(buf);
2201           fclose(fp);
2202           return true;
2203         }
2204       }
2205     }
2206     fclose(fp);
2207   }
2208 #endif // x86 platforms
2209   return false;
2210 }
2211 
2212 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2213   // Only print the model name if the platform provides this as a summary
2214   if (!print_model_name_and_flags(st, buf, buflen)) {
2215     st->print("\n/proc/cpuinfo:\n");
2216     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2217       st->print_cr("  <Not Available>");
2218     }
2219   }
2220 }
2221 
2222 #if defined(AMD64) || defined(IA32) || defined(X32)
2223 const char* search_string = "model name";
2224 #elif defined(SPARC)
2225 const char* search_string = "cpu";
2226 #elif defined(PPC64)
2227 const char* search_string = "cpu";
2228 #else
2229 const char* search_string = "Processor";
2230 #endif
2231 
2232 // Parses the cpuinfo file for string representing the model name.
2233 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2234   FILE* fp = fopen("/proc/cpuinfo", "r");
2235   if (fp != NULL) {
2236     while (!feof(fp)) {
2237       char buf[256];
2238       if (fgets(buf, sizeof(buf), fp)) {
2239         char* start = strstr(buf, search_string);
2240         if (start != NULL) {
2241           char *ptr = start + strlen(search_string);
2242           char *end = buf + strlen(buf);
2243           while (ptr != end) {
2244              // skip whitespace and colon for the rest of the name.
2245              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2246                break;
2247              }
2248              ptr++;
2249           }
2250           if (ptr != end) {
2251             // reasonable string, get rid of newline and keep the rest
2252             char* nl = strchr(buf, '\n');
2253             if (nl != NULL) *nl = '\0';
2254             strncpy(cpuinfo, ptr, length);
2255             fclose(fp);
2256             return;
2257           }
2258         }
2259       }
2260     }
2261     fclose(fp);
2262   }
2263   // cpuinfo not found or parsing failed, just print generic string.  The entire
2264   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2265 #if defined(AMD64)
2266   strncpy(cpuinfo, "x86_64", length);
2267 #elif defined(IA32)
2268   strncpy(cpuinfo, "x86_32", length);
2269 #elif defined(IA64)
2270   strncpy(cpuinfo, "IA64", length);
2271 #elif defined(SPARC)
2272   strncpy(cpuinfo, "sparcv9", length);
2273 #elif defined(AARCH64)
2274   strncpy(cpuinfo, "AArch64", length);
2275 #elif defined(ARM)
2276   strncpy(cpuinfo, "ARM", length);
2277 #elif defined(PPC)
2278   strncpy(cpuinfo, "PPC64", length);
2279 #elif defined(ZERO_LIBARCH)
2280   strncpy(cpuinfo, ZERO_LIBARCH, length);
2281 #else
2282   strncpy(cpuinfo, "unknown", length);
2283 #endif
2284 }
2285 
2286 static void print_signal_handler(outputStream* st, int sig,
2287                                  char* buf, size_t buflen);
2288 
2289 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2290   st->print_cr("Signal Handlers:");
2291   print_signal_handler(st, SIGSEGV, buf, buflen);
2292   print_signal_handler(st, SIGBUS , buf, buflen);
2293   print_signal_handler(st, SIGFPE , buf, buflen);
2294   print_signal_handler(st, SIGPIPE, buf, buflen);
2295   print_signal_handler(st, SIGXFSZ, buf, buflen);
2296   print_signal_handler(st, SIGILL , buf, buflen);
2297   print_signal_handler(st, SR_signum, buf, buflen);
2298   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2299   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2300   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2301   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2302 #if defined(PPC64)
2303   print_signal_handler(st, SIGTRAP, buf, buflen);
2304 #endif
2305 }
2306 
2307 static char saved_jvm_path[MAXPATHLEN] = {0};
2308 
2309 // Find the full path to the current module, libjvm.so
2310 void os::jvm_path(char *buf, jint buflen) {
2311   // Error checking.
2312   if (buflen < MAXPATHLEN) {
2313     assert(false, "must use a large-enough buffer");
2314     buf[0] = '\0';
2315     return;
2316   }
2317   // Lazy resolve the path to current module.
2318   if (saved_jvm_path[0] != 0) {
2319     strcpy(buf, saved_jvm_path);
2320     return;
2321   }
2322 
2323   char dli_fname[MAXPATHLEN];
2324   bool ret = dll_address_to_library_name(
2325                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2326                                          dli_fname, sizeof(dli_fname), NULL);
2327   assert(ret, "cannot locate libjvm");
2328   char *rp = NULL;
2329   if (ret && dli_fname[0] != '\0') {
2330     rp = realpath(dli_fname, buf);
2331   }
2332   if (rp == NULL) {
2333     return;
2334   }
2335 
2336   if (Arguments::sun_java_launcher_is_altjvm()) {
2337     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2338     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2339     // If "/jre/lib/" appears at the right place in the string, then
2340     // assume we are installed in a JDK and we're done. Otherwise, check
2341     // for a JAVA_HOME environment variable and fix up the path so it
2342     // looks like libjvm.so is installed there (append a fake suffix
2343     // hotspot/libjvm.so).
2344     const char *p = buf + strlen(buf) - 1;
2345     for (int count = 0; p > buf && count < 5; ++count) {
2346       for (--p; p > buf && *p != '/'; --p)
2347         /* empty */ ;
2348     }
2349 
2350     if (strncmp(p, "/jre/lib/", 9) != 0) {
2351       // Look for JAVA_HOME in the environment.
2352       char* java_home_var = ::getenv("JAVA_HOME");
2353       if (java_home_var != NULL && java_home_var[0] != 0) {
2354         char* jrelib_p;
2355         int len;
2356 
2357         // Check the current module name "libjvm.so".
2358         p = strrchr(buf, '/');
2359         if (p == NULL) {
2360           return;
2361         }
2362         assert(strstr(p, "/libjvm") == p, "invalid library name");
2363 
2364         rp = realpath(java_home_var, buf);
2365         if (rp == NULL) {
2366           return;
2367         }
2368 
2369         // determine if this is a legacy image or modules image
2370         // modules image doesn't have "jre" subdirectory
2371         len = strlen(buf);
2372         assert(len < buflen, "Ran out of buffer room");
2373         jrelib_p = buf + len;
2374         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2375         if (0 != access(buf, F_OK)) {
2376           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2377         }
2378 
2379         if (0 == access(buf, F_OK)) {
2380           // Use current module name "libjvm.so"
2381           len = strlen(buf);
2382           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2383         } else {
2384           // Go back to path of .so
2385           rp = realpath(dli_fname, buf);
2386           if (rp == NULL) {
2387             return;
2388           }
2389         }
2390       }
2391     }
2392   }
2393 
2394   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2395   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2396 }
2397 
2398 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2399   // no prefix required, not even "_"
2400 }
2401 
2402 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2403   // no suffix required
2404 }
2405 
2406 ////////////////////////////////////////////////////////////////////////////////
2407 // sun.misc.Signal support
2408 
2409 static volatile jint sigint_count = 0;
2410 
2411 static void UserHandler(int sig, void *siginfo, void *context) {
2412   // 4511530 - sem_post is serialized and handled by the manager thread. When
2413   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2414   // don't want to flood the manager thread with sem_post requests.
2415   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2416     return;
2417   }
2418 
2419   // Ctrl-C is pressed during error reporting, likely because the error
2420   // handler fails to abort. Let VM die immediately.
2421   if (sig == SIGINT && is_error_reported()) {
2422     os::die();
2423   }
2424 
2425   os::signal_notify(sig);
2426 }
2427 
2428 void* os::user_handler() {
2429   return CAST_FROM_FN_PTR(void*, UserHandler);
2430 }
2431 
2432 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2433   struct timespec ts;
2434   // Semaphore's are always associated with CLOCK_REALTIME
2435   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2436   // see unpackTime for discussion on overflow checking
2437   if (sec >= MAX_SECS) {
2438     ts.tv_sec += MAX_SECS;
2439     ts.tv_nsec = 0;
2440   } else {
2441     ts.tv_sec += sec;
2442     ts.tv_nsec += nsec;
2443     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2444       ts.tv_nsec -= NANOSECS_PER_SEC;
2445       ++ts.tv_sec; // note: this must be <= max_secs
2446     }
2447   }
2448 
2449   return ts;
2450 }
2451 
2452 extern "C" {
2453   typedef void (*sa_handler_t)(int);
2454   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2455 }
2456 
2457 void* os::signal(int signal_number, void* handler) {
2458   struct sigaction sigAct, oldSigAct;
2459 
2460   sigfillset(&(sigAct.sa_mask));
2461   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2462   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2463 
2464   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2465     // -1 means registration failed
2466     return (void *)-1;
2467   }
2468 
2469   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2470 }
2471 
2472 void os::signal_raise(int signal_number) {
2473   ::raise(signal_number);
2474 }
2475 
2476 // The following code is moved from os.cpp for making this
2477 // code platform specific, which it is by its very nature.
2478 
2479 // Will be modified when max signal is changed to be dynamic
2480 int os::sigexitnum_pd() {
2481   return NSIG;
2482 }
2483 
2484 // a counter for each possible signal value
2485 static volatile jint pending_signals[NSIG+1] = { 0 };
2486 
2487 // Linux(POSIX) specific hand shaking semaphore.
2488 static sem_t sig_sem;
2489 static PosixSemaphore sr_semaphore;
2490 
2491 void os::signal_init_pd() {
2492   // Initialize signal structures
2493   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2494 
2495   // Initialize signal semaphore
2496   ::sem_init(&sig_sem, 0, 0);
2497 }
2498 
2499 void os::signal_notify(int sig) {
2500   Atomic::inc(&pending_signals[sig]);
2501   ::sem_post(&sig_sem);
2502 }
2503 
2504 static int check_pending_signals(bool wait) {
2505   Atomic::store(0, &sigint_count);
2506   for (;;) {
2507     for (int i = 0; i < NSIG + 1; i++) {
2508       jint n = pending_signals[i];
2509       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2510         return i;
2511       }
2512     }
2513     if (!wait) {
2514       return -1;
2515     }
2516     JavaThread *thread = JavaThread::current();
2517     ThreadBlockInVM tbivm(thread);
2518 
2519     bool threadIsSuspended;
2520     do {
2521       thread->set_suspend_equivalent();
2522       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2523       ::sem_wait(&sig_sem);
2524 
2525       // were we externally suspended while we were waiting?
2526       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2527       if (threadIsSuspended) {
2528         // The semaphore has been incremented, but while we were waiting
2529         // another thread suspended us. We don't want to continue running
2530         // while suspended because that would surprise the thread that
2531         // suspended us.
2532         ::sem_post(&sig_sem);
2533 
2534         thread->java_suspend_self();
2535       }
2536     } while (threadIsSuspended);
2537   }
2538 }
2539 
2540 int os::signal_lookup() {
2541   return check_pending_signals(false);
2542 }
2543 
2544 int os::signal_wait() {
2545   return check_pending_signals(true);
2546 }
2547 
2548 ////////////////////////////////////////////////////////////////////////////////
2549 // Virtual Memory
2550 
2551 int os::vm_page_size() {
2552   // Seems redundant as all get out
2553   assert(os::Linux::page_size() != -1, "must call os::init");
2554   return os::Linux::page_size();
2555 }
2556 
2557 // Solaris allocates memory by pages.
2558 int os::vm_allocation_granularity() {
2559   assert(os::Linux::page_size() != -1, "must call os::init");
2560   return os::Linux::page_size();
2561 }
2562 
2563 // Rationale behind this function:
2564 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2565 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2566 //  samples for JITted code. Here we create private executable mapping over the code cache
2567 //  and then we can use standard (well, almost, as mapping can change) way to provide
2568 //  info for the reporting script by storing timestamp and location of symbol
2569 void linux_wrap_code(char* base, size_t size) {
2570   static volatile jint cnt = 0;
2571 
2572   if (!UseOprofile) {
2573     return;
2574   }
2575 
2576   char buf[PATH_MAX+1];
2577   int num = Atomic::add(1, &cnt);
2578 
2579   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2580            os::get_temp_directory(), os::current_process_id(), num);
2581   unlink(buf);
2582 
2583   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2584 
2585   if (fd != -1) {
2586     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2587     if (rv != (off_t)-1) {
2588       if (::write(fd, "", 1) == 1) {
2589         mmap(base, size,
2590              PROT_READ|PROT_WRITE|PROT_EXEC,
2591              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2592       }
2593     }
2594     ::close(fd);
2595     unlink(buf);
2596   }
2597 }
2598 
2599 static bool recoverable_mmap_error(int err) {
2600   // See if the error is one we can let the caller handle. This
2601   // list of errno values comes from JBS-6843484. I can't find a
2602   // Linux man page that documents this specific set of errno
2603   // values so while this list currently matches Solaris, it may
2604   // change as we gain experience with this failure mode.
2605   switch (err) {
2606   case EBADF:
2607   case EINVAL:
2608   case ENOTSUP:
2609     // let the caller deal with these errors
2610     return true;
2611 
2612   default:
2613     // Any remaining errors on this OS can cause our reserved mapping
2614     // to be lost. That can cause confusion where different data
2615     // structures think they have the same memory mapped. The worst
2616     // scenario is if both the VM and a library think they have the
2617     // same memory mapped.
2618     return false;
2619   }
2620 }
2621 
2622 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2623                                     int err) {
2624   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2625           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2626           os::strerror(err), err);
2627 }
2628 
2629 static void warn_fail_commit_memory(char* addr, size_t size,
2630                                     size_t alignment_hint, bool exec,
2631                                     int err) {
2632   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2633           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2634           alignment_hint, exec, os::strerror(err), err);
2635 }
2636 
2637 // NOTE: Linux kernel does not really reserve the pages for us.
2638 //       All it does is to check if there are enough free pages
2639 //       left at the time of mmap(). This could be a potential
2640 //       problem.
2641 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2642   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2643   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2644                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2645   if (res != (uintptr_t) MAP_FAILED) {
2646     if (UseNUMAInterleaving) {
2647       numa_make_global(addr, size);
2648     }
2649     return 0;
2650   }
2651 
2652   int err = errno;  // save errno from mmap() call above
2653 
2654   if (!recoverable_mmap_error(err)) {
2655     warn_fail_commit_memory(addr, size, exec, err);
2656     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2657   }
2658 
2659   return err;
2660 }
2661 
2662 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2663   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2664 }
2665 
2666 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2667                                   const char* mesg) {
2668   assert(mesg != NULL, "mesg must be specified");
2669   int err = os::Linux::commit_memory_impl(addr, size, exec);
2670   if (err != 0) {
2671     // the caller wants all commit errors to exit with the specified mesg:
2672     warn_fail_commit_memory(addr, size, exec, err);
2673     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2674   }
2675 }
2676 
2677 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2678 #ifndef MAP_HUGETLB
2679   #define MAP_HUGETLB 0x40000
2680 #endif
2681 
2682 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2683 #ifndef MADV_HUGEPAGE
2684   #define MADV_HUGEPAGE 14
2685 #endif
2686 
2687 int os::Linux::commit_memory_impl(char* addr, size_t size,
2688                                   size_t alignment_hint, bool exec) {
2689   int err = os::Linux::commit_memory_impl(addr, size, exec);
2690   if (err == 0) {
2691     realign_memory(addr, size, alignment_hint);
2692   }
2693   return err;
2694 }
2695 
2696 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2697                           bool exec) {
2698   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2699 }
2700 
2701 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2702                                   size_t alignment_hint, bool exec,
2703                                   const char* mesg) {
2704   assert(mesg != NULL, "mesg must be specified");
2705   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2706   if (err != 0) {
2707     // the caller wants all commit errors to exit with the specified mesg:
2708     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2709     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2710   }
2711 }
2712 
2713 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2714   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2715     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2716     // be supported or the memory may already be backed by huge pages.
2717     ::madvise(addr, bytes, MADV_HUGEPAGE);
2718   }
2719 }
2720 
2721 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2722   // This method works by doing an mmap over an existing mmaping and effectively discarding
2723   // the existing pages. However it won't work for SHM-based large pages that cannot be
2724   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2725   // small pages on top of the SHM segment. This method always works for small pages, so we
2726   // allow that in any case.
2727   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2728     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2729   }
2730 }
2731 
2732 void os::numa_make_global(char *addr, size_t bytes) {
2733   Linux::numa_interleave_memory(addr, bytes);
2734 }
2735 
2736 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2737 // bind policy to MPOL_PREFERRED for the current thread.
2738 #define USE_MPOL_PREFERRED 0
2739 
2740 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2741   // To make NUMA and large pages more robust when both enabled, we need to ease
2742   // the requirements on where the memory should be allocated. MPOL_BIND is the
2743   // default policy and it will force memory to be allocated on the specified
2744   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2745   // the specified node, but will not force it. Using this policy will prevent
2746   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2747   // free large pages.
2748   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2749   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2750 }
2751 
2752 bool os::numa_topology_changed() { return false; }
2753 
2754 size_t os::numa_get_groups_num() {
2755   int max_node = Linux::numa_max_node();
2756   return max_node > 0 ? max_node + 1 : 1;
2757 }
2758 
2759 int os::numa_get_group_id() {
2760   int cpu_id = Linux::sched_getcpu();
2761   if (cpu_id != -1) {
2762     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2763     if (lgrp_id != -1) {
2764       return lgrp_id;
2765     }
2766   }
2767   return 0;
2768 }
2769 
2770 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2771   for (size_t i = 0; i < size; i++) {
2772     ids[i] = i;
2773   }
2774   return size;
2775 }
2776 
2777 bool os::get_page_info(char *start, page_info* info) {
2778   return false;
2779 }
2780 
2781 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2782                      page_info* page_found) {
2783   return end;
2784 }
2785 
2786 
2787 int os::Linux::sched_getcpu_syscall(void) {
2788   unsigned int cpu = 0;
2789   int retval = -1;
2790 
2791 #if defined(IA32)
2792   #ifndef SYS_getcpu
2793     #define SYS_getcpu 318
2794   #endif
2795   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2796 #elif defined(AMD64)
2797 // Unfortunately we have to bring all these macros here from vsyscall.h
2798 // to be able to compile on old linuxes.
2799   #define __NR_vgetcpu 2
2800   #define VSYSCALL_START (-10UL << 20)
2801   #define VSYSCALL_SIZE 1024
2802   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2803   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2804   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2805   retval = vgetcpu(&cpu, NULL, NULL);
2806 #endif
2807 
2808   return (retval == -1) ? retval : cpu;
2809 }
2810 
2811 // Something to do with the numa-aware allocator needs these symbols
2812 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2813 extern "C" JNIEXPORT void numa_error(char *where) { }
2814 
2815 
2816 // If we are running with libnuma version > 2, then we should
2817 // be trying to use symbols with versions 1.1
2818 // If we are running with earlier version, which did not have symbol versions,
2819 // we should use the base version.
2820 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2821   void *f = dlvsym(handle, name, "libnuma_1.1");
2822   if (f == NULL) {
2823     f = dlsym(handle, name);
2824   }
2825   return f;
2826 }
2827 
2828 bool os::Linux::libnuma_init() {
2829   // sched_getcpu() should be in libc.
2830   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2831                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2832 
2833   // If it's not, try a direct syscall.
2834   if (sched_getcpu() == -1) {
2835     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2836                                     (void*)&sched_getcpu_syscall));
2837   }
2838 
2839   if (sched_getcpu() != -1) { // Does it work?
2840     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2841     if (handle != NULL) {
2842       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2843                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2844       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2845                                        libnuma_dlsym(handle, "numa_max_node")));
2846       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2847                                         libnuma_dlsym(handle, "numa_available")));
2848       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2849                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2850       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2851                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2852       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2853                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2854 
2855 
2856       if (numa_available() != -1) {
2857         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2858         // Create a cpu -> node mapping
2859         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2860         rebuild_cpu_to_node_map();
2861         return true;
2862       }
2863     }
2864   }
2865   return false;
2866 }
2867 
2868 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2869 // The table is later used in get_node_by_cpu().
2870 void os::Linux::rebuild_cpu_to_node_map() {
2871   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2872                               // in libnuma (possible values are starting from 16,
2873                               // and continuing up with every other power of 2, but less
2874                               // than the maximum number of CPUs supported by kernel), and
2875                               // is a subject to change (in libnuma version 2 the requirements
2876                               // are more reasonable) we'll just hardcode the number they use
2877                               // in the library.
2878   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2879 
2880   size_t cpu_num = os::active_processor_count();
2881   size_t cpu_map_size = NCPUS / BitsPerCLong;
2882   size_t cpu_map_valid_size =
2883     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2884 
2885   cpu_to_node()->clear();
2886   cpu_to_node()->at_grow(cpu_num - 1);
2887   size_t node_num = numa_get_groups_num();
2888 
2889   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2890   for (size_t i = 0; i < node_num; i++) {
2891     if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2892       for (size_t j = 0; j < cpu_map_valid_size; j++) {
2893         if (cpu_map[j] != 0) {
2894           for (size_t k = 0; k < BitsPerCLong; k++) {
2895             if (cpu_map[j] & (1UL << k)) {
2896               cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2897             }
2898           }
2899         }
2900       }
2901     }
2902   }
2903   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2904 }
2905 
2906 int os::Linux::get_node_by_cpu(int cpu_id) {
2907   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2908     return cpu_to_node()->at(cpu_id);
2909   }
2910   return -1;
2911 }
2912 
2913 GrowableArray<int>* os::Linux::_cpu_to_node;
2914 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2915 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2916 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2917 os::Linux::numa_available_func_t os::Linux::_numa_available;
2918 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2919 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2920 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2921 unsigned long* os::Linux::_numa_all_nodes;
2922 
2923 bool os::pd_uncommit_memory(char* addr, size_t size) {
2924   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2925                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2926   return res  != (uintptr_t) MAP_FAILED;
2927 }
2928 
2929 static address get_stack_commited_bottom(address bottom, size_t size) {
2930   address nbot = bottom;
2931   address ntop = bottom + size;
2932 
2933   size_t page_sz = os::vm_page_size();
2934   unsigned pages = size / page_sz;
2935 
2936   unsigned char vec[1];
2937   unsigned imin = 1, imax = pages + 1, imid;
2938   int mincore_return_value = 0;
2939 
2940   assert(imin <= imax, "Unexpected page size");
2941 
2942   while (imin < imax) {
2943     imid = (imax + imin) / 2;
2944     nbot = ntop - (imid * page_sz);
2945 
2946     // Use a trick with mincore to check whether the page is mapped or not.
2947     // mincore sets vec to 1 if page resides in memory and to 0 if page
2948     // is swapped output but if page we are asking for is unmapped
2949     // it returns -1,ENOMEM
2950     mincore_return_value = mincore(nbot, page_sz, vec);
2951 
2952     if (mincore_return_value == -1) {
2953       // Page is not mapped go up
2954       // to find first mapped page
2955       if (errno != EAGAIN) {
2956         assert(errno == ENOMEM, "Unexpected mincore errno");
2957         imax = imid;
2958       }
2959     } else {
2960       // Page is mapped go down
2961       // to find first not mapped page
2962       imin = imid + 1;
2963     }
2964   }
2965 
2966   nbot = nbot + page_sz;
2967 
2968   // Adjust stack bottom one page up if last checked page is not mapped
2969   if (mincore_return_value == -1) {
2970     nbot = nbot + page_sz;
2971   }
2972 
2973   return nbot;
2974 }
2975 
2976 
2977 // Linux uses a growable mapping for the stack, and if the mapping for
2978 // the stack guard pages is not removed when we detach a thread the
2979 // stack cannot grow beyond the pages where the stack guard was
2980 // mapped.  If at some point later in the process the stack expands to
2981 // that point, the Linux kernel cannot expand the stack any further
2982 // because the guard pages are in the way, and a segfault occurs.
2983 //
2984 // However, it's essential not to split the stack region by unmapping
2985 // a region (leaving a hole) that's already part of the stack mapping,
2986 // so if the stack mapping has already grown beyond the guard pages at
2987 // the time we create them, we have to truncate the stack mapping.
2988 // So, we need to know the extent of the stack mapping when
2989 // create_stack_guard_pages() is called.
2990 
2991 // We only need this for stacks that are growable: at the time of
2992 // writing thread stacks don't use growable mappings (i.e. those
2993 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2994 // only applies to the main thread.
2995 
2996 // If the (growable) stack mapping already extends beyond the point
2997 // where we're going to put our guard pages, truncate the mapping at
2998 // that point by munmap()ping it.  This ensures that when we later
2999 // munmap() the guard pages we don't leave a hole in the stack
3000 // mapping. This only affects the main/initial thread
3001 
3002 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3003   if (os::Linux::is_initial_thread()) {
3004     // As we manually grow stack up to bottom inside create_attached_thread(),
3005     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3006     // we don't need to do anything special.
3007     // Check it first, before calling heavy function.
3008     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3009     unsigned char vec[1];
3010 
3011     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3012       // Fallback to slow path on all errors, including EAGAIN
3013       stack_extent = (uintptr_t) get_stack_commited_bottom(
3014                                                            os::Linux::initial_thread_stack_bottom(),
3015                                                            (size_t)addr - stack_extent);
3016     }
3017 
3018     if (stack_extent < (uintptr_t)addr) {
3019       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3020     }
3021   }
3022 
3023   return os::commit_memory(addr, size, !ExecMem);
3024 }
3025 
3026 // If this is a growable mapping, remove the guard pages entirely by
3027 // munmap()ping them.  If not, just call uncommit_memory(). This only
3028 // affects the main/initial thread, but guard against future OS changes
3029 // It's safe to always unmap guard pages for initial thread because we
3030 // always place it right after end of the mapped region
3031 
3032 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3033   uintptr_t stack_extent, stack_base;
3034 
3035   if (os::Linux::is_initial_thread()) {
3036     return ::munmap(addr, size) == 0;
3037   }
3038 
3039   return os::uncommit_memory(addr, size);
3040 }
3041 
3042 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3043 // at 'requested_addr'. If there are existing memory mappings at the same
3044 // location, however, they will be overwritten. If 'fixed' is false,
3045 // 'requested_addr' is only treated as a hint, the return value may or
3046 // may not start from the requested address. Unlike Linux mmap(), this
3047 // function returns NULL to indicate failure.
3048 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3049   char * addr;
3050   int flags;
3051 
3052   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3053   if (fixed) {
3054     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3055     flags |= MAP_FIXED;
3056   }
3057 
3058   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3059   // touch an uncommitted page. Otherwise, the read/write might
3060   // succeed if we have enough swap space to back the physical page.
3061   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3062                        flags, -1, 0);
3063 
3064   return addr == MAP_FAILED ? NULL : addr;
3065 }
3066 
3067 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3068 //   (req_addr != NULL) or with a given alignment.
3069 //  - bytes shall be a multiple of alignment.
3070 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3071 //  - alignment sets the alignment at which memory shall be allocated.
3072 //     It must be a multiple of allocation granularity.
3073 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3074 //  req_addr or NULL.
3075 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3076 
3077   size_t extra_size = bytes;
3078   if (req_addr == NULL && alignment > 0) {
3079     extra_size += alignment;
3080   }
3081 
3082   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3083     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3084     -1, 0);
3085   if (start == MAP_FAILED) {
3086     start = NULL;
3087   } else {
3088     if (req_addr != NULL) {
3089       if (start != req_addr) {
3090         ::munmap(start, extra_size);
3091         start = NULL;
3092       }
3093     } else {
3094       char* const start_aligned = (char*) align_ptr_up(start, alignment);
3095       char* const end_aligned = start_aligned + bytes;
3096       char* const end = start + extra_size;
3097       if (start_aligned > start) {
3098         ::munmap(start, start_aligned - start);
3099       }
3100       if (end_aligned < end) {
3101         ::munmap(end_aligned, end - end_aligned);
3102       }
3103       start = start_aligned;
3104     }
3105   }
3106   return start;
3107 }
3108 
3109 static int anon_munmap(char * addr, size_t size) {
3110   return ::munmap(addr, size) == 0;
3111 }
3112 
3113 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3114                             size_t alignment_hint, int file_desc) {
3115   if (file_desc != -1) {
3116     return map_memory_to_file(requested_addr, bytes, file_desc);
3117   }
3118   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3119 }
3120 
3121 bool os::pd_release_memory(char* addr, size_t size) {
3122   return anon_munmap(addr, size);
3123 }
3124 
3125 static bool linux_mprotect(char* addr, size_t size, int prot) {
3126   // Linux wants the mprotect address argument to be page aligned.
3127   char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3128 
3129   // According to SUSv3, mprotect() should only be used with mappings
3130   // established by mmap(), and mmap() always maps whole pages. Unaligned
3131   // 'addr' likely indicates problem in the VM (e.g. trying to change
3132   // protection of malloc'ed or statically allocated memory). Check the
3133   // caller if you hit this assert.
3134   assert(addr == bottom, "sanity check");
3135 
3136   size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3137   return ::mprotect(bottom, size, prot) == 0;
3138 }
3139 
3140 // Set protections specified
3141 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3142                         bool is_committed) {
3143   unsigned int p = 0;
3144   switch (prot) {
3145   case MEM_PROT_NONE: p = PROT_NONE; break;
3146   case MEM_PROT_READ: p = PROT_READ; break;
3147   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3148   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3149   default:
3150     ShouldNotReachHere();
3151   }
3152   // is_committed is unused.
3153   return linux_mprotect(addr, bytes, p);
3154 }
3155 
3156 bool os::guard_memory(char* addr, size_t size) {
3157   return linux_mprotect(addr, size, PROT_NONE);
3158 }
3159 
3160 bool os::unguard_memory(char* addr, size_t size) {
3161   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3162 }
3163 
3164 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3165                                                     size_t page_size) {
3166   bool result = false;
3167   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3168                  MAP_ANONYMOUS|MAP_PRIVATE,
3169                  -1, 0);
3170   if (p != MAP_FAILED) {
3171     void *aligned_p = align_ptr_up(p, page_size);
3172 
3173     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3174 
3175     munmap(p, page_size * 2);
3176   }
3177 
3178   if (warn && !result) {
3179     warning("TransparentHugePages is not supported by the operating system.");
3180   }
3181 
3182   return result;
3183 }
3184 
3185 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3186   bool result = false;
3187   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3188                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3189                  -1, 0);
3190 
3191   if (p != MAP_FAILED) {
3192     // We don't know if this really is a huge page or not.
3193     FILE *fp = fopen("/proc/self/maps", "r");
3194     if (fp) {
3195       while (!feof(fp)) {
3196         char chars[257];
3197         long x = 0;
3198         if (fgets(chars, sizeof(chars), fp)) {
3199           if (sscanf(chars, "%lx-%*x", &x) == 1
3200               && x == (long)p) {
3201             if (strstr (chars, "hugepage")) {
3202               result = true;
3203               break;
3204             }
3205           }
3206         }
3207       }
3208       fclose(fp);
3209     }
3210     munmap(p, page_size);
3211   }
3212 
3213   if (warn && !result) {
3214     warning("HugeTLBFS is not supported by the operating system.");
3215   }
3216 
3217   return result;
3218 }
3219 
3220 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3221 //
3222 // From the coredump_filter documentation:
3223 //
3224 // - (bit 0) anonymous private memory
3225 // - (bit 1) anonymous shared memory
3226 // - (bit 2) file-backed private memory
3227 // - (bit 3) file-backed shared memory
3228 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3229 //           effective only if the bit 2 is cleared)
3230 // - (bit 5) hugetlb private memory
3231 // - (bit 6) hugetlb shared memory
3232 //
3233 static void set_coredump_filter(void) {
3234   FILE *f;
3235   long cdm;
3236 
3237   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3238     return;
3239   }
3240 
3241   if (fscanf(f, "%lx", &cdm) != 1) {
3242     fclose(f);
3243     return;
3244   }
3245 
3246   rewind(f);
3247 
3248   if ((cdm & LARGEPAGES_BIT) == 0) {
3249     cdm |= LARGEPAGES_BIT;
3250     fprintf(f, "%#lx", cdm);
3251   }
3252 
3253   fclose(f);
3254 }
3255 
3256 // Large page support
3257 
3258 static size_t _large_page_size = 0;
3259 
3260 size_t os::Linux::find_large_page_size() {
3261   size_t large_page_size = 0;
3262 
3263   // large_page_size on Linux is used to round up heap size. x86 uses either
3264   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3265   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3266   // page as large as 256M.
3267   //
3268   // Here we try to figure out page size by parsing /proc/meminfo and looking
3269   // for a line with the following format:
3270   //    Hugepagesize:     2048 kB
3271   //
3272   // If we can't determine the value (e.g. /proc is not mounted, or the text
3273   // format has been changed), we'll use the largest page size supported by
3274   // the processor.
3275 
3276 #ifndef ZERO
3277   large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3278                      ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3279 #endif // ZERO
3280 
3281   FILE *fp = fopen("/proc/meminfo", "r");
3282   if (fp) {
3283     while (!feof(fp)) {
3284       int x = 0;
3285       char buf[16];
3286       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3287         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3288           large_page_size = x * K;
3289           break;
3290         }
3291       } else {
3292         // skip to next line
3293         for (;;) {
3294           int ch = fgetc(fp);
3295           if (ch == EOF || ch == (int)'\n') break;
3296         }
3297       }
3298     }
3299     fclose(fp);
3300   }
3301 
3302   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3303     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3304             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3305             proper_unit_for_byte_size(large_page_size));
3306   }
3307 
3308   return large_page_size;
3309 }
3310 
3311 size_t os::Linux::setup_large_page_size() {
3312   _large_page_size = Linux::find_large_page_size();
3313   const size_t default_page_size = (size_t)Linux::page_size();
3314   if (_large_page_size > default_page_size) {
3315     _page_sizes[0] = _large_page_size;
3316     _page_sizes[1] = default_page_size;
3317     _page_sizes[2] = 0;
3318   }
3319 
3320   return _large_page_size;
3321 }
3322 
3323 bool os::Linux::setup_large_page_type(size_t page_size) {
3324   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3325       FLAG_IS_DEFAULT(UseSHM) &&
3326       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3327 
3328     // The type of large pages has not been specified by the user.
3329 
3330     // Try UseHugeTLBFS and then UseSHM.
3331     UseHugeTLBFS = UseSHM = true;
3332 
3333     // Don't try UseTransparentHugePages since there are known
3334     // performance issues with it turned on. This might change in the future.
3335     UseTransparentHugePages = false;
3336   }
3337 
3338   if (UseTransparentHugePages) {
3339     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3340     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3341       UseHugeTLBFS = false;
3342       UseSHM = false;
3343       return true;
3344     }
3345     UseTransparentHugePages = false;
3346   }
3347 
3348   if (UseHugeTLBFS) {
3349     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3350     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3351       UseSHM = false;
3352       return true;
3353     }
3354     UseHugeTLBFS = false;
3355   }
3356 
3357   return UseSHM;
3358 }
3359 
3360 void os::large_page_init() {
3361   if (!UseLargePages &&
3362       !UseTransparentHugePages &&
3363       !UseHugeTLBFS &&
3364       !UseSHM) {
3365     // Not using large pages.
3366     return;
3367   }
3368 
3369   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3370     // The user explicitly turned off large pages.
3371     // Ignore the rest of the large pages flags.
3372     UseTransparentHugePages = false;
3373     UseHugeTLBFS = false;
3374     UseSHM = false;
3375     return;
3376   }
3377 
3378   size_t large_page_size = Linux::setup_large_page_size();
3379   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3380 
3381   set_coredump_filter();
3382 }
3383 
3384 #ifndef SHM_HUGETLB
3385   #define SHM_HUGETLB 04000
3386 #endif
3387 
3388 #define shm_warning_format(format, ...)              \
3389   do {                                               \
3390     if (UseLargePages &&                             \
3391         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3392          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3393          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3394       warning(format, __VA_ARGS__);                  \
3395     }                                                \
3396   } while (0)
3397 
3398 #define shm_warning(str) shm_warning_format("%s", str)
3399 
3400 #define shm_warning_with_errno(str)                \
3401   do {                                             \
3402     int err = errno;                               \
3403     shm_warning_format(str " (error = %d)", err);  \
3404   } while (0)
3405 
3406 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3407   assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3408 
3409   if (!is_size_aligned(alignment, SHMLBA)) {
3410     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3411     return NULL;
3412   }
3413 
3414   // To ensure that we get 'alignment' aligned memory from shmat,
3415   // we pre-reserve aligned virtual memory and then attach to that.
3416 
3417   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3418   if (pre_reserved_addr == NULL) {
3419     // Couldn't pre-reserve aligned memory.
3420     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3421     return NULL;
3422   }
3423 
3424   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3425   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3426 
3427   if ((intptr_t)addr == -1) {
3428     int err = errno;
3429     shm_warning_with_errno("Failed to attach shared memory.");
3430 
3431     assert(err != EACCES, "Unexpected error");
3432     assert(err != EIDRM,  "Unexpected error");
3433     assert(err != EINVAL, "Unexpected error");
3434 
3435     // Since we don't know if the kernel unmapped the pre-reserved memory area
3436     // we can't unmap it, since that would potentially unmap memory that was
3437     // mapped from other threads.
3438     return NULL;
3439   }
3440 
3441   return addr;
3442 }
3443 
3444 static char* shmat_at_address(int shmid, char* req_addr) {
3445   if (!is_ptr_aligned(req_addr, SHMLBA)) {
3446     assert(false, "Requested address needs to be SHMLBA aligned");
3447     return NULL;
3448   }
3449 
3450   char* addr = (char*)shmat(shmid, req_addr, 0);
3451 
3452   if ((intptr_t)addr == -1) {
3453     shm_warning_with_errno("Failed to attach shared memory.");
3454     return NULL;
3455   }
3456 
3457   return addr;
3458 }
3459 
3460 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3461   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3462   if (req_addr != NULL) {
3463     assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3464     assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3465     return shmat_at_address(shmid, req_addr);
3466   }
3467 
3468   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3469   // return large page size aligned memory addresses when req_addr == NULL.
3470   // However, if the alignment is larger than the large page size, we have
3471   // to manually ensure that the memory returned is 'alignment' aligned.
3472   if (alignment > os::large_page_size()) {
3473     assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3474     return shmat_with_alignment(shmid, bytes, alignment);
3475   } else {
3476     return shmat_at_address(shmid, NULL);
3477   }
3478 }
3479 
3480 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3481                                             char* req_addr, bool exec) {
3482   // "exec" is passed in but not used.  Creating the shared image for
3483   // the code cache doesn't have an SHM_X executable permission to check.
3484   assert(UseLargePages && UseSHM, "only for SHM large pages");
3485   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3486   assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3487 
3488   if (!is_size_aligned(bytes, os::large_page_size())) {
3489     return NULL; // Fallback to small pages.
3490   }
3491 
3492   // Create a large shared memory region to attach to based on size.
3493   // Currently, size is the total size of the heap.
3494   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3495   if (shmid == -1) {
3496     // Possible reasons for shmget failure:
3497     // 1. shmmax is too small for Java heap.
3498     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3499     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3500     // 2. not enough large page memory.
3501     //    > check available large pages: cat /proc/meminfo
3502     //    > increase amount of large pages:
3503     //          echo new_value > /proc/sys/vm/nr_hugepages
3504     //      Note 1: different Linux may use different name for this property,
3505     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3506     //      Note 2: it's possible there's enough physical memory available but
3507     //            they are so fragmented after a long run that they can't
3508     //            coalesce into large pages. Try to reserve large pages when
3509     //            the system is still "fresh".
3510     shm_warning_with_errno("Failed to reserve shared memory.");
3511     return NULL;
3512   }
3513 
3514   // Attach to the region.
3515   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3516 
3517   // Remove shmid. If shmat() is successful, the actual shared memory segment
3518   // will be deleted when it's detached by shmdt() or when the process
3519   // terminates. If shmat() is not successful this will remove the shared
3520   // segment immediately.
3521   shmctl(shmid, IPC_RMID, NULL);
3522 
3523   return addr;
3524 }
3525 
3526 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3527                                         int error) {
3528   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3529 
3530   bool warn_on_failure = UseLargePages &&
3531       (!FLAG_IS_DEFAULT(UseLargePages) ||
3532        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3533        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3534 
3535   if (warn_on_failure) {
3536     char msg[128];
3537     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3538                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3539     warning("%s", msg);
3540   }
3541 }
3542 
3543 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3544                                                         char* req_addr,
3545                                                         bool exec) {
3546   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3547   assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3548   assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3549 
3550   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3551   char* addr = (char*)::mmap(req_addr, bytes, prot,
3552                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3553                              -1, 0);
3554 
3555   if (addr == MAP_FAILED) {
3556     warn_on_large_pages_failure(req_addr, bytes, errno);
3557     return NULL;
3558   }
3559 
3560   assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3561 
3562   return addr;
3563 }
3564 
3565 // Helper function to create a temp file in the given directory
3566 int os::create_file_for_heap(const char* dir, size_t size) {
3567 
3568   char name_template[] = "/jvmheap.XXXXXX";
3569 
3570   char *fullname = (char*)alloca(strlen(dir) + sizeof(name_template));
3571   (void)strcpy(fullname, dir);
3572   (void)strcat(fullname, name_template);
3573 
3574   sigset_t set, oldset;
3575   sigfillset(&set);
3576 
3577   // block all signals while we do the file operation
3578   (void)sigprocmask(SIG_BLOCK, &set, &oldset);
3579 
3580   // set the file creation mask
3581   mode_t new_mask = S_IRUSR | S_IWUSR;
3582   mode_t prev_umask = umask(new_mask);
3583 
3584   // create a new file
3585   int fd = mkstemp(fullname);
3586 
3587   // reset the file creation mask
3588   umask(prev_umask);
3589 
3590   if (fd < 0) {
3591     warning("Could not create file for heap");
3592     return -1;
3593   }
3594 
3595   // delete the name from the filesystem. When 'fd' is closed, the file (and space) will be deleted
3596   (void)unlink(fullname);
3597 
3598   // reset the signal mask
3599   (void)sigprocmask(SIG_SETMASK, &oldset, NULL);
3600 
3601   return fd;
3602 }
3603 
3604 // Map the given address range to the provided file descriptor
3605 char* os::map_memory_to_file(char* base, size_t size, int fd) {
3606   assert(fd != -1, "File descriptor is not valid");
3607 
3608   // allocate space for the file
3609   if ((errno = posix_fallocate(fd, 0, (off_t)size)) != 0) {
3610     vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3611     return NULL;
3612   }
3613 
3614   int prot = PROT_READ | PROT_WRITE;
3615   int flags = MAP_SHARED;
3616   if( base != NULL) {
3617     flags |= MAP_FIXED;
3618   }
3619   char* addr = (char*)mmap(base, size, prot, flags, fd, 0);
3620 
3621   if (addr == MAP_FAILED || (base != NULL && addr != base)) {
3622     if (addr != MAP_FAILED) {
3623       if (!pd_release_memory(addr, size)) {
3624         warning("Could not release memory on unsuccessful file mapping");
3625       }
3626     }
3627     return NULL;
3628   }
3629 
3630   return addr;
3631 }
3632 
3633 // Reserve memory using mmap(MAP_HUGETLB).
3634 //  - bytes shall be a multiple of alignment.
3635 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3636 //  - alignment sets the alignment at which memory shall be allocated.
3637 //     It must be a multiple of allocation granularity.
3638 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3639 //  req_addr or NULL.
3640 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3641                                                          size_t alignment,
3642                                                          char* req_addr,
3643                                                          bool exec) {
3644   size_t large_page_size = os::large_page_size();
3645   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3646 
3647   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3648   assert(is_size_aligned(bytes, alignment), "Must be");
3649 
3650   // First reserve - but not commit - the address range in small pages.
3651   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3652 
3653   if (start == NULL) {
3654     return NULL;
3655   }
3656 
3657   assert(is_ptr_aligned(start, alignment), "Must be");
3658 
3659   char* end = start + bytes;
3660 
3661   // Find the regions of the allocated chunk that can be promoted to large pages.
3662   char* lp_start = (char*)align_ptr_up(start, large_page_size);
3663   char* lp_end   = (char*)align_ptr_down(end, large_page_size);
3664 
3665   size_t lp_bytes = lp_end - lp_start;
3666 
3667   assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3668 
3669   if (lp_bytes == 0) {
3670     // The mapped region doesn't even span the start and the end of a large page.
3671     // Fall back to allocate a non-special area.
3672     ::munmap(start, end - start);
3673     return NULL;
3674   }
3675 
3676   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3677 
3678   void* result;
3679 
3680   // Commit small-paged leading area.
3681   if (start != lp_start) {
3682     result = ::mmap(start, lp_start - start, prot,
3683                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3684                     -1, 0);
3685     if (result == MAP_FAILED) {
3686       ::munmap(lp_start, end - lp_start);
3687       return NULL;
3688     }
3689   }
3690 
3691   // Commit large-paged area.
3692   result = ::mmap(lp_start, lp_bytes, prot,
3693                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3694                   -1, 0);
3695   if (result == MAP_FAILED) {
3696     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3697     // If the mmap above fails, the large pages region will be unmapped and we
3698     // have regions before and after with small pages. Release these regions.
3699     //
3700     // |  mapped  |  unmapped  |  mapped  |
3701     // ^          ^            ^          ^
3702     // start      lp_start     lp_end     end
3703     //
3704     ::munmap(start, lp_start - start);
3705     ::munmap(lp_end, end - lp_end);
3706     return NULL;
3707   }
3708 
3709   // Commit small-paged trailing area.
3710   if (lp_end != end) {
3711     result = ::mmap(lp_end, end - lp_end, prot,
3712                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3713                     -1, 0);
3714     if (result == MAP_FAILED) {
3715       ::munmap(start, lp_end - start);
3716       return NULL;
3717     }
3718   }
3719 
3720   return start;
3721 }
3722 
3723 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3724                                                    size_t alignment,
3725                                                    char* req_addr,
3726                                                    bool exec) {
3727   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3728   assert(is_ptr_aligned(req_addr, alignment), "Must be");
3729   assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3730   assert(is_power_of_2(os::large_page_size()), "Must be");
3731   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3732 
3733   if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3734     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3735   } else {
3736     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3737   }
3738 }
3739 
3740 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3741                                  char* req_addr, bool exec) {
3742   assert(UseLargePages, "only for large pages");
3743 
3744   char* addr;
3745   if (UseSHM) {
3746     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3747   } else {
3748     assert(UseHugeTLBFS, "must be");
3749     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3750   }
3751 
3752   if (addr != NULL) {
3753     if (UseNUMAInterleaving) {
3754       numa_make_global(addr, bytes);
3755     }
3756 
3757     // The memory is committed
3758     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3759   }
3760 
3761   return addr;
3762 }
3763 
3764 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3765   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3766   return shmdt(base) == 0;
3767 }
3768 
3769 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3770   return pd_release_memory(base, bytes);
3771 }
3772 
3773 bool os::release_memory_special(char* base, size_t bytes) {
3774   bool res;
3775   if (MemTracker::tracking_level() > NMT_minimal) {
3776     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3777     res = os::Linux::release_memory_special_impl(base, bytes);
3778     if (res) {
3779       tkr.record((address)base, bytes);
3780     }
3781 
3782   } else {
3783     res = os::Linux::release_memory_special_impl(base, bytes);
3784   }
3785   return res;
3786 }
3787 
3788 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3789   assert(UseLargePages, "only for large pages");
3790   bool res;
3791 
3792   if (UseSHM) {
3793     res = os::Linux::release_memory_special_shm(base, bytes);
3794   } else {
3795     assert(UseHugeTLBFS, "must be");
3796     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3797   }
3798   return res;
3799 }
3800 
3801 size_t os::large_page_size() {
3802   return _large_page_size;
3803 }
3804 
3805 // With SysV SHM the entire memory region must be allocated as shared
3806 // memory.
3807 // HugeTLBFS allows application to commit large page memory on demand.
3808 // However, when committing memory with HugeTLBFS fails, the region
3809 // that was supposed to be committed will lose the old reservation
3810 // and allow other threads to steal that memory region. Because of this
3811 // behavior we can't commit HugeTLBFS memory.
3812 bool os::can_commit_large_page_memory() {
3813   return UseTransparentHugePages;
3814 }
3815 
3816 bool os::can_execute_large_page_memory() {
3817   return UseTransparentHugePages || UseHugeTLBFS;
3818 }
3819 
3820 // Reserve memory at an arbitrary address, only if that area is
3821 // available (and not reserved for something else).
3822 
3823 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3824   const int max_tries = 10;
3825   char* base[max_tries];
3826   size_t size[max_tries];
3827   const size_t gap = 0x000000;
3828 
3829   // Assert only that the size is a multiple of the page size, since
3830   // that's all that mmap requires, and since that's all we really know
3831   // about at this low abstraction level.  If we need higher alignment,
3832   // we can either pass an alignment to this method or verify alignment
3833   // in one of the methods further up the call chain.  See bug 5044738.
3834   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3835 
3836   // Repeatedly allocate blocks until the block is allocated at the
3837   // right spot.
3838 
3839   // Linux mmap allows caller to pass an address as hint; give it a try first,
3840   // if kernel honors the hint then we can return immediately.
3841   char * addr = anon_mmap(requested_addr, bytes, false);
3842   if (addr == requested_addr) {
3843     if (file_desc == -1 || map_memory_to_file(requested_addr, bytes, file_desc) != NULL) {
3844       return requested_addr;
3845     }
3846   }
3847 
3848   if (addr != NULL) {
3849     // mmap() is successful but it fails to reserve at the requested address
3850     anon_munmap(addr, bytes);
3851   }
3852 
3853   int i;
3854   for (i = 0; i < max_tries; ++i) {
3855     base[i] = reserve_memory(bytes);
3856 
3857     if (base[i] != NULL) {
3858       // Is this the block we wanted?
3859       if (base[i] == requested_addr) {
3860         size[i] = bytes;
3861         break;
3862       }
3863 
3864       // Does this overlap the block we wanted? Give back the overlapped
3865       // parts and try again.
3866 
3867       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3868       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3869         unmap_memory(base[i], top_overlap);
3870         base[i] += top_overlap;
3871         size[i] = bytes - top_overlap;
3872       } else {
3873         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3874         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3875           unmap_memory(requested_addr, bottom_overlap);
3876           size[i] = bytes - bottom_overlap;
3877         } else {
3878           size[i] = bytes;
3879         }
3880       }
3881     }
3882   }
3883 
3884   // Give back the unused reserved pieces.
3885 
3886   for (int j = 0; j < i; ++j) {
3887     if (base[j] != NULL) {
3888       unmap_memory(base[j], size[j]);
3889     }
3890   }
3891 
3892   if (i < max_tries) {
3893     if (file_desc != -1 && map_memory_to_file(requested_addr, bytes, file_desc) == NULL) {
3894       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3895     }
3896     return requested_addr;
3897   } else {
3898     return NULL;
3899   }
3900 }
3901 
3902 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3903   return ::read(fd, buf, nBytes);
3904 }
3905 
3906 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3907   return ::pread(fd, buf, nBytes, offset);
3908 }
3909 
3910 // Short sleep, direct OS call.
3911 //
3912 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3913 // sched_yield(2) will actually give up the CPU:
3914 //
3915 //   * Alone on this pariticular CPU, keeps running.
3916 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3917 //     (pre 2.6.39).
3918 //
3919 // So calling this with 0 is an alternative.
3920 //
3921 void os::naked_short_sleep(jlong ms) {
3922   struct timespec req;
3923 
3924   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3925   req.tv_sec = 0;
3926   if (ms > 0) {
3927     req.tv_nsec = (ms % 1000) * 1000000;
3928   } else {
3929     req.tv_nsec = 1;
3930   }
3931 
3932   nanosleep(&req, NULL);
3933 
3934   return;
3935 }
3936 
3937 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3938 void os::infinite_sleep() {
3939   while (true) {    // sleep forever ...
3940     ::sleep(100);   // ... 100 seconds at a time
3941   }
3942 }
3943 
3944 // Used to convert frequent JVM_Yield() to nops
3945 bool os::dont_yield() {
3946   return DontYieldALot;
3947 }
3948 
3949 void os::naked_yield() {
3950   sched_yield();
3951 }
3952 
3953 ////////////////////////////////////////////////////////////////////////////////
3954 // thread priority support
3955 
3956 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3957 // only supports dynamic priority, static priority must be zero. For real-time
3958 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3959 // However, for large multi-threaded applications, SCHED_RR is not only slower
3960 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3961 // of 5 runs - Sep 2005).
3962 //
3963 // The following code actually changes the niceness of kernel-thread/LWP. It
3964 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3965 // not the entire user process, and user level threads are 1:1 mapped to kernel
3966 // threads. It has always been the case, but could change in the future. For
3967 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3968 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3969 
3970 int os::java_to_os_priority[CriticalPriority + 1] = {
3971   19,              // 0 Entry should never be used
3972 
3973    4,              // 1 MinPriority
3974    3,              // 2
3975    2,              // 3
3976 
3977    1,              // 4
3978    0,              // 5 NormPriority
3979   -1,              // 6
3980 
3981   -2,              // 7
3982   -3,              // 8
3983   -4,              // 9 NearMaxPriority
3984 
3985   -5,              // 10 MaxPriority
3986 
3987   -5               // 11 CriticalPriority
3988 };
3989 
3990 static int prio_init() {
3991   if (ThreadPriorityPolicy == 1) {
3992     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3993     // if effective uid is not root. Perhaps, a more elegant way of doing
3994     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3995     if (geteuid() != 0) {
3996       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3997         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3998       }
3999       ThreadPriorityPolicy = 0;
4000     }
4001   }
4002   if (UseCriticalJavaThreadPriority) {
4003     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4004   }
4005   return 0;
4006 }
4007 
4008 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4009   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4010 
4011   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4012   return (ret == 0) ? OS_OK : OS_ERR;
4013 }
4014 
4015 OSReturn os::get_native_priority(const Thread* const thread,
4016                                  int *priority_ptr) {
4017   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4018     *priority_ptr = java_to_os_priority[NormPriority];
4019     return OS_OK;
4020   }
4021 
4022   errno = 0;
4023   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4024   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4025 }
4026 
4027 // Hint to the underlying OS that a task switch would not be good.
4028 // Void return because it's a hint and can fail.
4029 void os::hint_no_preempt() {}
4030 
4031 ////////////////////////////////////////////////////////////////////////////////
4032 // suspend/resume support
4033 
4034 //  the low-level signal-based suspend/resume support is a remnant from the
4035 //  old VM-suspension that used to be for java-suspension, safepoints etc,
4036 //  within hotspot. Now there is a single use-case for this:
4037 //    - calling get_thread_pc() on the VMThread by the flat-profiler task
4038 //      that runs in the watcher thread.
4039 //  The remaining code is greatly simplified from the more general suspension
4040 //  code that used to be used.
4041 //
4042 //  The protocol is quite simple:
4043 //  - suspend:
4044 //      - sends a signal to the target thread
4045 //      - polls the suspend state of the osthread using a yield loop
4046 //      - target thread signal handler (SR_handler) sets suspend state
4047 //        and blocks in sigsuspend until continued
4048 //  - resume:
4049 //      - sets target osthread state to continue
4050 //      - sends signal to end the sigsuspend loop in the SR_handler
4051 //
4052 //  Note that the SR_lock plays no role in this suspend/resume protocol.
4053 
4054 static void resume_clear_context(OSThread *osthread) {
4055   osthread->set_ucontext(NULL);
4056   osthread->set_siginfo(NULL);
4057 }
4058 
4059 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4060                                  ucontext_t* context) {
4061   osthread->set_ucontext(context);
4062   osthread->set_siginfo(siginfo);
4063 }
4064 
4065 // Handler function invoked when a thread's execution is suspended or
4066 // resumed. We have to be careful that only async-safe functions are
4067 // called here (Note: most pthread functions are not async safe and
4068 // should be avoided.)
4069 //
4070 // Note: sigwait() is a more natural fit than sigsuspend() from an
4071 // interface point of view, but sigwait() prevents the signal hander
4072 // from being run. libpthread would get very confused by not having
4073 // its signal handlers run and prevents sigwait()'s use with the
4074 // mutex granting granting signal.
4075 //
4076 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4077 //
4078 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4079   // Save and restore errno to avoid confusing native code with EINTR
4080   // after sigsuspend.
4081   int old_errno = errno;
4082 
4083   Thread* thread = Thread::current_or_null_safe();
4084   assert(thread != NULL, "Missing current thread in SR_handler");
4085   OSThread* osthread = thread->osthread();
4086   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4087 
4088   os::SuspendResume::State current = osthread->sr.state();
4089   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4090     suspend_save_context(osthread, siginfo, context);
4091 
4092     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4093     os::SuspendResume::State state = osthread->sr.suspended();
4094     if (state == os::SuspendResume::SR_SUSPENDED) {
4095       sigset_t suspend_set;  // signals for sigsuspend()
4096       sigemptyset(&suspend_set);
4097       // get current set of blocked signals and unblock resume signal
4098       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4099       sigdelset(&suspend_set, SR_signum);
4100 
4101       sr_semaphore.signal();
4102       // wait here until we are resumed
4103       while (1) {
4104         sigsuspend(&suspend_set);
4105 
4106         os::SuspendResume::State result = osthread->sr.running();
4107         if (result == os::SuspendResume::SR_RUNNING) {
4108           sr_semaphore.signal();
4109           break;
4110         }
4111       }
4112 
4113     } else if (state == os::SuspendResume::SR_RUNNING) {
4114       // request was cancelled, continue
4115     } else {
4116       ShouldNotReachHere();
4117     }
4118 
4119     resume_clear_context(osthread);
4120   } else if (current == os::SuspendResume::SR_RUNNING) {
4121     // request was cancelled, continue
4122   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4123     // ignore
4124   } else {
4125     // ignore
4126   }
4127 
4128   errno = old_errno;
4129 }
4130 
4131 static int SR_initialize() {
4132   struct sigaction act;
4133   char *s;
4134 
4135   // Get signal number to use for suspend/resume
4136   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4137     int sig = ::strtol(s, 0, 10);
4138     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4139         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4140       SR_signum = sig;
4141     } else {
4142       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4143               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4144     }
4145   }
4146 
4147   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4148          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4149 
4150   sigemptyset(&SR_sigset);
4151   sigaddset(&SR_sigset, SR_signum);
4152 
4153   // Set up signal handler for suspend/resume
4154   act.sa_flags = SA_RESTART|SA_SIGINFO;
4155   act.sa_handler = (void (*)(int)) SR_handler;
4156 
4157   // SR_signum is blocked by default.
4158   // 4528190 - We also need to block pthread restart signal (32 on all
4159   // supported Linux platforms). Note that LinuxThreads need to block
4160   // this signal for all threads to work properly. So we don't have
4161   // to use hard-coded signal number when setting up the mask.
4162   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4163 
4164   if (sigaction(SR_signum, &act, 0) == -1) {
4165     return -1;
4166   }
4167 
4168   // Save signal flag
4169   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4170   return 0;
4171 }
4172 
4173 static int sr_notify(OSThread* osthread) {
4174   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4175   assert_status(status == 0, status, "pthread_kill");
4176   return status;
4177 }
4178 
4179 // "Randomly" selected value for how long we want to spin
4180 // before bailing out on suspending a thread, also how often
4181 // we send a signal to a thread we want to resume
4182 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4183 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4184 
4185 // returns true on success and false on error - really an error is fatal
4186 // but this seems the normal response to library errors
4187 static bool do_suspend(OSThread* osthread) {
4188   assert(osthread->sr.is_running(), "thread should be running");
4189   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4190 
4191   // mark as suspended and send signal
4192   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4193     // failed to switch, state wasn't running?
4194     ShouldNotReachHere();
4195     return false;
4196   }
4197 
4198   if (sr_notify(osthread) != 0) {
4199     ShouldNotReachHere();
4200   }
4201 
4202   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4203   while (true) {
4204     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4205       break;
4206     } else {
4207       // timeout
4208       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4209       if (cancelled == os::SuspendResume::SR_RUNNING) {
4210         return false;
4211       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4212         // make sure that we consume the signal on the semaphore as well
4213         sr_semaphore.wait();
4214         break;
4215       } else {
4216         ShouldNotReachHere();
4217         return false;
4218       }
4219     }
4220   }
4221 
4222   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4223   return true;
4224 }
4225 
4226 static void do_resume(OSThread* osthread) {
4227   assert(osthread->sr.is_suspended(), "thread should be suspended");
4228   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4229 
4230   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4231     // failed to switch to WAKEUP_REQUEST
4232     ShouldNotReachHere();
4233     return;
4234   }
4235 
4236   while (true) {
4237     if (sr_notify(osthread) == 0) {
4238       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4239         if (osthread->sr.is_running()) {
4240           return;
4241         }
4242       }
4243     } else {
4244       ShouldNotReachHere();
4245     }
4246   }
4247 
4248   guarantee(osthread->sr.is_running(), "Must be running!");
4249 }
4250 
4251 ///////////////////////////////////////////////////////////////////////////////////
4252 // signal handling (except suspend/resume)
4253 
4254 // This routine may be used by user applications as a "hook" to catch signals.
4255 // The user-defined signal handler must pass unrecognized signals to this
4256 // routine, and if it returns true (non-zero), then the signal handler must
4257 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4258 // routine will never retun false (zero), but instead will execute a VM panic
4259 // routine kill the process.
4260 //
4261 // If this routine returns false, it is OK to call it again.  This allows
4262 // the user-defined signal handler to perform checks either before or after
4263 // the VM performs its own checks.  Naturally, the user code would be making
4264 // a serious error if it tried to handle an exception (such as a null check
4265 // or breakpoint) that the VM was generating for its own correct operation.
4266 //
4267 // This routine may recognize any of the following kinds of signals:
4268 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4269 // It should be consulted by handlers for any of those signals.
4270 //
4271 // The caller of this routine must pass in the three arguments supplied
4272 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4273 // field of the structure passed to sigaction().  This routine assumes that
4274 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4275 //
4276 // Note that the VM will print warnings if it detects conflicting signal
4277 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4278 //
4279 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4280                                                  siginfo_t* siginfo,
4281                                                  void* ucontext,
4282                                                  int abort_if_unrecognized);
4283 
4284 void signalHandler(int sig, siginfo_t* info, void* uc) {
4285   assert(info != NULL && uc != NULL, "it must be old kernel");
4286   int orig_errno = errno;  // Preserve errno value over signal handler.
4287   JVM_handle_linux_signal(sig, info, uc, true);
4288   errno = orig_errno;
4289 }
4290 
4291 
4292 // This boolean allows users to forward their own non-matching signals
4293 // to JVM_handle_linux_signal, harmlessly.
4294 bool os::Linux::signal_handlers_are_installed = false;
4295 
4296 // For signal-chaining
4297 struct sigaction sigact[NSIG];
4298 uint64_t sigs = 0;
4299 #if (64 < NSIG-1)
4300 #error "Not all signals can be encoded in sigs. Adapt its type!"
4301 #endif
4302 bool os::Linux::libjsig_is_loaded = false;
4303 typedef struct sigaction *(*get_signal_t)(int);
4304 get_signal_t os::Linux::get_signal_action = NULL;
4305 
4306 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4307   struct sigaction *actp = NULL;
4308 
4309   if (libjsig_is_loaded) {
4310     // Retrieve the old signal handler from libjsig
4311     actp = (*get_signal_action)(sig);
4312   }
4313   if (actp == NULL) {
4314     // Retrieve the preinstalled signal handler from jvm
4315     actp = get_preinstalled_handler(sig);
4316   }
4317 
4318   return actp;
4319 }
4320 
4321 static bool call_chained_handler(struct sigaction *actp, int sig,
4322                                  siginfo_t *siginfo, void *context) {
4323   // Call the old signal handler
4324   if (actp->sa_handler == SIG_DFL) {
4325     // It's more reasonable to let jvm treat it as an unexpected exception
4326     // instead of taking the default action.
4327     return false;
4328   } else if (actp->sa_handler != SIG_IGN) {
4329     if ((actp->sa_flags & SA_NODEFER) == 0) {
4330       // automaticlly block the signal
4331       sigaddset(&(actp->sa_mask), sig);
4332     }
4333 
4334     sa_handler_t hand = NULL;
4335     sa_sigaction_t sa = NULL;
4336     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4337     // retrieve the chained handler
4338     if (siginfo_flag_set) {
4339       sa = actp->sa_sigaction;
4340     } else {
4341       hand = actp->sa_handler;
4342     }
4343 
4344     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4345       actp->sa_handler = SIG_DFL;
4346     }
4347 
4348     // try to honor the signal mask
4349     sigset_t oset;
4350     sigemptyset(&oset);
4351     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4352 
4353     // call into the chained handler
4354     if (siginfo_flag_set) {
4355       (*sa)(sig, siginfo, context);
4356     } else {
4357       (*hand)(sig);
4358     }
4359 
4360     // restore the signal mask
4361     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4362   }
4363   // Tell jvm's signal handler the signal is taken care of.
4364   return true;
4365 }
4366 
4367 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4368   bool chained = false;
4369   // signal-chaining
4370   if (UseSignalChaining) {
4371     struct sigaction *actp = get_chained_signal_action(sig);
4372     if (actp != NULL) {
4373       chained = call_chained_handler(actp, sig, siginfo, context);
4374     }
4375   }
4376   return chained;
4377 }
4378 
4379 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4380   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4381     return &sigact[sig];
4382   }
4383   return NULL;
4384 }
4385 
4386 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4387   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4388   sigact[sig] = oldAct;
4389   sigs |= (uint64_t)1 << (sig-1);
4390 }
4391 
4392 // for diagnostic
4393 int sigflags[NSIG];
4394 
4395 int os::Linux::get_our_sigflags(int sig) {
4396   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4397   return sigflags[sig];
4398 }
4399 
4400 void os::Linux::set_our_sigflags(int sig, int flags) {
4401   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4402   if (sig > 0 && sig < NSIG) {
4403     sigflags[sig] = flags;
4404   }
4405 }
4406 
4407 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4408   // Check for overwrite.
4409   struct sigaction oldAct;
4410   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4411 
4412   void* oldhand = oldAct.sa_sigaction
4413                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4414                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4415   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4416       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4417       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4418     if (AllowUserSignalHandlers || !set_installed) {
4419       // Do not overwrite; user takes responsibility to forward to us.
4420       return;
4421     } else if (UseSignalChaining) {
4422       // save the old handler in jvm
4423       save_preinstalled_handler(sig, oldAct);
4424       // libjsig also interposes the sigaction() call below and saves the
4425       // old sigaction on it own.
4426     } else {
4427       fatal("Encountered unexpected pre-existing sigaction handler "
4428             "%#lx for signal %d.", (long)oldhand, sig);
4429     }
4430   }
4431 
4432   struct sigaction sigAct;
4433   sigfillset(&(sigAct.sa_mask));
4434   sigAct.sa_handler = SIG_DFL;
4435   if (!set_installed) {
4436     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4437   } else {
4438     sigAct.sa_sigaction = signalHandler;
4439     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4440   }
4441   // Save flags, which are set by ours
4442   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4443   sigflags[sig] = sigAct.sa_flags;
4444 
4445   int ret = sigaction(sig, &sigAct, &oldAct);
4446   assert(ret == 0, "check");
4447 
4448   void* oldhand2  = oldAct.sa_sigaction
4449                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4450                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4451   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4452 }
4453 
4454 // install signal handlers for signals that HotSpot needs to
4455 // handle in order to support Java-level exception handling.
4456 
4457 void os::Linux::install_signal_handlers() {
4458   if (!signal_handlers_are_installed) {
4459     signal_handlers_are_installed = true;
4460 
4461     // signal-chaining
4462     typedef void (*signal_setting_t)();
4463     signal_setting_t begin_signal_setting = NULL;
4464     signal_setting_t end_signal_setting = NULL;
4465     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4466                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4467     if (begin_signal_setting != NULL) {
4468       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4469                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4470       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4471                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4472       libjsig_is_loaded = true;
4473       assert(UseSignalChaining, "should enable signal-chaining");
4474     }
4475     if (libjsig_is_loaded) {
4476       // Tell libjsig jvm is setting signal handlers
4477       (*begin_signal_setting)();
4478     }
4479 
4480     set_signal_handler(SIGSEGV, true);
4481     set_signal_handler(SIGPIPE, true);
4482     set_signal_handler(SIGBUS, true);
4483     set_signal_handler(SIGILL, true);
4484     set_signal_handler(SIGFPE, true);
4485 #if defined(PPC64)
4486     set_signal_handler(SIGTRAP, true);
4487 #endif
4488     set_signal_handler(SIGXFSZ, true);
4489 
4490     if (libjsig_is_loaded) {
4491       // Tell libjsig jvm finishes setting signal handlers
4492       (*end_signal_setting)();
4493     }
4494 
4495     // We don't activate signal checker if libjsig is in place, we trust ourselves
4496     // and if UserSignalHandler is installed all bets are off.
4497     // Log that signal checking is off only if -verbose:jni is specified.
4498     if (CheckJNICalls) {
4499       if (libjsig_is_loaded) {
4500         if (PrintJNIResolving) {
4501           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4502         }
4503         check_signals = false;
4504       }
4505       if (AllowUserSignalHandlers) {
4506         if (PrintJNIResolving) {
4507           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4508         }
4509         check_signals = false;
4510       }
4511     }
4512   }
4513 }
4514 
4515 // This is the fastest way to get thread cpu time on Linux.
4516 // Returns cpu time (user+sys) for any thread, not only for current.
4517 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4518 // It might work on 2.6.10+ with a special kernel/glibc patch.
4519 // For reference, please, see IEEE Std 1003.1-2004:
4520 //   http://www.unix.org/single_unix_specification
4521 
4522 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4523   struct timespec tp;
4524   int rc = os::Linux::clock_gettime(clockid, &tp);
4525   assert(rc == 0, "clock_gettime is expected to return 0 code");
4526 
4527   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4528 }
4529 
4530 void os::Linux::initialize_os_info() {
4531   assert(_os_version == 0, "OS info already initialized");
4532 
4533   struct utsname _uname;
4534 
4535   uint32_t major;
4536   uint32_t minor;
4537   uint32_t fix;
4538 
4539   int rc;
4540 
4541   // Kernel version is unknown if
4542   // verification below fails.
4543   _os_version = 0x01000000;
4544 
4545   rc = uname(&_uname);
4546   if (rc != -1) {
4547 
4548     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4549     if (rc == 3) {
4550 
4551       if (major < 256 && minor < 256 && fix < 256) {
4552         // Kernel version format is as expected,
4553         // set it overriding unknown state.
4554         _os_version = (major << 16) |
4555                       (minor << 8 ) |
4556                       (fix   << 0 ) ;
4557       }
4558     }
4559   }
4560 }
4561 
4562 uint32_t os::Linux::os_version() {
4563   assert(_os_version != 0, "not initialized");
4564   return _os_version & 0x00FFFFFF;
4565 }
4566 
4567 bool os::Linux::os_version_is_known() {
4568   assert(_os_version != 0, "not initialized");
4569   return _os_version & 0x01000000 ? false : true;
4570 }
4571 
4572 /////
4573 // glibc on Linux platform uses non-documented flag
4574 // to indicate, that some special sort of signal
4575 // trampoline is used.
4576 // We will never set this flag, and we should
4577 // ignore this flag in our diagnostic
4578 #ifdef SIGNIFICANT_SIGNAL_MASK
4579   #undef SIGNIFICANT_SIGNAL_MASK
4580 #endif
4581 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4582 
4583 static const char* get_signal_handler_name(address handler,
4584                                            char* buf, int buflen) {
4585   int offset = 0;
4586   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4587   if (found) {
4588     // skip directory names
4589     const char *p1, *p2;
4590     p1 = buf;
4591     size_t len = strlen(os::file_separator());
4592     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4593     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4594   } else {
4595     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4596   }
4597   return buf;
4598 }
4599 
4600 static void print_signal_handler(outputStream* st, int sig,
4601                                  char* buf, size_t buflen) {
4602   struct sigaction sa;
4603 
4604   sigaction(sig, NULL, &sa);
4605 
4606   // See comment for SIGNIFICANT_SIGNAL_MASK define
4607   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4608 
4609   st->print("%s: ", os::exception_name(sig, buf, buflen));
4610 
4611   address handler = (sa.sa_flags & SA_SIGINFO)
4612     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4613     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4614 
4615   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4616     st->print("SIG_DFL");
4617   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4618     st->print("SIG_IGN");
4619   } else {
4620     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4621   }
4622 
4623   st->print(", sa_mask[0]=");
4624   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4625 
4626   address rh = VMError::get_resetted_sighandler(sig);
4627   // May be, handler was resetted by VMError?
4628   if (rh != NULL) {
4629     handler = rh;
4630     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4631   }
4632 
4633   st->print(", sa_flags=");
4634   os::Posix::print_sa_flags(st, sa.sa_flags);
4635 
4636   // Check: is it our handler?
4637   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4638       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4639     // It is our signal handler
4640     // check for flags, reset system-used one!
4641     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4642       st->print(
4643                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4644                 os::Linux::get_our_sigflags(sig));
4645     }
4646   }
4647   st->cr();
4648 }
4649 
4650 
4651 #define DO_SIGNAL_CHECK(sig)                      \
4652   do {                                            \
4653     if (!sigismember(&check_signal_done, sig)) {  \
4654       os::Linux::check_signal_handler(sig);       \
4655     }                                             \
4656   } while (0)
4657 
4658 // This method is a periodic task to check for misbehaving JNI applications
4659 // under CheckJNI, we can add any periodic checks here
4660 
4661 void os::run_periodic_checks() {
4662   if (check_signals == false) return;
4663 
4664   // SEGV and BUS if overridden could potentially prevent
4665   // generation of hs*.log in the event of a crash, debugging
4666   // such a case can be very challenging, so we absolutely
4667   // check the following for a good measure:
4668   DO_SIGNAL_CHECK(SIGSEGV);
4669   DO_SIGNAL_CHECK(SIGILL);
4670   DO_SIGNAL_CHECK(SIGFPE);
4671   DO_SIGNAL_CHECK(SIGBUS);
4672   DO_SIGNAL_CHECK(SIGPIPE);
4673   DO_SIGNAL_CHECK(SIGXFSZ);
4674 #if defined(PPC64)
4675   DO_SIGNAL_CHECK(SIGTRAP);
4676 #endif
4677 
4678   // ReduceSignalUsage allows the user to override these handlers
4679   // see comments at the very top and jvm_solaris.h
4680   if (!ReduceSignalUsage) {
4681     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4682     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4683     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4684     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4685   }
4686 
4687   DO_SIGNAL_CHECK(SR_signum);
4688 }
4689 
4690 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4691 
4692 static os_sigaction_t os_sigaction = NULL;
4693 
4694 void os::Linux::check_signal_handler(int sig) {
4695   char buf[O_BUFLEN];
4696   address jvmHandler = NULL;
4697 
4698 
4699   struct sigaction act;
4700   if (os_sigaction == NULL) {
4701     // only trust the default sigaction, in case it has been interposed
4702     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4703     if (os_sigaction == NULL) return;
4704   }
4705 
4706   os_sigaction(sig, (struct sigaction*)NULL, &act);
4707 
4708 
4709   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4710 
4711   address thisHandler = (act.sa_flags & SA_SIGINFO)
4712     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4713     : CAST_FROM_FN_PTR(address, act.sa_handler);
4714 
4715 
4716   switch (sig) {
4717   case SIGSEGV:
4718   case SIGBUS:
4719   case SIGFPE:
4720   case SIGPIPE:
4721   case SIGILL:
4722   case SIGXFSZ:
4723     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4724     break;
4725 
4726   case SHUTDOWN1_SIGNAL:
4727   case SHUTDOWN2_SIGNAL:
4728   case SHUTDOWN3_SIGNAL:
4729   case BREAK_SIGNAL:
4730     jvmHandler = (address)user_handler();
4731     break;
4732 
4733   default:
4734     if (sig == SR_signum) {
4735       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4736     } else {
4737       return;
4738     }
4739     break;
4740   }
4741 
4742   if (thisHandler != jvmHandler) {
4743     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4744     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4745     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4746     // No need to check this sig any longer
4747     sigaddset(&check_signal_done, sig);
4748     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4749     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4750       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4751                     exception_name(sig, buf, O_BUFLEN));
4752     }
4753   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4754     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4755     tty->print("expected:");
4756     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4757     tty->cr();
4758     tty->print("  found:");
4759     os::Posix::print_sa_flags(tty, act.sa_flags);
4760     tty->cr();
4761     // No need to check this sig any longer
4762     sigaddset(&check_signal_done, sig);
4763   }
4764 
4765   // Dump all the signal
4766   if (sigismember(&check_signal_done, sig)) {
4767     print_signal_handlers(tty, buf, O_BUFLEN);
4768   }
4769 }
4770 
4771 extern void report_error(char* file_name, int line_no, char* title,
4772                          char* format, ...);
4773 
4774 // this is called _before_ the most of global arguments have been parsed
4775 void os::init(void) {
4776   char dummy;   // used to get a guess on initial stack address
4777 //  first_hrtime = gethrtime();
4778 
4779   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4780 
4781   init_random(1234567);
4782 
4783   ThreadCritical::initialize();
4784 
4785   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4786   if (Linux::page_size() == -1) {
4787     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4788           os::strerror(errno));
4789   }
4790   init_page_sizes((size_t) Linux::page_size());
4791 
4792   Linux::initialize_system_info();
4793 
4794   Linux::initialize_os_info();
4795 
4796   // main_thread points to the aboriginal thread
4797   Linux::_main_thread = pthread_self();
4798 
4799   Linux::clock_init();
4800   initial_time_count = javaTimeNanos();
4801 
4802   // pthread_condattr initialization for monotonic clock
4803   int status;
4804   pthread_condattr_t* _condattr = os::Linux::condAttr();
4805   if ((status = pthread_condattr_init(_condattr)) != 0) {
4806     fatal("pthread_condattr_init: %s", os::strerror(status));
4807   }
4808   // Only set the clock if CLOCK_MONOTONIC is available
4809   if (os::supports_monotonic_clock()) {
4810     if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4811       if (status == EINVAL) {
4812         warning("Unable to use monotonic clock with relative timed-waits" \
4813                 " - changes to the time-of-day clock may have adverse affects");
4814       } else {
4815         fatal("pthread_condattr_setclock: %s", os::strerror(status));
4816       }
4817     }
4818   }
4819   // else it defaults to CLOCK_REALTIME
4820 
4821   // retrieve entry point for pthread_setname_np
4822   Linux::_pthread_setname_np =
4823     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4824 
4825 }
4826 
4827 // To install functions for atexit system call
4828 extern "C" {
4829   static void perfMemory_exit_helper() {
4830     perfMemory_exit();
4831   }
4832 }
4833 
4834 // this is called _after_ the global arguments have been parsed
4835 jint os::init_2(void) {
4836   Linux::fast_thread_clock_init();
4837 
4838   // Allocate a single page and mark it as readable for safepoint polling
4839   address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4840   guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4841 
4842   os::set_polling_page(polling_page);
4843   log_info(os)("SafePoint Polling address: " INTPTR_FORMAT, p2i(polling_page));
4844 
4845   if (!UseMembar) {
4846     address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4847     guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4848     os::set_memory_serialize_page(mem_serialize_page);
4849     log_info(os)("Memory Serialize Page address: " INTPTR_FORMAT, p2i(mem_serialize_page));
4850   }
4851 
4852   // initialize suspend/resume support - must do this before signal_sets_init()
4853   if (SR_initialize() != 0) {
4854     perror("SR_initialize failed");
4855     return JNI_ERR;
4856   }
4857 
4858   Linux::signal_sets_init();
4859   Linux::install_signal_handlers();
4860 
4861   // Check minimum allowable stack size for thread creation and to initialize
4862   // the java system classes, including StackOverflowError - depends on page
4863   // size.  Add a page for compiler2 recursion in main thread.
4864   // Add in 2*BytesPerWord times page size to account for VM stack during
4865   // class initialization depending on 32 or 64 bit VM.
4866   os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4867                                       JavaThread::stack_guard_zone_size() +
4868                                       JavaThread::stack_shadow_zone_size() +
4869                                       (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4870 
4871   size_t threadStackSizeInBytes = ThreadStackSize * K;
4872   if (threadStackSizeInBytes != 0 &&
4873       threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4874     tty->print_cr("\nThe stack size specified is too small, "
4875                   "Specify at least " SIZE_FORMAT "k",
4876                   os::Linux::min_stack_allowed/ K);
4877     return JNI_ERR;
4878   }
4879 
4880   // Make the stack size a multiple of the page size so that
4881   // the yellow/red zones can be guarded.
4882   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4883                                                 vm_page_size()));
4884 
4885   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4886 
4887 #if defined(IA32)
4888   workaround_expand_exec_shield_cs_limit();
4889 #endif
4890 
4891   Linux::libpthread_init();
4892   log_info(os)("HotSpot is running with %s, %s",
4893                Linux::glibc_version(), Linux::libpthread_version());
4894 
4895   if (UseNUMA) {
4896     if (!Linux::libnuma_init()) {
4897       UseNUMA = false;
4898     } else {
4899       if ((Linux::numa_max_node() < 1)) {
4900         // There's only one node(they start from 0), disable NUMA.
4901         UseNUMA = false;
4902       }
4903     }
4904     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4905     // we can make the adaptive lgrp chunk resizing work. If the user specified
4906     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4907     // disable adaptive resizing.
4908     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4909       if (FLAG_IS_DEFAULT(UseNUMA)) {
4910         UseNUMA = false;
4911       } else {
4912         if (FLAG_IS_DEFAULT(UseLargePages) &&
4913             FLAG_IS_DEFAULT(UseSHM) &&
4914             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4915           UseLargePages = false;
4916         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4917           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4918           UseAdaptiveSizePolicy = false;
4919           UseAdaptiveNUMAChunkSizing = false;
4920         }
4921       }
4922     }
4923     if (!UseNUMA && ForceNUMA) {
4924       UseNUMA = true;
4925     }
4926   }
4927 
4928   if (MaxFDLimit) {
4929     // set the number of file descriptors to max. print out error
4930     // if getrlimit/setrlimit fails but continue regardless.
4931     struct rlimit nbr_files;
4932     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4933     if (status != 0) {
4934       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4935     } else {
4936       nbr_files.rlim_cur = nbr_files.rlim_max;
4937       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4938       if (status != 0) {
4939         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4940       }
4941     }
4942   }
4943 
4944   // Initialize lock used to serialize thread creation (see os::create_thread)
4945   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4946 
4947   // at-exit methods are called in the reverse order of their registration.
4948   // atexit functions are called on return from main or as a result of a
4949   // call to exit(3C). There can be only 32 of these functions registered
4950   // and atexit() does not set errno.
4951 
4952   if (PerfAllowAtExitRegistration) {
4953     // only register atexit functions if PerfAllowAtExitRegistration is set.
4954     // atexit functions can be delayed until process exit time, which
4955     // can be problematic for embedded VM situations. Embedded VMs should
4956     // call DestroyJavaVM() to assure that VM resources are released.
4957 
4958     // note: perfMemory_exit_helper atexit function may be removed in
4959     // the future if the appropriate cleanup code can be added to the
4960     // VM_Exit VMOperation's doit method.
4961     if (atexit(perfMemory_exit_helper) != 0) {
4962       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4963     }
4964   }
4965 
4966   // initialize thread priority policy
4967   prio_init();
4968 
4969   return JNI_OK;
4970 }
4971 
4972 // Mark the polling page as unreadable
4973 void os::make_polling_page_unreadable(void) {
4974   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4975     fatal("Could not disable polling page");
4976   }
4977 }
4978 
4979 // Mark the polling page as readable
4980 void os::make_polling_page_readable(void) {
4981   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4982     fatal("Could not enable polling page");
4983   }
4984 }
4985 
4986 // older glibc versions don't have this macro (which expands to
4987 // an optimized bit-counting function) so we have to roll our own
4988 #ifndef CPU_COUNT
4989 
4990 static int _cpu_count(const cpu_set_t* cpus) {
4991   int count = 0;
4992   // only look up to the number of configured processors
4993   for (int i = 0; i < os::processor_count(); i++) {
4994     if (CPU_ISSET(i, cpus)) {
4995       count++;
4996     }
4997   }
4998   return count;
4999 }
5000 
5001 #define CPU_COUNT(cpus) _cpu_count(cpus)
5002 
5003 #endif // CPU_COUNT
5004 
5005 // Get the current number of available processors for this process.
5006 // This value can change at any time during a process's lifetime.
5007 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5008 // If it appears there may be more than 1024 processors then we do a
5009 // dynamic check - see 6515172 for details.
5010 // If anything goes wrong we fallback to returning the number of online
5011 // processors - which can be greater than the number available to the process.
5012 int os::active_processor_count() {
5013   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
5014   cpu_set_t* cpus_p = &cpus;
5015   int cpus_size = sizeof(cpu_set_t);
5016 
5017   int configured_cpus = processor_count();  // upper bound on available cpus
5018   int cpu_count = 0;
5019 
5020 // old build platforms may not support dynamic cpu sets
5021 #ifdef CPU_ALLOC
5022 
5023   // To enable easy testing of the dynamic path on different platforms we
5024   // introduce a diagnostic flag: UseCpuAllocPath
5025   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5026     // kernel may use a mask bigger than cpu_set_t
5027     log_trace(os)("active_processor_count: using dynamic path %s"
5028                   "- configured processors: %d",
5029                   UseCpuAllocPath ? "(forced) " : "",
5030                   configured_cpus);
5031     cpus_p = CPU_ALLOC(configured_cpus);
5032     if (cpus_p != NULL) {
5033       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5034       // zero it just to be safe
5035       CPU_ZERO_S(cpus_size, cpus_p);
5036     }
5037     else {
5038        // failed to allocate so fallback to online cpus
5039        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5040        log_trace(os)("active_processor_count: "
5041                      "CPU_ALLOC failed (%s) - using "
5042                      "online processor count: %d",
5043                      os::strerror(errno), online_cpus);
5044        return online_cpus;
5045     }
5046   }
5047   else {
5048     log_trace(os)("active_processor_count: using static path - configured processors: %d",
5049                   configured_cpus);
5050   }
5051 #else // CPU_ALLOC
5052 // these stubs won't be executed
5053 #define CPU_COUNT_S(size, cpus) -1
5054 #define CPU_FREE(cpus)
5055 
5056   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5057                 configured_cpus);
5058 #endif // CPU_ALLOC
5059 
5060   // pid 0 means the current thread - which we have to assume represents the process
5061   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5062     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5063       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5064     }
5065     else {
5066       cpu_count = CPU_COUNT(cpus_p);
5067     }
5068     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5069   }
5070   else {
5071     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5072     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5073             "which may exceed available processors", os::strerror(errno), cpu_count);
5074   }
5075 
5076   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5077     CPU_FREE(cpus_p);
5078   }
5079 
5080   assert(cpu_count > 0 && cpu_count <= processor_count(), "sanity check");
5081   return cpu_count;
5082 }
5083 
5084 void os::set_native_thread_name(const char *name) {
5085   if (Linux::_pthread_setname_np) {
5086     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5087     snprintf(buf, sizeof(buf), "%s", name);
5088     buf[sizeof(buf) - 1] = '\0';
5089     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5090     // ERANGE should not happen; all other errors should just be ignored.
5091     assert(rc != ERANGE, "pthread_setname_np failed");
5092   }
5093 }
5094 
5095 bool os::distribute_processes(uint length, uint* distribution) {
5096   // Not yet implemented.
5097   return false;
5098 }
5099 
5100 bool os::bind_to_processor(uint processor_id) {
5101   // Not yet implemented.
5102   return false;
5103 }
5104 
5105 ///
5106 
5107 void os::SuspendedThreadTask::internal_do_task() {
5108   if (do_suspend(_thread->osthread())) {
5109     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5110     do_task(context);
5111     do_resume(_thread->osthread());
5112   }
5113 }
5114 
5115 class PcFetcher : public os::SuspendedThreadTask {
5116  public:
5117   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5118   ExtendedPC result();
5119  protected:
5120   void do_task(const os::SuspendedThreadTaskContext& context);
5121  private:
5122   ExtendedPC _epc;
5123 };
5124 
5125 ExtendedPC PcFetcher::result() {
5126   guarantee(is_done(), "task is not done yet.");
5127   return _epc;
5128 }
5129 
5130 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5131   Thread* thread = context.thread();
5132   OSThread* osthread = thread->osthread();
5133   if (osthread->ucontext() != NULL) {
5134     _epc = os::Linux::ucontext_get_pc((const ucontext_t *) context.ucontext());
5135   } else {
5136     // NULL context is unexpected, double-check this is the VMThread
5137     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5138   }
5139 }
5140 
5141 // Suspends the target using the signal mechanism and then grabs the PC before
5142 // resuming the target. Used by the flat-profiler only
5143 ExtendedPC os::get_thread_pc(Thread* thread) {
5144   // Make sure that it is called by the watcher for the VMThread
5145   assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5146   assert(thread->is_VM_thread(), "Can only be called for VMThread");
5147 
5148   PcFetcher fetcher(thread);
5149   fetcher.run();
5150   return fetcher.result();
5151 }
5152 
5153 ////////////////////////////////////////////////////////////////////////////////
5154 // debug support
5155 
5156 bool os::find(address addr, outputStream* st) {
5157   Dl_info dlinfo;
5158   memset(&dlinfo, 0, sizeof(dlinfo));
5159   if (dladdr(addr, &dlinfo) != 0) {
5160     st->print(PTR_FORMAT ": ", p2i(addr));
5161     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5162       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5163                 p2i(addr) - p2i(dlinfo.dli_saddr));
5164     } else if (dlinfo.dli_fbase != NULL) {
5165       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5166     } else {
5167       st->print("<absolute address>");
5168     }
5169     if (dlinfo.dli_fname != NULL) {
5170       st->print(" in %s", dlinfo.dli_fname);
5171     }
5172     if (dlinfo.dli_fbase != NULL) {
5173       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5174     }
5175     st->cr();
5176 
5177     if (Verbose) {
5178       // decode some bytes around the PC
5179       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5180       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5181       address       lowest = (address) dlinfo.dli_sname;
5182       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5183       if (begin < lowest)  begin = lowest;
5184       Dl_info dlinfo2;
5185       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5186           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5187         end = (address) dlinfo2.dli_saddr;
5188       }
5189       Disassembler::decode(begin, end, st);
5190     }
5191     return true;
5192   }
5193   return false;
5194 }
5195 
5196 ////////////////////////////////////////////////////////////////////////////////
5197 // misc
5198 
5199 // This does not do anything on Linux. This is basically a hook for being
5200 // able to use structured exception handling (thread-local exception filters)
5201 // on, e.g., Win32.
5202 void
5203 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5204                          JavaCallArguments* args, Thread* thread) {
5205   f(value, method, args, thread);
5206 }
5207 
5208 void os::print_statistics() {
5209 }
5210 
5211 bool os::message_box(const char* title, const char* message) {
5212   int i;
5213   fdStream err(defaultStream::error_fd());
5214   for (i = 0; i < 78; i++) err.print_raw("=");
5215   err.cr();
5216   err.print_raw_cr(title);
5217   for (i = 0; i < 78; i++) err.print_raw("-");
5218   err.cr();
5219   err.print_raw_cr(message);
5220   for (i = 0; i < 78; i++) err.print_raw("=");
5221   err.cr();
5222 
5223   char buf[16];
5224   // Prevent process from exiting upon "read error" without consuming all CPU
5225   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5226 
5227   return buf[0] == 'y' || buf[0] == 'Y';
5228 }
5229 
5230 int os::stat(const char *path, struct stat *sbuf) {
5231   char pathbuf[MAX_PATH];
5232   if (strlen(path) > MAX_PATH - 1) {
5233     errno = ENAMETOOLONG;
5234     return -1;
5235   }
5236   os::native_path(strcpy(pathbuf, path));
5237   return ::stat(pathbuf, sbuf);
5238 }
5239 
5240 bool os::check_heap(bool force) {
5241   return true;
5242 }
5243 
5244 // Is a (classpath) directory empty?
5245 bool os::dir_is_empty(const char* path) {
5246   DIR *dir = NULL;
5247   struct dirent *ptr;
5248 
5249   dir = opendir(path);
5250   if (dir == NULL) return true;
5251 
5252   // Scan the directory
5253   bool result = true;
5254   char buf[sizeof(struct dirent) + MAX_PATH];
5255   while (result && (ptr = ::readdir(dir)) != NULL) {
5256     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5257       result = false;
5258     }
5259   }
5260   closedir(dir);
5261   return result;
5262 }
5263 
5264 // This code originates from JDK's sysOpen and open64_w
5265 // from src/solaris/hpi/src/system_md.c
5266 
5267 int os::open(const char *path, int oflag, int mode) {
5268   if (strlen(path) > MAX_PATH - 1) {
5269     errno = ENAMETOOLONG;
5270     return -1;
5271   }
5272 
5273   // All file descriptors that are opened in the Java process and not
5274   // specifically destined for a subprocess should have the close-on-exec
5275   // flag set.  If we don't set it, then careless 3rd party native code
5276   // might fork and exec without closing all appropriate file descriptors
5277   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5278   // turn might:
5279   //
5280   // - cause end-of-file to fail to be detected on some file
5281   //   descriptors, resulting in mysterious hangs, or
5282   //
5283   // - might cause an fopen in the subprocess to fail on a system
5284   //   suffering from bug 1085341.
5285   //
5286   // (Yes, the default setting of the close-on-exec flag is a Unix
5287   // design flaw)
5288   //
5289   // See:
5290   // 1085341: 32-bit stdio routines should support file descriptors >255
5291   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5292   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5293   //
5294   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5295   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5296   // because it saves a system call and removes a small window where the flag
5297   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5298   // and we fall back to using FD_CLOEXEC (see below).
5299 #ifdef O_CLOEXEC
5300   oflag |= O_CLOEXEC;
5301 #endif
5302 
5303   int fd = ::open64(path, oflag, mode);
5304   if (fd == -1) return -1;
5305 
5306   //If the open succeeded, the file might still be a directory
5307   {
5308     struct stat64 buf64;
5309     int ret = ::fstat64(fd, &buf64);
5310     int st_mode = buf64.st_mode;
5311 
5312     if (ret != -1) {
5313       if ((st_mode & S_IFMT) == S_IFDIR) {
5314         errno = EISDIR;
5315         ::close(fd);
5316         return -1;
5317       }
5318     } else {
5319       ::close(fd);
5320       return -1;
5321     }
5322   }
5323 
5324 #ifdef FD_CLOEXEC
5325   // Validate that the use of the O_CLOEXEC flag on open above worked.
5326   // With recent kernels, we will perform this check exactly once.
5327   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5328   if (!O_CLOEXEC_is_known_to_work) {
5329     int flags = ::fcntl(fd, F_GETFD);
5330     if (flags != -1) {
5331       if ((flags & FD_CLOEXEC) != 0)
5332         O_CLOEXEC_is_known_to_work = 1;
5333       else
5334         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5335     }
5336   }
5337 #endif
5338 
5339   return fd;
5340 }
5341 
5342 
5343 // create binary file, rewriting existing file if required
5344 int os::create_binary_file(const char* path, bool rewrite_existing) {
5345   int oflags = O_WRONLY | O_CREAT;
5346   if (!rewrite_existing) {
5347     oflags |= O_EXCL;
5348   }
5349   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5350 }
5351 
5352 // return current position of file pointer
5353 jlong os::current_file_offset(int fd) {
5354   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5355 }
5356 
5357 // move file pointer to the specified offset
5358 jlong os::seek_to_file_offset(int fd, jlong offset) {
5359   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5360 }
5361 
5362 // This code originates from JDK's sysAvailable
5363 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5364 
5365 int os::available(int fd, jlong *bytes) {
5366   jlong cur, end;
5367   int mode;
5368   struct stat64 buf64;
5369 
5370   if (::fstat64(fd, &buf64) >= 0) {
5371     mode = buf64.st_mode;
5372     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5373       int n;
5374       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5375         *bytes = n;
5376         return 1;
5377       }
5378     }
5379   }
5380   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5381     return 0;
5382   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5383     return 0;
5384   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5385     return 0;
5386   }
5387   *bytes = end - cur;
5388   return 1;
5389 }
5390 
5391 // Map a block of memory.
5392 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5393                         char *addr, size_t bytes, bool read_only,
5394                         bool allow_exec) {
5395   int prot;
5396   int flags = MAP_PRIVATE;
5397 
5398   if (read_only) {
5399     prot = PROT_READ;
5400   } else {
5401     prot = PROT_READ | PROT_WRITE;
5402   }
5403 
5404   if (allow_exec) {
5405     prot |= PROT_EXEC;
5406   }
5407 
5408   if (addr != NULL) {
5409     flags |= MAP_FIXED;
5410   }
5411 
5412   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5413                                      fd, file_offset);
5414   if (mapped_address == MAP_FAILED) {
5415     return NULL;
5416   }
5417   return mapped_address;
5418 }
5419 
5420 
5421 // Remap a block of memory.
5422 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5423                           char *addr, size_t bytes, bool read_only,
5424                           bool allow_exec) {
5425   // same as map_memory() on this OS
5426   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5427                         allow_exec);
5428 }
5429 
5430 
5431 // Unmap a block of memory.
5432 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5433   return munmap(addr, bytes) == 0;
5434 }
5435 
5436 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5437 
5438 static clockid_t thread_cpu_clockid(Thread* thread) {
5439   pthread_t tid = thread->osthread()->pthread_id();
5440   clockid_t clockid;
5441 
5442   // Get thread clockid
5443   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5444   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5445   return clockid;
5446 }
5447 
5448 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5449 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5450 // of a thread.
5451 //
5452 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5453 // the fast estimate available on the platform.
5454 
5455 jlong os::current_thread_cpu_time() {
5456   if (os::Linux::supports_fast_thread_cpu_time()) {
5457     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5458   } else {
5459     // return user + sys since the cost is the same
5460     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5461   }
5462 }
5463 
5464 jlong os::thread_cpu_time(Thread* thread) {
5465   // consistent with what current_thread_cpu_time() returns
5466   if (os::Linux::supports_fast_thread_cpu_time()) {
5467     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5468   } else {
5469     return slow_thread_cpu_time(thread, true /* user + sys */);
5470   }
5471 }
5472 
5473 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5474   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5475     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5476   } else {
5477     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5478   }
5479 }
5480 
5481 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5482   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5483     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5484   } else {
5485     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5486   }
5487 }
5488 
5489 //  -1 on error.
5490 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5491   pid_t  tid = thread->osthread()->thread_id();
5492   char *s;
5493   char stat[2048];
5494   int statlen;
5495   char proc_name[64];
5496   int count;
5497   long sys_time, user_time;
5498   char cdummy;
5499   int idummy;
5500   long ldummy;
5501   FILE *fp;
5502 
5503   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5504   fp = fopen(proc_name, "r");
5505   if (fp == NULL) return -1;
5506   statlen = fread(stat, 1, 2047, fp);
5507   stat[statlen] = '\0';
5508   fclose(fp);
5509 
5510   // Skip pid and the command string. Note that we could be dealing with
5511   // weird command names, e.g. user could decide to rename java launcher
5512   // to "java 1.4.2 :)", then the stat file would look like
5513   //                1234 (java 1.4.2 :)) R ... ...
5514   // We don't really need to know the command string, just find the last
5515   // occurrence of ")" and then start parsing from there. See bug 4726580.
5516   s = strrchr(stat, ')');
5517   if (s == NULL) return -1;
5518 
5519   // Skip blank chars
5520   do { s++; } while (s && isspace(*s));
5521 
5522   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5523                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5524                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5525                  &user_time, &sys_time);
5526   if (count != 13) return -1;
5527   if (user_sys_cpu_time) {
5528     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5529   } else {
5530     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5531   }
5532 }
5533 
5534 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5535   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5536   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5537   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5538   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5539 }
5540 
5541 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5542   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5543   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5544   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5545   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5546 }
5547 
5548 bool os::is_thread_cpu_time_supported() {
5549   return true;
5550 }
5551 
5552 // System loadavg support.  Returns -1 if load average cannot be obtained.
5553 // Linux doesn't yet have a (official) notion of processor sets,
5554 // so just return the system wide load average.
5555 int os::loadavg(double loadavg[], int nelem) {
5556   return ::getloadavg(loadavg, nelem);
5557 }
5558 
5559 void os::pause() {
5560   char filename[MAX_PATH];
5561   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5562     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5563   } else {
5564     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5565   }
5566 
5567   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5568   if (fd != -1) {
5569     struct stat buf;
5570     ::close(fd);
5571     while (::stat(filename, &buf) == 0) {
5572       (void)::poll(NULL, 0, 100);
5573     }
5574   } else {
5575     jio_fprintf(stderr,
5576                 "Could not open pause file '%s', continuing immediately.\n", filename);
5577   }
5578 }
5579 
5580 
5581 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5582 // comment paragraphs are worth repeating here:
5583 //
5584 // Assumption:
5585 //    Only one parker can exist on an event, which is why we allocate
5586 //    them per-thread. Multiple unparkers can coexist.
5587 //
5588 // _Event serves as a restricted-range semaphore.
5589 //   -1 : thread is blocked, i.e. there is a waiter
5590 //    0 : neutral: thread is running or ready,
5591 //        could have been signaled after a wait started
5592 //    1 : signaled - thread is running or ready
5593 //
5594 
5595 // utility to compute the abstime argument to timedwait:
5596 // millis is the relative timeout time
5597 // abstime will be the absolute timeout time
5598 // TODO: replace compute_abstime() with unpackTime()
5599 
5600 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5601   if (millis < 0)  millis = 0;
5602 
5603   jlong seconds = millis / 1000;
5604   millis %= 1000;
5605   if (seconds > 50000000) { // see man cond_timedwait(3T)
5606     seconds = 50000000;
5607   }
5608 
5609   if (os::supports_monotonic_clock()) {
5610     struct timespec now;
5611     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5612     assert_status(status == 0, status, "clock_gettime");
5613     abstime->tv_sec = now.tv_sec  + seconds;
5614     long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5615     if (nanos >= NANOSECS_PER_SEC) {
5616       abstime->tv_sec += 1;
5617       nanos -= NANOSECS_PER_SEC;
5618     }
5619     abstime->tv_nsec = nanos;
5620   } else {
5621     struct timeval now;
5622     int status = gettimeofday(&now, NULL);
5623     assert(status == 0, "gettimeofday");
5624     abstime->tv_sec = now.tv_sec  + seconds;
5625     long usec = now.tv_usec + millis * 1000;
5626     if (usec >= 1000000) {
5627       abstime->tv_sec += 1;
5628       usec -= 1000000;
5629     }
5630     abstime->tv_nsec = usec * 1000;
5631   }
5632   return abstime;
5633 }
5634 
5635 void os::PlatformEvent::park() {       // AKA "down()"
5636   // Transitions for _Event:
5637   //   -1 => -1 : illegal
5638   //    1 =>  0 : pass - return immediately
5639   //    0 => -1 : block; then set _Event to 0 before returning
5640 
5641   // Invariant: Only the thread associated with the Event/PlatformEvent
5642   // may call park().
5643   // TODO: assert that _Assoc != NULL or _Assoc == Self
5644   assert(_nParked == 0, "invariant");
5645 
5646   int v;
5647   for (;;) {
5648     v = _Event;
5649     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5650   }
5651   guarantee(v >= 0, "invariant");
5652   if (v == 0) {
5653     // Do this the hard way by blocking ...
5654     int status = pthread_mutex_lock(_mutex);
5655     assert_status(status == 0, status, "mutex_lock");
5656     guarantee(_nParked == 0, "invariant");
5657     ++_nParked;
5658     while (_Event < 0) {
5659       status = pthread_cond_wait(_cond, _mutex);
5660       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5661       // Treat this the same as if the wait was interrupted
5662       if (status == ETIME) { status = EINTR; }
5663       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5664     }
5665     --_nParked;
5666 
5667     _Event = 0;
5668     status = pthread_mutex_unlock(_mutex);
5669     assert_status(status == 0, status, "mutex_unlock");
5670     // Paranoia to ensure our locked and lock-free paths interact
5671     // correctly with each other.
5672     OrderAccess::fence();
5673   }
5674   guarantee(_Event >= 0, "invariant");
5675 }
5676 
5677 int os::PlatformEvent::park(jlong millis) {
5678   // Transitions for _Event:
5679   //   -1 => -1 : illegal
5680   //    1 =>  0 : pass - return immediately
5681   //    0 => -1 : block; then set _Event to 0 before returning
5682 
5683   guarantee(_nParked == 0, "invariant");
5684 
5685   int v;
5686   for (;;) {
5687     v = _Event;
5688     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5689   }
5690   guarantee(v >= 0, "invariant");
5691   if (v != 0) return OS_OK;
5692 
5693   // We do this the hard way, by blocking the thread.
5694   // Consider enforcing a minimum timeout value.
5695   struct timespec abst;
5696   compute_abstime(&abst, millis);
5697 
5698   int ret = OS_TIMEOUT;
5699   int status = pthread_mutex_lock(_mutex);
5700   assert_status(status == 0, status, "mutex_lock");
5701   guarantee(_nParked == 0, "invariant");
5702   ++_nParked;
5703 
5704   // Object.wait(timo) will return because of
5705   // (a) notification
5706   // (b) timeout
5707   // (c) thread.interrupt
5708   //
5709   // Thread.interrupt and object.notify{All} both call Event::set.
5710   // That is, we treat thread.interrupt as a special case of notification.
5711   // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5712   // We assume all ETIME returns are valid.
5713   //
5714   // TODO: properly differentiate simultaneous notify+interrupt.
5715   // In that case, we should propagate the notify to another waiter.
5716 
5717   while (_Event < 0) {
5718     status = pthread_cond_timedwait(_cond, _mutex, &abst);
5719     assert_status(status == 0 || status == EINTR ||
5720                   status == ETIME || status == ETIMEDOUT,
5721                   status, "cond_timedwait");
5722     if (!FilterSpuriousWakeups) break;                 // previous semantics
5723     if (status == ETIME || status == ETIMEDOUT) break;
5724     // We consume and ignore EINTR and spurious wakeups.
5725   }
5726   --_nParked;
5727   if (_Event >= 0) {
5728     ret = OS_OK;
5729   }
5730   _Event = 0;
5731   status = pthread_mutex_unlock(_mutex);
5732   assert_status(status == 0, status, "mutex_unlock");
5733   assert(_nParked == 0, "invariant");
5734   // Paranoia to ensure our locked and lock-free paths interact
5735   // correctly with each other.
5736   OrderAccess::fence();
5737   return ret;
5738 }
5739 
5740 void os::PlatformEvent::unpark() {
5741   // Transitions for _Event:
5742   //    0 => 1 : just return
5743   //    1 => 1 : just return
5744   //   -1 => either 0 or 1; must signal target thread
5745   //         That is, we can safely transition _Event from -1 to either
5746   //         0 or 1.
5747   // See also: "Semaphores in Plan 9" by Mullender & Cox
5748   //
5749   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5750   // that it will take two back-to-back park() calls for the owning
5751   // thread to block. This has the benefit of forcing a spurious return
5752   // from the first park() call after an unpark() call which will help
5753   // shake out uses of park() and unpark() without condition variables.
5754 
5755   if (Atomic::xchg(1, &_Event) >= 0) return;
5756 
5757   // Wait for the thread associated with the event to vacate
5758   int status = pthread_mutex_lock(_mutex);
5759   assert_status(status == 0, status, "mutex_lock");
5760   int AnyWaiters = _nParked;
5761   assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5762   status = pthread_mutex_unlock(_mutex);
5763   assert_status(status == 0, status, "mutex_unlock");
5764   if (AnyWaiters != 0) {
5765     // Note that we signal() *after* dropping the lock for "immortal" Events.
5766     // This is safe and avoids a common class of  futile wakeups.  In rare
5767     // circumstances this can cause a thread to return prematurely from
5768     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5769     // will simply re-test the condition and re-park itself.
5770     // This provides particular benefit if the underlying platform does not
5771     // provide wait morphing.
5772     status = pthread_cond_signal(_cond);
5773     assert_status(status == 0, status, "cond_signal");
5774   }
5775 }
5776 
5777 
5778 // JSR166
5779 // -------------------------------------------------------
5780 
5781 // The solaris and linux implementations of park/unpark are fairly
5782 // conservative for now, but can be improved. They currently use a
5783 // mutex/condvar pair, plus a a count.
5784 // Park decrements count if > 0, else does a condvar wait.  Unpark
5785 // sets count to 1 and signals condvar.  Only one thread ever waits
5786 // on the condvar. Contention seen when trying to park implies that someone
5787 // is unparking you, so don't wait. And spurious returns are fine, so there
5788 // is no need to track notifications.
5789 
5790 // This code is common to linux and solaris and will be moved to a
5791 // common place in dolphin.
5792 //
5793 // The passed in time value is either a relative time in nanoseconds
5794 // or an absolute time in milliseconds. Either way it has to be unpacked
5795 // into suitable seconds and nanoseconds components and stored in the
5796 // given timespec structure.
5797 // Given time is a 64-bit value and the time_t used in the timespec is only
5798 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5799 // overflow if times way in the future are given. Further on Solaris versions
5800 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5801 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5802 // As it will be 28 years before "now + 100000000" will overflow we can
5803 // ignore overflow and just impose a hard-limit on seconds using the value
5804 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5805 // years from "now".
5806 
5807 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5808   assert(time > 0, "convertTime");
5809   time_t max_secs = 0;
5810 
5811   if (!os::supports_monotonic_clock() || isAbsolute) {
5812     struct timeval now;
5813     int status = gettimeofday(&now, NULL);
5814     assert(status == 0, "gettimeofday");
5815 
5816     max_secs = now.tv_sec + MAX_SECS;
5817 
5818     if (isAbsolute) {
5819       jlong secs = time / 1000;
5820       if (secs > max_secs) {
5821         absTime->tv_sec = max_secs;
5822       } else {
5823         absTime->tv_sec = secs;
5824       }
5825       absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5826     } else {
5827       jlong secs = time / NANOSECS_PER_SEC;
5828       if (secs >= MAX_SECS) {
5829         absTime->tv_sec = max_secs;
5830         absTime->tv_nsec = 0;
5831       } else {
5832         absTime->tv_sec = now.tv_sec + secs;
5833         absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5834         if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5835           absTime->tv_nsec -= NANOSECS_PER_SEC;
5836           ++absTime->tv_sec; // note: this must be <= max_secs
5837         }
5838       }
5839     }
5840   } else {
5841     // must be relative using monotonic clock
5842     struct timespec now;
5843     int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5844     assert_status(status == 0, status, "clock_gettime");
5845     max_secs = now.tv_sec + MAX_SECS;
5846     jlong secs = time / NANOSECS_PER_SEC;
5847     if (secs >= MAX_SECS) {
5848       absTime->tv_sec = max_secs;
5849       absTime->tv_nsec = 0;
5850     } else {
5851       absTime->tv_sec = now.tv_sec + secs;
5852       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5853       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5854         absTime->tv_nsec -= NANOSECS_PER_SEC;
5855         ++absTime->tv_sec; // note: this must be <= max_secs
5856       }
5857     }
5858   }
5859   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5860   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5861   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5862   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5863 }
5864 
5865 void Parker::park(bool isAbsolute, jlong time) {
5866   // Ideally we'd do something useful while spinning, such
5867   // as calling unpackTime().
5868 
5869   // Optional fast-path check:
5870   // Return immediately if a permit is available.
5871   // We depend on Atomic::xchg() having full barrier semantics
5872   // since we are doing a lock-free update to _counter.
5873   if (Atomic::xchg(0, &_counter) > 0) return;
5874 
5875   Thread* thread = Thread::current();
5876   assert(thread->is_Java_thread(), "Must be JavaThread");
5877   JavaThread *jt = (JavaThread *)thread;
5878 
5879   // Optional optimization -- avoid state transitions if there's an interrupt pending.
5880   // Check interrupt before trying to wait
5881   if (Thread::is_interrupted(thread, false)) {
5882     return;
5883   }
5884 
5885   // Next, demultiplex/decode time arguments
5886   timespec absTime;
5887   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5888     return;
5889   }
5890   if (time > 0) {
5891     unpackTime(&absTime, isAbsolute, time);
5892   }
5893 
5894 
5895   // Enter safepoint region
5896   // Beware of deadlocks such as 6317397.
5897   // The per-thread Parker:: mutex is a classic leaf-lock.
5898   // In particular a thread must never block on the Threads_lock while
5899   // holding the Parker:: mutex.  If safepoints are pending both the
5900   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5901   ThreadBlockInVM tbivm(jt);
5902 
5903   // Don't wait if cannot get lock since interference arises from
5904   // unblocking.  Also. check interrupt before trying wait
5905   if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5906     return;
5907   }
5908 
5909   int status;
5910   if (_counter > 0)  { // no wait needed
5911     _counter = 0;
5912     status = pthread_mutex_unlock(_mutex);
5913     assert_status(status == 0, status, "invariant");
5914     // Paranoia to ensure our locked and lock-free paths interact
5915     // correctly with each other and Java-level accesses.
5916     OrderAccess::fence();
5917     return;
5918   }
5919 
5920 #ifdef ASSERT
5921   // Don't catch signals while blocked; let the running threads have the signals.
5922   // (This allows a debugger to break into the running thread.)
5923   sigset_t oldsigs;
5924   sigemptyset(&oldsigs);
5925   sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5926   pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5927 #endif
5928 
5929   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5930   jt->set_suspend_equivalent();
5931   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5932 
5933   assert(_cur_index == -1, "invariant");
5934   if (time == 0) {
5935     _cur_index = REL_INDEX; // arbitrary choice when not timed
5936     status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5937   } else {
5938     _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5939     status = pthread_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5940   }
5941   _cur_index = -1;
5942   assert_status(status == 0 || status == EINTR ||
5943                 status == ETIME || status == ETIMEDOUT,
5944                 status, "cond_timedwait");
5945 
5946 #ifdef ASSERT
5947   pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5948 #endif
5949 
5950   _counter = 0;
5951   status = pthread_mutex_unlock(_mutex);
5952   assert_status(status == 0, status, "invariant");
5953   // Paranoia to ensure our locked and lock-free paths interact
5954   // correctly with each other and Java-level accesses.
5955   OrderAccess::fence();
5956 
5957   // If externally suspended while waiting, re-suspend
5958   if (jt->handle_special_suspend_equivalent_condition()) {
5959     jt->java_suspend_self();
5960   }
5961 }
5962 
5963 void Parker::unpark() {
5964   int status = pthread_mutex_lock(_mutex);
5965   assert_status(status == 0, status, "invariant");
5966   const int s = _counter;
5967   _counter = 1;
5968   // must capture correct index before unlocking
5969   int index = _cur_index;
5970   status = pthread_mutex_unlock(_mutex);
5971   assert_status(status == 0, status, "invariant");
5972   if (s < 1 && index != -1) {
5973     // thread is definitely parked
5974     status = pthread_cond_signal(&_cond[index]);
5975     assert_status(status == 0, status, "invariant");
5976   }
5977 }
5978 
5979 
5980 extern char** environ;
5981 
5982 // Run the specified command in a separate process. Return its exit value,
5983 // or -1 on failure (e.g. can't fork a new process).
5984 // Unlike system(), this function can be called from signal handler. It
5985 // doesn't block SIGINT et al.
5986 int os::fork_and_exec(char* cmd) {
5987   const char * argv[4] = {"sh", "-c", cmd, NULL};
5988 
5989   pid_t pid = fork();
5990 
5991   if (pid < 0) {
5992     // fork failed
5993     return -1;
5994 
5995   } else if (pid == 0) {
5996     // child process
5997 
5998     execve("/bin/sh", (char* const*)argv, environ);
5999 
6000     // execve failed
6001     _exit(-1);
6002 
6003   } else  {
6004     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6005     // care about the actual exit code, for now.
6006 
6007     int status;
6008 
6009     // Wait for the child process to exit.  This returns immediately if
6010     // the child has already exited. */
6011     while (waitpid(pid, &status, 0) < 0) {
6012       switch (errno) {
6013       case ECHILD: return 0;
6014       case EINTR: break;
6015       default: return -1;
6016       }
6017     }
6018 
6019     if (WIFEXITED(status)) {
6020       // The child exited normally; get its exit code.
6021       return WEXITSTATUS(status);
6022     } else if (WIFSIGNALED(status)) {
6023       // The child exited because of a signal
6024       // The best value to return is 0x80 + signal number,
6025       // because that is what all Unix shells do, and because
6026       // it allows callers to distinguish between process exit and
6027       // process death by signal.
6028       return 0x80 + WTERMSIG(status);
6029     } else {
6030       // Unknown exit code; pass it through
6031       return status;
6032     }
6033   }
6034 }
6035 
6036 // is_headless_jre()
6037 //
6038 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6039 // in order to report if we are running in a headless jre
6040 //
6041 // Since JDK8 xawt/libmawt.so was moved into the same directory
6042 // as libawt.so, and renamed libawt_xawt.so
6043 //
6044 bool os::is_headless_jre() {
6045   struct stat statbuf;
6046   char buf[MAXPATHLEN];
6047   char libmawtpath[MAXPATHLEN];
6048   const char *xawtstr  = "/xawt/libmawt.so";
6049   const char *new_xawtstr = "/libawt_xawt.so";
6050   char *p;
6051 
6052   // Get path to libjvm.so
6053   os::jvm_path(buf, sizeof(buf));
6054 
6055   // Get rid of libjvm.so
6056   p = strrchr(buf, '/');
6057   if (p == NULL) {
6058     return false;
6059   } else {
6060     *p = '\0';
6061   }
6062 
6063   // Get rid of client or server
6064   p = strrchr(buf, '/');
6065   if (p == NULL) {
6066     return false;
6067   } else {
6068     *p = '\0';
6069   }
6070 
6071   // check xawt/libmawt.so
6072   strcpy(libmawtpath, buf);
6073   strcat(libmawtpath, xawtstr);
6074   if (::stat(libmawtpath, &statbuf) == 0) return false;
6075 
6076   // check libawt_xawt.so
6077   strcpy(libmawtpath, buf);
6078   strcat(libmawtpath, new_xawtstr);
6079   if (::stat(libmawtpath, &statbuf) == 0) return false;
6080 
6081   return true;
6082 }
6083 
6084 // Get the default path to the core file
6085 // Returns the length of the string
6086 int os::get_core_path(char* buffer, size_t bufferSize) {
6087   /*
6088    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
6089    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
6090    */
6091   const int core_pattern_len = 129;
6092   char core_pattern[core_pattern_len] = {0};
6093 
6094   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
6095   if (core_pattern_file == -1) {
6096     return -1;
6097   }
6098 
6099   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6100   ::close(core_pattern_file);
6101   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
6102     return -1;
6103   }
6104   if (core_pattern[ret-1] == '\n') {
6105     core_pattern[ret-1] = '\0';
6106   } else {
6107     core_pattern[ret] = '\0';
6108   }
6109 
6110   char *pid_pos = strstr(core_pattern, "%p");
6111   int written;
6112 
6113   if (core_pattern[0] == '/') {
6114     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6115   } else {
6116     char cwd[PATH_MAX];
6117 
6118     const char* p = get_current_directory(cwd, PATH_MAX);
6119     if (p == NULL) {
6120       return -1;
6121     }
6122 
6123     if (core_pattern[0] == '|') {
6124       written = jio_snprintf(buffer, bufferSize,
6125                              "\"%s\" (or dumping to %s/core.%d)",
6126                              &core_pattern[1], p, current_process_id());
6127     } else {
6128       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6129     }
6130   }
6131 
6132   if (written < 0) {
6133     return -1;
6134   }
6135 
6136   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6137     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6138 
6139     if (core_uses_pid_file != -1) {
6140       char core_uses_pid = 0;
6141       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6142       ::close(core_uses_pid_file);
6143 
6144       if (core_uses_pid == '1') {
6145         jio_snprintf(buffer + written, bufferSize - written,
6146                                           ".%d", current_process_id());
6147       }
6148     }
6149   }
6150 
6151   return strlen(buffer);
6152 }
6153 
6154 bool os::start_debugging(char *buf, int buflen) {
6155   int len = (int)strlen(buf);
6156   char *p = &buf[len];
6157 
6158   jio_snprintf(p, buflen-len,
6159                "\n\n"
6160                "Do you want to debug the problem?\n\n"
6161                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6162                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6163                "Otherwise, press RETURN to abort...",
6164                os::current_process_id(), os::current_process_id(),
6165                os::current_thread_id(), os::current_thread_id());
6166 
6167   bool yes = os::message_box("Unexpected Error", buf);
6168 
6169   if (yes) {
6170     // yes, user asked VM to launch debugger
6171     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6172                  os::current_process_id(), os::current_process_id());
6173 
6174     os::fork_and_exec(buf);
6175     yes = false;
6176   }
6177   return yes;
6178 }
6179 
6180 static inline struct timespec get_mtime(const char* filename) {
6181   struct stat st;
6182   int ret = os::stat(filename, &st);
6183   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
6184   return st.st_mtim;
6185 }
6186 
6187 int os::compare_file_modified_times(const char* file1, const char* file2) {
6188   struct timespec filetime1 = get_mtime(file1);
6189   struct timespec filetime2 = get_mtime(file2);
6190   int diff = filetime1.tv_sec - filetime2.tv_sec;
6191   if (diff == 0) {
6192     return filetime1.tv_nsec - filetime2.tv_nsec;
6193   }
6194   return diff;
6195 }
6196 
6197 /////////////// Unit tests ///////////////
6198 
6199 #ifndef PRODUCT
6200 
6201 #define test_log(...)              \
6202   do {                             \
6203     if (VerboseInternalVMTests) {  \
6204       tty->print_cr(__VA_ARGS__);  \
6205       tty->flush();                \
6206     }                              \
6207   } while (false)
6208 
6209 class TestReserveMemorySpecial : AllStatic {
6210  public:
6211   static void small_page_write(void* addr, size_t size) {
6212     size_t page_size = os::vm_page_size();
6213 
6214     char* end = (char*)addr + size;
6215     for (char* p = (char*)addr; p < end; p += page_size) {
6216       *p = 1;
6217     }
6218   }
6219 
6220   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6221     if (!UseHugeTLBFS) {
6222       return;
6223     }
6224 
6225     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6226 
6227     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6228 
6229     if (addr != NULL) {
6230       small_page_write(addr, size);
6231 
6232       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6233     }
6234   }
6235 
6236   static void test_reserve_memory_special_huge_tlbfs_only() {
6237     if (!UseHugeTLBFS) {
6238       return;
6239     }
6240 
6241     size_t lp = os::large_page_size();
6242 
6243     for (size_t size = lp; size <= lp * 10; size += lp) {
6244       test_reserve_memory_special_huge_tlbfs_only(size);
6245     }
6246   }
6247 
6248   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6249     size_t lp = os::large_page_size();
6250     size_t ag = os::vm_allocation_granularity();
6251 
6252     // sizes to test
6253     const size_t sizes[] = {
6254       lp, lp + ag, lp + lp / 2, lp * 2,
6255       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6256       lp * 10, lp * 10 + lp / 2
6257     };
6258     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6259 
6260     // For each size/alignment combination, we test three scenarios:
6261     // 1) with req_addr == NULL
6262     // 2) with a non-null req_addr at which we expect to successfully allocate
6263     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6264     //    expect the allocation to either fail or to ignore req_addr
6265 
6266     // Pre-allocate two areas; they shall be as large as the largest allocation
6267     //  and aligned to the largest alignment we will be testing.
6268     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6269     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6270       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6271       -1, 0);
6272     assert(mapping1 != MAP_FAILED, "should work");
6273 
6274     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6275       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6276       -1, 0);
6277     assert(mapping2 != MAP_FAILED, "should work");
6278 
6279     // Unmap the first mapping, but leave the second mapping intact: the first
6280     // mapping will serve as a value for a "good" req_addr (case 2). The second
6281     // mapping, still intact, as "bad" req_addr (case 3).
6282     ::munmap(mapping1, mapping_size);
6283 
6284     // Case 1
6285     test_log("%s, req_addr NULL:", __FUNCTION__);
6286     test_log("size            align           result");
6287 
6288     for (int i = 0; i < num_sizes; i++) {
6289       const size_t size = sizes[i];
6290       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6291         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6292         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6293                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6294         if (p != NULL) {
6295           assert(is_ptr_aligned(p, alignment), "must be");
6296           small_page_write(p, size);
6297           os::Linux::release_memory_special_huge_tlbfs(p, size);
6298         }
6299       }
6300     }
6301 
6302     // Case 2
6303     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6304     test_log("size            align           req_addr         result");
6305 
6306     for (int i = 0; i < num_sizes; i++) {
6307       const size_t size = sizes[i];
6308       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6309         char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6310         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6311         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6312                  size, alignment, p2i(req_addr), p2i(p),
6313                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6314         if (p != NULL) {
6315           assert(p == req_addr, "must be");
6316           small_page_write(p, size);
6317           os::Linux::release_memory_special_huge_tlbfs(p, size);
6318         }
6319       }
6320     }
6321 
6322     // Case 3
6323     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6324     test_log("size            align           req_addr         result");
6325 
6326     for (int i = 0; i < num_sizes; i++) {
6327       const size_t size = sizes[i];
6328       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6329         char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6330         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6331         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6332                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6333         // as the area around req_addr contains already existing mappings, the API should always
6334         // return NULL (as per contract, it cannot return another address)
6335         assert(p == NULL, "must be");
6336       }
6337     }
6338 
6339     ::munmap(mapping2, mapping_size);
6340 
6341   }
6342 
6343   static void test_reserve_memory_special_huge_tlbfs() {
6344     if (!UseHugeTLBFS) {
6345       return;
6346     }
6347 
6348     test_reserve_memory_special_huge_tlbfs_only();
6349     test_reserve_memory_special_huge_tlbfs_mixed();
6350   }
6351 
6352   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6353     if (!UseSHM) {
6354       return;
6355     }
6356 
6357     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6358 
6359     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6360 
6361     if (addr != NULL) {
6362       assert(is_ptr_aligned(addr, alignment), "Check");
6363       assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6364 
6365       small_page_write(addr, size);
6366 
6367       os::Linux::release_memory_special_shm(addr, size);
6368     }
6369   }
6370 
6371   static void test_reserve_memory_special_shm() {
6372     size_t lp = os::large_page_size();
6373     size_t ag = os::vm_allocation_granularity();
6374 
6375     for (size_t size = ag; size < lp * 3; size += ag) {
6376       for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6377         test_reserve_memory_special_shm(size, alignment);
6378       }
6379     }
6380   }
6381 
6382   static void test() {
6383     test_reserve_memory_special_huge_tlbfs();
6384     test_reserve_memory_special_shm();
6385   }
6386 };
6387 
6388 void TestReserveMemorySpecial_test() {
6389   TestReserveMemorySpecial::test();
6390 }
6391 
6392 #endif