1 /*
   2  * Copyright (c) 1999, 2018, 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 "jvm.h"
  27 #include "classfile/classLoader.hpp"
  28 #include "classfile/systemDictionary.hpp"
  29 #include "classfile/vmSymbols.hpp"
  30 #include "code/icBuffer.hpp"
  31 #include "code/vtableStubs.hpp"
  32 #include "compiler/compileBroker.hpp"
  33 #include "compiler/disassembler.hpp"
  34 #include "interpreter/interpreter.hpp"
  35 #include "logging/log.hpp"
  36 #include "memory/allocation.inline.hpp"
  37 #include "memory/filemap.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_linux.inline.hpp"
  40 #include "os_share_linux.hpp"
  41 #include "osContainer_linux.hpp"
  42 #include "prims/jniFastGetField.hpp"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.inline.hpp"
  49 #include "runtime/init.hpp"
  50 #include "runtime/java.hpp"
  51 #include "runtime/javaCalls.hpp"
  52 #include "runtime/mutexLocker.hpp"
  53 #include "runtime/objectMonitor.hpp"
  54 #include "runtime/orderAccess.inline.hpp"
  55 #include "runtime/osThread.hpp"
  56 #include "runtime/perfMemory.hpp"
  57 #include "runtime/sharedRuntime.hpp"
  58 #include "runtime/statSampler.hpp"
  59 #include "runtime/stubRoutines.hpp"
  60 #include "runtime/thread.inline.hpp"
  61 #include "runtime/threadCritical.hpp"
  62 #include "runtime/threadSMR.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/align.hpp"
  69 #include "utilities/decoder.hpp"
  70 #include "utilities/defaultStream.hpp"
  71 #include "utilities/events.hpp"
  72 #include "utilities/elfFile.hpp"
  73 #include "utilities/growableArray.hpp"
  74 #include "utilities/macros.hpp"
  75 #include "utilities/vmError.hpp"
  76 
  77 // put OS-includes here
  78 # include <sys/types.h>
  79 # include <sys/mman.h>
  80 # include <sys/stat.h>
  81 # include <sys/select.h>
  82 # include <pthread.h>
  83 # include <signal.h>
  84 # include <errno.h>
  85 # include <dlfcn.h>
  86 # include <stdio.h>
  87 # include <unistd.h>
  88 # include <sys/resource.h>
  89 # include <pthread.h>
  90 # include <sys/stat.h>
  91 # include <sys/time.h>
  92 # include <sys/times.h>
  93 # include <sys/utsname.h>
  94 # include <sys/socket.h>
  95 # include <sys/wait.h>
  96 # include <pwd.h>
  97 # include <poll.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 #define DAX_SHARED_BIT (1 << 8)
 133 ////////////////////////////////////////////////////////////////////////////////
 134 // global variables
 135 julong os::Linux::_physical_memory = 0;
 136 
 137 address   os::Linux::_initial_thread_stack_bottom = NULL;
 138 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 139 
 140 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 141 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 142 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 143 Mutex* os::Linux::_createThread_lock = NULL;
 144 pthread_t os::Linux::_main_thread;
 145 int os::Linux::_page_size = -1;
 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 
 151 static jlong initial_time_count=0;
 152 
 153 static int clock_tics_per_sec = 100;
 154 
 155 // If the VM might have been created on the primordial thread, we need to resolve the
 156 // primordial thread stack bounds and check if the current thread might be the
 157 // primordial thread in places. If we know that the primordial thread is never used,
 158 // such as when the VM was created by one of the standard java launchers, we can
 159 // avoid this
 160 static bool suppress_primordial_thread_resolution = false;
 161 
 162 // For diagnostics to print a message once. see run_periodic_checks
 163 static sigset_t check_signal_done;
 164 static bool check_signals = true;
 165 
 166 // Signal number used to suspend/resume a thread
 167 
 168 // do not use any signal number less than SIGSEGV, see 4355769
 169 static int SR_signum = SIGUSR2;
 170 sigset_t SR_sigset;
 171 
 172 // utility functions
 173 
 174 static int SR_initialize();
 175 
 176 julong os::available_memory() {
 177   return Linux::available_memory();
 178 }
 179 
 180 julong os::Linux::available_memory() {
 181   // values in struct sysinfo are "unsigned long"
 182   struct sysinfo si;
 183   julong avail_mem;
 184 
 185   if (OSContainer::is_containerized()) {
 186     jlong mem_limit, mem_usage;
 187     if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
 188       log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
 189                              mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
 190     }
 191     if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
 192       log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
 193     }
 194     if (mem_limit > 0 && mem_usage > 0 ) {
 195       avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
 196       log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
 197       return avail_mem;
 198     }
 199   }
 200 
 201   sysinfo(&si);
 202   avail_mem = (julong)si.freeram * si.mem_unit;
 203   log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
 204   return avail_mem;
 205 }
 206 
 207 julong os::physical_memory() {
 208   jlong phys_mem = 0;
 209   if (OSContainer::is_containerized()) {
 210     jlong mem_limit;
 211     if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
 212       log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
 213       return mem_limit;
 214     }
 215     log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
 216                             mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
 217   }
 218 
 219   phys_mem = Linux::physical_memory();
 220   log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
 221   return phys_mem;
 222 }
 223 
 224 // Return true if user is running as root.
 225 
 226 bool os::have_special_privileges() {
 227   static bool init = false;
 228   static bool privileges = false;
 229   if (!init) {
 230     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 231     init = true;
 232   }
 233   return privileges;
 234 }
 235 
 236 
 237 #ifndef SYS_gettid
 238 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 239   #ifdef __ia64__
 240     #define SYS_gettid 1105
 241   #else
 242     #ifdef __i386__
 243       #define SYS_gettid 224
 244     #else
 245       #ifdef __amd64__
 246         #define SYS_gettid 186
 247       #else
 248         #ifdef __sparc__
 249           #define SYS_gettid 143
 250         #else
 251           #error define gettid for the arch
 252         #endif
 253       #endif
 254     #endif
 255   #endif
 256 #endif
 257 
 258 
 259 // pid_t gettid()
 260 //
 261 // Returns the kernel thread id of the currently running thread. Kernel
 262 // thread id is used to access /proc.
 263 pid_t os::Linux::gettid() {
 264   int rslt = syscall(SYS_gettid);
 265   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 266   return (pid_t)rslt;
 267 }
 268 
 269 // Most versions of linux have a bug where the number of processors are
 270 // determined by looking at the /proc file system.  In a chroot environment,
 271 // the system call returns 1.  This causes the VM to act as if it is
 272 // a single processor and elide locking (see is_MP() call).
 273 static bool unsafe_chroot_detected = false;
 274 static const char *unstable_chroot_error = "/proc file system not found.\n"
 275                      "Java may be unstable running multithreaded in a chroot "
 276                      "environment on Linux when /proc filesystem is not mounted.";
 277 
 278 void os::Linux::initialize_system_info() {
 279   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 280   if (processor_count() == 1) {
 281     pid_t pid = os::Linux::gettid();
 282     char fname[32];
 283     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 284     FILE *fp = fopen(fname, "r");
 285     if (fp == NULL) {
 286       unsafe_chroot_detected = true;
 287     } else {
 288       fclose(fp);
 289     }
 290   }
 291   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 292   assert(processor_count() > 0, "linux error");
 293 }
 294 
 295 void os::init_system_properties_values() {
 296   // The next steps are taken in the product version:
 297   //
 298   // Obtain the JAVA_HOME value from the location of libjvm.so.
 299   // This library should be located at:
 300   // <JAVA_HOME>/lib/{client|server}/libjvm.so.
 301   //
 302   // If "/jre/lib/" appears at the right place in the path, then we
 303   // assume libjvm.so is installed in a JDK and we use this path.
 304   //
 305   // Otherwise exit with message: "Could not create the Java virtual machine."
 306   //
 307   // The following extra steps are taken in the debugging version:
 308   //
 309   // If "/jre/lib/" does NOT appear at the right place in the path
 310   // instead of exit check for $JAVA_HOME environment variable.
 311   //
 312   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 313   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 314   // it looks like libjvm.so is installed there
 315   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 316   //
 317   // Otherwise exit.
 318   //
 319   // Important note: if the location of libjvm.so changes this
 320   // code needs to be changed accordingly.
 321 
 322   // See ld(1):
 323   //      The linker uses the following search paths to locate required
 324   //      shared libraries:
 325   //        1: ...
 326   //        ...
 327   //        7: The default directories, normally /lib and /usr/lib.
 328 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
 329   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 330 #else
 331   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 332 #endif
 333 
 334 // Base path of extensions installed on the system.
 335 #define SYS_EXT_DIR     "/usr/java/packages"
 336 #define EXTENSIONS_DIR  "/lib/ext"
 337 
 338   // Buffer that fits several sprintfs.
 339   // Note that the space for the colon and the trailing null are provided
 340   // by the nulls included by the sizeof operator.
 341   const size_t bufsize =
 342     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 343          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 344   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 345 
 346   // sysclasspath, java_home, dll_dir
 347   {
 348     char *pslash;
 349     os::jvm_path(buf, bufsize);
 350 
 351     // Found the full path to libjvm.so.
 352     // Now cut the path to <java_home>/jre if we can.
 353     pslash = strrchr(buf, '/');
 354     if (pslash != NULL) {
 355       *pslash = '\0';            // Get rid of /libjvm.so.
 356     }
 357     pslash = strrchr(buf, '/');
 358     if (pslash != NULL) {
 359       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 360     }
 361     Arguments::set_dll_dir(buf);
 362 
 363     if (pslash != NULL) {
 364       pslash = strrchr(buf, '/');
 365       if (pslash != NULL) {
 366         *pslash = '\0';        // Get rid of /lib.
 367       }
 368     }
 369     Arguments::set_java_home(buf);
 370     set_boot_path('/', ':');
 371   }
 372 
 373   // Where to look for native libraries.
 374   //
 375   // Note: Due to a legacy implementation, most of the library path
 376   // is set in the launcher. This was to accomodate linking restrictions
 377   // on legacy Linux implementations (which are no longer supported).
 378   // Eventually, all the library path setting will be done here.
 379   //
 380   // However, to prevent the proliferation of improperly built native
 381   // libraries, the new path component /usr/java/packages is added here.
 382   // Eventually, all the library path setting will be done here.
 383   {
 384     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 385     // should always exist (until the legacy problem cited above is
 386     // addressed).
 387     const char *v = ::getenv("LD_LIBRARY_PATH");
 388     const char *v_colon = ":";
 389     if (v == NULL) { v = ""; v_colon = ""; }
 390     // That's +1 for the colon and +1 for the trailing '\0'.
 391     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 392                                                      strlen(v) + 1 +
 393                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
 394                                                      mtInternal);
 395     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
 396     Arguments::set_library_path(ld_library_path);
 397     FREE_C_HEAP_ARRAY(char, ld_library_path);
 398   }
 399 
 400   // Extensions directories.
 401   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 402   Arguments::set_ext_dirs(buf);
 403 
 404   FREE_C_HEAP_ARRAY(char, buf);
 405 
 406 #undef DEFAULT_LIBPATH
 407 #undef SYS_EXT_DIR
 408 #undef EXTENSIONS_DIR
 409 }
 410 
 411 ////////////////////////////////////////////////////////////////////////////////
 412 // breakpoint support
 413 
 414 void os::breakpoint() {
 415   BREAKPOINT;
 416 }
 417 
 418 extern "C" void breakpoint() {
 419   // use debugger to set breakpoint here
 420 }
 421 
 422 ////////////////////////////////////////////////////////////////////////////////
 423 // signal support
 424 
 425 debug_only(static bool signal_sets_initialized = false);
 426 static sigset_t unblocked_sigs, vm_sigs;
 427 
 428 bool os::Linux::is_sig_ignored(int sig) {
 429   struct sigaction oact;
 430   sigaction(sig, (struct sigaction*)NULL, &oact);
 431   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 432                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 433   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 434     return true;
 435   } else {
 436     return false;
 437   }
 438 }
 439 
 440 void os::Linux::signal_sets_init() {
 441   // Should also have an assertion stating we are still single-threaded.
 442   assert(!signal_sets_initialized, "Already initialized");
 443   // Fill in signals that are necessarily unblocked for all threads in
 444   // the VM. Currently, we unblock the following signals:
 445   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 446   //                         by -Xrs (=ReduceSignalUsage));
 447   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 448   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 449   // the dispositions or masks wrt these signals.
 450   // Programs embedding the VM that want to use the above signals for their
 451   // own purposes must, at this time, use the "-Xrs" option to prevent
 452   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 453   // (See bug 4345157, and other related bugs).
 454   // In reality, though, unblocking these signals is really a nop, since
 455   // these signals are not blocked by default.
 456   sigemptyset(&unblocked_sigs);
 457   sigaddset(&unblocked_sigs, SIGILL);
 458   sigaddset(&unblocked_sigs, SIGSEGV);
 459   sigaddset(&unblocked_sigs, SIGBUS);
 460   sigaddset(&unblocked_sigs, SIGFPE);
 461 #if defined(PPC64)
 462   sigaddset(&unblocked_sigs, SIGTRAP);
 463 #endif
 464   sigaddset(&unblocked_sigs, SR_signum);
 465 
 466   if (!ReduceSignalUsage) {
 467     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 468       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 469     }
 470     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 471       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 472     }
 473     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 474       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 475     }
 476   }
 477   // Fill in signals that are blocked by all but the VM thread.
 478   sigemptyset(&vm_sigs);
 479   if (!ReduceSignalUsage) {
 480     sigaddset(&vm_sigs, BREAK_SIGNAL);
 481   }
 482   debug_only(signal_sets_initialized = true);
 483 
 484 }
 485 
 486 // These are signals that are unblocked while a thread is running Java.
 487 // (For some reason, they get blocked by default.)
 488 sigset_t* os::Linux::unblocked_signals() {
 489   assert(signal_sets_initialized, "Not initialized");
 490   return &unblocked_sigs;
 491 }
 492 
 493 // These are the signals that are blocked while a (non-VM) thread is
 494 // running Java. Only the VM thread handles these signals.
 495 sigset_t* os::Linux::vm_signals() {
 496   assert(signal_sets_initialized, "Not initialized");
 497   return &vm_sigs;
 498 }
 499 
 500 void os::Linux::hotspot_sigmask(Thread* thread) {
 501 
 502   //Save caller's signal mask before setting VM signal mask
 503   sigset_t caller_sigmask;
 504   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 505 
 506   OSThread* osthread = thread->osthread();
 507   osthread->set_caller_sigmask(caller_sigmask);
 508 
 509   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 510 
 511   if (!ReduceSignalUsage) {
 512     if (thread->is_VM_thread()) {
 513       // Only the VM thread handles BREAK_SIGNAL ...
 514       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 515     } else {
 516       // ... all other threads block BREAK_SIGNAL
 517       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 518     }
 519   }
 520 }
 521 
 522 //////////////////////////////////////////////////////////////////////////////
 523 // detecting pthread library
 524 
 525 void os::Linux::libpthread_init() {
 526   // Save glibc and pthread version strings.
 527 #if !defined(_CS_GNU_LIBC_VERSION) || \
 528     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 529   #error "glibc too old (< 2.3.2)"
 530 #endif
 531 
 532   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 533   assert(n > 0, "cannot retrieve glibc version");
 534   char *str = (char *)malloc(n, mtInternal);
 535   confstr(_CS_GNU_LIBC_VERSION, str, n);
 536   os::Linux::set_glibc_version(str);
 537 
 538   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 539   assert(n > 0, "cannot retrieve pthread version");
 540   str = (char *)malloc(n, mtInternal);
 541   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 542   os::Linux::set_libpthread_version(str);
 543 }
 544 
 545 /////////////////////////////////////////////////////////////////////////////
 546 // thread stack expansion
 547 
 548 // os::Linux::manually_expand_stack() takes care of expanding the thread
 549 // stack. Note that this is normally not needed: pthread stacks allocate
 550 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 551 // committed. Therefore it is not necessary to expand the stack manually.
 552 //
 553 // Manually expanding the stack was historically needed on LinuxThreads
 554 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 555 // it is kept to deal with very rare corner cases:
 556 //
 557 // For one, user may run the VM on an own implementation of threads
 558 // whose stacks are - like the old LinuxThreads - implemented using
 559 // mmap(MAP_GROWSDOWN).
 560 //
 561 // Also, this coding may be needed if the VM is running on the primordial
 562 // thread. Normally we avoid running on the primordial thread; however,
 563 // user may still invoke the VM on the primordial thread.
 564 //
 565 // The following historical comment describes the details about running
 566 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 567 
 568 
 569 // Force Linux kernel to expand current thread stack. If "bottom" is close
 570 // to the stack guard, caller should block all signals.
 571 //
 572 // MAP_GROWSDOWN:
 573 //   A special mmap() flag that is used to implement thread stacks. It tells
 574 //   kernel that the memory region should extend downwards when needed. This
 575 //   allows early versions of LinuxThreads to only mmap the first few pages
 576 //   when creating a new thread. Linux kernel will automatically expand thread
 577 //   stack as needed (on page faults).
 578 //
 579 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 580 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 581 //   region, it's hard to tell if the fault is due to a legitimate stack
 582 //   access or because of reading/writing non-exist memory (e.g. buffer
 583 //   overrun). As a rule, if the fault happens below current stack pointer,
 584 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 585 //   application (see Linux kernel fault.c).
 586 //
 587 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 588 //   stack overflow detection.
 589 //
 590 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 591 //   not use MAP_GROWSDOWN.
 592 //
 593 // To get around the problem and allow stack banging on Linux, we need to
 594 // manually expand thread stack after receiving the SIGSEGV.
 595 //
 596 // There are two ways to expand thread stack to address "bottom", we used
 597 // both of them in JVM before 1.5:
 598 //   1. adjust stack pointer first so that it is below "bottom", and then
 599 //      touch "bottom"
 600 //   2. mmap() the page in question
 601 //
 602 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 603 // if current sp is already near the lower end of page 101, and we need to
 604 // call mmap() to map page 100, it is possible that part of the mmap() frame
 605 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 606 // That will destroy the mmap() frame and cause VM to crash.
 607 //
 608 // The following code works by adjusting sp first, then accessing the "bottom"
 609 // page to force a page fault. Linux kernel will then automatically expand the
 610 // stack mapping.
 611 //
 612 // _expand_stack_to() assumes its frame size is less than page size, which
 613 // should always be true if the function is not inlined.
 614 
 615 static void NOINLINE _expand_stack_to(address bottom) {
 616   address sp;
 617   size_t size;
 618   volatile char *p;
 619 
 620   // Adjust bottom to point to the largest address within the same page, it
 621   // gives us a one-page buffer if alloca() allocates slightly more memory.
 622   bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
 623   bottom += os::Linux::page_size() - 1;
 624 
 625   // sp might be slightly above current stack pointer; if that's the case, we
 626   // will alloca() a little more space than necessary, which is OK. Don't use
 627   // os::current_stack_pointer(), as its result can be slightly below current
 628   // stack pointer, causing us to not alloca enough to reach "bottom".
 629   sp = (address)&sp;
 630 
 631   if (sp > bottom) {
 632     size = sp - bottom;
 633     p = (volatile char *)alloca(size);
 634     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 635     p[0] = '\0';
 636   }
 637 }
 638 
 639 void os::Linux::expand_stack_to(address bottom) {
 640   _expand_stack_to(bottom);
 641 }
 642 
 643 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 644   assert(t!=NULL, "just checking");
 645   assert(t->osthread()->expanding_stack(), "expand should be set");
 646   assert(t->stack_base() != NULL, "stack_base was not initialized");
 647 
 648   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 649     sigset_t mask_all, old_sigset;
 650     sigfillset(&mask_all);
 651     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 652     _expand_stack_to(addr);
 653     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 654     return true;
 655   }
 656   return false;
 657 }
 658 
 659 //////////////////////////////////////////////////////////////////////////////
 660 // create new thread
 661 
 662 // Thread start routine for all newly created threads
 663 static void *thread_native_entry(Thread *thread) {
 664   // Try to randomize the cache line index of hot stack frames.
 665   // This helps when threads of the same stack traces evict each other's
 666   // cache lines. The threads can be either from the same JVM instance, or
 667   // from different JVM instances. The benefit is especially true for
 668   // processors with hyperthreading technology.
 669   static int counter = 0;
 670   int pid = os::current_process_id();
 671   alloca(((pid ^ counter++) & 7) * 128);
 672 
 673   thread->initialize_thread_current();
 674 
 675   OSThread* osthread = thread->osthread();
 676   Monitor* sync = osthread->startThread_lock();
 677 
 678   osthread->set_thread_id(os::current_thread_id());
 679 
 680   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 681     os::current_thread_id(), (uintx) pthread_self());
 682 
 683   if (UseNUMA) {
 684     int lgrp_id = os::numa_get_group_id();
 685     if (lgrp_id != -1) {
 686       thread->set_lgrp_id(lgrp_id);
 687     }
 688   }
 689   // initialize signal mask for this thread
 690   os::Linux::hotspot_sigmask(thread);
 691 
 692   // initialize floating point control register
 693   os::Linux::init_thread_fpu_state();
 694 
 695   // handshaking with parent thread
 696   {
 697     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 698 
 699     // notify parent thread
 700     osthread->set_state(INITIALIZED);
 701     sync->notify_all();
 702 
 703     // wait until os::start_thread()
 704     while (osthread->get_state() == INITIALIZED) {
 705       sync->wait(Mutex::_no_safepoint_check_flag);
 706     }
 707   }
 708 
 709   // call one more level start routine
 710   thread->run();
 711 
 712   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 713     os::current_thread_id(), (uintx) pthread_self());
 714 
 715   // If a thread has not deleted itself ("delete this") as part of its
 716   // termination sequence, we have to ensure thread-local-storage is
 717   // cleared before we actually terminate. No threads should ever be
 718   // deleted asynchronously with respect to their termination.
 719   if (Thread::current_or_null_safe() != NULL) {
 720     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 721     thread->clear_thread_current();
 722   }
 723 
 724   return 0;
 725 }
 726 
 727 bool os::create_thread(Thread* thread, ThreadType thr_type,
 728                        size_t req_stack_size) {
 729   assert(thread->osthread() == NULL, "caller responsible");
 730 
 731   // Allocate the OSThread object
 732   OSThread* osthread = new OSThread(NULL, NULL);
 733   if (osthread == NULL) {
 734     return false;
 735   }
 736 
 737   // set the correct thread state
 738   osthread->set_thread_type(thr_type);
 739 
 740   // Initial state is ALLOCATED but not INITIALIZED
 741   osthread->set_state(ALLOCATED);
 742 
 743   thread->set_osthread(osthread);
 744 
 745   // init thread attributes
 746   pthread_attr_t attr;
 747   pthread_attr_init(&attr);
 748   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 749 
 750   // Calculate stack size if it's not specified by caller.
 751   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 752   // In the Linux NPTL pthread implementation the guard size mechanism
 753   // is not implemented properly. The posix standard requires adding
 754   // the size of the guard pages to the stack size, instead Linux
 755   // takes the space out of 'stacksize'. Thus we adapt the requested
 756   // stack_size by the size of the guard pages to mimick proper
 757   // behaviour. However, be careful not to end up with a size
 758   // of zero due to overflow. Don't add the guard page in that case.
 759   size_t guard_size = os::Linux::default_guard_size(thr_type);
 760   if (stack_size <= SIZE_MAX - guard_size) {
 761     stack_size += guard_size;
 762   }
 763   assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
 764 
 765   int status = pthread_attr_setstacksize(&attr, stack_size);
 766   assert_status(status == 0, status, "pthread_attr_setstacksize");
 767 
 768   // Configure glibc guard page.
 769   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 770 
 771   ThreadState state;
 772 
 773   {
 774     pthread_t tid;
 775     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 776 
 777     char buf[64];
 778     if (ret == 0) {
 779       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 780         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 781     } else {
 782       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 783         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 784     }
 785 
 786     pthread_attr_destroy(&attr);
 787 
 788     if (ret != 0) {
 789       // Need to clean up stuff we've allocated so far
 790       thread->set_osthread(NULL);
 791       delete osthread;
 792       return false;
 793     }
 794 
 795     // Store pthread info into the OSThread
 796     osthread->set_pthread_id(tid);
 797 
 798     // Wait until child thread is either initialized or aborted
 799     {
 800       Monitor* sync_with_child = osthread->startThread_lock();
 801       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 802       while ((state = osthread->get_state()) == ALLOCATED) {
 803         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 804       }
 805     }
 806   }
 807 
 808   // Aborted due to thread limit being reached
 809   if (state == ZOMBIE) {
 810     thread->set_osthread(NULL);
 811     delete osthread;
 812     return false;
 813   }
 814 
 815   // The thread is returned suspended (in state INITIALIZED),
 816   // and is started higher up in the call chain
 817   assert(state == INITIALIZED, "race condition");
 818   return true;
 819 }
 820 
 821 /////////////////////////////////////////////////////////////////////////////
 822 // attach existing thread
 823 
 824 // bootstrap the main thread
 825 bool os::create_main_thread(JavaThread* thread) {
 826   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 827   return create_attached_thread(thread);
 828 }
 829 
 830 bool os::create_attached_thread(JavaThread* thread) {
 831 #ifdef ASSERT
 832   thread->verify_not_published();
 833 #endif
 834 
 835   // Allocate the OSThread object
 836   OSThread* osthread = new OSThread(NULL, NULL);
 837 
 838   if (osthread == NULL) {
 839     return false;
 840   }
 841 
 842   // Store pthread info into the OSThread
 843   osthread->set_thread_id(os::Linux::gettid());
 844   osthread->set_pthread_id(::pthread_self());
 845 
 846   // initialize floating point control register
 847   os::Linux::init_thread_fpu_state();
 848 
 849   // Initial thread state is RUNNABLE
 850   osthread->set_state(RUNNABLE);
 851 
 852   thread->set_osthread(osthread);
 853 
 854   if (UseNUMA) {
 855     int lgrp_id = os::numa_get_group_id();
 856     if (lgrp_id != -1) {
 857       thread->set_lgrp_id(lgrp_id);
 858     }
 859   }
 860 
 861   if (os::is_primordial_thread()) {
 862     // If current thread is primordial thread, its stack is mapped on demand,
 863     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 864     // the entire stack region to avoid SEGV in stack banging.
 865     // It is also useful to get around the heap-stack-gap problem on SuSE
 866     // kernel (see 4821821 for details). We first expand stack to the top
 867     // of yellow zone, then enable stack yellow zone (order is significant,
 868     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 869     // is no gap between the last two virtual memory regions.
 870 
 871     JavaThread *jt = (JavaThread *)thread;
 872     address addr = jt->stack_reserved_zone_base();
 873     assert(addr != NULL, "initialization problem?");
 874     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 875 
 876     osthread->set_expanding_stack();
 877     os::Linux::manually_expand_stack(jt, addr);
 878     osthread->clear_expanding_stack();
 879   }
 880 
 881   // initialize signal mask for this thread
 882   // and save the caller's signal mask
 883   os::Linux::hotspot_sigmask(thread);
 884 
 885   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 886     os::current_thread_id(), (uintx) pthread_self());
 887 
 888   return true;
 889 }
 890 
 891 void os::pd_start_thread(Thread* thread) {
 892   OSThread * osthread = thread->osthread();
 893   assert(osthread->get_state() != INITIALIZED, "just checking");
 894   Monitor* sync_with_child = osthread->startThread_lock();
 895   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 896   sync_with_child->notify();
 897 }
 898 
 899 // Free Linux resources related to the OSThread
 900 void os::free_thread(OSThread* osthread) {
 901   assert(osthread != NULL, "osthread not set");
 902 
 903   // We are told to free resources of the argument thread,
 904   // but we can only really operate on the current thread.
 905   assert(Thread::current()->osthread() == osthread,
 906          "os::free_thread but not current thread");
 907 
 908 #ifdef ASSERT
 909   sigset_t current;
 910   sigemptyset(&current);
 911   pthread_sigmask(SIG_SETMASK, NULL, &current);
 912   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 913 #endif
 914 
 915   // Restore caller's signal mask
 916   sigset_t sigmask = osthread->caller_sigmask();
 917   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 918 
 919   delete osthread;
 920 }
 921 
 922 //////////////////////////////////////////////////////////////////////////////
 923 // primordial thread
 924 
 925 // Check if current thread is the primordial thread, similar to Solaris thr_main.
 926 bool os::is_primordial_thread(void) {
 927   if (suppress_primordial_thread_resolution) {
 928     return false;
 929   }
 930   char dummy;
 931   // If called before init complete, thread stack bottom will be null.
 932   // Can be called if fatal error occurs before initialization.
 933   if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
 934   assert(os::Linux::initial_thread_stack_bottom() != NULL &&
 935          os::Linux::initial_thread_stack_size()   != 0,
 936          "os::init did not locate primordial thread's stack region");
 937   if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
 938       (address)&dummy < os::Linux::initial_thread_stack_bottom() +
 939                         os::Linux::initial_thread_stack_size()) {
 940     return true;
 941   } else {
 942     return false;
 943   }
 944 }
 945 
 946 // Find the virtual memory area that contains addr
 947 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 948   FILE *fp = fopen("/proc/self/maps", "r");
 949   if (fp) {
 950     address low, high;
 951     while (!feof(fp)) {
 952       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 953         if (low <= addr && addr < high) {
 954           if (vma_low)  *vma_low  = low;
 955           if (vma_high) *vma_high = high;
 956           fclose(fp);
 957           return true;
 958         }
 959       }
 960       for (;;) {
 961         int ch = fgetc(fp);
 962         if (ch == EOF || ch == (int)'\n') break;
 963       }
 964     }
 965     fclose(fp);
 966   }
 967   return false;
 968 }
 969 
 970 // Locate primordial thread stack. This special handling of primordial thread stack
 971 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 972 // bogus value for the primordial process thread. While the launcher has created
 973 // the VM in a new thread since JDK 6, we still have to allow for the use of the
 974 // JNI invocation API from a primordial thread.
 975 void os::Linux::capture_initial_stack(size_t max_size) {
 976 
 977   // max_size is either 0 (which means accept OS default for thread stacks) or
 978   // a user-specified value known to be at least the minimum needed. If we
 979   // are actually on the primordial thread we can make it appear that we have a
 980   // smaller max_size stack by inserting the guard pages at that location. But we
 981   // cannot do anything to emulate a larger stack than what has been provided by
 982   // the OS or threading library. In fact if we try to use a stack greater than
 983   // what is set by rlimit then we will crash the hosting process.
 984 
 985   // Maximum stack size is the easy part, get it from RLIMIT_STACK.
 986   // If this is "unlimited" then it will be a huge value.
 987   struct rlimit rlim;
 988   getrlimit(RLIMIT_STACK, &rlim);
 989   size_t stack_size = rlim.rlim_cur;
 990 
 991   // 6308388: a bug in ld.so will relocate its own .data section to the
 992   //   lower end of primordial stack; reduce ulimit -s value a little bit
 993   //   so we won't install guard page on ld.so's data section.
 994   //   But ensure we don't underflow the stack size - allow 1 page spare
 995   if (stack_size >= (size_t)(3 * page_size())) {
 996     stack_size -= 2 * page_size();
 997   }
 998 
 999   // Try to figure out where the stack base (top) is. This is harder.
1000   //
1001   // When an application is started, glibc saves the initial stack pointer in
1002   // a global variable "__libc_stack_end", which is then used by system
1003   // libraries. __libc_stack_end should be pretty close to stack top. The
1004   // variable is available since the very early days. However, because it is
1005   // a private interface, it could disappear in the future.
1006   //
1007   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1008   // to __libc_stack_end, it is very close to stack top, but isn't the real
1009   // stack top. Note that /proc may not exist if VM is running as a chroot
1010   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1011   // /proc/<pid>/stat could change in the future (though unlikely).
1012   //
1013   // We try __libc_stack_end first. If that doesn't work, look for
1014   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1015   // as a hint, which should work well in most cases.
1016 
1017   uintptr_t stack_start;
1018 
1019   // try __libc_stack_end first
1020   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1021   if (p && *p) {
1022     stack_start = *p;
1023   } else {
1024     // see if we can get the start_stack field from /proc/self/stat
1025     FILE *fp;
1026     int pid;
1027     char state;
1028     int ppid;
1029     int pgrp;
1030     int session;
1031     int nr;
1032     int tpgrp;
1033     unsigned long flags;
1034     unsigned long minflt;
1035     unsigned long cminflt;
1036     unsigned long majflt;
1037     unsigned long cmajflt;
1038     unsigned long utime;
1039     unsigned long stime;
1040     long cutime;
1041     long cstime;
1042     long prio;
1043     long nice;
1044     long junk;
1045     long it_real;
1046     uintptr_t start;
1047     uintptr_t vsize;
1048     intptr_t rss;
1049     uintptr_t rsslim;
1050     uintptr_t scodes;
1051     uintptr_t ecode;
1052     int i;
1053 
1054     // Figure what the primordial thread stack base is. Code is inspired
1055     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1056     // followed by command name surrounded by parentheses, state, etc.
1057     char stat[2048];
1058     int statlen;
1059 
1060     fp = fopen("/proc/self/stat", "r");
1061     if (fp) {
1062       statlen = fread(stat, 1, 2047, fp);
1063       stat[statlen] = '\0';
1064       fclose(fp);
1065 
1066       // Skip pid and the command string. Note that we could be dealing with
1067       // weird command names, e.g. user could decide to rename java launcher
1068       // to "java 1.4.2 :)", then the stat file would look like
1069       //                1234 (java 1.4.2 :)) R ... ...
1070       // We don't really need to know the command string, just find the last
1071       // occurrence of ")" and then start parsing from there. See bug 4726580.
1072       char * s = strrchr(stat, ')');
1073 
1074       i = 0;
1075       if (s) {
1076         // Skip blank chars
1077         do { s++; } while (s && isspace(*s));
1078 
1079 #define _UFM UINTX_FORMAT
1080 #define _DFM INTX_FORMAT
1081 
1082         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1083         //              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
1084         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,
1085                    &state,          // 3  %c
1086                    &ppid,           // 4  %d
1087                    &pgrp,           // 5  %d
1088                    &session,        // 6  %d
1089                    &nr,             // 7  %d
1090                    &tpgrp,          // 8  %d
1091                    &flags,          // 9  %lu
1092                    &minflt,         // 10 %lu
1093                    &cminflt,        // 11 %lu
1094                    &majflt,         // 12 %lu
1095                    &cmajflt,        // 13 %lu
1096                    &utime,          // 14 %lu
1097                    &stime,          // 15 %lu
1098                    &cutime,         // 16 %ld
1099                    &cstime,         // 17 %ld
1100                    &prio,           // 18 %ld
1101                    &nice,           // 19 %ld
1102                    &junk,           // 20 %ld
1103                    &it_real,        // 21 %ld
1104                    &start,          // 22 UINTX_FORMAT
1105                    &vsize,          // 23 UINTX_FORMAT
1106                    &rss,            // 24 INTX_FORMAT
1107                    &rsslim,         // 25 UINTX_FORMAT
1108                    &scodes,         // 26 UINTX_FORMAT
1109                    &ecode,          // 27 UINTX_FORMAT
1110                    &stack_start);   // 28 UINTX_FORMAT
1111       }
1112 
1113 #undef _UFM
1114 #undef _DFM
1115 
1116       if (i != 28 - 2) {
1117         assert(false, "Bad conversion from /proc/self/stat");
1118         // product mode - assume we are the primordial thread, good luck in the
1119         // embedded case.
1120         warning("Can't detect primordial thread stack location - bad conversion");
1121         stack_start = (uintptr_t) &rlim;
1122       }
1123     } else {
1124       // For some reason we can't open /proc/self/stat (for example, running on
1125       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1126       // most cases, so don't abort:
1127       warning("Can't detect primordial thread stack location - no /proc/self/stat");
1128       stack_start = (uintptr_t) &rlim;
1129     }
1130   }
1131 
1132   // Now we have a pointer (stack_start) very close to the stack top, the
1133   // next thing to do is to figure out the exact location of stack top. We
1134   // can find out the virtual memory area that contains stack_start by
1135   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1136   // and its upper limit is the real stack top. (again, this would fail if
1137   // running inside chroot, because /proc may not exist.)
1138 
1139   uintptr_t stack_top;
1140   address low, high;
1141   if (find_vma((address)stack_start, &low, &high)) {
1142     // success, "high" is the true stack top. (ignore "low", because initial
1143     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1144     stack_top = (uintptr_t)high;
1145   } else {
1146     // failed, likely because /proc/self/maps does not exist
1147     warning("Can't detect primordial thread stack location - find_vma failed");
1148     // best effort: stack_start is normally within a few pages below the real
1149     // stack top, use it as stack top, and reduce stack size so we won't put
1150     // guard page outside stack.
1151     stack_top = stack_start;
1152     stack_size -= 16 * page_size();
1153   }
1154 
1155   // stack_top could be partially down the page so align it
1156   stack_top = align_up(stack_top, page_size());
1157 
1158   // Allowed stack value is minimum of max_size and what we derived from rlimit
1159   if (max_size > 0) {
1160     _initial_thread_stack_size = MIN2(max_size, stack_size);
1161   } else {
1162     // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1163     // clamp it at 8MB as we do on Solaris
1164     _initial_thread_stack_size = MIN2(stack_size, 8*M);
1165   }
1166   _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1167   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1168 
1169   assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1170 
1171   if (log_is_enabled(Info, os, thread)) {
1172     // See if we seem to be on primordial process thread
1173     bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1174                       uintptr_t(&rlim) < stack_top;
1175 
1176     log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1177                          SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1178                          primordial ? "primordial" : "user", max_size / K,  _initial_thread_stack_size / K,
1179                          stack_top, intptr_t(_initial_thread_stack_bottom));
1180   }
1181 }
1182 
1183 ////////////////////////////////////////////////////////////////////////////////
1184 // time support
1185 
1186 // Time since start-up in seconds to a fine granularity.
1187 // Used by VMSelfDestructTimer and the MemProfiler.
1188 double os::elapsedTime() {
1189 
1190   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1191 }
1192 
1193 jlong os::elapsed_counter() {
1194   return javaTimeNanos() - initial_time_count;
1195 }
1196 
1197 jlong os::elapsed_frequency() {
1198   return NANOSECS_PER_SEC; // nanosecond resolution
1199 }
1200 
1201 bool os::supports_vtime() { return true; }
1202 bool os::enable_vtime()   { return false; }
1203 bool os::vtime_enabled()  { return false; }
1204 
1205 double os::elapsedVTime() {
1206   struct rusage usage;
1207   int retval = getrusage(RUSAGE_THREAD, &usage);
1208   if (retval == 0) {
1209     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);
1210   } else {
1211     // better than nothing, but not much
1212     return elapsedTime();
1213   }
1214 }
1215 
1216 jlong os::javaTimeMillis() {
1217   timeval time;
1218   int status = gettimeofday(&time, NULL);
1219   assert(status != -1, "linux error");
1220   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1221 }
1222 
1223 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1224   timeval time;
1225   int status = gettimeofday(&time, NULL);
1226   assert(status != -1, "linux error");
1227   seconds = jlong(time.tv_sec);
1228   nanos = jlong(time.tv_usec) * 1000;
1229 }
1230 
1231 
1232 #ifndef CLOCK_MONOTONIC
1233   #define CLOCK_MONOTONIC (1)
1234 #endif
1235 
1236 void os::Linux::clock_init() {
1237   // we do dlopen's in this particular order due to bug in linux
1238   // dynamical loader (see 6348968) leading to crash on exit
1239   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1240   if (handle == NULL) {
1241     handle = dlopen("librt.so", RTLD_LAZY);
1242   }
1243 
1244   if (handle) {
1245     int (*clock_getres_func)(clockid_t, struct timespec*) =
1246            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1247     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1248            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1249     if (clock_getres_func && clock_gettime_func) {
1250       // See if monotonic clock is supported by the kernel. Note that some
1251       // early implementations simply return kernel jiffies (updated every
1252       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1253       // for nano time (though the monotonic property is still nice to have).
1254       // It's fixed in newer kernels, however clock_getres() still returns
1255       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1256       // resolution for now. Hopefully as people move to new kernels, this
1257       // won't be a problem.
1258       struct timespec res;
1259       struct timespec tp;
1260       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1261           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1262         // yes, monotonic clock is supported
1263         _clock_gettime = clock_gettime_func;
1264         return;
1265       } else {
1266         // close librt if there is no monotonic clock
1267         dlclose(handle);
1268       }
1269     }
1270   }
1271   warning("No monotonic clock was available - timed services may " \
1272           "be adversely affected if the time-of-day clock changes");
1273 }
1274 
1275 #ifndef SYS_clock_getres
1276   #if defined(X86) || defined(PPC64) || defined(S390)
1277     #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1278     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1279   #else
1280     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1281     #define sys_clock_getres(x,y)  -1
1282   #endif
1283 #else
1284   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1285 #endif
1286 
1287 void os::Linux::fast_thread_clock_init() {
1288   if (!UseLinuxPosixThreadCPUClocks) {
1289     return;
1290   }
1291   clockid_t clockid;
1292   struct timespec tp;
1293   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1294       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1295 
1296   // Switch to using fast clocks for thread cpu time if
1297   // the sys_clock_getres() returns 0 error code.
1298   // Note, that some kernels may support the current thread
1299   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1300   // returned by the pthread_getcpuclockid().
1301   // If the fast Posix clocks are supported then the sys_clock_getres()
1302   // must return at least tp.tv_sec == 0 which means a resolution
1303   // better than 1 sec. This is extra check for reliability.
1304 
1305   if (pthread_getcpuclockid_func &&
1306       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1307       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1308     _supports_fast_thread_cpu_time = true;
1309     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1310   }
1311 }
1312 
1313 jlong os::javaTimeNanos() {
1314   if (os::supports_monotonic_clock()) {
1315     struct timespec tp;
1316     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1317     assert(status == 0, "gettime error");
1318     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1319     return result;
1320   } else {
1321     timeval time;
1322     int status = gettimeofday(&time, NULL);
1323     assert(status != -1, "linux error");
1324     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1325     return 1000 * usecs;
1326   }
1327 }
1328 
1329 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1330   if (os::supports_monotonic_clock()) {
1331     info_ptr->max_value = ALL_64_BITS;
1332 
1333     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1334     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1335     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1336   } else {
1337     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1338     info_ptr->max_value = ALL_64_BITS;
1339 
1340     // gettimeofday is a real time clock so it skips
1341     info_ptr->may_skip_backward = true;
1342     info_ptr->may_skip_forward = true;
1343   }
1344 
1345   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1346 }
1347 
1348 // Return the real, user, and system times in seconds from an
1349 // arbitrary fixed point in the past.
1350 bool os::getTimesSecs(double* process_real_time,
1351                       double* process_user_time,
1352                       double* process_system_time) {
1353   struct tms ticks;
1354   clock_t real_ticks = times(&ticks);
1355 
1356   if (real_ticks == (clock_t) (-1)) {
1357     return false;
1358   } else {
1359     double ticks_per_second = (double) clock_tics_per_sec;
1360     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1361     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1362     *process_real_time = ((double) real_ticks) / ticks_per_second;
1363 
1364     return true;
1365   }
1366 }
1367 
1368 
1369 char * os::local_time_string(char *buf, size_t buflen) {
1370   struct tm t;
1371   time_t long_time;
1372   time(&long_time);
1373   localtime_r(&long_time, &t);
1374   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1375                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1376                t.tm_hour, t.tm_min, t.tm_sec);
1377   return buf;
1378 }
1379 
1380 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1381   return localtime_r(clock, res);
1382 }
1383 
1384 ////////////////////////////////////////////////////////////////////////////////
1385 // runtime exit support
1386 
1387 // Note: os::shutdown() might be called very early during initialization, or
1388 // called from signal handler. Before adding something to os::shutdown(), make
1389 // sure it is async-safe and can handle partially initialized VM.
1390 void os::shutdown() {
1391 
1392   // allow PerfMemory to attempt cleanup of any persistent resources
1393   perfMemory_exit();
1394 
1395   // needs to remove object in file system
1396   AttachListener::abort();
1397 
1398   // flush buffered output, finish log files
1399   ostream_abort();
1400 
1401   // Check for abort hook
1402   abort_hook_t abort_hook = Arguments::abort_hook();
1403   if (abort_hook != NULL) {
1404     abort_hook();
1405   }
1406 
1407 }
1408 
1409 // Note: os::abort() might be called very early during initialization, or
1410 // called from signal handler. Before adding something to os::abort(), make
1411 // sure it is async-safe and can handle partially initialized VM.
1412 void os::abort(bool dump_core, void* siginfo, const void* context) {
1413   os::shutdown();
1414   if (dump_core) {
1415 #ifndef PRODUCT
1416     fdStream out(defaultStream::output_fd());
1417     out.print_raw("Current thread is ");
1418     char buf[16];
1419     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1420     out.print_raw_cr(buf);
1421     out.print_raw_cr("Dumping core ...");
1422 #endif
1423     ::abort(); // dump core
1424   }
1425 
1426   ::exit(1);
1427 }
1428 
1429 // Die immediately, no exit hook, no abort hook, no cleanup.
1430 void os::die() {
1431   ::abort();
1432 }
1433 
1434 
1435 // This method is a copy of JDK's sysGetLastErrorString
1436 // from src/solaris/hpi/src/system_md.c
1437 
1438 size_t os::lasterror(char *buf, size_t len) {
1439   if (errno == 0)  return 0;
1440 
1441   const char *s = os::strerror(errno);
1442   size_t n = ::strlen(s);
1443   if (n >= len) {
1444     n = len - 1;
1445   }
1446   ::strncpy(buf, s, n);
1447   buf[n] = '\0';
1448   return n;
1449 }
1450 
1451 // thread_id is kernel thread id (similar to Solaris LWP id)
1452 intx os::current_thread_id() { return os::Linux::gettid(); }
1453 int os::current_process_id() {
1454   return ::getpid();
1455 }
1456 
1457 // DLL functions
1458 
1459 const char* os::dll_file_extension() { return ".so"; }
1460 
1461 // This must be hard coded because it's the system's temporary
1462 // directory not the java application's temp directory, ala java.io.tmpdir.
1463 const char* os::get_temp_directory() { return "/tmp"; }
1464 
1465 static bool file_exists(const char* filename) {
1466   struct stat statbuf;
1467   if (filename == NULL || strlen(filename) == 0) {
1468     return false;
1469   }
1470   return os::stat(filename, &statbuf) == 0;
1471 }
1472 
1473 // check if addr is inside libjvm.so
1474 bool os::address_is_in_vm(address addr) {
1475   static address libjvm_base_addr;
1476   Dl_info dlinfo;
1477 
1478   if (libjvm_base_addr == NULL) {
1479     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1480       libjvm_base_addr = (address)dlinfo.dli_fbase;
1481     }
1482     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1483   }
1484 
1485   if (dladdr((void *)addr, &dlinfo) != 0) {
1486     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1487   }
1488 
1489   return false;
1490 }
1491 
1492 bool os::dll_address_to_function_name(address addr, char *buf,
1493                                       int buflen, int *offset,
1494                                       bool demangle) {
1495   // buf is not optional, but offset is optional
1496   assert(buf != NULL, "sanity check");
1497 
1498   Dl_info dlinfo;
1499 
1500   if (dladdr((void*)addr, &dlinfo) != 0) {
1501     // see if we have a matching symbol
1502     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1503       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1504         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1505       }
1506       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1507       return true;
1508     }
1509     // no matching symbol so try for just file info
1510     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1511       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1512                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1513         return true;
1514       }
1515     }
1516   }
1517 
1518   buf[0] = '\0';
1519   if (offset != NULL) *offset = -1;
1520   return false;
1521 }
1522 
1523 struct _address_to_library_name {
1524   address addr;          // input : memory address
1525   size_t  buflen;        //         size of fname
1526   char*   fname;         // output: library name
1527   address base;          //         library base addr
1528 };
1529 
1530 static int address_to_library_name_callback(struct dl_phdr_info *info,
1531                                             size_t size, void *data) {
1532   int i;
1533   bool found = false;
1534   address libbase = NULL;
1535   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1536 
1537   // iterate through all loadable segments
1538   for (i = 0; i < info->dlpi_phnum; i++) {
1539     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1540     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1541       // base address of a library is the lowest address of its loaded
1542       // segments.
1543       if (libbase == NULL || libbase > segbase) {
1544         libbase = segbase;
1545       }
1546       // see if 'addr' is within current segment
1547       if (segbase <= d->addr &&
1548           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1549         found = true;
1550       }
1551     }
1552   }
1553 
1554   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1555   // so dll_address_to_library_name() can fall through to use dladdr() which
1556   // can figure out executable name from argv[0].
1557   if (found && info->dlpi_name && info->dlpi_name[0]) {
1558     d->base = libbase;
1559     if (d->fname) {
1560       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1561     }
1562     return 1;
1563   }
1564   return 0;
1565 }
1566 
1567 bool os::dll_address_to_library_name(address addr, char* buf,
1568                                      int buflen, int* offset) {
1569   // buf is not optional, but offset is optional
1570   assert(buf != NULL, "sanity check");
1571 
1572   Dl_info dlinfo;
1573   struct _address_to_library_name data;
1574 
1575   // There is a bug in old glibc dladdr() implementation that it could resolve
1576   // to wrong library name if the .so file has a base address != NULL. Here
1577   // we iterate through the program headers of all loaded libraries to find
1578   // out which library 'addr' really belongs to. This workaround can be
1579   // removed once the minimum requirement for glibc is moved to 2.3.x.
1580   data.addr = addr;
1581   data.fname = buf;
1582   data.buflen = buflen;
1583   data.base = NULL;
1584   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1585 
1586   if (rslt) {
1587     // buf already contains library name
1588     if (offset) *offset = addr - data.base;
1589     return true;
1590   }
1591   if (dladdr((void*)addr, &dlinfo) != 0) {
1592     if (dlinfo.dli_fname != NULL) {
1593       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1594     }
1595     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1596       *offset = addr - (address)dlinfo.dli_fbase;
1597     }
1598     return true;
1599   }
1600 
1601   buf[0] = '\0';
1602   if (offset) *offset = -1;
1603   return false;
1604 }
1605 
1606 // Loads .dll/.so and
1607 // in case of error it checks if .dll/.so was built for the
1608 // same architecture as Hotspot is running on
1609 
1610 
1611 // Remember the stack's state. The Linux dynamic linker will change
1612 // the stack to 'executable' at most once, so we must safepoint only once.
1613 bool os::Linux::_stack_is_executable = false;
1614 
1615 // VM operation that loads a library.  This is necessary if stack protection
1616 // of the Java stacks can be lost during loading the library.  If we
1617 // do not stop the Java threads, they can stack overflow before the stacks
1618 // are protected again.
1619 class VM_LinuxDllLoad: public VM_Operation {
1620  private:
1621   const char *_filename;
1622   char *_ebuf;
1623   int _ebuflen;
1624   void *_lib;
1625  public:
1626   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1627     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1628   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1629   void doit() {
1630     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1631     os::Linux::_stack_is_executable = true;
1632   }
1633   void* loaded_library() { return _lib; }
1634 };
1635 
1636 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1637   void * result = NULL;
1638   bool load_attempted = false;
1639 
1640   // Check whether the library to load might change execution rights
1641   // of the stack. If they are changed, the protection of the stack
1642   // guard pages will be lost. We need a safepoint to fix this.
1643   //
1644   // See Linux man page execstack(8) for more info.
1645   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1646     if (!ElfFile::specifies_noexecstack(filename)) {
1647       if (!is_init_completed()) {
1648         os::Linux::_stack_is_executable = true;
1649         // This is OK - No Java threads have been created yet, and hence no
1650         // stack guard pages to fix.
1651         //
1652         // This should happen only when you are building JDK7 using a very
1653         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1654         //
1655         // Dynamic loader will make all stacks executable after
1656         // this function returns, and will not do that again.
1657         assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1658       } else {
1659         warning("You have loaded library %s which might have disabled stack guard. "
1660                 "The VM will try to fix the stack guard now.\n"
1661                 "It's highly recommended that you fix the library with "
1662                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1663                 filename);
1664 
1665         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1666         JavaThread *jt = JavaThread::current();
1667         if (jt->thread_state() != _thread_in_native) {
1668           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1669           // that requires ExecStack. Cannot enter safe point. Let's give up.
1670           warning("Unable to fix stack guard. Giving up.");
1671         } else {
1672           if (!LoadExecStackDllInVMThread) {
1673             // This is for the case where the DLL has an static
1674             // constructor function that executes JNI code. We cannot
1675             // load such DLLs in the VMThread.
1676             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1677           }
1678 
1679           ThreadInVMfromNative tiv(jt);
1680           debug_only(VMNativeEntryWrapper vew;)
1681 
1682           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1683           VMThread::execute(&op);
1684           if (LoadExecStackDllInVMThread) {
1685             result = op.loaded_library();
1686           }
1687           load_attempted = true;
1688         }
1689       }
1690     }
1691   }
1692 
1693   if (!load_attempted) {
1694     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1695   }
1696 
1697   if (result != NULL) {
1698     // Successful loading
1699     return result;
1700   }
1701 
1702   Elf32_Ehdr elf_head;
1703   int diag_msg_max_length=ebuflen-strlen(ebuf);
1704   char* diag_msg_buf=ebuf+strlen(ebuf);
1705 
1706   if (diag_msg_max_length==0) {
1707     // No more space in ebuf for additional diagnostics message
1708     return NULL;
1709   }
1710 
1711 
1712   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1713 
1714   if (file_descriptor < 0) {
1715     // Can't open library, report dlerror() message
1716     return NULL;
1717   }
1718 
1719   bool failed_to_read_elf_head=
1720     (sizeof(elf_head)!=
1721      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1722 
1723   ::close(file_descriptor);
1724   if (failed_to_read_elf_head) {
1725     // file i/o error - report dlerror() msg
1726     return NULL;
1727   }
1728 
1729   typedef struct {
1730     Elf32_Half    code;         // Actual value as defined in elf.h
1731     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1732     unsigned char elf_class;    // 32 or 64 bit
1733     unsigned char endianess;    // MSB or LSB
1734     char*         name;         // String representation
1735   } arch_t;
1736 
1737 #ifndef EM_486
1738   #define EM_486          6               /* Intel 80486 */
1739 #endif
1740 #ifndef EM_AARCH64
1741   #define EM_AARCH64    183               /* ARM AARCH64 */
1742 #endif
1743 
1744   static const arch_t arch_array[]={
1745     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1746     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1747     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1748     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1749     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1750     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1751     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1752     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1753 #if defined(VM_LITTLE_ENDIAN)
1754     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1755     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1756 #else
1757     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1758     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1759 #endif
1760     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1761     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1762     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1763     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1764     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1765     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1766     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1767     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1768   };
1769 
1770 #if  (defined IA32)
1771   static  Elf32_Half running_arch_code=EM_386;
1772 #elif   (defined AMD64)
1773   static  Elf32_Half running_arch_code=EM_X86_64;
1774 #elif  (defined IA64)
1775   static  Elf32_Half running_arch_code=EM_IA_64;
1776 #elif  (defined __sparc) && (defined _LP64)
1777   static  Elf32_Half running_arch_code=EM_SPARCV9;
1778 #elif  (defined __sparc) && (!defined _LP64)
1779   static  Elf32_Half running_arch_code=EM_SPARC;
1780 #elif  (defined __powerpc64__)
1781   static  Elf32_Half running_arch_code=EM_PPC64;
1782 #elif  (defined __powerpc__)
1783   static  Elf32_Half running_arch_code=EM_PPC;
1784 #elif  (defined AARCH64)
1785   static  Elf32_Half running_arch_code=EM_AARCH64;
1786 #elif  (defined ARM)
1787   static  Elf32_Half running_arch_code=EM_ARM;
1788 #elif  (defined S390)
1789   static  Elf32_Half running_arch_code=EM_S390;
1790 #elif  (defined ALPHA)
1791   static  Elf32_Half running_arch_code=EM_ALPHA;
1792 #elif  (defined MIPSEL)
1793   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1794 #elif  (defined PARISC)
1795   static  Elf32_Half running_arch_code=EM_PARISC;
1796 #elif  (defined MIPS)
1797   static  Elf32_Half running_arch_code=EM_MIPS;
1798 #elif  (defined M68K)
1799   static  Elf32_Half running_arch_code=EM_68K;
1800 #elif  (defined SH)
1801   static  Elf32_Half running_arch_code=EM_SH;
1802 #else
1803     #error Method os::dll_load requires that one of following is defined:\
1804         AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1805 #endif
1806 
1807   // Identify compatability class for VM's architecture and library's architecture
1808   // Obtain string descriptions for architectures
1809 
1810   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1811   int running_arch_index=-1;
1812 
1813   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1814     if (running_arch_code == arch_array[i].code) {
1815       running_arch_index    = i;
1816     }
1817     if (lib_arch.code == arch_array[i].code) {
1818       lib_arch.compat_class = arch_array[i].compat_class;
1819       lib_arch.name         = arch_array[i].name;
1820     }
1821   }
1822 
1823   assert(running_arch_index != -1,
1824          "Didn't find running architecture code (running_arch_code) in arch_array");
1825   if (running_arch_index == -1) {
1826     // Even though running architecture detection failed
1827     // we may still continue with reporting dlerror() message
1828     return NULL;
1829   }
1830 
1831   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1832     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1833     return NULL;
1834   }
1835 
1836 #ifndef S390
1837   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1838     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1839     return NULL;
1840   }
1841 #endif // !S390
1842 
1843   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1844     if (lib_arch.name!=NULL) {
1845       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1846                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1847                  lib_arch.name, arch_array[running_arch_index].name);
1848     } else {
1849       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1850                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1851                  lib_arch.code,
1852                  arch_array[running_arch_index].name);
1853     }
1854   }
1855 
1856   return NULL;
1857 }
1858 
1859 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1860                                 int ebuflen) {
1861   void * result = ::dlopen(filename, RTLD_LAZY);
1862   if (result == NULL) {
1863     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1864     ebuf[ebuflen-1] = '\0';
1865   }
1866   return result;
1867 }
1868 
1869 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1870                                        int ebuflen) {
1871   void * result = NULL;
1872   if (LoadExecStackDllInVMThread) {
1873     result = dlopen_helper(filename, ebuf, ebuflen);
1874   }
1875 
1876   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1877   // library that requires an executable stack, or which does not have this
1878   // stack attribute set, dlopen changes the stack attribute to executable. The
1879   // read protection of the guard pages gets lost.
1880   //
1881   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1882   // may have been queued at the same time.
1883 
1884   if (!_stack_is_executable) {
1885     for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1886       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1887           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1888         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1889           warning("Attempt to reguard stack yellow zone failed.");
1890         }
1891       }
1892     }
1893   }
1894 
1895   return result;
1896 }
1897 
1898 void* os::dll_lookup(void* handle, const char* name) {
1899   void* res = dlsym(handle, name);
1900   return res;
1901 }
1902 
1903 void* os::get_default_process_handle() {
1904   return (void*)::dlopen(NULL, RTLD_LAZY);
1905 }
1906 
1907 static bool _print_ascii_file(const char* filename, outputStream* st) {
1908   int fd = ::open(filename, O_RDONLY);
1909   if (fd == -1) {
1910     return false;
1911   }
1912 
1913   char buf[33];
1914   int bytes;
1915   buf[32] = '\0';
1916   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1917     st->print_raw(buf, bytes);
1918   }
1919 
1920   ::close(fd);
1921 
1922   return true;
1923 }
1924 
1925 void os::print_dll_info(outputStream *st) {
1926   st->print_cr("Dynamic libraries:");
1927 
1928   char fname[32];
1929   pid_t pid = os::Linux::gettid();
1930 
1931   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1932 
1933   if (!_print_ascii_file(fname, st)) {
1934     st->print("Can not get library information for pid = %d\n", pid);
1935   }
1936 }
1937 
1938 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1939   FILE *procmapsFile = NULL;
1940 
1941   // Open the procfs maps file for the current process
1942   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1943     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1944     char line[PATH_MAX + 100];
1945 
1946     // Read line by line from 'file'
1947     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1948       u8 base, top, offset, inode;
1949       char permissions[5];
1950       char device[6];
1951       char name[PATH_MAX + 1];
1952 
1953       // Parse fields from line
1954       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1955              &base, &top, permissions, &offset, device, &inode, name);
1956 
1957       // Filter by device id '00:00' so that we only get file system mapped files.
1958       if (strcmp(device, "00:00") != 0) {
1959 
1960         // Call callback with the fields of interest
1961         if(callback(name, (address)base, (address)top, param)) {
1962           // Oops abort, callback aborted
1963           fclose(procmapsFile);
1964           return 1;
1965         }
1966       }
1967     }
1968     fclose(procmapsFile);
1969   }
1970   return 0;
1971 }
1972 
1973 void os::print_os_info_brief(outputStream* st) {
1974   os::Linux::print_distro_info(st);
1975 
1976   os::Posix::print_uname_info(st);
1977 
1978   os::Linux::print_libversion_info(st);
1979 
1980 }
1981 
1982 void os::print_os_info(outputStream* st) {
1983   st->print("OS:");
1984 
1985   os::Linux::print_distro_info(st);
1986 
1987   os::Posix::print_uname_info(st);
1988 
1989   // Print warning if unsafe chroot environment detected
1990   if (unsafe_chroot_detected) {
1991     st->print("WARNING!! ");
1992     st->print_cr("%s", unstable_chroot_error);
1993   }
1994 
1995   os::Linux::print_libversion_info(st);
1996 
1997   os::Posix::print_rlimit_info(st);
1998 
1999   os::Posix::print_load_average(st);
2000 
2001   os::Linux::print_full_memory_info(st);
2002 
2003   os::Linux::print_container_info(st);
2004 }
2005 
2006 // Try to identify popular distros.
2007 // Most Linux distributions have a /etc/XXX-release file, which contains
2008 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2009 // file that also contains the OS version string. Some have more than one
2010 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2011 // /etc/redhat-release.), so the order is important.
2012 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2013 // their own specific XXX-release file as well as a redhat-release file.
2014 // Because of this the XXX-release file needs to be searched for before the
2015 // redhat-release file.
2016 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2017 // search for redhat-release / SuSE-release needs to be before lsb-release.
2018 // Since the lsb-release file is the new standard it needs to be searched
2019 // before the older style release files.
2020 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2021 // next to last resort.  The os-release file is a new standard that contains
2022 // distribution information and the system-release file seems to be an old
2023 // standard that has been replaced by the lsb-release and os-release files.
2024 // Searching for the debian_version file is the last resort.  It contains
2025 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2026 // "Debian " is printed before the contents of the debian_version file.
2027 
2028 const char* distro_files[] = {
2029   "/etc/oracle-release",
2030   "/etc/mandriva-release",
2031   "/etc/mandrake-release",
2032   "/etc/sun-release",
2033   "/etc/redhat-release",
2034   "/etc/SuSE-release",
2035   "/etc/lsb-release",
2036   "/etc/turbolinux-release",
2037   "/etc/gentoo-release",
2038   "/etc/ltib-release",
2039   "/etc/angstrom-version",
2040   "/etc/system-release",
2041   "/etc/os-release",
2042   NULL };
2043 
2044 void os::Linux::print_distro_info(outputStream* st) {
2045   for (int i = 0;; i++) {
2046     const char* file = distro_files[i];
2047     if (file == NULL) {
2048       break;  // done
2049     }
2050     // If file prints, we found it.
2051     if (_print_ascii_file(file, st)) {
2052       return;
2053     }
2054   }
2055 
2056   if (file_exists("/etc/debian_version")) {
2057     st->print("Debian ");
2058     _print_ascii_file("/etc/debian_version", st);
2059   } else {
2060     st->print("Linux");
2061   }
2062   st->cr();
2063 }
2064 
2065 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2066   char buf[256];
2067   while (fgets(buf, sizeof(buf), fp)) {
2068     // Edit out extra stuff in expected format
2069     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2070       char* ptr = strstr(buf, "\"");  // the name is in quotes
2071       if (ptr != NULL) {
2072         ptr++; // go beyond first quote
2073         char* nl = strchr(ptr, '\"');
2074         if (nl != NULL) *nl = '\0';
2075         strncpy(distro, ptr, length);
2076       } else {
2077         ptr = strstr(buf, "=");
2078         ptr++; // go beyond equals then
2079         char* nl = strchr(ptr, '\n');
2080         if (nl != NULL) *nl = '\0';
2081         strncpy(distro, ptr, length);
2082       }
2083       return;
2084     } else if (get_first_line) {
2085       char* nl = strchr(buf, '\n');
2086       if (nl != NULL) *nl = '\0';
2087       strncpy(distro, buf, length);
2088       return;
2089     }
2090   }
2091   // print last line and close
2092   char* nl = strchr(buf, '\n');
2093   if (nl != NULL) *nl = '\0';
2094   strncpy(distro, buf, length);
2095 }
2096 
2097 static void parse_os_info(char* distro, size_t length, const char* file) {
2098   FILE* fp = fopen(file, "r");
2099   if (fp != NULL) {
2100     // if suse format, print out first line
2101     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2102     parse_os_info_helper(fp, distro, length, get_first_line);
2103     fclose(fp);
2104   }
2105 }
2106 
2107 void os::get_summary_os_info(char* buf, size_t buflen) {
2108   for (int i = 0;; i++) {
2109     const char* file = distro_files[i];
2110     if (file == NULL) {
2111       break; // ran out of distro_files
2112     }
2113     if (file_exists(file)) {
2114       parse_os_info(buf, buflen, file);
2115       return;
2116     }
2117   }
2118   // special case for debian
2119   if (file_exists("/etc/debian_version")) {
2120     strncpy(buf, "Debian ", buflen);
2121     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2122   } else {
2123     strncpy(buf, "Linux", buflen);
2124   }
2125 }
2126 
2127 void os::Linux::print_libversion_info(outputStream* st) {
2128   // libc, pthread
2129   st->print("libc:");
2130   st->print("%s ", os::Linux::glibc_version());
2131   st->print("%s ", os::Linux::libpthread_version());
2132   st->cr();
2133 }
2134 
2135 void os::Linux::print_full_memory_info(outputStream* st) {
2136   st->print("\n/proc/meminfo:\n");
2137   _print_ascii_file("/proc/meminfo", st);
2138   st->cr();
2139 }
2140 
2141 void os::Linux::print_container_info(outputStream* st) {
2142   if (!OSContainer::is_containerized()) {
2143     return;
2144   }
2145 
2146   st->print("container (cgroup) information:\n");
2147 
2148   const char *p_ct = OSContainer::container_type();
2149   st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2150 
2151   char *p = OSContainer::cpu_cpuset_cpus();
2152   st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2153   free(p);
2154 
2155   p = OSContainer::cpu_cpuset_memory_nodes();
2156   st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2157   free(p);
2158 
2159   int i = OSContainer::active_processor_count();
2160   if (i > 0) {
2161     st->print("active_processor_count: %d\n", i);
2162   } else {
2163     st->print("active_processor_count: failed\n");
2164   }
2165 
2166   i = OSContainer::cpu_quota();
2167   st->print("cpu_quota: %d\n", i);
2168 
2169   i = OSContainer::cpu_period();
2170   st->print("cpu_period: %d\n", i);
2171 
2172   i = OSContainer::cpu_shares();
2173   st->print("cpu_shares: %d\n", i);
2174 
2175   jlong j = OSContainer::memory_limit_in_bytes();
2176   st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2177 
2178   j = OSContainer::memory_and_swap_limit_in_bytes();
2179   st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2180 
2181   j = OSContainer::memory_soft_limit_in_bytes();
2182   st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2183 
2184   j = OSContainer::OSContainer::memory_usage_in_bytes();
2185   st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2186 
2187   j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2188   st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2189   st->cr();
2190 }
2191 
2192 void os::print_memory_info(outputStream* st) {
2193 
2194   st->print("Memory:");
2195   st->print(" %dk page", os::vm_page_size()>>10);
2196 
2197   // values in struct sysinfo are "unsigned long"
2198   struct sysinfo si;
2199   sysinfo(&si);
2200 
2201   st->print(", physical " UINT64_FORMAT "k",
2202             os::physical_memory() >> 10);
2203   st->print("(" UINT64_FORMAT "k free)",
2204             os::available_memory() >> 10);
2205   st->print(", swap " UINT64_FORMAT "k",
2206             ((jlong)si.totalswap * si.mem_unit) >> 10);
2207   st->print("(" UINT64_FORMAT "k free)",
2208             ((jlong)si.freeswap * si.mem_unit) >> 10);
2209   st->cr();
2210 }
2211 
2212 // Print the first "model name" line and the first "flags" line
2213 // that we find and nothing more. We assume "model name" comes
2214 // before "flags" so if we find a second "model name", then the
2215 // "flags" field is considered missing.
2216 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2217 #if defined(IA32) || defined(AMD64)
2218   // Other platforms have less repetitive cpuinfo files
2219   FILE *fp = fopen("/proc/cpuinfo", "r");
2220   if (fp) {
2221     while (!feof(fp)) {
2222       if (fgets(buf, buflen, fp)) {
2223         // Assume model name comes before flags
2224         bool model_name_printed = false;
2225         if (strstr(buf, "model name") != NULL) {
2226           if (!model_name_printed) {
2227             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2228             st->print_raw(buf);
2229             model_name_printed = true;
2230           } else {
2231             // model name printed but not flags?  Odd, just return
2232             fclose(fp);
2233             return true;
2234           }
2235         }
2236         // print the flags line too
2237         if (strstr(buf, "flags") != NULL) {
2238           st->print_raw(buf);
2239           fclose(fp);
2240           return true;
2241         }
2242       }
2243     }
2244     fclose(fp);
2245   }
2246 #endif // x86 platforms
2247   return false;
2248 }
2249 
2250 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2251   // Only print the model name if the platform provides this as a summary
2252   if (!print_model_name_and_flags(st, buf, buflen)) {
2253     st->print("\n/proc/cpuinfo:\n");
2254     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2255       st->print_cr("  <Not Available>");
2256     }
2257   }
2258 }
2259 
2260 #if defined(AMD64) || defined(IA32) || defined(X32)
2261 const char* search_string = "model name";
2262 #elif defined(M68K)
2263 const char* search_string = "CPU";
2264 #elif defined(PPC64)
2265 const char* search_string = "cpu";
2266 #elif defined(S390)
2267 const char* search_string = "processor";
2268 #elif defined(SPARC)
2269 const char* search_string = "cpu";
2270 #else
2271 const char* search_string = "Processor";
2272 #endif
2273 
2274 // Parses the cpuinfo file for string representing the model name.
2275 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2276   FILE* fp = fopen("/proc/cpuinfo", "r");
2277   if (fp != NULL) {
2278     while (!feof(fp)) {
2279       char buf[256];
2280       if (fgets(buf, sizeof(buf), fp)) {
2281         char* start = strstr(buf, search_string);
2282         if (start != NULL) {
2283           char *ptr = start + strlen(search_string);
2284           char *end = buf + strlen(buf);
2285           while (ptr != end) {
2286              // skip whitespace and colon for the rest of the name.
2287              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2288                break;
2289              }
2290              ptr++;
2291           }
2292           if (ptr != end) {
2293             // reasonable string, get rid of newline and keep the rest
2294             char* nl = strchr(buf, '\n');
2295             if (nl != NULL) *nl = '\0';
2296             strncpy(cpuinfo, ptr, length);
2297             fclose(fp);
2298             return;
2299           }
2300         }
2301       }
2302     }
2303     fclose(fp);
2304   }
2305   // cpuinfo not found or parsing failed, just print generic string.  The entire
2306   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2307 #if   defined(AARCH64)
2308   strncpy(cpuinfo, "AArch64", length);
2309 #elif defined(AMD64)
2310   strncpy(cpuinfo, "x86_64", length);
2311 #elif defined(ARM)  // Order wrt. AARCH64 is relevant!
2312   strncpy(cpuinfo, "ARM", length);
2313 #elif defined(IA32)
2314   strncpy(cpuinfo, "x86_32", length);
2315 #elif defined(IA64)
2316   strncpy(cpuinfo, "IA64", length);
2317 #elif defined(PPC)
2318   strncpy(cpuinfo, "PPC64", length);
2319 #elif defined(S390)
2320   strncpy(cpuinfo, "S390", length);
2321 #elif defined(SPARC)
2322   strncpy(cpuinfo, "sparcv9", length);
2323 #elif defined(ZERO_LIBARCH)
2324   strncpy(cpuinfo, ZERO_LIBARCH, length);
2325 #else
2326   strncpy(cpuinfo, "unknown", length);
2327 #endif
2328 }
2329 
2330 static void print_signal_handler(outputStream* st, int sig,
2331                                  char* buf, size_t buflen);
2332 
2333 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2334   st->print_cr("Signal Handlers:");
2335   print_signal_handler(st, SIGSEGV, buf, buflen);
2336   print_signal_handler(st, SIGBUS , buf, buflen);
2337   print_signal_handler(st, SIGFPE , buf, buflen);
2338   print_signal_handler(st, SIGPIPE, buf, buflen);
2339   print_signal_handler(st, SIGXFSZ, buf, buflen);
2340   print_signal_handler(st, SIGILL , buf, buflen);
2341   print_signal_handler(st, SR_signum, buf, buflen);
2342   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2343   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2344   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2345   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2346 #if defined(PPC64)
2347   print_signal_handler(st, SIGTRAP, buf, buflen);
2348 #endif
2349 }
2350 
2351 static char saved_jvm_path[MAXPATHLEN] = {0};
2352 
2353 // Find the full path to the current module, libjvm.so
2354 void os::jvm_path(char *buf, jint buflen) {
2355   // Error checking.
2356   if (buflen < MAXPATHLEN) {
2357     assert(false, "must use a large-enough buffer");
2358     buf[0] = '\0';
2359     return;
2360   }
2361   // Lazy resolve the path to current module.
2362   if (saved_jvm_path[0] != 0) {
2363     strcpy(buf, saved_jvm_path);
2364     return;
2365   }
2366 
2367   char dli_fname[MAXPATHLEN];
2368   bool ret = dll_address_to_library_name(
2369                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2370                                          dli_fname, sizeof(dli_fname), NULL);
2371   assert(ret, "cannot locate libjvm");
2372   char *rp = NULL;
2373   if (ret && dli_fname[0] != '\0') {
2374     rp = os::Posix::realpath(dli_fname, buf, buflen);
2375   }
2376   if (rp == NULL) {
2377     return;
2378   }
2379 
2380   if (Arguments::sun_java_launcher_is_altjvm()) {
2381     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2382     // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2383     // If "/jre/lib/" appears at the right place in the string, then
2384     // assume we are installed in a JDK and we're done. Otherwise, check
2385     // for a JAVA_HOME environment variable and fix up the path so it
2386     // looks like libjvm.so is installed there (append a fake suffix
2387     // hotspot/libjvm.so).
2388     const char *p = buf + strlen(buf) - 1;
2389     for (int count = 0; p > buf && count < 5; ++count) {
2390       for (--p; p > buf && *p != '/'; --p)
2391         /* empty */ ;
2392     }
2393 
2394     if (strncmp(p, "/jre/lib/", 9) != 0) {
2395       // Look for JAVA_HOME in the environment.
2396       char* java_home_var = ::getenv("JAVA_HOME");
2397       if (java_home_var != NULL && java_home_var[0] != 0) {
2398         char* jrelib_p;
2399         int len;
2400 
2401         // Check the current module name "libjvm.so".
2402         p = strrchr(buf, '/');
2403         if (p == NULL) {
2404           return;
2405         }
2406         assert(strstr(p, "/libjvm") == p, "invalid library name");
2407 
2408         rp = os::Posix::realpath(java_home_var, buf, buflen);
2409         if (rp == NULL) {
2410           return;
2411         }
2412 
2413         // determine if this is a legacy image or modules image
2414         // modules image doesn't have "jre" subdirectory
2415         len = strlen(buf);
2416         assert(len < buflen, "Ran out of buffer room");
2417         jrelib_p = buf + len;
2418         snprintf(jrelib_p, buflen-len, "/jre/lib");
2419         if (0 != access(buf, F_OK)) {
2420           snprintf(jrelib_p, buflen-len, "/lib");
2421         }
2422 
2423         if (0 == access(buf, F_OK)) {
2424           // Use current module name "libjvm.so"
2425           len = strlen(buf);
2426           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2427         } else {
2428           // Go back to path of .so
2429           rp = os::Posix::realpath(dli_fname, buf, buflen);
2430           if (rp == NULL) {
2431             return;
2432           }
2433         }
2434       }
2435     }
2436   }
2437 
2438   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2439   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2440 }
2441 
2442 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2443   // no prefix required, not even "_"
2444 }
2445 
2446 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2447   // no suffix required
2448 }
2449 
2450 ////////////////////////////////////////////////////////////////////////////////
2451 // sun.misc.Signal support
2452 
2453 static volatile jint sigint_count = 0;
2454 
2455 static void UserHandler(int sig, void *siginfo, void *context) {
2456   // 4511530 - sem_post is serialized and handled by the manager thread. When
2457   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2458   // don't want to flood the manager thread with sem_post requests.
2459   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2460     return;
2461   }
2462 
2463   // Ctrl-C is pressed during error reporting, likely because the error
2464   // handler fails to abort. Let VM die immediately.
2465   if (sig == SIGINT && VMError::is_error_reported()) {
2466     os::die();
2467   }
2468 
2469   os::signal_notify(sig);
2470 }
2471 
2472 void* os::user_handler() {
2473   return CAST_FROM_FN_PTR(void*, UserHandler);
2474 }
2475 
2476 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2477   struct timespec ts;
2478   // Semaphore's are always associated with CLOCK_REALTIME
2479   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2480   // see os_posix.cpp for discussion on overflow checking
2481   if (sec >= MAX_SECS) {
2482     ts.tv_sec += MAX_SECS;
2483     ts.tv_nsec = 0;
2484   } else {
2485     ts.tv_sec += sec;
2486     ts.tv_nsec += nsec;
2487     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2488       ts.tv_nsec -= NANOSECS_PER_SEC;
2489       ++ts.tv_sec; // note: this must be <= max_secs
2490     }
2491   }
2492 
2493   return ts;
2494 }
2495 
2496 extern "C" {
2497   typedef void (*sa_handler_t)(int);
2498   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2499 }
2500 
2501 void* os::signal(int signal_number, void* handler) {
2502   struct sigaction sigAct, oldSigAct;
2503 
2504   sigfillset(&(sigAct.sa_mask));
2505   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2506   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2507 
2508   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2509     // -1 means registration failed
2510     return (void *)-1;
2511   }
2512 
2513   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2514 }
2515 
2516 void os::signal_raise(int signal_number) {
2517   ::raise(signal_number);
2518 }
2519 
2520 // The following code is moved from os.cpp for making this
2521 // code platform specific, which it is by its very nature.
2522 
2523 // Will be modified when max signal is changed to be dynamic
2524 int os::sigexitnum_pd() {
2525   return NSIG;
2526 }
2527 
2528 // a counter for each possible signal value
2529 static volatile jint pending_signals[NSIG+1] = { 0 };
2530 
2531 // Linux(POSIX) specific hand shaking semaphore.
2532 static Semaphore* sig_sem = NULL;
2533 static PosixSemaphore sr_semaphore;
2534 
2535 static void jdk_misc_signal_init() {
2536   // Initialize signal structures
2537   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2538 
2539   // Initialize signal semaphore
2540   sig_sem = new Semaphore();
2541 }
2542 
2543 void os::signal_notify(int sig) {
2544   if (sig_sem != NULL) {
2545     Atomic::inc(&pending_signals[sig]);
2546     sig_sem->signal();
2547   } else {
2548     // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2549     // initialization isn't called.
2550     assert(ReduceSignalUsage, "signal semaphore should be created");
2551   }
2552 }
2553 
2554 static int check_pending_signals() {
2555   Atomic::store(0, &sigint_count);
2556   for (;;) {
2557     for (int i = 0; i < NSIG + 1; i++) {
2558       jint n = pending_signals[i];
2559       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2560         return i;
2561       }
2562     }
2563     JavaThread *thread = JavaThread::current();
2564     ThreadBlockInVM tbivm(thread);
2565 
2566     bool threadIsSuspended;
2567     do {
2568       thread->set_suspend_equivalent();
2569       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2570       sig_sem->wait();
2571 
2572       // were we externally suspended while we were waiting?
2573       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2574       if (threadIsSuspended) {
2575         // The semaphore has been incremented, but while we were waiting
2576         // another thread suspended us. We don't want to continue running
2577         // while suspended because that would surprise the thread that
2578         // suspended us.
2579         sig_sem->signal();
2580 
2581         thread->java_suspend_self();
2582       }
2583     } while (threadIsSuspended);
2584   }
2585 }
2586 
2587 int os::signal_wait() {
2588   return check_pending_signals();
2589 }
2590 
2591 ////////////////////////////////////////////////////////////////////////////////
2592 // Virtual Memory
2593 
2594 int os::vm_page_size() {
2595   // Seems redundant as all get out
2596   assert(os::Linux::page_size() != -1, "must call os::init");
2597   return os::Linux::page_size();
2598 }
2599 
2600 // Solaris allocates memory by pages.
2601 int os::vm_allocation_granularity() {
2602   assert(os::Linux::page_size() != -1, "must call os::init");
2603   return os::Linux::page_size();
2604 }
2605 
2606 // Rationale behind this function:
2607 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2608 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2609 //  samples for JITted code. Here we create private executable mapping over the code cache
2610 //  and then we can use standard (well, almost, as mapping can change) way to provide
2611 //  info for the reporting script by storing timestamp and location of symbol
2612 void linux_wrap_code(char* base, size_t size) {
2613   static volatile jint cnt = 0;
2614 
2615   if (!UseOprofile) {
2616     return;
2617   }
2618 
2619   char buf[PATH_MAX+1];
2620   int num = Atomic::add(1, &cnt);
2621 
2622   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2623            os::get_temp_directory(), os::current_process_id(), num);
2624   unlink(buf);
2625 
2626   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2627 
2628   if (fd != -1) {
2629     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2630     if (rv != (off_t)-1) {
2631       if (::write(fd, "", 1) == 1) {
2632         mmap(base, size,
2633              PROT_READ|PROT_WRITE|PROT_EXEC,
2634              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2635       }
2636     }
2637     ::close(fd);
2638     unlink(buf);
2639   }
2640 }
2641 
2642 static bool recoverable_mmap_error(int err) {
2643   // See if the error is one we can let the caller handle. This
2644   // list of errno values comes from JBS-6843484. I can't find a
2645   // Linux man page that documents this specific set of errno
2646   // values so while this list currently matches Solaris, it may
2647   // change as we gain experience with this failure mode.
2648   switch (err) {
2649   case EBADF:
2650   case EINVAL:
2651   case ENOTSUP:
2652     // let the caller deal with these errors
2653     return true;
2654 
2655   default:
2656     // Any remaining errors on this OS can cause our reserved mapping
2657     // to be lost. That can cause confusion where different data
2658     // structures think they have the same memory mapped. The worst
2659     // scenario is if both the VM and a library think they have the
2660     // same memory mapped.
2661     return false;
2662   }
2663 }
2664 
2665 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2666                                     int err) {
2667   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2668           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2669           os::strerror(err), err);
2670 }
2671 
2672 static void warn_fail_commit_memory(char* addr, size_t size,
2673                                     size_t alignment_hint, bool exec,
2674                                     int err) {
2675   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2676           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2677           alignment_hint, exec, os::strerror(err), err);
2678 }
2679 
2680 // NOTE: Linux kernel does not really reserve the pages for us.
2681 //       All it does is to check if there are enough free pages
2682 //       left at the time of mmap(). This could be a potential
2683 //       problem.
2684 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2685   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2686   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2687                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2688   if (res != (uintptr_t) MAP_FAILED) {
2689     if (UseNUMAInterleaving) {
2690       numa_make_global(addr, size);
2691     }
2692     return 0;
2693   }
2694 
2695   int err = errno;  // save errno from mmap() call above
2696 
2697   if (!recoverable_mmap_error(err)) {
2698     warn_fail_commit_memory(addr, size, exec, err);
2699     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2700   }
2701 
2702   return err;
2703 }
2704 
2705 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2706   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2707 }
2708 
2709 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2710                                   const char* mesg) {
2711   assert(mesg != NULL, "mesg must be specified");
2712   int err = os::Linux::commit_memory_impl(addr, size, exec);
2713   if (err != 0) {
2714     // the caller wants all commit errors to exit with the specified mesg:
2715     warn_fail_commit_memory(addr, size, exec, err);
2716     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2717   }
2718 }
2719 
2720 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2721 #ifndef MAP_HUGETLB
2722   #define MAP_HUGETLB 0x40000
2723 #endif
2724 
2725 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2726 #ifndef MADV_HUGEPAGE
2727   #define MADV_HUGEPAGE 14
2728 #endif
2729 
2730 int os::Linux::commit_memory_impl(char* addr, size_t size,
2731                                   size_t alignment_hint, bool exec) {
2732   int err = os::Linux::commit_memory_impl(addr, size, exec);
2733   if (err == 0) {
2734     realign_memory(addr, size, alignment_hint);
2735   }
2736   return err;
2737 }
2738 
2739 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2740                           bool exec) {
2741   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2742 }
2743 
2744 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2745                                   size_t alignment_hint, bool exec,
2746                                   const char* mesg) {
2747   assert(mesg != NULL, "mesg must be specified");
2748   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2749   if (err != 0) {
2750     // the caller wants all commit errors to exit with the specified mesg:
2751     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2752     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2753   }
2754 }
2755 
2756 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2757   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2758     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2759     // be supported or the memory may already be backed by huge pages.
2760     ::madvise(addr, bytes, MADV_HUGEPAGE);
2761   }
2762 }
2763 
2764 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2765   // This method works by doing an mmap over an existing mmaping and effectively discarding
2766   // the existing pages. However it won't work for SHM-based large pages that cannot be
2767   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2768   // small pages on top of the SHM segment. This method always works for small pages, so we
2769   // allow that in any case.
2770   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2771     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2772   }
2773 }
2774 
2775 void os::numa_make_global(char *addr, size_t bytes) {
2776   Linux::numa_interleave_memory(addr, bytes);
2777 }
2778 
2779 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2780 // bind policy to MPOL_PREFERRED for the current thread.
2781 #define USE_MPOL_PREFERRED 0
2782 
2783 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2784   // To make NUMA and large pages more robust when both enabled, we need to ease
2785   // the requirements on where the memory should be allocated. MPOL_BIND is the
2786   // default policy and it will force memory to be allocated on the specified
2787   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2788   // the specified node, but will not force it. Using this policy will prevent
2789   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2790   // free large pages.
2791   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2792   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2793 }
2794 
2795 bool os::numa_topology_changed() { return false; }
2796 
2797 size_t os::numa_get_groups_num() {
2798   // Return just the number of nodes in which it's possible to allocate memory
2799   // (in numa terminology, configured nodes).
2800   return Linux::numa_num_configured_nodes();
2801 }
2802 
2803 int os::numa_get_group_id() {
2804   int cpu_id = Linux::sched_getcpu();
2805   if (cpu_id != -1) {
2806     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2807     if (lgrp_id != -1) {
2808       return lgrp_id;
2809     }
2810   }
2811   return 0;
2812 }
2813 
2814 int os::Linux::get_existing_num_nodes() {
2815   size_t node;
2816   size_t highest_node_number = Linux::numa_max_node();
2817   int num_nodes = 0;
2818 
2819   // Get the total number of nodes in the system including nodes without memory.
2820   for (node = 0; node <= highest_node_number; node++) {
2821     if (isnode_in_existing_nodes(node)) {
2822       num_nodes++;
2823     }
2824   }
2825   return num_nodes;
2826 }
2827 
2828 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2829   size_t highest_node_number = Linux::numa_max_node();
2830   size_t i = 0;
2831 
2832   // Map all node ids in which is possible to allocate memory. Also nodes are
2833   // not always consecutively available, i.e. available from 0 to the highest
2834   // node number.
2835   for (size_t node = 0; node <= highest_node_number; node++) {
2836     if (Linux::isnode_in_configured_nodes(node)) {
2837       ids[i++] = node;
2838     }
2839   }
2840   return i;
2841 }
2842 
2843 bool os::get_page_info(char *start, page_info* info) {
2844   return false;
2845 }
2846 
2847 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2848                      page_info* page_found) {
2849   return end;
2850 }
2851 
2852 
2853 int os::Linux::sched_getcpu_syscall(void) {
2854   unsigned int cpu = 0;
2855   int retval = -1;
2856 
2857 #if defined(IA32)
2858   #ifndef SYS_getcpu
2859     #define SYS_getcpu 318
2860   #endif
2861   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2862 #elif defined(AMD64)
2863 // Unfortunately we have to bring all these macros here from vsyscall.h
2864 // to be able to compile on old linuxes.
2865   #define __NR_vgetcpu 2
2866   #define VSYSCALL_START (-10UL << 20)
2867   #define VSYSCALL_SIZE 1024
2868   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2869   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2870   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2871   retval = vgetcpu(&cpu, NULL, NULL);
2872 #endif
2873 
2874   return (retval == -1) ? retval : cpu;
2875 }
2876 
2877 void os::Linux::sched_getcpu_init() {
2878   // sched_getcpu() should be in libc.
2879   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2880                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2881 
2882   // If it's not, try a direct syscall.
2883   if (sched_getcpu() == -1) {
2884     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2885                                     (void*)&sched_getcpu_syscall));
2886   }
2887 }
2888 
2889 // Something to do with the numa-aware allocator needs these symbols
2890 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2891 extern "C" JNIEXPORT void numa_error(char *where) { }
2892 
2893 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2894 // load symbol from base version instead.
2895 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2896   void *f = dlvsym(handle, name, "libnuma_1.1");
2897   if (f == NULL) {
2898     f = dlsym(handle, name);
2899   }
2900   return f;
2901 }
2902 
2903 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2904 // Return NULL if the symbol is not defined in this particular version.
2905 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2906   return dlvsym(handle, name, "libnuma_1.2");
2907 }
2908 
2909 bool os::Linux::libnuma_init() {
2910   if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2911     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2912     if (handle != NULL) {
2913       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2914                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2915       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2916                                        libnuma_dlsym(handle, "numa_max_node")));
2917       set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2918                                                    libnuma_dlsym(handle, "numa_num_configured_nodes")));
2919       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2920                                         libnuma_dlsym(handle, "numa_available")));
2921       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2922                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2923       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2924                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2925       set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2926                                                 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2927       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2928                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2929       set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2930                                                libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2931       set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2932                                        libnuma_dlsym(handle, "numa_distance")));
2933 
2934       if (numa_available() != -1) {
2935         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2936         set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2937         set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2938         // Create an index -> node mapping, since nodes are not always consecutive
2939         _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2940         rebuild_nindex_to_node_map();
2941         // Create a cpu -> node mapping
2942         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2943         rebuild_cpu_to_node_map();
2944         return true;
2945       }
2946     }
2947   }
2948   return false;
2949 }
2950 
2951 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2952   // Creating guard page is very expensive. Java thread has HotSpot
2953   // guard pages, only enable glibc guard page for non-Java threads.
2954   // (Remember: compiler thread is a Java thread, too!)
2955   return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2956 }
2957 
2958 void os::Linux::rebuild_nindex_to_node_map() {
2959   int highest_node_number = Linux::numa_max_node();
2960 
2961   nindex_to_node()->clear();
2962   for (int node = 0; node <= highest_node_number; node++) {
2963     if (Linux::isnode_in_existing_nodes(node)) {
2964       nindex_to_node()->append(node);
2965     }
2966   }
2967 }
2968 
2969 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2970 // The table is later used in get_node_by_cpu().
2971 void os::Linux::rebuild_cpu_to_node_map() {
2972   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2973                               // in libnuma (possible values are starting from 16,
2974                               // and continuing up with every other power of 2, but less
2975                               // than the maximum number of CPUs supported by kernel), and
2976                               // is a subject to change (in libnuma version 2 the requirements
2977                               // are more reasonable) we'll just hardcode the number they use
2978                               // in the library.
2979   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2980 
2981   size_t cpu_num = processor_count();
2982   size_t cpu_map_size = NCPUS / BitsPerCLong;
2983   size_t cpu_map_valid_size =
2984     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2985 
2986   cpu_to_node()->clear();
2987   cpu_to_node()->at_grow(cpu_num - 1);
2988 
2989   size_t node_num = get_existing_num_nodes();
2990 
2991   int distance = 0;
2992   int closest_distance = INT_MAX;
2993   int closest_node = 0;
2994   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2995   for (size_t i = 0; i < node_num; i++) {
2996     // Check if node is configured (not a memory-less node). If it is not, find
2997     // the closest configured node.
2998     if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
2999       closest_distance = INT_MAX;
3000       // Check distance from all remaining nodes in the system. Ignore distance
3001       // from itself and from another non-configured node.
3002       for (size_t m = 0; m < node_num; m++) {
3003         if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
3004           distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3005           // If a closest node is found, update. There is always at least one
3006           // configured node in the system so there is always at least one node
3007           // close.
3008           if (distance != 0 && distance < closest_distance) {
3009             closest_distance = distance;
3010             closest_node = nindex_to_node()->at(m);
3011           }
3012         }
3013       }
3014      } else {
3015        // Current node is already a configured node.
3016        closest_node = nindex_to_node()->at(i);
3017      }
3018 
3019     // Get cpus from the original node and map them to the closest node. If node
3020     // is a configured node (not a memory-less node), then original node and
3021     // closest node are the same.
3022     if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3023       for (size_t j = 0; j < cpu_map_valid_size; j++) {
3024         if (cpu_map[j] != 0) {
3025           for (size_t k = 0; k < BitsPerCLong; k++) {
3026             if (cpu_map[j] & (1UL << k)) {
3027               cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3028             }
3029           }
3030         }
3031       }
3032     }
3033   }
3034   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3035 }
3036 
3037 int os::Linux::get_node_by_cpu(int cpu_id) {
3038   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3039     return cpu_to_node()->at(cpu_id);
3040   }
3041   return -1;
3042 }
3043 
3044 GrowableArray<int>* os::Linux::_cpu_to_node;
3045 GrowableArray<int>* os::Linux::_nindex_to_node;
3046 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3047 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3048 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3049 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3050 os::Linux::numa_available_func_t os::Linux::_numa_available;
3051 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3052 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3053 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3054 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3055 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3056 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3057 unsigned long* os::Linux::_numa_all_nodes;
3058 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3059 struct bitmask* os::Linux::_numa_nodes_ptr;
3060 
3061 bool os::pd_uncommit_memory(char* addr, size_t size) {
3062   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3063                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3064   return res  != (uintptr_t) MAP_FAILED;
3065 }
3066 
3067 static address get_stack_commited_bottom(address bottom, size_t size) {
3068   address nbot = bottom;
3069   address ntop = bottom + size;
3070 
3071   size_t page_sz = os::vm_page_size();
3072   unsigned pages = size / page_sz;
3073 
3074   unsigned char vec[1];
3075   unsigned imin = 1, imax = pages + 1, imid;
3076   int mincore_return_value = 0;
3077 
3078   assert(imin <= imax, "Unexpected page size");
3079 
3080   while (imin < imax) {
3081     imid = (imax + imin) / 2;
3082     nbot = ntop - (imid * page_sz);
3083 
3084     // Use a trick with mincore to check whether the page is mapped or not.
3085     // mincore sets vec to 1 if page resides in memory and to 0 if page
3086     // is swapped output but if page we are asking for is unmapped
3087     // it returns -1,ENOMEM
3088     mincore_return_value = mincore(nbot, page_sz, vec);
3089 
3090     if (mincore_return_value == -1) {
3091       // Page is not mapped go up
3092       // to find first mapped page
3093       if (errno != EAGAIN) {
3094         assert(errno == ENOMEM, "Unexpected mincore errno");
3095         imax = imid;
3096       }
3097     } else {
3098       // Page is mapped go down
3099       // to find first not mapped page
3100       imin = imid + 1;
3101     }
3102   }
3103 
3104   nbot = nbot + page_sz;
3105 
3106   // Adjust stack bottom one page up if last checked page is not mapped
3107   if (mincore_return_value == -1) {
3108     nbot = nbot + page_sz;
3109   }
3110 
3111   return nbot;
3112 }
3113 
3114 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3115   int mincore_return_value;
3116   const size_t stripe = 1024;  // query this many pages each time
3117   unsigned char vec[stripe];
3118   const size_t page_sz = os::vm_page_size();
3119   size_t pages = size / page_sz;
3120 
3121   assert(is_aligned(start, page_sz), "Start address must be page aligned");
3122   assert(is_aligned(size, page_sz), "Size must be page aligned");
3123 
3124   committed_start = NULL;
3125 
3126   int loops = (pages + stripe - 1) / stripe;
3127   int committed_pages = 0;
3128   address loop_base = start;
3129   for (int index = 0; index < loops; index ++) {
3130     assert(pages > 0, "Nothing to do");
3131     int pages_to_query = (pages >= stripe) ? stripe : pages;
3132     pages -= pages_to_query;
3133 
3134     // Get stable read
3135     while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3136 
3137     // During shutdown, some memory goes away without properly notifying NMT,
3138     // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3139     // Bailout and return as not committed for now.
3140     if (mincore_return_value == -1 && errno == ENOMEM) {
3141       return false;
3142     }
3143 
3144     assert(mincore_return_value == 0, "Range must be valid");
3145     // Process this stripe
3146     for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3147       if ((vec[vecIdx] & 0x01) == 0) { // not committed
3148         // End of current contiguous region
3149         if (committed_start != NULL) {
3150           break;
3151         }
3152       } else { // committed
3153         // Start of region
3154         if (committed_start == NULL) {
3155           committed_start = loop_base + page_sz * vecIdx;
3156         }
3157         committed_pages ++;
3158       }
3159     }
3160 
3161     loop_base += pages_to_query * page_sz;
3162   }
3163 
3164   if (committed_start != NULL) {
3165     assert(committed_pages > 0, "Must have committed region");
3166     assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3167     assert(committed_start >= start && committed_start < start + size, "Out of range");
3168     committed_size = page_sz * committed_pages;
3169     return true;
3170   } else {
3171     assert(committed_pages == 0, "Should not have committed region");
3172     return false;
3173   }
3174 }
3175 
3176 
3177 // Linux uses a growable mapping for the stack, and if the mapping for
3178 // the stack guard pages is not removed when we detach a thread the
3179 // stack cannot grow beyond the pages where the stack guard was
3180 // mapped.  If at some point later in the process the stack expands to
3181 // that point, the Linux kernel cannot expand the stack any further
3182 // because the guard pages are in the way, and a segfault occurs.
3183 //
3184 // However, it's essential not to split the stack region by unmapping
3185 // a region (leaving a hole) that's already part of the stack mapping,
3186 // so if the stack mapping has already grown beyond the guard pages at
3187 // the time we create them, we have to truncate the stack mapping.
3188 // So, we need to know the extent of the stack mapping when
3189 // create_stack_guard_pages() is called.
3190 
3191 // We only need this for stacks that are growable: at the time of
3192 // writing thread stacks don't use growable mappings (i.e. those
3193 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3194 // only applies to the main thread.
3195 
3196 // If the (growable) stack mapping already extends beyond the point
3197 // where we're going to put our guard pages, truncate the mapping at
3198 // that point by munmap()ping it.  This ensures that when we later
3199 // munmap() the guard pages we don't leave a hole in the stack
3200 // mapping. This only affects the main/primordial thread
3201 
3202 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3203   if (os::is_primordial_thread()) {
3204     // As we manually grow stack up to bottom inside create_attached_thread(),
3205     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3206     // we don't need to do anything special.
3207     // Check it first, before calling heavy function.
3208     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3209     unsigned char vec[1];
3210 
3211     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3212       // Fallback to slow path on all errors, including EAGAIN
3213       stack_extent = (uintptr_t) get_stack_commited_bottom(
3214                                                            os::Linux::initial_thread_stack_bottom(),
3215                                                            (size_t)addr - stack_extent);
3216     }
3217 
3218     if (stack_extent < (uintptr_t)addr) {
3219       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3220     }
3221   }
3222 
3223   return os::commit_memory(addr, size, !ExecMem);
3224 }
3225 
3226 // If this is a growable mapping, remove the guard pages entirely by
3227 // munmap()ping them.  If not, just call uncommit_memory(). This only
3228 // affects the main/primordial thread, but guard against future OS changes.
3229 // It's safe to always unmap guard pages for primordial thread because we
3230 // always place it right after end of the mapped region.
3231 
3232 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3233   uintptr_t stack_extent, stack_base;
3234 
3235   if (os::is_primordial_thread()) {
3236     return ::munmap(addr, size) == 0;
3237   }
3238 
3239   return os::uncommit_memory(addr, size);
3240 }
3241 
3242 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3243 // at 'requested_addr'. If there are existing memory mappings at the same
3244 // location, however, they will be overwritten. If 'fixed' is false,
3245 // 'requested_addr' is only treated as a hint, the return value may or
3246 // may not start from the requested address. Unlike Linux mmap(), this
3247 // function returns NULL to indicate failure.
3248 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3249   char * addr;
3250   int flags;
3251 
3252   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3253   if (fixed) {
3254     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3255     flags |= MAP_FIXED;
3256   }
3257 
3258   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3259   // touch an uncommitted page. Otherwise, the read/write might
3260   // succeed if we have enough swap space to back the physical page.
3261   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3262                        flags, -1, 0);
3263 
3264   return addr == MAP_FAILED ? NULL : addr;
3265 }
3266 
3267 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3268 //   (req_addr != NULL) or with a given alignment.
3269 //  - bytes shall be a multiple of alignment.
3270 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3271 //  - alignment sets the alignment at which memory shall be allocated.
3272 //     It must be a multiple of allocation granularity.
3273 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3274 //  req_addr or NULL.
3275 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3276 
3277   size_t extra_size = bytes;
3278   if (req_addr == NULL && alignment > 0) {
3279     extra_size += alignment;
3280   }
3281 
3282   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3283     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3284     -1, 0);
3285   if (start == MAP_FAILED) {
3286     start = NULL;
3287   } else {
3288     if (req_addr != NULL) {
3289       if (start != req_addr) {
3290         ::munmap(start, extra_size);
3291         start = NULL;
3292       }
3293     } else {
3294       char* const start_aligned = align_up(start, alignment);
3295       char* const end_aligned = start_aligned + bytes;
3296       char* const end = start + extra_size;
3297       if (start_aligned > start) {
3298         ::munmap(start, start_aligned - start);
3299       }
3300       if (end_aligned < end) {
3301         ::munmap(end_aligned, end - end_aligned);
3302       }
3303       start = start_aligned;
3304     }
3305   }
3306   return start;
3307 }
3308 
3309 static int anon_munmap(char * addr, size_t size) {
3310   return ::munmap(addr, size) == 0;
3311 }
3312 
3313 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3314                             size_t alignment_hint) {
3315   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3316 }
3317 
3318 bool os::pd_release_memory(char* addr, size_t size) {
3319   return anon_munmap(addr, size);
3320 }
3321 
3322 static bool linux_mprotect(char* addr, size_t size, int prot) {
3323   // Linux wants the mprotect address argument to be page aligned.
3324   char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3325 
3326   // According to SUSv3, mprotect() should only be used with mappings
3327   // established by mmap(), and mmap() always maps whole pages. Unaligned
3328   // 'addr' likely indicates problem in the VM (e.g. trying to change
3329   // protection of malloc'ed or statically allocated memory). Check the
3330   // caller if you hit this assert.
3331   assert(addr == bottom, "sanity check");
3332 
3333   size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3334   return ::mprotect(bottom, size, prot) == 0;
3335 }
3336 
3337 // Set protections specified
3338 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3339                         bool is_committed) {
3340   unsigned int p = 0;
3341   switch (prot) {
3342   case MEM_PROT_NONE: p = PROT_NONE; break;
3343   case MEM_PROT_READ: p = PROT_READ; break;
3344   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3345   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3346   default:
3347     ShouldNotReachHere();
3348   }
3349   // is_committed is unused.
3350   return linux_mprotect(addr, bytes, p);
3351 }
3352 
3353 bool os::guard_memory(char* addr, size_t size) {
3354   return linux_mprotect(addr, size, PROT_NONE);
3355 }
3356 
3357 bool os::unguard_memory(char* addr, size_t size) {
3358   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3359 }
3360 
3361 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3362                                                     size_t page_size) {
3363   bool result = false;
3364   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3365                  MAP_ANONYMOUS|MAP_PRIVATE,
3366                  -1, 0);
3367   if (p != MAP_FAILED) {
3368     void *aligned_p = align_up(p, page_size);
3369 
3370     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3371 
3372     munmap(p, page_size * 2);
3373   }
3374 
3375   if (warn && !result) {
3376     warning("TransparentHugePages is not supported by the operating system.");
3377   }
3378 
3379   return result;
3380 }
3381 
3382 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3383   bool result = false;
3384   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3385                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3386                  -1, 0);
3387 
3388   if (p != MAP_FAILED) {
3389     // We don't know if this really is a huge page or not.
3390     FILE *fp = fopen("/proc/self/maps", "r");
3391     if (fp) {
3392       while (!feof(fp)) {
3393         char chars[257];
3394         long x = 0;
3395         if (fgets(chars, sizeof(chars), fp)) {
3396           if (sscanf(chars, "%lx-%*x", &x) == 1
3397               && x == (long)p) {
3398             if (strstr (chars, "hugepage")) {
3399               result = true;
3400               break;
3401             }
3402           }
3403         }
3404       }
3405       fclose(fp);
3406     }
3407     munmap(p, page_size);
3408   }
3409 
3410   if (warn && !result) {
3411     warning("HugeTLBFS is not supported by the operating system.");
3412   }
3413 
3414   return result;
3415 }
3416 
3417 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3418 //
3419 // From the coredump_filter documentation:
3420 //
3421 // - (bit 0) anonymous private memory
3422 // - (bit 1) anonymous shared memory
3423 // - (bit 2) file-backed private memory
3424 // - (bit 3) file-backed shared memory
3425 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3426 //           effective only if the bit 2 is cleared)
3427 // - (bit 5) hugetlb private memory
3428 // - (bit 6) hugetlb shared memory
3429 // - (bit 7) dax private memory
3430 // - (bit 8) dax shared memory
3431 //
3432 static void set_coredump_filter(bool largepages, bool dax_shared) {
3433   FILE *f;
3434   long cdm;
3435   bool filter_changed = false;
3436 
3437   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3438     return;
3439   }
3440 
3441   if (fscanf(f, "%lx", &cdm) != 1) {
3442     fclose(f);
3443     return;
3444   }
3445 
3446   rewind(f);
3447 
3448   if (largepages && (cdm & LARGEPAGES_BIT) == 0) {
3449     cdm |= LARGEPAGES_BIT;
3450     filter_changed = true;
3451   }
3452   if (dax_shared && (cdm & DAX_SHARED_BIT) == 0) {
3453     cdm |= DAX_SHARED_BIT;
3454     filter_changed = true;
3455   }
3456   if (filter_changed) {
3457     fprintf(f, "%#lx", cdm);
3458   }
3459 
3460   fclose(f);
3461 }
3462 
3463 // Large page support
3464 
3465 static size_t _large_page_size = 0;
3466 
3467 size_t os::Linux::find_large_page_size() {
3468   size_t large_page_size = 0;
3469 
3470   // large_page_size on Linux is used to round up heap size. x86 uses either
3471   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3472   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3473   // page as large as 256M.
3474   //
3475   // Here we try to figure out page size by parsing /proc/meminfo and looking
3476   // for a line with the following format:
3477   //    Hugepagesize:     2048 kB
3478   //
3479   // If we can't determine the value (e.g. /proc is not mounted, or the text
3480   // format has been changed), we'll use the largest page size supported by
3481   // the processor.
3482 
3483 #ifndef ZERO
3484   large_page_size =
3485     AARCH64_ONLY(2 * M)
3486     AMD64_ONLY(2 * M)
3487     ARM32_ONLY(2 * M)
3488     IA32_ONLY(4 * M)
3489     IA64_ONLY(256 * M)
3490     PPC_ONLY(4 * M)
3491     S390_ONLY(1 * M)
3492     SPARC_ONLY(4 * M);
3493 #endif // ZERO
3494 
3495   FILE *fp = fopen("/proc/meminfo", "r");
3496   if (fp) {
3497     while (!feof(fp)) {
3498       int x = 0;
3499       char buf[16];
3500       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3501         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3502           large_page_size = x * K;
3503           break;
3504         }
3505       } else {
3506         // skip to next line
3507         for (;;) {
3508           int ch = fgetc(fp);
3509           if (ch == EOF || ch == (int)'\n') break;
3510         }
3511       }
3512     }
3513     fclose(fp);
3514   }
3515 
3516   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3517     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3518             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3519             proper_unit_for_byte_size(large_page_size));
3520   }
3521 
3522   return large_page_size;
3523 }
3524 
3525 size_t os::Linux::setup_large_page_size() {
3526   _large_page_size = Linux::find_large_page_size();
3527   const size_t default_page_size = (size_t)Linux::page_size();
3528   if (_large_page_size > default_page_size) {
3529     _page_sizes[0] = _large_page_size;
3530     _page_sizes[1] = default_page_size;
3531     _page_sizes[2] = 0;
3532   }
3533 
3534   return _large_page_size;
3535 }
3536 
3537 bool os::Linux::setup_large_page_type(size_t page_size) {
3538   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3539       FLAG_IS_DEFAULT(UseSHM) &&
3540       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3541 
3542     // The type of large pages has not been specified by the user.
3543 
3544     // Try UseHugeTLBFS and then UseSHM.
3545     UseHugeTLBFS = UseSHM = true;
3546 
3547     // Don't try UseTransparentHugePages since there are known
3548     // performance issues with it turned on. This might change in the future.
3549     UseTransparentHugePages = false;
3550   }
3551 
3552   if (UseTransparentHugePages) {
3553     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3554     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3555       UseHugeTLBFS = false;
3556       UseSHM = false;
3557       return true;
3558     }
3559     UseTransparentHugePages = false;
3560   }
3561 
3562   if (UseHugeTLBFS) {
3563     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3564     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3565       UseSHM = false;
3566       return true;
3567     }
3568     UseHugeTLBFS = false;
3569   }
3570 
3571   return UseSHM;
3572 }
3573 
3574 void os::large_page_init() {
3575   if (!UseLargePages &&
3576       !UseTransparentHugePages &&
3577       !UseHugeTLBFS &&
3578       !UseSHM) {
3579     // Not using large pages.
3580     return;
3581   }
3582 
3583   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3584     // The user explicitly turned off large pages.
3585     // Ignore the rest of the large pages flags.
3586     UseTransparentHugePages = false;
3587     UseHugeTLBFS = false;
3588     UseSHM = false;
3589     return;
3590   }
3591 
3592   size_t large_page_size = Linux::setup_large_page_size();
3593   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3594 
3595   set_coredump_filter(true /*largepages*/, false /*dax_shared*/);
3596 }
3597 
3598 #ifndef SHM_HUGETLB
3599   #define SHM_HUGETLB 04000
3600 #endif
3601 
3602 #define shm_warning_format(format, ...)              \
3603   do {                                               \
3604     if (UseLargePages &&                             \
3605         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3606          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3607          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3608       warning(format, __VA_ARGS__);                  \
3609     }                                                \
3610   } while (0)
3611 
3612 #define shm_warning(str) shm_warning_format("%s", str)
3613 
3614 #define shm_warning_with_errno(str)                \
3615   do {                                             \
3616     int err = errno;                               \
3617     shm_warning_format(str " (error = %d)", err);  \
3618   } while (0)
3619 
3620 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3621   assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3622 
3623   if (!is_aligned(alignment, SHMLBA)) {
3624     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3625     return NULL;
3626   }
3627 
3628   // To ensure that we get 'alignment' aligned memory from shmat,
3629   // we pre-reserve aligned virtual memory and then attach to that.
3630 
3631   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3632   if (pre_reserved_addr == NULL) {
3633     // Couldn't pre-reserve aligned memory.
3634     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3635     return NULL;
3636   }
3637 
3638   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3639   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3640 
3641   if ((intptr_t)addr == -1) {
3642     int err = errno;
3643     shm_warning_with_errno("Failed to attach shared memory.");
3644 
3645     assert(err != EACCES, "Unexpected error");
3646     assert(err != EIDRM,  "Unexpected error");
3647     assert(err != EINVAL, "Unexpected error");
3648 
3649     // Since we don't know if the kernel unmapped the pre-reserved memory area
3650     // we can't unmap it, since that would potentially unmap memory that was
3651     // mapped from other threads.
3652     return NULL;
3653   }
3654 
3655   return addr;
3656 }
3657 
3658 static char* shmat_at_address(int shmid, char* req_addr) {
3659   if (!is_aligned(req_addr, SHMLBA)) {
3660     assert(false, "Requested address needs to be SHMLBA aligned");
3661     return NULL;
3662   }
3663 
3664   char* addr = (char*)shmat(shmid, req_addr, 0);
3665 
3666   if ((intptr_t)addr == -1) {
3667     shm_warning_with_errno("Failed to attach shared memory.");
3668     return NULL;
3669   }
3670 
3671   return addr;
3672 }
3673 
3674 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3675   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3676   if (req_addr != NULL) {
3677     assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3678     assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3679     return shmat_at_address(shmid, req_addr);
3680   }
3681 
3682   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3683   // return large page size aligned memory addresses when req_addr == NULL.
3684   // However, if the alignment is larger than the large page size, we have
3685   // to manually ensure that the memory returned is 'alignment' aligned.
3686   if (alignment > os::large_page_size()) {
3687     assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3688     return shmat_with_alignment(shmid, bytes, alignment);
3689   } else {
3690     return shmat_at_address(shmid, NULL);
3691   }
3692 }
3693 
3694 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3695                                             char* req_addr, bool exec) {
3696   // "exec" is passed in but not used.  Creating the shared image for
3697   // the code cache doesn't have an SHM_X executable permission to check.
3698   assert(UseLargePages && UseSHM, "only for SHM large pages");
3699   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3700   assert(is_aligned(req_addr, alignment), "Unaligned address");
3701 
3702   if (!is_aligned(bytes, os::large_page_size())) {
3703     return NULL; // Fallback to small pages.
3704   }
3705 
3706   // Create a large shared memory region to attach to based on size.
3707   // Currently, size is the total size of the heap.
3708   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3709   if (shmid == -1) {
3710     // Possible reasons for shmget failure:
3711     // 1. shmmax is too small for Java heap.
3712     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3713     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3714     // 2. not enough large page memory.
3715     //    > check available large pages: cat /proc/meminfo
3716     //    > increase amount of large pages:
3717     //          echo new_value > /proc/sys/vm/nr_hugepages
3718     //      Note 1: different Linux may use different name for this property,
3719     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3720     //      Note 2: it's possible there's enough physical memory available but
3721     //            they are so fragmented after a long run that they can't
3722     //            coalesce into large pages. Try to reserve large pages when
3723     //            the system is still "fresh".
3724     shm_warning_with_errno("Failed to reserve shared memory.");
3725     return NULL;
3726   }
3727 
3728   // Attach to the region.
3729   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3730 
3731   // Remove shmid. If shmat() is successful, the actual shared memory segment
3732   // will be deleted when it's detached by shmdt() or when the process
3733   // terminates. If shmat() is not successful this will remove the shared
3734   // segment immediately.
3735   shmctl(shmid, IPC_RMID, NULL);
3736 
3737   return addr;
3738 }
3739 
3740 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3741                                         int error) {
3742   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3743 
3744   bool warn_on_failure = UseLargePages &&
3745       (!FLAG_IS_DEFAULT(UseLargePages) ||
3746        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3747        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3748 
3749   if (warn_on_failure) {
3750     char msg[128];
3751     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3752                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3753     warning("%s", msg);
3754   }
3755 }
3756 
3757 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3758                                                         char* req_addr,
3759                                                         bool exec) {
3760   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3761   assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3762   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3763 
3764   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3765   char* addr = (char*)::mmap(req_addr, bytes, prot,
3766                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3767                              -1, 0);
3768 
3769   if (addr == MAP_FAILED) {
3770     warn_on_large_pages_failure(req_addr, bytes, errno);
3771     return NULL;
3772   }
3773 
3774   assert(is_aligned(addr, os::large_page_size()), "Must be");
3775 
3776   return addr;
3777 }
3778 
3779 // Reserve memory using mmap(MAP_HUGETLB).
3780 //  - bytes shall be a multiple of alignment.
3781 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3782 //  - alignment sets the alignment at which memory shall be allocated.
3783 //     It must be a multiple of allocation granularity.
3784 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3785 //  req_addr or NULL.
3786 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3787                                                          size_t alignment,
3788                                                          char* req_addr,
3789                                                          bool exec) {
3790   size_t large_page_size = os::large_page_size();
3791   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3792 
3793   assert(is_aligned(req_addr, alignment), "Must be");
3794   assert(is_aligned(bytes, alignment), "Must be");
3795 
3796   // First reserve - but not commit - the address range in small pages.
3797   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3798 
3799   if (start == NULL) {
3800     return NULL;
3801   }
3802 
3803   assert(is_aligned(start, alignment), "Must be");
3804 
3805   char* end = start + bytes;
3806 
3807   // Find the regions of the allocated chunk that can be promoted to large pages.
3808   char* lp_start = align_up(start, large_page_size);
3809   char* lp_end   = align_down(end, large_page_size);
3810 
3811   size_t lp_bytes = lp_end - lp_start;
3812 
3813   assert(is_aligned(lp_bytes, large_page_size), "Must be");
3814 
3815   if (lp_bytes == 0) {
3816     // The mapped region doesn't even span the start and the end of a large page.
3817     // Fall back to allocate a non-special area.
3818     ::munmap(start, end - start);
3819     return NULL;
3820   }
3821 
3822   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3823 
3824   void* result;
3825 
3826   // Commit small-paged leading area.
3827   if (start != lp_start) {
3828     result = ::mmap(start, lp_start - start, prot,
3829                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3830                     -1, 0);
3831     if (result == MAP_FAILED) {
3832       ::munmap(lp_start, end - lp_start);
3833       return NULL;
3834     }
3835   }
3836 
3837   // Commit large-paged area.
3838   result = ::mmap(lp_start, lp_bytes, prot,
3839                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3840                   -1, 0);
3841   if (result == MAP_FAILED) {
3842     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3843     // If the mmap above fails, the large pages region will be unmapped and we
3844     // have regions before and after with small pages. Release these regions.
3845     //
3846     // |  mapped  |  unmapped  |  mapped  |
3847     // ^          ^            ^          ^
3848     // start      lp_start     lp_end     end
3849     //
3850     ::munmap(start, lp_start - start);
3851     ::munmap(lp_end, end - lp_end);
3852     return NULL;
3853   }
3854 
3855   // Commit small-paged trailing area.
3856   if (lp_end != end) {
3857     result = ::mmap(lp_end, end - lp_end, prot,
3858                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3859                     -1, 0);
3860     if (result == MAP_FAILED) {
3861       ::munmap(start, lp_end - start);
3862       return NULL;
3863     }
3864   }
3865 
3866   return start;
3867 }
3868 
3869 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3870                                                    size_t alignment,
3871                                                    char* req_addr,
3872                                                    bool exec) {
3873   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3874   assert(is_aligned(req_addr, alignment), "Must be");
3875   assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3876   assert(is_power_of_2(os::large_page_size()), "Must be");
3877   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3878 
3879   if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3880     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3881   } else {
3882     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3883   }
3884 }
3885 
3886 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3887                                  char* req_addr, bool exec) {
3888   assert(UseLargePages, "only for large pages");
3889 
3890   char* addr;
3891   if (UseSHM) {
3892     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3893   } else {
3894     assert(UseHugeTLBFS, "must be");
3895     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3896   }
3897 
3898   if (addr != NULL) {
3899     if (UseNUMAInterleaving) {
3900       numa_make_global(addr, bytes);
3901     }
3902 
3903     // The memory is committed
3904     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3905   }
3906 
3907   return addr;
3908 }
3909 
3910 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3911   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3912   return shmdt(base) == 0;
3913 }
3914 
3915 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3916   return pd_release_memory(base, bytes);
3917 }
3918 
3919 bool os::release_memory_special(char* base, size_t bytes) {
3920   bool res;
3921   if (MemTracker::tracking_level() > NMT_minimal) {
3922     Tracker tkr(Tracker::release);
3923     res = os::Linux::release_memory_special_impl(base, bytes);
3924     if (res) {
3925       tkr.record((address)base, bytes);
3926     }
3927 
3928   } else {
3929     res = os::Linux::release_memory_special_impl(base, bytes);
3930   }
3931   return res;
3932 }
3933 
3934 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3935   assert(UseLargePages, "only for large pages");
3936   bool res;
3937 
3938   if (UseSHM) {
3939     res = os::Linux::release_memory_special_shm(base, bytes);
3940   } else {
3941     assert(UseHugeTLBFS, "must be");
3942     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3943   }
3944   return res;
3945 }
3946 
3947 size_t os::large_page_size() {
3948   return _large_page_size;
3949 }
3950 
3951 // With SysV SHM the entire memory region must be allocated as shared
3952 // memory.
3953 // HugeTLBFS allows application to commit large page memory on demand.
3954 // However, when committing memory with HugeTLBFS fails, the region
3955 // that was supposed to be committed will lose the old reservation
3956 // and allow other threads to steal that memory region. Because of this
3957 // behavior we can't commit HugeTLBFS memory.
3958 bool os::can_commit_large_page_memory() {
3959   return UseTransparentHugePages;
3960 }
3961 
3962 bool os::can_execute_large_page_memory() {
3963   return UseTransparentHugePages || UseHugeTLBFS;
3964 }
3965 
3966 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3967   assert(file_desc >= 0, "file_desc is not valid");
3968   char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3969   if (result != NULL) {
3970     if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3971       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3972     }
3973   }
3974   return result;
3975 }
3976 
3977 // Reserve memory at an arbitrary address, only if that area is
3978 // available (and not reserved for something else).
3979 
3980 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3981   const int max_tries = 10;
3982   char* base[max_tries];
3983   size_t size[max_tries];
3984   const size_t gap = 0x000000;
3985 
3986   // Assert only that the size is a multiple of the page size, since
3987   // that's all that mmap requires, and since that's all we really know
3988   // about at this low abstraction level.  If we need higher alignment,
3989   // we can either pass an alignment to this method or verify alignment
3990   // in one of the methods further up the call chain.  See bug 5044738.
3991   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3992 
3993   // Repeatedly allocate blocks until the block is allocated at the
3994   // right spot.
3995 
3996   // Linux mmap allows caller to pass an address as hint; give it a try first,
3997   // if kernel honors the hint then we can return immediately.
3998   char * addr = anon_mmap(requested_addr, bytes, false);
3999   if (addr == requested_addr) {
4000     return requested_addr;
4001   }
4002 
4003   if (addr != NULL) {
4004     // mmap() is successful but it fails to reserve at the requested address
4005     anon_munmap(addr, bytes);
4006   }
4007 
4008   int i;
4009   for (i = 0; i < max_tries; ++i) {
4010     base[i] = reserve_memory(bytes);
4011 
4012     if (base[i] != NULL) {
4013       // Is this the block we wanted?
4014       if (base[i] == requested_addr) {
4015         size[i] = bytes;
4016         break;
4017       }
4018 
4019       // Does this overlap the block we wanted? Give back the overlapped
4020       // parts and try again.
4021 
4022       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
4023       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
4024         unmap_memory(base[i], top_overlap);
4025         base[i] += top_overlap;
4026         size[i] = bytes - top_overlap;
4027       } else {
4028         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
4029         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
4030           unmap_memory(requested_addr, bottom_overlap);
4031           size[i] = bytes - bottom_overlap;
4032         } else {
4033           size[i] = bytes;
4034         }
4035       }
4036     }
4037   }
4038 
4039   // Give back the unused reserved pieces.
4040 
4041   for (int j = 0; j < i; ++j) {
4042     if (base[j] != NULL) {
4043       unmap_memory(base[j], size[j]);
4044     }
4045   }
4046 
4047   if (i < max_tries) {
4048     return requested_addr;
4049   } else {
4050     return NULL;
4051   }
4052 }
4053 
4054 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4055   return ::read(fd, buf, nBytes);
4056 }
4057 
4058 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4059   return ::pread(fd, buf, nBytes, offset);
4060 }
4061 
4062 // Short sleep, direct OS call.
4063 //
4064 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4065 // sched_yield(2) will actually give up the CPU:
4066 //
4067 //   * Alone on this pariticular CPU, keeps running.
4068 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
4069 //     (pre 2.6.39).
4070 //
4071 // So calling this with 0 is an alternative.
4072 //
4073 void os::naked_short_sleep(jlong ms) {
4074   struct timespec req;
4075 
4076   assert(ms < 1000, "Un-interruptable sleep, short time use only");
4077   req.tv_sec = 0;
4078   if (ms > 0) {
4079     req.tv_nsec = (ms % 1000) * 1000000;
4080   } else {
4081     req.tv_nsec = 1;
4082   }
4083 
4084   nanosleep(&req, NULL);
4085 
4086   return;
4087 }
4088 
4089 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4090 void os::infinite_sleep() {
4091   while (true) {    // sleep forever ...
4092     ::sleep(100);   // ... 100 seconds at a time
4093   }
4094 }
4095 
4096 // Used to convert frequent JVM_Yield() to nops
4097 bool os::dont_yield() {
4098   return DontYieldALot;
4099 }
4100 
4101 void os::naked_yield() {
4102   sched_yield();
4103 }
4104 
4105 ////////////////////////////////////////////////////////////////////////////////
4106 // thread priority support
4107 
4108 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4109 // only supports dynamic priority, static priority must be zero. For real-time
4110 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4111 // However, for large multi-threaded applications, SCHED_RR is not only slower
4112 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4113 // of 5 runs - Sep 2005).
4114 //
4115 // The following code actually changes the niceness of kernel-thread/LWP. It
4116 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4117 // not the entire user process, and user level threads are 1:1 mapped to kernel
4118 // threads. It has always been the case, but could change in the future. For
4119 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4120 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4121 
4122 int os::java_to_os_priority[CriticalPriority + 1] = {
4123   19,              // 0 Entry should never be used
4124 
4125    4,              // 1 MinPriority
4126    3,              // 2
4127    2,              // 3
4128 
4129    1,              // 4
4130    0,              // 5 NormPriority
4131   -1,              // 6
4132 
4133   -2,              // 7
4134   -3,              // 8
4135   -4,              // 9 NearMaxPriority
4136 
4137   -5,              // 10 MaxPriority
4138 
4139   -5               // 11 CriticalPriority
4140 };
4141 
4142 static int prio_init() {
4143   if (ThreadPriorityPolicy == 1) {
4144     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4145     // if effective uid is not root. Perhaps, a more elegant way of doing
4146     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4147     if (geteuid() != 0) {
4148       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4149         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4150       }
4151       ThreadPriorityPolicy = 0;
4152     }
4153   }
4154   if (UseCriticalJavaThreadPriority) {
4155     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4156   }
4157   return 0;
4158 }
4159 
4160 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4161   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4162 
4163   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4164   return (ret == 0) ? OS_OK : OS_ERR;
4165 }
4166 
4167 OSReturn os::get_native_priority(const Thread* const thread,
4168                                  int *priority_ptr) {
4169   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4170     *priority_ptr = java_to_os_priority[NormPriority];
4171     return OS_OK;
4172   }
4173 
4174   errno = 0;
4175   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4176   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4177 }
4178 
4179 // Hint to the underlying OS that a task switch would not be good.
4180 // Void return because it's a hint and can fail.
4181 void os::hint_no_preempt() {}
4182 
4183 ////////////////////////////////////////////////////////////////////////////////
4184 // suspend/resume support
4185 
4186 //  The low-level signal-based suspend/resume support is a remnant from the
4187 //  old VM-suspension that used to be for java-suspension, safepoints etc,
4188 //  within hotspot. Currently used by JFR's OSThreadSampler
4189 //
4190 //  The remaining code is greatly simplified from the more general suspension
4191 //  code that used to be used.
4192 //
4193 //  The protocol is quite simple:
4194 //  - suspend:
4195 //      - sends a signal to the target thread
4196 //      - polls the suspend state of the osthread using a yield loop
4197 //      - target thread signal handler (SR_handler) sets suspend state
4198 //        and blocks in sigsuspend until continued
4199 //  - resume:
4200 //      - sets target osthread state to continue
4201 //      - sends signal to end the sigsuspend loop in the SR_handler
4202 //
4203 //  Note that the SR_lock plays no role in this suspend/resume protocol,
4204 //  but is checked for NULL in SR_handler as a thread termination indicator.
4205 //  The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4206 //
4207 //  Note that resume_clear_context() and suspend_save_context() are needed
4208 //  by SR_handler(), so that fetch_frame_from_ucontext() works,
4209 //  which in part is used by:
4210 //    - Forte Analyzer: AsyncGetCallTrace()
4211 //    - StackBanging: get_frame_at_stack_banging_point()
4212 
4213 static void resume_clear_context(OSThread *osthread) {
4214   osthread->set_ucontext(NULL);
4215   osthread->set_siginfo(NULL);
4216 }
4217 
4218 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4219                                  ucontext_t* context) {
4220   osthread->set_ucontext(context);
4221   osthread->set_siginfo(siginfo);
4222 }
4223 
4224 // Handler function invoked when a thread's execution is suspended or
4225 // resumed. We have to be careful that only async-safe functions are
4226 // called here (Note: most pthread functions are not async safe and
4227 // should be avoided.)
4228 //
4229 // Note: sigwait() is a more natural fit than sigsuspend() from an
4230 // interface point of view, but sigwait() prevents the signal hander
4231 // from being run. libpthread would get very confused by not having
4232 // its signal handlers run and prevents sigwait()'s use with the
4233 // mutex granting granting signal.
4234 //
4235 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4236 //
4237 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4238   // Save and restore errno to avoid confusing native code with EINTR
4239   // after sigsuspend.
4240   int old_errno = errno;
4241 
4242   Thread* thread = Thread::current_or_null_safe();
4243   assert(thread != NULL, "Missing current thread in SR_handler");
4244 
4245   // On some systems we have seen signal delivery get "stuck" until the signal
4246   // mask is changed as part of thread termination. Check that the current thread
4247   // has not already terminated (via SR_lock()) - else the following assertion
4248   // will fail because the thread is no longer a JavaThread as the ~JavaThread
4249   // destructor has completed.
4250 
4251   if (thread->SR_lock() == NULL) {
4252     return;
4253   }
4254 
4255   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4256 
4257   OSThread* osthread = thread->osthread();
4258 
4259   os::SuspendResume::State current = osthread->sr.state();
4260   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4261     suspend_save_context(osthread, siginfo, context);
4262 
4263     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4264     os::SuspendResume::State state = osthread->sr.suspended();
4265     if (state == os::SuspendResume::SR_SUSPENDED) {
4266       sigset_t suspend_set;  // signals for sigsuspend()
4267       sigemptyset(&suspend_set);
4268       // get current set of blocked signals and unblock resume signal
4269       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4270       sigdelset(&suspend_set, SR_signum);
4271 
4272       sr_semaphore.signal();
4273       // wait here until we are resumed
4274       while (1) {
4275         sigsuspend(&suspend_set);
4276 
4277         os::SuspendResume::State result = osthread->sr.running();
4278         if (result == os::SuspendResume::SR_RUNNING) {
4279           sr_semaphore.signal();
4280           break;
4281         }
4282       }
4283 
4284     } else if (state == os::SuspendResume::SR_RUNNING) {
4285       // request was cancelled, continue
4286     } else {
4287       ShouldNotReachHere();
4288     }
4289 
4290     resume_clear_context(osthread);
4291   } else if (current == os::SuspendResume::SR_RUNNING) {
4292     // request was cancelled, continue
4293   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4294     // ignore
4295   } else {
4296     // ignore
4297   }
4298 
4299   errno = old_errno;
4300 }
4301 
4302 static int SR_initialize() {
4303   struct sigaction act;
4304   char *s;
4305 
4306   // Get signal number to use for suspend/resume
4307   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4308     int sig = ::strtol(s, 0, 10);
4309     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4310         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4311       SR_signum = sig;
4312     } else {
4313       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4314               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4315     }
4316   }
4317 
4318   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4319          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4320 
4321   sigemptyset(&SR_sigset);
4322   sigaddset(&SR_sigset, SR_signum);
4323 
4324   // Set up signal handler for suspend/resume
4325   act.sa_flags = SA_RESTART|SA_SIGINFO;
4326   act.sa_handler = (void (*)(int)) SR_handler;
4327 
4328   // SR_signum is blocked by default.
4329   // 4528190 - We also need to block pthread restart signal (32 on all
4330   // supported Linux platforms). Note that LinuxThreads need to block
4331   // this signal for all threads to work properly. So we don't have
4332   // to use hard-coded signal number when setting up the mask.
4333   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4334 
4335   if (sigaction(SR_signum, &act, 0) == -1) {
4336     return -1;
4337   }
4338 
4339   // Save signal flag
4340   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4341   return 0;
4342 }
4343 
4344 static int sr_notify(OSThread* osthread) {
4345   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4346   assert_status(status == 0, status, "pthread_kill");
4347   return status;
4348 }
4349 
4350 // "Randomly" selected value for how long we want to spin
4351 // before bailing out on suspending a thread, also how often
4352 // we send a signal to a thread we want to resume
4353 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4354 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4355 
4356 // returns true on success and false on error - really an error is fatal
4357 // but this seems the normal response to library errors
4358 static bool do_suspend(OSThread* osthread) {
4359   assert(osthread->sr.is_running(), "thread should be running");
4360   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4361 
4362   // mark as suspended and send signal
4363   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4364     // failed to switch, state wasn't running?
4365     ShouldNotReachHere();
4366     return false;
4367   }
4368 
4369   if (sr_notify(osthread) != 0) {
4370     ShouldNotReachHere();
4371   }
4372 
4373   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4374   while (true) {
4375     if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4376       break;
4377     } else {
4378       // timeout
4379       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4380       if (cancelled == os::SuspendResume::SR_RUNNING) {
4381         return false;
4382       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4383         // make sure that we consume the signal on the semaphore as well
4384         sr_semaphore.wait();
4385         break;
4386       } else {
4387         ShouldNotReachHere();
4388         return false;
4389       }
4390     }
4391   }
4392 
4393   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4394   return true;
4395 }
4396 
4397 static void do_resume(OSThread* osthread) {
4398   assert(osthread->sr.is_suspended(), "thread should be suspended");
4399   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4400 
4401   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4402     // failed to switch to WAKEUP_REQUEST
4403     ShouldNotReachHere();
4404     return;
4405   }
4406 
4407   while (true) {
4408     if (sr_notify(osthread) == 0) {
4409       if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4410         if (osthread->sr.is_running()) {
4411           return;
4412         }
4413       }
4414     } else {
4415       ShouldNotReachHere();
4416     }
4417   }
4418 
4419   guarantee(osthread->sr.is_running(), "Must be running!");
4420 }
4421 
4422 ///////////////////////////////////////////////////////////////////////////////////
4423 // signal handling (except suspend/resume)
4424 
4425 // This routine may be used by user applications as a "hook" to catch signals.
4426 // The user-defined signal handler must pass unrecognized signals to this
4427 // routine, and if it returns true (non-zero), then the signal handler must
4428 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4429 // routine will never retun false (zero), but instead will execute a VM panic
4430 // routine kill the process.
4431 //
4432 // If this routine returns false, it is OK to call it again.  This allows
4433 // the user-defined signal handler to perform checks either before or after
4434 // the VM performs its own checks.  Naturally, the user code would be making
4435 // a serious error if it tried to handle an exception (such as a null check
4436 // or breakpoint) that the VM was generating for its own correct operation.
4437 //
4438 // This routine may recognize any of the following kinds of signals:
4439 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4440 // It should be consulted by handlers for any of those signals.
4441 //
4442 // The caller of this routine must pass in the three arguments supplied
4443 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4444 // field of the structure passed to sigaction().  This routine assumes that
4445 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4446 //
4447 // Note that the VM will print warnings if it detects conflicting signal
4448 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4449 //
4450 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4451                                                  siginfo_t* siginfo,
4452                                                  void* ucontext,
4453                                                  int abort_if_unrecognized);
4454 
4455 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4456   assert(info != NULL && uc != NULL, "it must be old kernel");
4457   int orig_errno = errno;  // Preserve errno value over signal handler.
4458   JVM_handle_linux_signal(sig, info, uc, true);
4459   errno = orig_errno;
4460 }
4461 
4462 
4463 // This boolean allows users to forward their own non-matching signals
4464 // to JVM_handle_linux_signal, harmlessly.
4465 bool os::Linux::signal_handlers_are_installed = false;
4466 
4467 // For signal-chaining
4468 struct sigaction sigact[NSIG];
4469 uint64_t sigs = 0;
4470 #if (64 < NSIG-1)
4471 #error "Not all signals can be encoded in sigs. Adapt its type!"
4472 #endif
4473 bool os::Linux::libjsig_is_loaded = false;
4474 typedef struct sigaction *(*get_signal_t)(int);
4475 get_signal_t os::Linux::get_signal_action = NULL;
4476 
4477 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4478   struct sigaction *actp = NULL;
4479 
4480   if (libjsig_is_loaded) {
4481     // Retrieve the old signal handler from libjsig
4482     actp = (*get_signal_action)(sig);
4483   }
4484   if (actp == NULL) {
4485     // Retrieve the preinstalled signal handler from jvm
4486     actp = get_preinstalled_handler(sig);
4487   }
4488 
4489   return actp;
4490 }
4491 
4492 static bool call_chained_handler(struct sigaction *actp, int sig,
4493                                  siginfo_t *siginfo, void *context) {
4494   // Call the old signal handler
4495   if (actp->sa_handler == SIG_DFL) {
4496     // It's more reasonable to let jvm treat it as an unexpected exception
4497     // instead of taking the default action.
4498     return false;
4499   } else if (actp->sa_handler != SIG_IGN) {
4500     if ((actp->sa_flags & SA_NODEFER) == 0) {
4501       // automaticlly block the signal
4502       sigaddset(&(actp->sa_mask), sig);
4503     }
4504 
4505     sa_handler_t hand = NULL;
4506     sa_sigaction_t sa = NULL;
4507     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4508     // retrieve the chained handler
4509     if (siginfo_flag_set) {
4510       sa = actp->sa_sigaction;
4511     } else {
4512       hand = actp->sa_handler;
4513     }
4514 
4515     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4516       actp->sa_handler = SIG_DFL;
4517     }
4518 
4519     // try to honor the signal mask
4520     sigset_t oset;
4521     sigemptyset(&oset);
4522     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4523 
4524     // call into the chained handler
4525     if (siginfo_flag_set) {
4526       (*sa)(sig, siginfo, context);
4527     } else {
4528       (*hand)(sig);
4529     }
4530 
4531     // restore the signal mask
4532     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4533   }
4534   // Tell jvm's signal handler the signal is taken care of.
4535   return true;
4536 }
4537 
4538 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4539   bool chained = false;
4540   // signal-chaining
4541   if (UseSignalChaining) {
4542     struct sigaction *actp = get_chained_signal_action(sig);
4543     if (actp != NULL) {
4544       chained = call_chained_handler(actp, sig, siginfo, context);
4545     }
4546   }
4547   return chained;
4548 }
4549 
4550 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4551   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4552     return &sigact[sig];
4553   }
4554   return NULL;
4555 }
4556 
4557 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4558   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4559   sigact[sig] = oldAct;
4560   sigs |= (uint64_t)1 << (sig-1);
4561 }
4562 
4563 // for diagnostic
4564 int sigflags[NSIG];
4565 
4566 int os::Linux::get_our_sigflags(int sig) {
4567   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4568   return sigflags[sig];
4569 }
4570 
4571 void os::Linux::set_our_sigflags(int sig, int flags) {
4572   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4573   if (sig > 0 && sig < NSIG) {
4574     sigflags[sig] = flags;
4575   }
4576 }
4577 
4578 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4579   // Check for overwrite.
4580   struct sigaction oldAct;
4581   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4582 
4583   void* oldhand = oldAct.sa_sigaction
4584                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4585                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4586   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4587       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4588       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4589     if (AllowUserSignalHandlers || !set_installed) {
4590       // Do not overwrite; user takes responsibility to forward to us.
4591       return;
4592     } else if (UseSignalChaining) {
4593       // save the old handler in jvm
4594       save_preinstalled_handler(sig, oldAct);
4595       // libjsig also interposes the sigaction() call below and saves the
4596       // old sigaction on it own.
4597     } else {
4598       fatal("Encountered unexpected pre-existing sigaction handler "
4599             "%#lx for signal %d.", (long)oldhand, sig);
4600     }
4601   }
4602 
4603   struct sigaction sigAct;
4604   sigfillset(&(sigAct.sa_mask));
4605   sigAct.sa_handler = SIG_DFL;
4606   if (!set_installed) {
4607     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4608   } else {
4609     sigAct.sa_sigaction = signalHandler;
4610     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4611   }
4612   // Save flags, which are set by ours
4613   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4614   sigflags[sig] = sigAct.sa_flags;
4615 
4616   int ret = sigaction(sig, &sigAct, &oldAct);
4617   assert(ret == 0, "check");
4618 
4619   void* oldhand2  = oldAct.sa_sigaction
4620                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4621                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4622   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4623 }
4624 
4625 // install signal handlers for signals that HotSpot needs to
4626 // handle in order to support Java-level exception handling.
4627 
4628 void os::Linux::install_signal_handlers() {
4629   if (!signal_handlers_are_installed) {
4630     signal_handlers_are_installed = true;
4631 
4632     // signal-chaining
4633     typedef void (*signal_setting_t)();
4634     signal_setting_t begin_signal_setting = NULL;
4635     signal_setting_t end_signal_setting = NULL;
4636     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4637                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4638     if (begin_signal_setting != NULL) {
4639       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4640                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4641       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4642                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4643       libjsig_is_loaded = true;
4644       assert(UseSignalChaining, "should enable signal-chaining");
4645     }
4646     if (libjsig_is_loaded) {
4647       // Tell libjsig jvm is setting signal handlers
4648       (*begin_signal_setting)();
4649     }
4650 
4651     set_signal_handler(SIGSEGV, true);
4652     set_signal_handler(SIGPIPE, true);
4653     set_signal_handler(SIGBUS, true);
4654     set_signal_handler(SIGILL, true);
4655     set_signal_handler(SIGFPE, true);
4656 #if defined(PPC64)
4657     set_signal_handler(SIGTRAP, true);
4658 #endif
4659     set_signal_handler(SIGXFSZ, true);
4660 
4661     if (libjsig_is_loaded) {
4662       // Tell libjsig jvm finishes setting signal handlers
4663       (*end_signal_setting)();
4664     }
4665 
4666     // We don't activate signal checker if libjsig is in place, we trust ourselves
4667     // and if UserSignalHandler is installed all bets are off.
4668     // Log that signal checking is off only if -verbose:jni is specified.
4669     if (CheckJNICalls) {
4670       if (libjsig_is_loaded) {
4671         if (PrintJNIResolving) {
4672           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4673         }
4674         check_signals = false;
4675       }
4676       if (AllowUserSignalHandlers) {
4677         if (PrintJNIResolving) {
4678           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4679         }
4680         check_signals = false;
4681       }
4682     }
4683   }
4684 }
4685 
4686 // This is the fastest way to get thread cpu time on Linux.
4687 // Returns cpu time (user+sys) for any thread, not only for current.
4688 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4689 // It might work on 2.6.10+ with a special kernel/glibc patch.
4690 // For reference, please, see IEEE Std 1003.1-2004:
4691 //   http://www.unix.org/single_unix_specification
4692 
4693 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4694   struct timespec tp;
4695   int rc = os::Linux::clock_gettime(clockid, &tp);
4696   assert(rc == 0, "clock_gettime is expected to return 0 code");
4697 
4698   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4699 }
4700 
4701 void os::Linux::initialize_os_info() {
4702   assert(_os_version == 0, "OS info already initialized");
4703 
4704   struct utsname _uname;
4705 
4706   uint32_t major;
4707   uint32_t minor;
4708   uint32_t fix;
4709 
4710   int rc;
4711 
4712   // Kernel version is unknown if
4713   // verification below fails.
4714   _os_version = 0x01000000;
4715 
4716   rc = uname(&_uname);
4717   if (rc != -1) {
4718 
4719     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4720     if (rc == 3) {
4721 
4722       if (major < 256 && minor < 256 && fix < 256) {
4723         // Kernel version format is as expected,
4724         // set it overriding unknown state.
4725         _os_version = (major << 16) |
4726                       (minor << 8 ) |
4727                       (fix   << 0 ) ;
4728       }
4729     }
4730   }
4731 }
4732 
4733 uint32_t os::Linux::os_version() {
4734   assert(_os_version != 0, "not initialized");
4735   return _os_version & 0x00FFFFFF;
4736 }
4737 
4738 bool os::Linux::os_version_is_known() {
4739   assert(_os_version != 0, "not initialized");
4740   return _os_version & 0x01000000 ? false : true;
4741 }
4742 
4743 /////
4744 // glibc on Linux platform uses non-documented flag
4745 // to indicate, that some special sort of signal
4746 // trampoline is used.
4747 // We will never set this flag, and we should
4748 // ignore this flag in our diagnostic
4749 #ifdef SIGNIFICANT_SIGNAL_MASK
4750   #undef SIGNIFICANT_SIGNAL_MASK
4751 #endif
4752 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4753 
4754 static const char* get_signal_handler_name(address handler,
4755                                            char* buf, int buflen) {
4756   int offset = 0;
4757   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4758   if (found) {
4759     // skip directory names
4760     const char *p1, *p2;
4761     p1 = buf;
4762     size_t len = strlen(os::file_separator());
4763     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4764     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4765   } else {
4766     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4767   }
4768   return buf;
4769 }
4770 
4771 static void print_signal_handler(outputStream* st, int sig,
4772                                  char* buf, size_t buflen) {
4773   struct sigaction sa;
4774 
4775   sigaction(sig, NULL, &sa);
4776 
4777   // See comment for SIGNIFICANT_SIGNAL_MASK define
4778   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4779 
4780   st->print("%s: ", os::exception_name(sig, buf, buflen));
4781 
4782   address handler = (sa.sa_flags & SA_SIGINFO)
4783     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4784     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4785 
4786   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4787     st->print("SIG_DFL");
4788   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4789     st->print("SIG_IGN");
4790   } else {
4791     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4792   }
4793 
4794   st->print(", sa_mask[0]=");
4795   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4796 
4797   address rh = VMError::get_resetted_sighandler(sig);
4798   // May be, handler was resetted by VMError?
4799   if (rh != NULL) {
4800     handler = rh;
4801     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4802   }
4803 
4804   st->print(", sa_flags=");
4805   os::Posix::print_sa_flags(st, sa.sa_flags);
4806 
4807   // Check: is it our handler?
4808   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4809       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4810     // It is our signal handler
4811     // check for flags, reset system-used one!
4812     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4813       st->print(
4814                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4815                 os::Linux::get_our_sigflags(sig));
4816     }
4817   }
4818   st->cr();
4819 }
4820 
4821 
4822 #define DO_SIGNAL_CHECK(sig)                      \
4823   do {                                            \
4824     if (!sigismember(&check_signal_done, sig)) {  \
4825       os::Linux::check_signal_handler(sig);       \
4826     }                                             \
4827   } while (0)
4828 
4829 // This method is a periodic task to check for misbehaving JNI applications
4830 // under CheckJNI, we can add any periodic checks here
4831 
4832 void os::run_periodic_checks() {
4833   if (check_signals == false) return;
4834 
4835   // SEGV and BUS if overridden could potentially prevent
4836   // generation of hs*.log in the event of a crash, debugging
4837   // such a case can be very challenging, so we absolutely
4838   // check the following for a good measure:
4839   DO_SIGNAL_CHECK(SIGSEGV);
4840   DO_SIGNAL_CHECK(SIGILL);
4841   DO_SIGNAL_CHECK(SIGFPE);
4842   DO_SIGNAL_CHECK(SIGBUS);
4843   DO_SIGNAL_CHECK(SIGPIPE);
4844   DO_SIGNAL_CHECK(SIGXFSZ);
4845 #if defined(PPC64)
4846   DO_SIGNAL_CHECK(SIGTRAP);
4847 #endif
4848 
4849   // ReduceSignalUsage allows the user to override these handlers
4850   // see comments at the very top and jvm_md.h
4851   if (!ReduceSignalUsage) {
4852     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4853     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4854     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4855     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4856   }
4857 
4858   DO_SIGNAL_CHECK(SR_signum);
4859 }
4860 
4861 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4862 
4863 static os_sigaction_t os_sigaction = NULL;
4864 
4865 void os::Linux::check_signal_handler(int sig) {
4866   char buf[O_BUFLEN];
4867   address jvmHandler = NULL;
4868 
4869 
4870   struct sigaction act;
4871   if (os_sigaction == NULL) {
4872     // only trust the default sigaction, in case it has been interposed
4873     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4874     if (os_sigaction == NULL) return;
4875   }
4876 
4877   os_sigaction(sig, (struct sigaction*)NULL, &act);
4878 
4879 
4880   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4881 
4882   address thisHandler = (act.sa_flags & SA_SIGINFO)
4883     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4884     : CAST_FROM_FN_PTR(address, act.sa_handler);
4885 
4886 
4887   switch (sig) {
4888   case SIGSEGV:
4889   case SIGBUS:
4890   case SIGFPE:
4891   case SIGPIPE:
4892   case SIGILL:
4893   case SIGXFSZ:
4894     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4895     break;
4896 
4897   case SHUTDOWN1_SIGNAL:
4898   case SHUTDOWN2_SIGNAL:
4899   case SHUTDOWN3_SIGNAL:
4900   case BREAK_SIGNAL:
4901     jvmHandler = (address)user_handler();
4902     break;
4903 
4904   default:
4905     if (sig == SR_signum) {
4906       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4907     } else {
4908       return;
4909     }
4910     break;
4911   }
4912 
4913   if (thisHandler != jvmHandler) {
4914     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4915     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4916     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4917     // No need to check this sig any longer
4918     sigaddset(&check_signal_done, sig);
4919     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4920     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4921       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4922                     exception_name(sig, buf, O_BUFLEN));
4923     }
4924   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4925     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4926     tty->print("expected:");
4927     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4928     tty->cr();
4929     tty->print("  found:");
4930     os::Posix::print_sa_flags(tty, act.sa_flags);
4931     tty->cr();
4932     // No need to check this sig any longer
4933     sigaddset(&check_signal_done, sig);
4934   }
4935 
4936   // Dump all the signal
4937   if (sigismember(&check_signal_done, sig)) {
4938     print_signal_handlers(tty, buf, O_BUFLEN);
4939   }
4940 }
4941 
4942 extern void report_error(char* file_name, int line_no, char* title,
4943                          char* format, ...);
4944 
4945 // this is called _before_ most of the global arguments have been parsed
4946 void os::init(void) {
4947   char dummy;   // used to get a guess on initial stack address
4948 
4949   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4950 
4951   init_random(1234567);
4952 
4953   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4954   if (Linux::page_size() == -1) {
4955     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4956           os::strerror(errno));
4957   }
4958   init_page_sizes((size_t) Linux::page_size());
4959 
4960   Linux::initialize_system_info();
4961 
4962   Linux::initialize_os_info();
4963 
4964   // _main_thread points to the thread that created/loaded the JVM.
4965   Linux::_main_thread = pthread_self();
4966 
4967   Linux::clock_init();
4968   initial_time_count = javaTimeNanos();
4969 
4970   // retrieve entry point for pthread_setname_np
4971   Linux::_pthread_setname_np =
4972     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4973 
4974   os::Posix::init();
4975 }
4976 
4977 // To install functions for atexit system call
4978 extern "C" {
4979   static void perfMemory_exit_helper() {
4980     perfMemory_exit();
4981   }
4982 }
4983 
4984 void os::pd_init_container_support() {
4985   OSContainer::init();
4986 }
4987 
4988 // this is called _after_ the global arguments have been parsed
4989 jint os::init_2(void) {
4990 
4991   os::Posix::init_2();
4992 
4993   Linux::fast_thread_clock_init();
4994 
4995   // initialize suspend/resume support - must do this before signal_sets_init()
4996   if (SR_initialize() != 0) {
4997     perror("SR_initialize failed");
4998     return JNI_ERR;
4999   }
5000 
5001   Linux::signal_sets_init();
5002   Linux::install_signal_handlers();
5003   // Initialize data for jdk.internal.misc.Signal
5004   if (!ReduceSignalUsage) {
5005     jdk_misc_signal_init();
5006   }
5007 
5008   // Check and sets minimum stack sizes against command line options
5009   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5010     return JNI_ERR;
5011   }
5012 
5013   suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5014   if (!suppress_primordial_thread_resolution) {
5015     Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5016   }
5017 
5018 #if defined(IA32)
5019   workaround_expand_exec_shield_cs_limit();
5020 #endif
5021 
5022   Linux::libpthread_init();
5023   Linux::sched_getcpu_init();
5024   log_info(os)("HotSpot is running with %s, %s",
5025                Linux::glibc_version(), Linux::libpthread_version());
5026 
5027   if (UseNUMA) {
5028     if (!Linux::libnuma_init()) {
5029       UseNUMA = false;
5030     } else {
5031       if ((Linux::numa_max_node() < 1)) {
5032         // There's only one node(they start from 0), disable NUMA.
5033         UseNUMA = false;
5034       }
5035     }
5036 
5037     if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5038       // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5039       // we can make the adaptive lgrp chunk resizing work. If the user specified both
5040       // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5041       // and disable adaptive resizing.
5042       if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5043         warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5044                 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5045         UseAdaptiveSizePolicy = false;
5046         UseAdaptiveNUMAChunkSizing = false;
5047       }
5048     }
5049 
5050     if (!UseNUMA && ForceNUMA) {
5051       UseNUMA = true;
5052     }
5053   }
5054 
5055   if (MaxFDLimit) {
5056     // set the number of file descriptors to max. print out error
5057     // if getrlimit/setrlimit fails but continue regardless.
5058     struct rlimit nbr_files;
5059     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5060     if (status != 0) {
5061       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5062     } else {
5063       nbr_files.rlim_cur = nbr_files.rlim_max;
5064       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5065       if (status != 0) {
5066         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5067       }
5068     }
5069   }
5070 
5071   // Initialize lock used to serialize thread creation (see os::create_thread)
5072   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5073 
5074   // at-exit methods are called in the reverse order of their registration.
5075   // atexit functions are called on return from main or as a result of a
5076   // call to exit(3C). There can be only 32 of these functions registered
5077   // and atexit() does not set errno.
5078 
5079   if (PerfAllowAtExitRegistration) {
5080     // only register atexit functions if PerfAllowAtExitRegistration is set.
5081     // atexit functions can be delayed until process exit time, which
5082     // can be problematic for embedded VM situations. Embedded VMs should
5083     // call DestroyJavaVM() to assure that VM resources are released.
5084 
5085     // note: perfMemory_exit_helper atexit function may be removed in
5086     // the future if the appropriate cleanup code can be added to the
5087     // VM_Exit VMOperation's doit method.
5088     if (atexit(perfMemory_exit_helper) != 0) {
5089       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5090     }
5091   }
5092 
5093   // initialize thread priority policy
5094   prio_init();
5095 
5096   if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5097     set_coredump_filter(false /*largepages*/, true /*dax_shared*/);
5098   }
5099   return JNI_OK;
5100 }
5101 
5102 // Mark the polling page as unreadable
5103 void os::make_polling_page_unreadable(void) {
5104   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5105     fatal("Could not disable polling page");
5106   }
5107 }
5108 
5109 // Mark the polling page as readable
5110 void os::make_polling_page_readable(void) {
5111   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5112     fatal("Could not enable polling page");
5113   }
5114 }
5115 
5116 // older glibc versions don't have this macro (which expands to
5117 // an optimized bit-counting function) so we have to roll our own
5118 #ifndef CPU_COUNT
5119 
5120 static int _cpu_count(const cpu_set_t* cpus) {
5121   int count = 0;
5122   // only look up to the number of configured processors
5123   for (int i = 0; i < os::processor_count(); i++) {
5124     if (CPU_ISSET(i, cpus)) {
5125       count++;
5126     }
5127   }
5128   return count;
5129 }
5130 
5131 #define CPU_COUNT(cpus) _cpu_count(cpus)
5132 
5133 #endif // CPU_COUNT
5134 
5135 // Get the current number of available processors for this process.
5136 // This value can change at any time during a process's lifetime.
5137 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5138 // If it appears there may be more than 1024 processors then we do a
5139 // dynamic check - see 6515172 for details.
5140 // If anything goes wrong we fallback to returning the number of online
5141 // processors - which can be greater than the number available to the process.
5142 int os::Linux::active_processor_count() {
5143   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
5144   cpu_set_t* cpus_p = &cpus;
5145   int cpus_size = sizeof(cpu_set_t);
5146 
5147   int configured_cpus = os::processor_count();  // upper bound on available cpus
5148   int cpu_count = 0;
5149 
5150 // old build platforms may not support dynamic cpu sets
5151 #ifdef CPU_ALLOC
5152 
5153   // To enable easy testing of the dynamic path on different platforms we
5154   // introduce a diagnostic flag: UseCpuAllocPath
5155   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5156     // kernel may use a mask bigger than cpu_set_t
5157     log_trace(os)("active_processor_count: using dynamic path %s"
5158                   "- configured processors: %d",
5159                   UseCpuAllocPath ? "(forced) " : "",
5160                   configured_cpus);
5161     cpus_p = CPU_ALLOC(configured_cpus);
5162     if (cpus_p != NULL) {
5163       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5164       // zero it just to be safe
5165       CPU_ZERO_S(cpus_size, cpus_p);
5166     }
5167     else {
5168        // failed to allocate so fallback to online cpus
5169        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5170        log_trace(os)("active_processor_count: "
5171                      "CPU_ALLOC failed (%s) - using "
5172                      "online processor count: %d",
5173                      os::strerror(errno), online_cpus);
5174        return online_cpus;
5175     }
5176   }
5177   else {
5178     log_trace(os)("active_processor_count: using static path - configured processors: %d",
5179                   configured_cpus);
5180   }
5181 #else // CPU_ALLOC
5182 // these stubs won't be executed
5183 #define CPU_COUNT_S(size, cpus) -1
5184 #define CPU_FREE(cpus)
5185 
5186   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5187                 configured_cpus);
5188 #endif // CPU_ALLOC
5189 
5190   // pid 0 means the current thread - which we have to assume represents the process
5191   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5192     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5193       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5194     }
5195     else {
5196       cpu_count = CPU_COUNT(cpus_p);
5197     }
5198     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5199   }
5200   else {
5201     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5202     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5203             "which may exceed available processors", os::strerror(errno), cpu_count);
5204   }
5205 
5206   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5207     CPU_FREE(cpus_p);
5208   }
5209 
5210   assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5211   return cpu_count;
5212 }
5213 
5214 // Determine the active processor count from one of
5215 // three different sources:
5216 //
5217 // 1. User option -XX:ActiveProcessorCount
5218 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5219 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5220 //
5221 // Option 1, if specified, will always override.
5222 // If the cgroup subsystem is active and configured, we
5223 // will return the min of the cgroup and option 2 results.
5224 // This is required since tools, such as numactl, that
5225 // alter cpu affinity do not update cgroup subsystem
5226 // cpuset configuration files.
5227 int os::active_processor_count() {
5228   // User has overridden the number of active processors
5229   if (ActiveProcessorCount > 0) {
5230     log_trace(os)("active_processor_count: "
5231                   "active processor count set by user : %d",
5232                   ActiveProcessorCount);
5233     return ActiveProcessorCount;
5234   }
5235 
5236   int active_cpus;
5237   if (OSContainer::is_containerized()) {
5238     active_cpus = OSContainer::active_processor_count();
5239     log_trace(os)("active_processor_count: determined by OSContainer: %d",
5240                    active_cpus);
5241   } else {
5242     active_cpus = os::Linux::active_processor_count();
5243   }
5244 
5245   return active_cpus;
5246 }
5247 
5248 void os::set_native_thread_name(const char *name) {
5249   if (Linux::_pthread_setname_np) {
5250     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5251     snprintf(buf, sizeof(buf), "%s", name);
5252     buf[sizeof(buf) - 1] = '\0';
5253     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5254     // ERANGE should not happen; all other errors should just be ignored.
5255     assert(rc != ERANGE, "pthread_setname_np failed");
5256   }
5257 }
5258 
5259 bool os::distribute_processes(uint length, uint* distribution) {
5260   // Not yet implemented.
5261   return false;
5262 }
5263 
5264 bool os::bind_to_processor(uint processor_id) {
5265   // Not yet implemented.
5266   return false;
5267 }
5268 
5269 ///
5270 
5271 void os::SuspendedThreadTask::internal_do_task() {
5272   if (do_suspend(_thread->osthread())) {
5273     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5274     do_task(context);
5275     do_resume(_thread->osthread());
5276   }
5277 }
5278 
5279 ////////////////////////////////////////////////////////////////////////////////
5280 // debug support
5281 
5282 bool os::find(address addr, outputStream* st) {
5283   Dl_info dlinfo;
5284   memset(&dlinfo, 0, sizeof(dlinfo));
5285   if (dladdr(addr, &dlinfo) != 0) {
5286     st->print(PTR_FORMAT ": ", p2i(addr));
5287     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5288       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5289                 p2i(addr) - p2i(dlinfo.dli_saddr));
5290     } else if (dlinfo.dli_fbase != NULL) {
5291       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5292     } else {
5293       st->print("<absolute address>");
5294     }
5295     if (dlinfo.dli_fname != NULL) {
5296       st->print(" in %s", dlinfo.dli_fname);
5297     }
5298     if (dlinfo.dli_fbase != NULL) {
5299       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5300     }
5301     st->cr();
5302 
5303     if (Verbose) {
5304       // decode some bytes around the PC
5305       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5306       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5307       address       lowest = (address) dlinfo.dli_sname;
5308       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5309       if (begin < lowest)  begin = lowest;
5310       Dl_info dlinfo2;
5311       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5312           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5313         end = (address) dlinfo2.dli_saddr;
5314       }
5315       Disassembler::decode(begin, end, st);
5316     }
5317     return true;
5318   }
5319   return false;
5320 }
5321 
5322 ////////////////////////////////////////////////////////////////////////////////
5323 // misc
5324 
5325 // This does not do anything on Linux. This is basically a hook for being
5326 // able to use structured exception handling (thread-local exception filters)
5327 // on, e.g., Win32.
5328 void
5329 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5330                          JavaCallArguments* args, Thread* thread) {
5331   f(value, method, args, thread);
5332 }
5333 
5334 void os::print_statistics() {
5335 }
5336 
5337 bool os::message_box(const char* title, const char* message) {
5338   int i;
5339   fdStream err(defaultStream::error_fd());
5340   for (i = 0; i < 78; i++) err.print_raw("=");
5341   err.cr();
5342   err.print_raw_cr(title);
5343   for (i = 0; i < 78; i++) err.print_raw("-");
5344   err.cr();
5345   err.print_raw_cr(message);
5346   for (i = 0; i < 78; i++) err.print_raw("=");
5347   err.cr();
5348 
5349   char buf[16];
5350   // Prevent process from exiting upon "read error" without consuming all CPU
5351   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5352 
5353   return buf[0] == 'y' || buf[0] == 'Y';
5354 }
5355 
5356 int os::stat(const char *path, struct stat *sbuf) {
5357   char pathbuf[MAX_PATH];
5358   if (strlen(path) > MAX_PATH - 1) {
5359     errno = ENAMETOOLONG;
5360     return -1;
5361   }
5362   os::native_path(strcpy(pathbuf, path));
5363   return ::stat(pathbuf, sbuf);
5364 }
5365 
5366 // Is a (classpath) directory empty?
5367 bool os::dir_is_empty(const char* path) {
5368   DIR *dir = NULL;
5369   struct dirent *ptr;
5370 
5371   dir = opendir(path);
5372   if (dir == NULL) return true;
5373 
5374   // Scan the directory
5375   bool result = true;
5376   char buf[sizeof(struct dirent) + MAX_PATH];
5377   while (result && (ptr = ::readdir(dir)) != NULL) {
5378     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5379       result = false;
5380     }
5381   }
5382   closedir(dir);
5383   return result;
5384 }
5385 
5386 // This code originates from JDK's sysOpen and open64_w
5387 // from src/solaris/hpi/src/system_md.c
5388 
5389 int os::open(const char *path, int oflag, int mode) {
5390   if (strlen(path) > MAX_PATH - 1) {
5391     errno = ENAMETOOLONG;
5392     return -1;
5393   }
5394 
5395   // All file descriptors that are opened in the Java process and not
5396   // specifically destined for a subprocess should have the close-on-exec
5397   // flag set.  If we don't set it, then careless 3rd party native code
5398   // might fork and exec without closing all appropriate file descriptors
5399   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5400   // turn might:
5401   //
5402   // - cause end-of-file to fail to be detected on some file
5403   //   descriptors, resulting in mysterious hangs, or
5404   //
5405   // - might cause an fopen in the subprocess to fail on a system
5406   //   suffering from bug 1085341.
5407   //
5408   // (Yes, the default setting of the close-on-exec flag is a Unix
5409   // design flaw)
5410   //
5411   // See:
5412   // 1085341: 32-bit stdio routines should support file descriptors >255
5413   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5414   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5415   //
5416   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5417   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5418   // because it saves a system call and removes a small window where the flag
5419   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5420   // and we fall back to using FD_CLOEXEC (see below).
5421 #ifdef O_CLOEXEC
5422   oflag |= O_CLOEXEC;
5423 #endif
5424 
5425   int fd = ::open64(path, oflag, mode);
5426   if (fd == -1) return -1;
5427 
5428   //If the open succeeded, the file might still be a directory
5429   {
5430     struct stat64 buf64;
5431     int ret = ::fstat64(fd, &buf64);
5432     int st_mode = buf64.st_mode;
5433 
5434     if (ret != -1) {
5435       if ((st_mode & S_IFMT) == S_IFDIR) {
5436         errno = EISDIR;
5437         ::close(fd);
5438         return -1;
5439       }
5440     } else {
5441       ::close(fd);
5442       return -1;
5443     }
5444   }
5445 
5446 #ifdef FD_CLOEXEC
5447   // Validate that the use of the O_CLOEXEC flag on open above worked.
5448   // With recent kernels, we will perform this check exactly once.
5449   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5450   if (!O_CLOEXEC_is_known_to_work) {
5451     int flags = ::fcntl(fd, F_GETFD);
5452     if (flags != -1) {
5453       if ((flags & FD_CLOEXEC) != 0)
5454         O_CLOEXEC_is_known_to_work = 1;
5455       else
5456         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5457     }
5458   }
5459 #endif
5460 
5461   return fd;
5462 }
5463 
5464 
5465 // create binary file, rewriting existing file if required
5466 int os::create_binary_file(const char* path, bool rewrite_existing) {
5467   int oflags = O_WRONLY | O_CREAT;
5468   if (!rewrite_existing) {
5469     oflags |= O_EXCL;
5470   }
5471   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5472 }
5473 
5474 // return current position of file pointer
5475 jlong os::current_file_offset(int fd) {
5476   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5477 }
5478 
5479 // move file pointer to the specified offset
5480 jlong os::seek_to_file_offset(int fd, jlong offset) {
5481   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5482 }
5483 
5484 // This code originates from JDK's sysAvailable
5485 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5486 
5487 int os::available(int fd, jlong *bytes) {
5488   jlong cur, end;
5489   int mode;
5490   struct stat64 buf64;
5491 
5492   if (::fstat64(fd, &buf64) >= 0) {
5493     mode = buf64.st_mode;
5494     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5495       int n;
5496       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5497         *bytes = n;
5498         return 1;
5499       }
5500     }
5501   }
5502   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5503     return 0;
5504   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5505     return 0;
5506   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5507     return 0;
5508   }
5509   *bytes = end - cur;
5510   return 1;
5511 }
5512 
5513 // Map a block of memory.
5514 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5515                         char *addr, size_t bytes, bool read_only,
5516                         bool allow_exec) {
5517   int prot;
5518   int flags = MAP_PRIVATE;
5519 
5520   if (read_only) {
5521     prot = PROT_READ;
5522   } else {
5523     prot = PROT_READ | PROT_WRITE;
5524   }
5525 
5526   if (allow_exec) {
5527     prot |= PROT_EXEC;
5528   }
5529 
5530   if (addr != NULL) {
5531     flags |= MAP_FIXED;
5532   }
5533 
5534   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5535                                      fd, file_offset);
5536   if (mapped_address == MAP_FAILED) {
5537     return NULL;
5538   }
5539   return mapped_address;
5540 }
5541 
5542 
5543 // Remap a block of memory.
5544 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5545                           char *addr, size_t bytes, bool read_only,
5546                           bool allow_exec) {
5547   // same as map_memory() on this OS
5548   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5549                         allow_exec);
5550 }
5551 
5552 
5553 // Unmap a block of memory.
5554 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5555   return munmap(addr, bytes) == 0;
5556 }
5557 
5558 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5559 
5560 static clockid_t thread_cpu_clockid(Thread* thread) {
5561   pthread_t tid = thread->osthread()->pthread_id();
5562   clockid_t clockid;
5563 
5564   // Get thread clockid
5565   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5566   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5567   return clockid;
5568 }
5569 
5570 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5571 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5572 // of a thread.
5573 //
5574 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5575 // the fast estimate available on the platform.
5576 
5577 jlong os::current_thread_cpu_time() {
5578   if (os::Linux::supports_fast_thread_cpu_time()) {
5579     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5580   } else {
5581     // return user + sys since the cost is the same
5582     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5583   }
5584 }
5585 
5586 jlong os::thread_cpu_time(Thread* thread) {
5587   // consistent with what current_thread_cpu_time() returns
5588   if (os::Linux::supports_fast_thread_cpu_time()) {
5589     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5590   } else {
5591     return slow_thread_cpu_time(thread, true /* user + sys */);
5592   }
5593 }
5594 
5595 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5596   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5597     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5598   } else {
5599     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5600   }
5601 }
5602 
5603 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5604   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5605     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5606   } else {
5607     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5608   }
5609 }
5610 
5611 //  -1 on error.
5612 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5613   pid_t  tid = thread->osthread()->thread_id();
5614   char *s;
5615   char stat[2048];
5616   int statlen;
5617   char proc_name[64];
5618   int count;
5619   long sys_time, user_time;
5620   char cdummy;
5621   int idummy;
5622   long ldummy;
5623   FILE *fp;
5624 
5625   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5626   fp = fopen(proc_name, "r");
5627   if (fp == NULL) return -1;
5628   statlen = fread(stat, 1, 2047, fp);
5629   stat[statlen] = '\0';
5630   fclose(fp);
5631 
5632   // Skip pid and the command string. Note that we could be dealing with
5633   // weird command names, e.g. user could decide to rename java launcher
5634   // to "java 1.4.2 :)", then the stat file would look like
5635   //                1234 (java 1.4.2 :)) R ... ...
5636   // We don't really need to know the command string, just find the last
5637   // occurrence of ")" and then start parsing from there. See bug 4726580.
5638   s = strrchr(stat, ')');
5639   if (s == NULL) return -1;
5640 
5641   // Skip blank chars
5642   do { s++; } while (s && isspace(*s));
5643 
5644   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5645                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5646                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5647                  &user_time, &sys_time);
5648   if (count != 13) return -1;
5649   if (user_sys_cpu_time) {
5650     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5651   } else {
5652     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5653   }
5654 }
5655 
5656 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5657   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5658   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5659   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5660   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5661 }
5662 
5663 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5664   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5665   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5666   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5667   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5668 }
5669 
5670 bool os::is_thread_cpu_time_supported() {
5671   return true;
5672 }
5673 
5674 // System loadavg support.  Returns -1 if load average cannot be obtained.
5675 // Linux doesn't yet have a (official) notion of processor sets,
5676 // so just return the system wide load average.
5677 int os::loadavg(double loadavg[], int nelem) {
5678   return ::getloadavg(loadavg, nelem);
5679 }
5680 
5681 void os::pause() {
5682   char filename[MAX_PATH];
5683   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5684     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5685   } else {
5686     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5687   }
5688 
5689   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5690   if (fd != -1) {
5691     struct stat buf;
5692     ::close(fd);
5693     while (::stat(filename, &buf) == 0) {
5694       (void)::poll(NULL, 0, 100);
5695     }
5696   } else {
5697     jio_fprintf(stderr,
5698                 "Could not open pause file '%s', continuing immediately.\n", filename);
5699   }
5700 }
5701 
5702 extern char** environ;
5703 
5704 // Run the specified command in a separate process. Return its exit value,
5705 // or -1 on failure (e.g. can't fork a new process).
5706 // Unlike system(), this function can be called from signal handler. It
5707 // doesn't block SIGINT et al.
5708 int os::fork_and_exec(char* cmd) {
5709   const char * argv[4] = {"sh", "-c", cmd, NULL};
5710 
5711   pid_t pid = fork();
5712 
5713   if (pid < 0) {
5714     // fork failed
5715     return -1;
5716 
5717   } else if (pid == 0) {
5718     // child process
5719 
5720     execve("/bin/sh", (char* const*)argv, environ);
5721 
5722     // execve failed
5723     _exit(-1);
5724 
5725   } else  {
5726     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5727     // care about the actual exit code, for now.
5728 
5729     int status;
5730 
5731     // Wait for the child process to exit.  This returns immediately if
5732     // the child has already exited. */
5733     while (waitpid(pid, &status, 0) < 0) {
5734       switch (errno) {
5735       case ECHILD: return 0;
5736       case EINTR: break;
5737       default: return -1;
5738       }
5739     }
5740 
5741     if (WIFEXITED(status)) {
5742       // The child exited normally; get its exit code.
5743       return WEXITSTATUS(status);
5744     } else if (WIFSIGNALED(status)) {
5745       // The child exited because of a signal
5746       // The best value to return is 0x80 + signal number,
5747       // because that is what all Unix shells do, and because
5748       // it allows callers to distinguish between process exit and
5749       // process death by signal.
5750       return 0x80 + WTERMSIG(status);
5751     } else {
5752       // Unknown exit code; pass it through
5753       return status;
5754     }
5755   }
5756 }
5757 
5758 // Get the default path to the core file
5759 // Returns the length of the string
5760 int os::get_core_path(char* buffer, size_t bufferSize) {
5761   /*
5762    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5763    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5764    */
5765   const int core_pattern_len = 129;
5766   char core_pattern[core_pattern_len] = {0};
5767 
5768   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5769   if (core_pattern_file == -1) {
5770     return -1;
5771   }
5772 
5773   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5774   ::close(core_pattern_file);
5775   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5776     return -1;
5777   }
5778   if (core_pattern[ret-1] == '\n') {
5779     core_pattern[ret-1] = '\0';
5780   } else {
5781     core_pattern[ret] = '\0';
5782   }
5783 
5784   char *pid_pos = strstr(core_pattern, "%p");
5785   int written;
5786 
5787   if (core_pattern[0] == '/') {
5788     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5789   } else {
5790     char cwd[PATH_MAX];
5791 
5792     const char* p = get_current_directory(cwd, PATH_MAX);
5793     if (p == NULL) {
5794       return -1;
5795     }
5796 
5797     if (core_pattern[0] == '|') {
5798       written = jio_snprintf(buffer, bufferSize,
5799                              "\"%s\" (or dumping to %s/core.%d)",
5800                              &core_pattern[1], p, current_process_id());
5801     } else {
5802       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5803     }
5804   }
5805 
5806   if (written < 0) {
5807     return -1;
5808   }
5809 
5810   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5811     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5812 
5813     if (core_uses_pid_file != -1) {
5814       char core_uses_pid = 0;
5815       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5816       ::close(core_uses_pid_file);
5817 
5818       if (core_uses_pid == '1') {
5819         jio_snprintf(buffer + written, bufferSize - written,
5820                                           ".%d", current_process_id());
5821       }
5822     }
5823   }
5824 
5825   return strlen(buffer);
5826 }
5827 
5828 bool os::start_debugging(char *buf, int buflen) {
5829   int len = (int)strlen(buf);
5830   char *p = &buf[len];
5831 
5832   jio_snprintf(p, buflen-len,
5833                "\n\n"
5834                "Do you want to debug the problem?\n\n"
5835                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5836                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5837                "Otherwise, press RETURN to abort...",
5838                os::current_process_id(), os::current_process_id(),
5839                os::current_thread_id(), os::current_thread_id());
5840 
5841   bool yes = os::message_box("Unexpected Error", buf);
5842 
5843   if (yes) {
5844     // yes, user asked VM to launch debugger
5845     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5846                  os::current_process_id(), os::current_process_id());
5847 
5848     os::fork_and_exec(buf);
5849     yes = false;
5850   }
5851   return yes;
5852 }
5853 
5854 
5855 // Java/Compiler thread:
5856 //
5857 //   Low memory addresses
5858 // P0 +------------------------+
5859 //    |                        |\  Java thread created by VM does not have glibc
5860 //    |    glibc guard page    | - guard page, attached Java thread usually has
5861 //    |                        |/  1 glibc guard page.
5862 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5863 //    |                        |\
5864 //    |  HotSpot Guard Pages   | - red, yellow and reserved pages
5865 //    |                        |/
5866 //    +------------------------+ JavaThread::stack_reserved_zone_base()
5867 //    |                        |\
5868 //    |      Normal Stack      | -
5869 //    |                        |/
5870 // P2 +------------------------+ Thread::stack_base()
5871 //
5872 // Non-Java thread:
5873 //
5874 //   Low memory addresses
5875 // P0 +------------------------+
5876 //    |                        |\
5877 //    |  glibc guard page      | - usually 1 page
5878 //    |                        |/
5879 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5880 //    |                        |\
5881 //    |      Normal Stack      | -
5882 //    |                        |/
5883 // P2 +------------------------+ Thread::stack_base()
5884 //
5885 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5886 //    returned from pthread_attr_getstack().
5887 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5888 //    of the stack size given in pthread_attr. We work around this for
5889 //    threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5890 //
5891 #ifndef ZERO
5892 static void current_stack_region(address * bottom, size_t * size) {
5893   if (os::is_primordial_thread()) {
5894     // primordial thread needs special handling because pthread_getattr_np()
5895     // may return bogus value.
5896     *bottom = os::Linux::initial_thread_stack_bottom();
5897     *size   = os::Linux::initial_thread_stack_size();
5898   } else {
5899     pthread_attr_t attr;
5900 
5901     int rslt = pthread_getattr_np(pthread_self(), &attr);
5902 
5903     // JVM needs to know exact stack location, abort if it fails
5904     if (rslt != 0) {
5905       if (rslt == ENOMEM) {
5906         vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5907       } else {
5908         fatal("pthread_getattr_np failed with error = %d", rslt);
5909       }
5910     }
5911 
5912     if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5913       fatal("Cannot locate current stack attributes!");
5914     }
5915 
5916     // Work around NPTL stack guard error.
5917     size_t guard_size = 0;
5918     rslt = pthread_attr_getguardsize(&attr, &guard_size);
5919     if (rslt != 0) {
5920       fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5921     }
5922     *bottom += guard_size;
5923     *size   -= guard_size;
5924 
5925     pthread_attr_destroy(&attr);
5926 
5927   }
5928   assert(os::current_stack_pointer() >= *bottom &&
5929          os::current_stack_pointer() < *bottom + *size, "just checking");
5930 }
5931 
5932 address os::current_stack_base() {
5933   address bottom;
5934   size_t size;
5935   current_stack_region(&bottom, &size);
5936   return (bottom + size);
5937 }
5938 
5939 size_t os::current_stack_size() {
5940   // This stack size includes the usable stack and HotSpot guard pages
5941   // (for the threads that have Hotspot guard pages).
5942   address bottom;
5943   size_t size;
5944   current_stack_region(&bottom, &size);
5945   return size;
5946 }
5947 #endif
5948 
5949 static inline struct timespec get_mtime(const char* filename) {
5950   struct stat st;
5951   int ret = os::stat(filename, &st);
5952   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5953   return st.st_mtim;
5954 }
5955 
5956 int os::compare_file_modified_times(const char* file1, const char* file2) {
5957   struct timespec filetime1 = get_mtime(file1);
5958   struct timespec filetime2 = get_mtime(file2);
5959   int diff = filetime1.tv_sec - filetime2.tv_sec;
5960   if (diff == 0) {
5961     return filetime1.tv_nsec - filetime2.tv_nsec;
5962   }
5963   return diff;
5964 }
5965 
5966 /////////////// Unit tests ///////////////
5967 
5968 #ifndef PRODUCT
5969 
5970 #define test_log(...)              \
5971   do {                             \
5972     if (VerboseInternalVMTests) {  \
5973       tty->print_cr(__VA_ARGS__);  \
5974       tty->flush();                \
5975     }                              \
5976   } while (false)
5977 
5978 class TestReserveMemorySpecial : AllStatic {
5979  public:
5980   static void small_page_write(void* addr, size_t size) {
5981     size_t page_size = os::vm_page_size();
5982 
5983     char* end = (char*)addr + size;
5984     for (char* p = (char*)addr; p < end; p += page_size) {
5985       *p = 1;
5986     }
5987   }
5988 
5989   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5990     if (!UseHugeTLBFS) {
5991       return;
5992     }
5993 
5994     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5995 
5996     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5997 
5998     if (addr != NULL) {
5999       small_page_write(addr, size);
6000 
6001       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6002     }
6003   }
6004 
6005   static void test_reserve_memory_special_huge_tlbfs_only() {
6006     if (!UseHugeTLBFS) {
6007       return;
6008     }
6009 
6010     size_t lp = os::large_page_size();
6011 
6012     for (size_t size = lp; size <= lp * 10; size += lp) {
6013       test_reserve_memory_special_huge_tlbfs_only(size);
6014     }
6015   }
6016 
6017   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6018     size_t lp = os::large_page_size();
6019     size_t ag = os::vm_allocation_granularity();
6020 
6021     // sizes to test
6022     const size_t sizes[] = {
6023       lp, lp + ag, lp + lp / 2, lp * 2,
6024       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6025       lp * 10, lp * 10 + lp / 2
6026     };
6027     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6028 
6029     // For each size/alignment combination, we test three scenarios:
6030     // 1) with req_addr == NULL
6031     // 2) with a non-null req_addr at which we expect to successfully allocate
6032     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6033     //    expect the allocation to either fail or to ignore req_addr
6034 
6035     // Pre-allocate two areas; they shall be as large as the largest allocation
6036     //  and aligned to the largest alignment we will be testing.
6037     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6038     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6039       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6040       -1, 0);
6041     assert(mapping1 != MAP_FAILED, "should work");
6042 
6043     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6044       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6045       -1, 0);
6046     assert(mapping2 != MAP_FAILED, "should work");
6047 
6048     // Unmap the first mapping, but leave the second mapping intact: the first
6049     // mapping will serve as a value for a "good" req_addr (case 2). The second
6050     // mapping, still intact, as "bad" req_addr (case 3).
6051     ::munmap(mapping1, mapping_size);
6052 
6053     // Case 1
6054     test_log("%s, req_addr NULL:", __FUNCTION__);
6055     test_log("size            align           result");
6056 
6057     for (int i = 0; i < num_sizes; i++) {
6058       const size_t size = sizes[i];
6059       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6060         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6061         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6062                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6063         if (p != NULL) {
6064           assert(is_aligned(p, alignment), "must be");
6065           small_page_write(p, size);
6066           os::Linux::release_memory_special_huge_tlbfs(p, size);
6067         }
6068       }
6069     }
6070 
6071     // Case 2
6072     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6073     test_log("size            align           req_addr         result");
6074 
6075     for (int i = 0; i < num_sizes; i++) {
6076       const size_t size = sizes[i];
6077       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6078         char* const req_addr = align_up(mapping1, alignment);
6079         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6080         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6081                  size, alignment, p2i(req_addr), p2i(p),
6082                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6083         if (p != NULL) {
6084           assert(p == req_addr, "must be");
6085           small_page_write(p, size);
6086           os::Linux::release_memory_special_huge_tlbfs(p, size);
6087         }
6088       }
6089     }
6090 
6091     // Case 3
6092     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6093     test_log("size            align           req_addr         result");
6094 
6095     for (int i = 0; i < num_sizes; i++) {
6096       const size_t size = sizes[i];
6097       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6098         char* const req_addr = align_up(mapping2, alignment);
6099         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6100         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6101                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6102         // as the area around req_addr contains already existing mappings, the API should always
6103         // return NULL (as per contract, it cannot return another address)
6104         assert(p == NULL, "must be");
6105       }
6106     }
6107 
6108     ::munmap(mapping2, mapping_size);
6109 
6110   }
6111 
6112   static void test_reserve_memory_special_huge_tlbfs() {
6113     if (!UseHugeTLBFS) {
6114       return;
6115     }
6116 
6117     test_reserve_memory_special_huge_tlbfs_only();
6118     test_reserve_memory_special_huge_tlbfs_mixed();
6119   }
6120 
6121   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6122     if (!UseSHM) {
6123       return;
6124     }
6125 
6126     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6127 
6128     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6129 
6130     if (addr != NULL) {
6131       assert(is_aligned(addr, alignment), "Check");
6132       assert(is_aligned(addr, os::large_page_size()), "Check");
6133 
6134       small_page_write(addr, size);
6135 
6136       os::Linux::release_memory_special_shm(addr, size);
6137     }
6138   }
6139 
6140   static void test_reserve_memory_special_shm() {
6141     size_t lp = os::large_page_size();
6142     size_t ag = os::vm_allocation_granularity();
6143 
6144     for (size_t size = ag; size < lp * 3; size += ag) {
6145       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6146         test_reserve_memory_special_shm(size, alignment);
6147       }
6148     }
6149   }
6150 
6151   static void test() {
6152     test_reserve_memory_special_huge_tlbfs();
6153     test_reserve_memory_special_shm();
6154   }
6155 };
6156 
6157 void TestReserveMemorySpecial_test() {
6158   TestReserveMemorySpecial::test();
6159 }
6160 
6161 #endif