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.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 void os::Linux::signal_sets_init() {
 429   // Should also have an assertion stating we are still single-threaded.
 430   assert(!signal_sets_initialized, "Already initialized");
 431   // Fill in signals that are necessarily unblocked for all threads in
 432   // the VM. Currently, we unblock the following signals:
 433   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 434   //                         by -Xrs (=ReduceSignalUsage));
 435   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 436   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 437   // the dispositions or masks wrt these signals.
 438   // Programs embedding the VM that want to use the above signals for their
 439   // own purposes must, at this time, use the "-Xrs" option to prevent
 440   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 441   // (See bug 4345157, and other related bugs).
 442   // In reality, though, unblocking these signals is really a nop, since
 443   // these signals are not blocked by default.
 444   sigemptyset(&unblocked_sigs);
 445   sigaddset(&unblocked_sigs, SIGILL);
 446   sigaddset(&unblocked_sigs, SIGSEGV);
 447   sigaddset(&unblocked_sigs, SIGBUS);
 448   sigaddset(&unblocked_sigs, SIGFPE);
 449 #if defined(PPC64)
 450   sigaddset(&unblocked_sigs, SIGTRAP);
 451 #endif
 452   sigaddset(&unblocked_sigs, SR_signum);
 453 
 454   if (!ReduceSignalUsage) {
 455     if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 456       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 457     }
 458     if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 459       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 460     }
 461     if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 462       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 463     }
 464   }
 465   // Fill in signals that are blocked by all but the VM thread.
 466   sigemptyset(&vm_sigs);
 467   if (!ReduceSignalUsage) {
 468     sigaddset(&vm_sigs, BREAK_SIGNAL);
 469   }
 470   debug_only(signal_sets_initialized = true);
 471 
 472 }
 473 
 474 // These are signals that are unblocked while a thread is running Java.
 475 // (For some reason, they get blocked by default.)
 476 sigset_t* os::Linux::unblocked_signals() {
 477   assert(signal_sets_initialized, "Not initialized");
 478   return &unblocked_sigs;
 479 }
 480 
 481 // These are the signals that are blocked while a (non-VM) thread is
 482 // running Java. Only the VM thread handles these signals.
 483 sigset_t* os::Linux::vm_signals() {
 484   assert(signal_sets_initialized, "Not initialized");
 485   return &vm_sigs;
 486 }
 487 
 488 void os::Linux::hotspot_sigmask(Thread* thread) {
 489 
 490   //Save caller's signal mask before setting VM signal mask
 491   sigset_t caller_sigmask;
 492   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 493 
 494   OSThread* osthread = thread->osthread();
 495   osthread->set_caller_sigmask(caller_sigmask);
 496 
 497   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 498 
 499   if (!ReduceSignalUsage) {
 500     if (thread->is_VM_thread()) {
 501       // Only the VM thread handles BREAK_SIGNAL ...
 502       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 503     } else {
 504       // ... all other threads block BREAK_SIGNAL
 505       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 506     }
 507   }
 508 }
 509 
 510 //////////////////////////////////////////////////////////////////////////////
 511 // detecting pthread library
 512 
 513 void os::Linux::libpthread_init() {
 514   // Save glibc and pthread version strings.
 515 #if !defined(_CS_GNU_LIBC_VERSION) || \
 516     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 517   #error "glibc too old (< 2.3.2)"
 518 #endif
 519 
 520   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 521   assert(n > 0, "cannot retrieve glibc version");
 522   char *str = (char *)malloc(n, mtInternal);
 523   confstr(_CS_GNU_LIBC_VERSION, str, n);
 524   os::Linux::set_glibc_version(str);
 525 
 526   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 527   assert(n > 0, "cannot retrieve pthread version");
 528   str = (char *)malloc(n, mtInternal);
 529   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 530   os::Linux::set_libpthread_version(str);
 531 }
 532 
 533 /////////////////////////////////////////////////////////////////////////////
 534 // thread stack expansion
 535 
 536 // os::Linux::manually_expand_stack() takes care of expanding the thread
 537 // stack. Note that this is normally not needed: pthread stacks allocate
 538 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 539 // committed. Therefore it is not necessary to expand the stack manually.
 540 //
 541 // Manually expanding the stack was historically needed on LinuxThreads
 542 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 543 // it is kept to deal with very rare corner cases:
 544 //
 545 // For one, user may run the VM on an own implementation of threads
 546 // whose stacks are - like the old LinuxThreads - implemented using
 547 // mmap(MAP_GROWSDOWN).
 548 //
 549 // Also, this coding may be needed if the VM is running on the primordial
 550 // thread. Normally we avoid running on the primordial thread; however,
 551 // user may still invoke the VM on the primordial thread.
 552 //
 553 // The following historical comment describes the details about running
 554 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 555 
 556 
 557 // Force Linux kernel to expand current thread stack. If "bottom" is close
 558 // to the stack guard, caller should block all signals.
 559 //
 560 // MAP_GROWSDOWN:
 561 //   A special mmap() flag that is used to implement thread stacks. It tells
 562 //   kernel that the memory region should extend downwards when needed. This
 563 //   allows early versions of LinuxThreads to only mmap the first few pages
 564 //   when creating a new thread. Linux kernel will automatically expand thread
 565 //   stack as needed (on page faults).
 566 //
 567 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 568 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 569 //   region, it's hard to tell if the fault is due to a legitimate stack
 570 //   access or because of reading/writing non-exist memory (e.g. buffer
 571 //   overrun). As a rule, if the fault happens below current stack pointer,
 572 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 573 //   application (see Linux kernel fault.c).
 574 //
 575 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 576 //   stack overflow detection.
 577 //
 578 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 579 //   not use MAP_GROWSDOWN.
 580 //
 581 // To get around the problem and allow stack banging on Linux, we need to
 582 // manually expand thread stack after receiving the SIGSEGV.
 583 //
 584 // There are two ways to expand thread stack to address "bottom", we used
 585 // both of them in JVM before 1.5:
 586 //   1. adjust stack pointer first so that it is below "bottom", and then
 587 //      touch "bottom"
 588 //   2. mmap() the page in question
 589 //
 590 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 591 // if current sp is already near the lower end of page 101, and we need to
 592 // call mmap() to map page 100, it is possible that part of the mmap() frame
 593 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 594 // That will destroy the mmap() frame and cause VM to crash.
 595 //
 596 // The following code works by adjusting sp first, then accessing the "bottom"
 597 // page to force a page fault. Linux kernel will then automatically expand the
 598 // stack mapping.
 599 //
 600 // _expand_stack_to() assumes its frame size is less than page size, which
 601 // should always be true if the function is not inlined.
 602 
 603 static void NOINLINE _expand_stack_to(address bottom) {
 604   address sp;
 605   size_t size;
 606   volatile char *p;
 607 
 608   // Adjust bottom to point to the largest address within the same page, it
 609   // gives us a one-page buffer if alloca() allocates slightly more memory.
 610   bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
 611   bottom += os::Linux::page_size() - 1;
 612 
 613   // sp might be slightly above current stack pointer; if that's the case, we
 614   // will alloca() a little more space than necessary, which is OK. Don't use
 615   // os::current_stack_pointer(), as its result can be slightly below current
 616   // stack pointer, causing us to not alloca enough to reach "bottom".
 617   sp = (address)&sp;
 618 
 619   if (sp > bottom) {
 620     size = sp - bottom;
 621     p = (volatile char *)alloca(size);
 622     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 623     p[0] = '\0';
 624   }
 625 }
 626 
 627 void os::Linux::expand_stack_to(address bottom) {
 628   _expand_stack_to(bottom);
 629 }
 630 
 631 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 632   assert(t!=NULL, "just checking");
 633   assert(t->osthread()->expanding_stack(), "expand should be set");
 634   assert(t->stack_base() != NULL, "stack_base was not initialized");
 635 
 636   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 637     sigset_t mask_all, old_sigset;
 638     sigfillset(&mask_all);
 639     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 640     _expand_stack_to(addr);
 641     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 642     return true;
 643   }
 644   return false;
 645 }
 646 
 647 //////////////////////////////////////////////////////////////////////////////
 648 // create new thread
 649 
 650 // Thread start routine for all newly created threads
 651 static void *thread_native_entry(Thread *thread) {
 652   // Try to randomize the cache line index of hot stack frames.
 653   // This helps when threads of the same stack traces evict each other's
 654   // cache lines. The threads can be either from the same JVM instance, or
 655   // from different JVM instances. The benefit is especially true for
 656   // processors with hyperthreading technology.
 657   static int counter = 0;
 658   int pid = os::current_process_id();
 659   alloca(((pid ^ counter++) & 7) * 128);
 660 
 661   thread->initialize_thread_current();
 662 
 663   OSThread* osthread = thread->osthread();
 664   Monitor* sync = osthread->startThread_lock();
 665 
 666   osthread->set_thread_id(os::current_thread_id());
 667 
 668   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 669     os::current_thread_id(), (uintx) pthread_self());
 670 
 671   if (UseNUMA) {
 672     int lgrp_id = os::numa_get_group_id();
 673     if (lgrp_id != -1) {
 674       thread->set_lgrp_id(lgrp_id);
 675     }
 676   }
 677   // initialize signal mask for this thread
 678   os::Linux::hotspot_sigmask(thread);
 679 
 680   // initialize floating point control register
 681   os::Linux::init_thread_fpu_state();
 682 
 683   // handshaking with parent thread
 684   {
 685     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 686 
 687     // notify parent thread
 688     osthread->set_state(INITIALIZED);
 689     sync->notify_all();
 690 
 691     // wait until os::start_thread()
 692     while (osthread->get_state() == INITIALIZED) {
 693       sync->wait(Mutex::_no_safepoint_check_flag);
 694     }
 695   }
 696 
 697   // call one more level start routine
 698   thread->run();
 699 
 700   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 701     os::current_thread_id(), (uintx) pthread_self());
 702 
 703   // If a thread has not deleted itself ("delete this") as part of its
 704   // termination sequence, we have to ensure thread-local-storage is
 705   // cleared before we actually terminate. No threads should ever be
 706   // deleted asynchronously with respect to their termination.
 707   if (Thread::current_or_null_safe() != NULL) {
 708     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 709     thread->clear_thread_current();
 710   }
 711 
 712   return 0;
 713 }
 714 
 715 bool os::create_thread(Thread* thread, ThreadType thr_type,
 716                        size_t req_stack_size) {
 717   assert(thread->osthread() == NULL, "caller responsible");
 718 
 719   // Allocate the OSThread object
 720   OSThread* osthread = new OSThread(NULL, NULL);
 721   if (osthread == NULL) {
 722     return false;
 723   }
 724 
 725   // set the correct thread state
 726   osthread->set_thread_type(thr_type);
 727 
 728   // Initial state is ALLOCATED but not INITIALIZED
 729   osthread->set_state(ALLOCATED);
 730 
 731   thread->set_osthread(osthread);
 732 
 733   // init thread attributes
 734   pthread_attr_t attr;
 735   pthread_attr_init(&attr);
 736   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 737 
 738   // Calculate stack size if it's not specified by caller.
 739   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 740   // In the Linux NPTL pthread implementation the guard size mechanism
 741   // is not implemented properly. The posix standard requires adding
 742   // the size of the guard pages to the stack size, instead Linux
 743   // takes the space out of 'stacksize'. Thus we adapt the requested
 744   // stack_size by the size of the guard pages to mimick proper
 745   // behaviour. However, be careful not to end up with a size
 746   // of zero due to overflow. Don't add the guard page in that case.
 747   size_t guard_size = os::Linux::default_guard_size(thr_type);
 748   if (stack_size <= SIZE_MAX - guard_size) {
 749     stack_size += guard_size;
 750   }
 751   assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
 752 
 753   int status = pthread_attr_setstacksize(&attr, stack_size);
 754   assert_status(status == 0, status, "pthread_attr_setstacksize");
 755 
 756   // Configure glibc guard page.
 757   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 758 
 759   ThreadState state;
 760 
 761   {
 762     pthread_t tid;
 763     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 764 
 765     char buf[64];
 766     if (ret == 0) {
 767       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 768         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 769     } else {
 770       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 771         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 772     }
 773 
 774     pthread_attr_destroy(&attr);
 775 
 776     if (ret != 0) {
 777       // Need to clean up stuff we've allocated so far
 778       thread->set_osthread(NULL);
 779       delete osthread;
 780       return false;
 781     }
 782 
 783     // Store pthread info into the OSThread
 784     osthread->set_pthread_id(tid);
 785 
 786     // Wait until child thread is either initialized or aborted
 787     {
 788       Monitor* sync_with_child = osthread->startThread_lock();
 789       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 790       while ((state = osthread->get_state()) == ALLOCATED) {
 791         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 792       }
 793     }
 794   }
 795 
 796   // Aborted due to thread limit being reached
 797   if (state == ZOMBIE) {
 798     thread->set_osthread(NULL);
 799     delete osthread;
 800     return false;
 801   }
 802 
 803   // The thread is returned suspended (in state INITIALIZED),
 804   // and is started higher up in the call chain
 805   assert(state == INITIALIZED, "race condition");
 806   return true;
 807 }
 808 
 809 /////////////////////////////////////////////////////////////////////////////
 810 // attach existing thread
 811 
 812 // bootstrap the main thread
 813 bool os::create_main_thread(JavaThread* thread) {
 814   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 815   return create_attached_thread(thread);
 816 }
 817 
 818 bool os::create_attached_thread(JavaThread* thread) {
 819 #ifdef ASSERT
 820   thread->verify_not_published();
 821 #endif
 822 
 823   // Allocate the OSThread object
 824   OSThread* osthread = new OSThread(NULL, NULL);
 825 
 826   if (osthread == NULL) {
 827     return false;
 828   }
 829 
 830   // Store pthread info into the OSThread
 831   osthread->set_thread_id(os::Linux::gettid());
 832   osthread->set_pthread_id(::pthread_self());
 833 
 834   // initialize floating point control register
 835   os::Linux::init_thread_fpu_state();
 836 
 837   // Initial thread state is RUNNABLE
 838   osthread->set_state(RUNNABLE);
 839 
 840   thread->set_osthread(osthread);
 841 
 842   if (UseNUMA) {
 843     int lgrp_id = os::numa_get_group_id();
 844     if (lgrp_id != -1) {
 845       thread->set_lgrp_id(lgrp_id);
 846     }
 847   }
 848 
 849   if (os::is_primordial_thread()) {
 850     // If current thread is primordial thread, its stack is mapped on demand,
 851     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 852     // the entire stack region to avoid SEGV in stack banging.
 853     // It is also useful to get around the heap-stack-gap problem on SuSE
 854     // kernel (see 4821821 for details). We first expand stack to the top
 855     // of yellow zone, then enable stack yellow zone (order is significant,
 856     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 857     // is no gap between the last two virtual memory regions.
 858 
 859     JavaThread *jt = (JavaThread *)thread;
 860     address addr = jt->stack_reserved_zone_base();
 861     assert(addr != NULL, "initialization problem?");
 862     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 863 
 864     osthread->set_expanding_stack();
 865     os::Linux::manually_expand_stack(jt, addr);
 866     osthread->clear_expanding_stack();
 867   }
 868 
 869   // initialize signal mask for this thread
 870   // and save the caller's signal mask
 871   os::Linux::hotspot_sigmask(thread);
 872 
 873   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 874     os::current_thread_id(), (uintx) pthread_self());
 875 
 876   return true;
 877 }
 878 
 879 void os::pd_start_thread(Thread* thread) {
 880   OSThread * osthread = thread->osthread();
 881   assert(osthread->get_state() != INITIALIZED, "just checking");
 882   Monitor* sync_with_child = osthread->startThread_lock();
 883   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 884   sync_with_child->notify();
 885 }
 886 
 887 // Free Linux resources related to the OSThread
 888 void os::free_thread(OSThread* osthread) {
 889   assert(osthread != NULL, "osthread not set");
 890 
 891   // We are told to free resources of the argument thread,
 892   // but we can only really operate on the current thread.
 893   assert(Thread::current()->osthread() == osthread,
 894          "os::free_thread but not current thread");
 895 
 896 #ifdef ASSERT
 897   sigset_t current;
 898   sigemptyset(&current);
 899   pthread_sigmask(SIG_SETMASK, NULL, &current);
 900   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 901 #endif
 902 
 903   // Restore caller's signal mask
 904   sigset_t sigmask = osthread->caller_sigmask();
 905   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 906 
 907   delete osthread;
 908 }
 909 
 910 //////////////////////////////////////////////////////////////////////////////
 911 // primordial thread
 912 
 913 // Check if current thread is the primordial thread, similar to Solaris thr_main.
 914 bool os::is_primordial_thread(void) {
 915   if (suppress_primordial_thread_resolution) {
 916     return false;
 917   }
 918   char dummy;
 919   // If called before init complete, thread stack bottom will be null.
 920   // Can be called if fatal error occurs before initialization.
 921   if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
 922   assert(os::Linux::initial_thread_stack_bottom() != NULL &&
 923          os::Linux::initial_thread_stack_size()   != 0,
 924          "os::init did not locate primordial thread's stack region");
 925   if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
 926       (address)&dummy < os::Linux::initial_thread_stack_bottom() +
 927                         os::Linux::initial_thread_stack_size()) {
 928     return true;
 929   } else {
 930     return false;
 931   }
 932 }
 933 
 934 // Find the virtual memory area that contains addr
 935 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 936   FILE *fp = fopen("/proc/self/maps", "r");
 937   if (fp) {
 938     address low, high;
 939     while (!feof(fp)) {
 940       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 941         if (low <= addr && addr < high) {
 942           if (vma_low)  *vma_low  = low;
 943           if (vma_high) *vma_high = high;
 944           fclose(fp);
 945           return true;
 946         }
 947       }
 948       for (;;) {
 949         int ch = fgetc(fp);
 950         if (ch == EOF || ch == (int)'\n') break;
 951       }
 952     }
 953     fclose(fp);
 954   }
 955   return false;
 956 }
 957 
 958 // Locate primordial thread stack. This special handling of primordial thread stack
 959 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 960 // bogus value for the primordial process thread. While the launcher has created
 961 // the VM in a new thread since JDK 6, we still have to allow for the use of the
 962 // JNI invocation API from a primordial thread.
 963 void os::Linux::capture_initial_stack(size_t max_size) {
 964 
 965   // max_size is either 0 (which means accept OS default for thread stacks) or
 966   // a user-specified value known to be at least the minimum needed. If we
 967   // are actually on the primordial thread we can make it appear that we have a
 968   // smaller max_size stack by inserting the guard pages at that location. But we
 969   // cannot do anything to emulate a larger stack than what has been provided by
 970   // the OS or threading library. In fact if we try to use a stack greater than
 971   // what is set by rlimit then we will crash the hosting process.
 972 
 973   // Maximum stack size is the easy part, get it from RLIMIT_STACK.
 974   // If this is "unlimited" then it will be a huge value.
 975   struct rlimit rlim;
 976   getrlimit(RLIMIT_STACK, &rlim);
 977   size_t stack_size = rlim.rlim_cur;
 978 
 979   // 6308388: a bug in ld.so will relocate its own .data section to the
 980   //   lower end of primordial stack; reduce ulimit -s value a little bit
 981   //   so we won't install guard page on ld.so's data section.
 982   //   But ensure we don't underflow the stack size - allow 1 page spare
 983   if (stack_size >= (size_t)(3 * page_size())) {
 984     stack_size -= 2 * page_size();
 985   }
 986 
 987   // Try to figure out where the stack base (top) is. This is harder.
 988   //
 989   // When an application is started, glibc saves the initial stack pointer in
 990   // a global variable "__libc_stack_end", which is then used by system
 991   // libraries. __libc_stack_end should be pretty close to stack top. The
 992   // variable is available since the very early days. However, because it is
 993   // a private interface, it could disappear in the future.
 994   //
 995   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 996   // to __libc_stack_end, it is very close to stack top, but isn't the real
 997   // stack top. Note that /proc may not exist if VM is running as a chroot
 998   // program, so reading /proc/<pid>/stat could fail. Also the contents of
 999   // /proc/<pid>/stat could change in the future (though unlikely).
1000   //
1001   // We try __libc_stack_end first. If that doesn't work, look for
1002   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1003   // as a hint, which should work well in most cases.
1004 
1005   uintptr_t stack_start;
1006 
1007   // try __libc_stack_end first
1008   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1009   if (p && *p) {
1010     stack_start = *p;
1011   } else {
1012     // see if we can get the start_stack field from /proc/self/stat
1013     FILE *fp;
1014     int pid;
1015     char state;
1016     int ppid;
1017     int pgrp;
1018     int session;
1019     int nr;
1020     int tpgrp;
1021     unsigned long flags;
1022     unsigned long minflt;
1023     unsigned long cminflt;
1024     unsigned long majflt;
1025     unsigned long cmajflt;
1026     unsigned long utime;
1027     unsigned long stime;
1028     long cutime;
1029     long cstime;
1030     long prio;
1031     long nice;
1032     long junk;
1033     long it_real;
1034     uintptr_t start;
1035     uintptr_t vsize;
1036     intptr_t rss;
1037     uintptr_t rsslim;
1038     uintptr_t scodes;
1039     uintptr_t ecode;
1040     int i;
1041 
1042     // Figure what the primordial thread stack base is. Code is inspired
1043     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1044     // followed by command name surrounded by parentheses, state, etc.
1045     char stat[2048];
1046     int statlen;
1047 
1048     fp = fopen("/proc/self/stat", "r");
1049     if (fp) {
1050       statlen = fread(stat, 1, 2047, fp);
1051       stat[statlen] = '\0';
1052       fclose(fp);
1053 
1054       // Skip pid and the command string. Note that we could be dealing with
1055       // weird command names, e.g. user could decide to rename java launcher
1056       // to "java 1.4.2 :)", then the stat file would look like
1057       //                1234 (java 1.4.2 :)) R ... ...
1058       // We don't really need to know the command string, just find the last
1059       // occurrence of ")" and then start parsing from there. See bug 4726580.
1060       char * s = strrchr(stat, ')');
1061 
1062       i = 0;
1063       if (s) {
1064         // Skip blank chars
1065         do { s++; } while (s && isspace(*s));
1066 
1067 #define _UFM UINTX_FORMAT
1068 #define _DFM INTX_FORMAT
1069 
1070         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1071         //              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
1072         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,
1073                    &state,          // 3  %c
1074                    &ppid,           // 4  %d
1075                    &pgrp,           // 5  %d
1076                    &session,        // 6  %d
1077                    &nr,             // 7  %d
1078                    &tpgrp,          // 8  %d
1079                    &flags,          // 9  %lu
1080                    &minflt,         // 10 %lu
1081                    &cminflt,        // 11 %lu
1082                    &majflt,         // 12 %lu
1083                    &cmajflt,        // 13 %lu
1084                    &utime,          // 14 %lu
1085                    &stime,          // 15 %lu
1086                    &cutime,         // 16 %ld
1087                    &cstime,         // 17 %ld
1088                    &prio,           // 18 %ld
1089                    &nice,           // 19 %ld
1090                    &junk,           // 20 %ld
1091                    &it_real,        // 21 %ld
1092                    &start,          // 22 UINTX_FORMAT
1093                    &vsize,          // 23 UINTX_FORMAT
1094                    &rss,            // 24 INTX_FORMAT
1095                    &rsslim,         // 25 UINTX_FORMAT
1096                    &scodes,         // 26 UINTX_FORMAT
1097                    &ecode,          // 27 UINTX_FORMAT
1098                    &stack_start);   // 28 UINTX_FORMAT
1099       }
1100 
1101 #undef _UFM
1102 #undef _DFM
1103 
1104       if (i != 28 - 2) {
1105         assert(false, "Bad conversion from /proc/self/stat");
1106         // product mode - assume we are the primordial thread, good luck in the
1107         // embedded case.
1108         warning("Can't detect primordial thread stack location - bad conversion");
1109         stack_start = (uintptr_t) &rlim;
1110       }
1111     } else {
1112       // For some reason we can't open /proc/self/stat (for example, running on
1113       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1114       // most cases, so don't abort:
1115       warning("Can't detect primordial thread stack location - no /proc/self/stat");
1116       stack_start = (uintptr_t) &rlim;
1117     }
1118   }
1119 
1120   // Now we have a pointer (stack_start) very close to the stack top, the
1121   // next thing to do is to figure out the exact location of stack top. We
1122   // can find out the virtual memory area that contains stack_start by
1123   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1124   // and its upper limit is the real stack top. (again, this would fail if
1125   // running inside chroot, because /proc may not exist.)
1126 
1127   uintptr_t stack_top;
1128   address low, high;
1129   if (find_vma((address)stack_start, &low, &high)) {
1130     // success, "high" is the true stack top. (ignore "low", because initial
1131     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1132     stack_top = (uintptr_t)high;
1133   } else {
1134     // failed, likely because /proc/self/maps does not exist
1135     warning("Can't detect primordial thread stack location - find_vma failed");
1136     // best effort: stack_start is normally within a few pages below the real
1137     // stack top, use it as stack top, and reduce stack size so we won't put
1138     // guard page outside stack.
1139     stack_top = stack_start;
1140     stack_size -= 16 * page_size();
1141   }
1142 
1143   // stack_top could be partially down the page so align it
1144   stack_top = align_up(stack_top, page_size());
1145 
1146   // Allowed stack value is minimum of max_size and what we derived from rlimit
1147   if (max_size > 0) {
1148     _initial_thread_stack_size = MIN2(max_size, stack_size);
1149   } else {
1150     // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1151     // clamp it at 8MB as we do on Solaris
1152     _initial_thread_stack_size = MIN2(stack_size, 8*M);
1153   }
1154   _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1155   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1156 
1157   assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1158 
1159   if (log_is_enabled(Info, os, thread)) {
1160     // See if we seem to be on primordial process thread
1161     bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1162                       uintptr_t(&rlim) < stack_top;
1163 
1164     log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1165                          SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1166                          primordial ? "primordial" : "user", max_size / K,  _initial_thread_stack_size / K,
1167                          stack_top, intptr_t(_initial_thread_stack_bottom));
1168   }
1169 }
1170 
1171 ////////////////////////////////////////////////////////////////////////////////
1172 // time support
1173 
1174 // Time since start-up in seconds to a fine granularity.
1175 // Used by VMSelfDestructTimer and the MemProfiler.
1176 double os::elapsedTime() {
1177 
1178   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1179 }
1180 
1181 jlong os::elapsed_counter() {
1182   return javaTimeNanos() - initial_time_count;
1183 }
1184 
1185 jlong os::elapsed_frequency() {
1186   return NANOSECS_PER_SEC; // nanosecond resolution
1187 }
1188 
1189 bool os::supports_vtime() { return true; }
1190 bool os::enable_vtime()   { return false; }
1191 bool os::vtime_enabled()  { return false; }
1192 
1193 double os::elapsedVTime() {
1194   struct rusage usage;
1195   int retval = getrusage(RUSAGE_THREAD, &usage);
1196   if (retval == 0) {
1197     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);
1198   } else {
1199     // better than nothing, but not much
1200     return elapsedTime();
1201   }
1202 }
1203 
1204 jlong os::javaTimeMillis() {
1205   timeval time;
1206   int status = gettimeofday(&time, NULL);
1207   assert(status != -1, "linux error");
1208   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1209 }
1210 
1211 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1212   timeval time;
1213   int status = gettimeofday(&time, NULL);
1214   assert(status != -1, "linux error");
1215   seconds = jlong(time.tv_sec);
1216   nanos = jlong(time.tv_usec) * 1000;
1217 }
1218 
1219 
1220 #ifndef CLOCK_MONOTONIC
1221   #define CLOCK_MONOTONIC (1)
1222 #endif
1223 
1224 void os::Linux::clock_init() {
1225   // we do dlopen's in this particular order due to bug in linux
1226   // dynamical loader (see 6348968) leading to crash on exit
1227   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1228   if (handle == NULL) {
1229     handle = dlopen("librt.so", RTLD_LAZY);
1230   }
1231 
1232   if (handle) {
1233     int (*clock_getres_func)(clockid_t, struct timespec*) =
1234            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1235     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1236            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1237     if (clock_getres_func && clock_gettime_func) {
1238       // See if monotonic clock is supported by the kernel. Note that some
1239       // early implementations simply return kernel jiffies (updated every
1240       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1241       // for nano time (though the monotonic property is still nice to have).
1242       // It's fixed in newer kernels, however clock_getres() still returns
1243       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1244       // resolution for now. Hopefully as people move to new kernels, this
1245       // won't be a problem.
1246       struct timespec res;
1247       struct timespec tp;
1248       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1249           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1250         // yes, monotonic clock is supported
1251         _clock_gettime = clock_gettime_func;
1252         return;
1253       } else {
1254         // close librt if there is no monotonic clock
1255         dlclose(handle);
1256       }
1257     }
1258   }
1259   warning("No monotonic clock was available - timed services may " \
1260           "be adversely affected if the time-of-day clock changes");
1261 }
1262 
1263 #ifndef SYS_clock_getres
1264   #if defined(X86) || defined(PPC64) || defined(S390)
1265     #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1266     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1267   #else
1268     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1269     #define sys_clock_getres(x,y)  -1
1270   #endif
1271 #else
1272   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1273 #endif
1274 
1275 void os::Linux::fast_thread_clock_init() {
1276   if (!UseLinuxPosixThreadCPUClocks) {
1277     return;
1278   }
1279   clockid_t clockid;
1280   struct timespec tp;
1281   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1282       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1283 
1284   // Switch to using fast clocks for thread cpu time if
1285   // the sys_clock_getres() returns 0 error code.
1286   // Note, that some kernels may support the current thread
1287   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1288   // returned by the pthread_getcpuclockid().
1289   // If the fast Posix clocks are supported then the sys_clock_getres()
1290   // must return at least tp.tv_sec == 0 which means a resolution
1291   // better than 1 sec. This is extra check for reliability.
1292 
1293   if (pthread_getcpuclockid_func &&
1294       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1295       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1296     _supports_fast_thread_cpu_time = true;
1297     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1298   }
1299 }
1300 
1301 jlong os::javaTimeNanos() {
1302   if (os::supports_monotonic_clock()) {
1303     struct timespec tp;
1304     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1305     assert(status == 0, "gettime error");
1306     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1307     return result;
1308   } else {
1309     timeval time;
1310     int status = gettimeofday(&time, NULL);
1311     assert(status != -1, "linux error");
1312     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1313     return 1000 * usecs;
1314   }
1315 }
1316 
1317 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1318   if (os::supports_monotonic_clock()) {
1319     info_ptr->max_value = ALL_64_BITS;
1320 
1321     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1322     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1323     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1324   } else {
1325     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1326     info_ptr->max_value = ALL_64_BITS;
1327 
1328     // gettimeofday is a real time clock so it skips
1329     info_ptr->may_skip_backward = true;
1330     info_ptr->may_skip_forward = true;
1331   }
1332 
1333   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1334 }
1335 
1336 // Return the real, user, and system times in seconds from an
1337 // arbitrary fixed point in the past.
1338 bool os::getTimesSecs(double* process_real_time,
1339                       double* process_user_time,
1340                       double* process_system_time) {
1341   struct tms ticks;
1342   clock_t real_ticks = times(&ticks);
1343 
1344   if (real_ticks == (clock_t) (-1)) {
1345     return false;
1346   } else {
1347     double ticks_per_second = (double) clock_tics_per_sec;
1348     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1349     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1350     *process_real_time = ((double) real_ticks) / ticks_per_second;
1351 
1352     return true;
1353   }
1354 }
1355 
1356 
1357 char * os::local_time_string(char *buf, size_t buflen) {
1358   struct tm t;
1359   time_t long_time;
1360   time(&long_time);
1361   localtime_r(&long_time, &t);
1362   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1363                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1364                t.tm_hour, t.tm_min, t.tm_sec);
1365   return buf;
1366 }
1367 
1368 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1369   return localtime_r(clock, res);
1370 }
1371 
1372 ////////////////////////////////////////////////////////////////////////////////
1373 // runtime exit support
1374 
1375 // Note: os::shutdown() might be called very early during initialization, or
1376 // called from signal handler. Before adding something to os::shutdown(), make
1377 // sure it is async-safe and can handle partially initialized VM.
1378 void os::shutdown() {
1379 
1380   // allow PerfMemory to attempt cleanup of any persistent resources
1381   perfMemory_exit();
1382 
1383   // needs to remove object in file system
1384   AttachListener::abort();
1385 
1386   // flush buffered output, finish log files
1387   ostream_abort();
1388 
1389   // Check for abort hook
1390   abort_hook_t abort_hook = Arguments::abort_hook();
1391   if (abort_hook != NULL) {
1392     abort_hook();
1393   }
1394 
1395 }
1396 
1397 // Note: os::abort() might be called very early during initialization, or
1398 // called from signal handler. Before adding something to os::abort(), make
1399 // sure it is async-safe and can handle partially initialized VM.
1400 void os::abort(bool dump_core, void* siginfo, const void* context) {
1401   os::shutdown();
1402   if (dump_core) {
1403 #ifndef PRODUCT
1404     fdStream out(defaultStream::output_fd());
1405     out.print_raw("Current thread is ");
1406     char buf[16];
1407     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1408     out.print_raw_cr(buf);
1409     out.print_raw_cr("Dumping core ...");
1410 #endif
1411     ::abort(); // dump core
1412   }
1413 
1414   ::exit(1);
1415 }
1416 
1417 // Die immediately, no exit hook, no abort hook, no cleanup.
1418 void os::die() {
1419   ::abort();
1420 }
1421 
1422 
1423 // This method is a copy of JDK's sysGetLastErrorString
1424 // from src/solaris/hpi/src/system_md.c
1425 
1426 size_t os::lasterror(char *buf, size_t len) {
1427   if (errno == 0)  return 0;
1428 
1429   const char *s = os::strerror(errno);
1430   size_t n = ::strlen(s);
1431   if (n >= len) {
1432     n = len - 1;
1433   }
1434   ::strncpy(buf, s, n);
1435   buf[n] = '\0';
1436   return n;
1437 }
1438 
1439 // thread_id is kernel thread id (similar to Solaris LWP id)
1440 intx os::current_thread_id() { return os::Linux::gettid(); }
1441 int os::current_process_id() {
1442   return ::getpid();
1443 }
1444 
1445 // DLL functions
1446 
1447 const char* os::dll_file_extension() { return ".so"; }
1448 
1449 // This must be hard coded because it's the system's temporary
1450 // directory not the java application's temp directory, ala java.io.tmpdir.
1451 const char* os::get_temp_directory() { return "/tmp"; }
1452 
1453 static bool file_exists(const char* filename) {
1454   struct stat statbuf;
1455   if (filename == NULL || strlen(filename) == 0) {
1456     return false;
1457   }
1458   return os::stat(filename, &statbuf) == 0;
1459 }
1460 
1461 // check if addr is inside libjvm.so
1462 bool os::address_is_in_vm(address addr) {
1463   static address libjvm_base_addr;
1464   Dl_info dlinfo;
1465 
1466   if (libjvm_base_addr == NULL) {
1467     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1468       libjvm_base_addr = (address)dlinfo.dli_fbase;
1469     }
1470     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1471   }
1472 
1473   if (dladdr((void *)addr, &dlinfo) != 0) {
1474     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1475   }
1476 
1477   return false;
1478 }
1479 
1480 bool os::dll_address_to_function_name(address addr, char *buf,
1481                                       int buflen, int *offset,
1482                                       bool demangle) {
1483   // buf is not optional, but offset is optional
1484   assert(buf != NULL, "sanity check");
1485 
1486   Dl_info dlinfo;
1487 
1488   if (dladdr((void*)addr, &dlinfo) != 0) {
1489     // see if we have a matching symbol
1490     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1491       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1492         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1493       }
1494       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1495       return true;
1496     }
1497     // no matching symbol so try for just file info
1498     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1499       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1500                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1501         return true;
1502       }
1503     }
1504   }
1505 
1506   buf[0] = '\0';
1507   if (offset != NULL) *offset = -1;
1508   return false;
1509 }
1510 
1511 struct _address_to_library_name {
1512   address addr;          // input : memory address
1513   size_t  buflen;        //         size of fname
1514   char*   fname;         // output: library name
1515   address base;          //         library base addr
1516 };
1517 
1518 static int address_to_library_name_callback(struct dl_phdr_info *info,
1519                                             size_t size, void *data) {
1520   int i;
1521   bool found = false;
1522   address libbase = NULL;
1523   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1524 
1525   // iterate through all loadable segments
1526   for (i = 0; i < info->dlpi_phnum; i++) {
1527     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1528     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1529       // base address of a library is the lowest address of its loaded
1530       // segments.
1531       if (libbase == NULL || libbase > segbase) {
1532         libbase = segbase;
1533       }
1534       // see if 'addr' is within current segment
1535       if (segbase <= d->addr &&
1536           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1537         found = true;
1538       }
1539     }
1540   }
1541 
1542   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1543   // so dll_address_to_library_name() can fall through to use dladdr() which
1544   // can figure out executable name from argv[0].
1545   if (found && info->dlpi_name && info->dlpi_name[0]) {
1546     d->base = libbase;
1547     if (d->fname) {
1548       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1549     }
1550     return 1;
1551   }
1552   return 0;
1553 }
1554 
1555 bool os::dll_address_to_library_name(address addr, char* buf,
1556                                      int buflen, int* offset) {
1557   // buf is not optional, but offset is optional
1558   assert(buf != NULL, "sanity check");
1559 
1560   Dl_info dlinfo;
1561   struct _address_to_library_name data;
1562 
1563   // There is a bug in old glibc dladdr() implementation that it could resolve
1564   // to wrong library name if the .so file has a base address != NULL. Here
1565   // we iterate through the program headers of all loaded libraries to find
1566   // out which library 'addr' really belongs to. This workaround can be
1567   // removed once the minimum requirement for glibc is moved to 2.3.x.
1568   data.addr = addr;
1569   data.fname = buf;
1570   data.buflen = buflen;
1571   data.base = NULL;
1572   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1573 
1574   if (rslt) {
1575     // buf already contains library name
1576     if (offset) *offset = addr - data.base;
1577     return true;
1578   }
1579   if (dladdr((void*)addr, &dlinfo) != 0) {
1580     if (dlinfo.dli_fname != NULL) {
1581       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1582     }
1583     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1584       *offset = addr - (address)dlinfo.dli_fbase;
1585     }
1586     return true;
1587   }
1588 
1589   buf[0] = '\0';
1590   if (offset) *offset = -1;
1591   return false;
1592 }
1593 
1594 // Loads .dll/.so and
1595 // in case of error it checks if .dll/.so was built for the
1596 // same architecture as Hotspot is running on
1597 
1598 
1599 // Remember the stack's state. The Linux dynamic linker will change
1600 // the stack to 'executable' at most once, so we must safepoint only once.
1601 bool os::Linux::_stack_is_executable = false;
1602 
1603 // VM operation that loads a library.  This is necessary if stack protection
1604 // of the Java stacks can be lost during loading the library.  If we
1605 // do not stop the Java threads, they can stack overflow before the stacks
1606 // are protected again.
1607 class VM_LinuxDllLoad: public VM_Operation {
1608  private:
1609   const char *_filename;
1610   char *_ebuf;
1611   int _ebuflen;
1612   void *_lib;
1613  public:
1614   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1615     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1616   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1617   void doit() {
1618     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1619     os::Linux::_stack_is_executable = true;
1620   }
1621   void* loaded_library() { return _lib; }
1622 };
1623 
1624 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1625   void * result = NULL;
1626   bool load_attempted = false;
1627 
1628   // Check whether the library to load might change execution rights
1629   // of the stack. If they are changed, the protection of the stack
1630   // guard pages will be lost. We need a safepoint to fix this.
1631   //
1632   // See Linux man page execstack(8) for more info.
1633   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1634     if (!ElfFile::specifies_noexecstack(filename)) {
1635       if (!is_init_completed()) {
1636         os::Linux::_stack_is_executable = true;
1637         // This is OK - No Java threads have been created yet, and hence no
1638         // stack guard pages to fix.
1639         //
1640         // This should happen only when you are building JDK7 using a very
1641         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1642         //
1643         // Dynamic loader will make all stacks executable after
1644         // this function returns, and will not do that again.
1645         assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1646       } else {
1647         warning("You have loaded library %s which might have disabled stack guard. "
1648                 "The VM will try to fix the stack guard now.\n"
1649                 "It's highly recommended that you fix the library with "
1650                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1651                 filename);
1652 
1653         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1654         JavaThread *jt = JavaThread::current();
1655         if (jt->thread_state() != _thread_in_native) {
1656           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1657           // that requires ExecStack. Cannot enter safe point. Let's give up.
1658           warning("Unable to fix stack guard. Giving up.");
1659         } else {
1660           if (!LoadExecStackDllInVMThread) {
1661             // This is for the case where the DLL has an static
1662             // constructor function that executes JNI code. We cannot
1663             // load such DLLs in the VMThread.
1664             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1665           }
1666 
1667           ThreadInVMfromNative tiv(jt);
1668           debug_only(VMNativeEntryWrapper vew;)
1669 
1670           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1671           VMThread::execute(&op);
1672           if (LoadExecStackDllInVMThread) {
1673             result = op.loaded_library();
1674           }
1675           load_attempted = true;
1676         }
1677       }
1678     }
1679   }
1680 
1681   if (!load_attempted) {
1682     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1683   }
1684 
1685   if (result != NULL) {
1686     // Successful loading
1687     return result;
1688   }
1689 
1690   Elf32_Ehdr elf_head;
1691   int diag_msg_max_length=ebuflen-strlen(ebuf);
1692   char* diag_msg_buf=ebuf+strlen(ebuf);
1693 
1694   if (diag_msg_max_length==0) {
1695     // No more space in ebuf for additional diagnostics message
1696     return NULL;
1697   }
1698 
1699 
1700   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1701 
1702   if (file_descriptor < 0) {
1703     // Can't open library, report dlerror() message
1704     return NULL;
1705   }
1706 
1707   bool failed_to_read_elf_head=
1708     (sizeof(elf_head)!=
1709      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1710 
1711   ::close(file_descriptor);
1712   if (failed_to_read_elf_head) {
1713     // file i/o error - report dlerror() msg
1714     return NULL;
1715   }
1716 
1717   typedef struct {
1718     Elf32_Half    code;         // Actual value as defined in elf.h
1719     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1720     unsigned char elf_class;    // 32 or 64 bit
1721     unsigned char endianess;    // MSB or LSB
1722     char*         name;         // String representation
1723   } arch_t;
1724 
1725 #ifndef EM_486
1726   #define EM_486          6               /* Intel 80486 */
1727 #endif
1728 #ifndef EM_AARCH64
1729   #define EM_AARCH64    183               /* ARM AARCH64 */
1730 #endif
1731 
1732   static const arch_t arch_array[]={
1733     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1734     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1735     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1736     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1737     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1738     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1739     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1740     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1741 #if defined(VM_LITTLE_ENDIAN)
1742     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1743     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1744 #else
1745     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1746     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1747 #endif
1748     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1749     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1750     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1751     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1752     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1753     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1754     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1755     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1756   };
1757 
1758 #if  (defined IA32)
1759   static  Elf32_Half running_arch_code=EM_386;
1760 #elif   (defined AMD64)
1761   static  Elf32_Half running_arch_code=EM_X86_64;
1762 #elif  (defined IA64)
1763   static  Elf32_Half running_arch_code=EM_IA_64;
1764 #elif  (defined __sparc) && (defined _LP64)
1765   static  Elf32_Half running_arch_code=EM_SPARCV9;
1766 #elif  (defined __sparc) && (!defined _LP64)
1767   static  Elf32_Half running_arch_code=EM_SPARC;
1768 #elif  (defined __powerpc64__)
1769   static  Elf32_Half running_arch_code=EM_PPC64;
1770 #elif  (defined __powerpc__)
1771   static  Elf32_Half running_arch_code=EM_PPC;
1772 #elif  (defined AARCH64)
1773   static  Elf32_Half running_arch_code=EM_AARCH64;
1774 #elif  (defined ARM)
1775   static  Elf32_Half running_arch_code=EM_ARM;
1776 #elif  (defined AARCH32)
1777   static  Elf32_Half running_arch_code=EM_ARM;
1778 #elif  (defined S390)
1779   static  Elf32_Half running_arch_code=EM_S390;
1780 #elif  (defined ALPHA)
1781   static  Elf32_Half running_arch_code=EM_ALPHA;
1782 #elif  (defined MIPSEL)
1783   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1784 #elif  (defined PARISC)
1785   static  Elf32_Half running_arch_code=EM_PARISC;
1786 #elif  (defined MIPS)
1787   static  Elf32_Half running_arch_code=EM_MIPS;
1788 #elif  (defined M68K)
1789   static  Elf32_Half running_arch_code=EM_68K;
1790 #elif  (defined SH)
1791   static  Elf32_Half running_arch_code=EM_SH;
1792 #else
1793     #error Method os::dll_load requires that one of following is defined:\
1794         AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1795 #endif
1796 
1797   // Identify compatability class for VM's architecture and library's architecture
1798   // Obtain string descriptions for architectures
1799 
1800   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1801   int running_arch_index=-1;
1802 
1803   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1804     if (running_arch_code == arch_array[i].code) {
1805       running_arch_index    = i;
1806     }
1807     if (lib_arch.code == arch_array[i].code) {
1808       lib_arch.compat_class = arch_array[i].compat_class;
1809       lib_arch.name         = arch_array[i].name;
1810     }
1811   }
1812 
1813   assert(running_arch_index != -1,
1814          "Didn't find running architecture code (running_arch_code) in arch_array");
1815   if (running_arch_index == -1) {
1816     // Even though running architecture detection failed
1817     // we may still continue with reporting dlerror() message
1818     return NULL;
1819   }
1820 
1821   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1822     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1823     return NULL;
1824   }
1825 
1826 #ifndef S390
1827   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1828     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1829     return NULL;
1830   }
1831 #endif // !S390
1832 
1833   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1834     if (lib_arch.name!=NULL) {
1835       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1836                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1837                  lib_arch.name, arch_array[running_arch_index].name);
1838     } else {
1839       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1840                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1841                  lib_arch.code,
1842                  arch_array[running_arch_index].name);
1843     }
1844   }
1845 
1846   return NULL;
1847 }
1848 
1849 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1850                                 int ebuflen) {
1851   void * result = ::dlopen(filename, RTLD_LAZY);
1852   if (result == NULL) {
1853     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1854     ebuf[ebuflen-1] = '\0';
1855   }
1856   return result;
1857 }
1858 
1859 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1860                                        int ebuflen) {
1861   void * result = NULL;
1862   if (LoadExecStackDllInVMThread) {
1863     result = dlopen_helper(filename, ebuf, ebuflen);
1864   }
1865 
1866   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1867   // library that requires an executable stack, or which does not have this
1868   // stack attribute set, dlopen changes the stack attribute to executable. The
1869   // read protection of the guard pages gets lost.
1870   //
1871   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1872   // may have been queued at the same time.
1873 
1874   if (!_stack_is_executable) {
1875     for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1876       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1877           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1878         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1879           warning("Attempt to reguard stack yellow zone failed.");
1880         }
1881       }
1882     }
1883   }
1884 
1885   return result;
1886 }
1887 
1888 void* os::dll_lookup(void* handle, const char* name) {
1889   void* res = dlsym(handle, name);
1890   return res;
1891 }
1892 
1893 void* os::get_default_process_handle() {
1894   return (void*)::dlopen(NULL, RTLD_LAZY);
1895 }
1896 
1897 static bool _print_ascii_file(const char* filename, outputStream* st) {
1898   int fd = ::open(filename, O_RDONLY);
1899   if (fd == -1) {
1900     return false;
1901   }
1902 
1903   char buf[33];
1904   int bytes;
1905   buf[32] = '\0';
1906   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1907     st->print_raw(buf, bytes);
1908   }
1909 
1910   ::close(fd);
1911 
1912   return true;
1913 }
1914 
1915 void os::print_dll_info(outputStream *st) {
1916   st->print_cr("Dynamic libraries:");
1917 
1918   char fname[32];
1919   pid_t pid = os::Linux::gettid();
1920 
1921   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1922 
1923   if (!_print_ascii_file(fname, st)) {
1924     st->print("Can not get library information for pid = %d\n", pid);
1925   }
1926 }
1927 
1928 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1929   FILE *procmapsFile = NULL;
1930 
1931   // Open the procfs maps file for the current process
1932   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1933     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1934     char line[PATH_MAX + 100];
1935 
1936     // Read line by line from 'file'
1937     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1938       u8 base, top, offset, inode;
1939       char permissions[5];
1940       char device[6];
1941       char name[PATH_MAX + 1];
1942 
1943       // Parse fields from line
1944       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1945              &base, &top, permissions, &offset, device, &inode, name);
1946 
1947       // Filter by device id '00:00' so that we only get file system mapped files.
1948       if (strcmp(device, "00:00") != 0) {
1949 
1950         // Call callback with the fields of interest
1951         if(callback(name, (address)base, (address)top, param)) {
1952           // Oops abort, callback aborted
1953           fclose(procmapsFile);
1954           return 1;
1955         }
1956       }
1957     }
1958     fclose(procmapsFile);
1959   }
1960   return 0;
1961 }
1962 
1963 void os::print_os_info_brief(outputStream* st) {
1964   os::Linux::print_distro_info(st);
1965 
1966   os::Posix::print_uname_info(st);
1967 
1968   os::Linux::print_libversion_info(st);
1969 
1970 }
1971 
1972 void os::print_os_info(outputStream* st) {
1973   st->print("OS:");
1974 
1975   os::Linux::print_distro_info(st);
1976 
1977   os::Posix::print_uname_info(st);
1978 
1979   // Print warning if unsafe chroot environment detected
1980   if (unsafe_chroot_detected) {
1981     st->print("WARNING!! ");
1982     st->print_cr("%s", unstable_chroot_error);
1983   }
1984 
1985   os::Linux::print_libversion_info(st);
1986 
1987   os::Posix::print_rlimit_info(st);
1988 
1989   os::Posix::print_load_average(st);
1990 
1991   os::Linux::print_full_memory_info(st);
1992 
1993   os::Linux::print_proc_sys_info(st);
1994 
1995   os::Linux::print_container_info(st);
1996 }
1997 
1998 // Try to identify popular distros.
1999 // Most Linux distributions have a /etc/XXX-release file, which contains
2000 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2001 // file that also contains the OS version string. Some have more than one
2002 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2003 // /etc/redhat-release.), so the order is important.
2004 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2005 // their own specific XXX-release file as well as a redhat-release file.
2006 // Because of this the XXX-release file needs to be searched for before the
2007 // redhat-release file.
2008 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2009 // search for redhat-release / SuSE-release needs to be before lsb-release.
2010 // Since the lsb-release file is the new standard it needs to be searched
2011 // before the older style release files.
2012 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2013 // next to last resort.  The os-release file is a new standard that contains
2014 // distribution information and the system-release file seems to be an old
2015 // standard that has been replaced by the lsb-release and os-release files.
2016 // Searching for the debian_version file is the last resort.  It contains
2017 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2018 // "Debian " is printed before the contents of the debian_version file.
2019 
2020 const char* distro_files[] = {
2021   "/etc/oracle-release",
2022   "/etc/mandriva-release",
2023   "/etc/mandrake-release",
2024   "/etc/sun-release",
2025   "/etc/redhat-release",
2026   "/etc/SuSE-release",
2027   "/etc/lsb-release",
2028   "/etc/turbolinux-release",
2029   "/etc/gentoo-release",
2030   "/etc/ltib-release",
2031   "/etc/angstrom-version",
2032   "/etc/system-release",
2033   "/etc/os-release",
2034   NULL };
2035 
2036 void os::Linux::print_distro_info(outputStream* st) {
2037   for (int i = 0;; i++) {
2038     const char* file = distro_files[i];
2039     if (file == NULL) {
2040       break;  // done
2041     }
2042     // If file prints, we found it.
2043     if (_print_ascii_file(file, st)) {
2044       return;
2045     }
2046   }
2047 
2048   if (file_exists("/etc/debian_version")) {
2049     st->print("Debian ");
2050     _print_ascii_file("/etc/debian_version", st);
2051   } else {
2052     st->print("Linux");
2053   }
2054   st->cr();
2055 }
2056 
2057 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2058   char buf[256];
2059   while (fgets(buf, sizeof(buf), fp)) {
2060     // Edit out extra stuff in expected format
2061     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2062       char* ptr = strstr(buf, "\"");  // the name is in quotes
2063       if (ptr != NULL) {
2064         ptr++; // go beyond first quote
2065         char* nl = strchr(ptr, '\"');
2066         if (nl != NULL) *nl = '\0';
2067         strncpy(distro, ptr, length);
2068       } else {
2069         ptr = strstr(buf, "=");
2070         ptr++; // go beyond equals then
2071         char* nl = strchr(ptr, '\n');
2072         if (nl != NULL) *nl = '\0';
2073         strncpy(distro, ptr, length);
2074       }
2075       return;
2076     } else if (get_first_line) {
2077       char* nl = strchr(buf, '\n');
2078       if (nl != NULL) *nl = '\0';
2079       strncpy(distro, buf, length);
2080       return;
2081     }
2082   }
2083   // print last line and close
2084   char* nl = strchr(buf, '\n');
2085   if (nl != NULL) *nl = '\0';
2086   strncpy(distro, buf, length);
2087 }
2088 
2089 static void parse_os_info(char* distro, size_t length, const char* file) {
2090   FILE* fp = fopen(file, "r");
2091   if (fp != NULL) {
2092     // if suse format, print out first line
2093     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2094     parse_os_info_helper(fp, distro, length, get_first_line);
2095     fclose(fp);
2096   }
2097 }
2098 
2099 void os::get_summary_os_info(char* buf, size_t buflen) {
2100   for (int i = 0;; i++) {
2101     const char* file = distro_files[i];
2102     if (file == NULL) {
2103       break; // ran out of distro_files
2104     }
2105     if (file_exists(file)) {
2106       parse_os_info(buf, buflen, file);
2107       return;
2108     }
2109   }
2110   // special case for debian
2111   if (file_exists("/etc/debian_version")) {
2112     strncpy(buf, "Debian ", buflen);
2113     if (buflen > 7) {
2114       parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2115     }
2116   } else {
2117     strncpy(buf, "Linux", buflen);
2118   }
2119 }
2120 
2121 void os::Linux::print_libversion_info(outputStream* st) {
2122   // libc, pthread
2123   st->print("libc:");
2124   st->print("%s ", os::Linux::glibc_version());
2125   st->print("%s ", os::Linux::libpthread_version());
2126   st->cr();
2127 }
2128 
2129 void os::Linux::print_proc_sys_info(outputStream* st) {
2130   st->cr();
2131   st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2132   _print_ascii_file("/proc/sys/kernel/threads-max", st);
2133   st->cr();
2134   st->cr();
2135 
2136   st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2137   _print_ascii_file("/proc/sys/vm/max_map_count", st);
2138   st->cr();
2139   st->cr();
2140 
2141   st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2142   _print_ascii_file("/proc/sys/kernel/pid_max", st);
2143   st->cr();
2144   st->cr();
2145 }
2146 
2147 void os::Linux::print_full_memory_info(outputStream* st) {
2148   st->print("\n/proc/meminfo:\n");
2149   _print_ascii_file("/proc/meminfo", st);
2150   st->cr();
2151 }
2152 
2153 void os::Linux::print_container_info(outputStream* st) {
2154   if (!OSContainer::is_containerized()) {
2155     return;
2156   }
2157 
2158   st->print("container (cgroup) information:\n");
2159 
2160   const char *p_ct = OSContainer::container_type();
2161   st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2162 
2163   char *p = OSContainer::cpu_cpuset_cpus();
2164   st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2165   free(p);
2166 
2167   p = OSContainer::cpu_cpuset_memory_nodes();
2168   st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2169   free(p);
2170 
2171   int i = OSContainer::active_processor_count();
2172   if (i > 0) {
2173     st->print("active_processor_count: %d\n", i);
2174   } else {
2175     st->print("active_processor_count: failed\n");
2176   }
2177 
2178   i = OSContainer::cpu_quota();
2179   st->print("cpu_quota: %d\n", i);
2180 
2181   i = OSContainer::cpu_period();
2182   st->print("cpu_period: %d\n", i);
2183 
2184   i = OSContainer::cpu_shares();
2185   st->print("cpu_shares: %d\n", i);
2186 
2187   jlong j = OSContainer::memory_limit_in_bytes();
2188   st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2189 
2190   j = OSContainer::memory_and_swap_limit_in_bytes();
2191   st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2192 
2193   j = OSContainer::memory_soft_limit_in_bytes();
2194   st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2195 
2196   j = OSContainer::OSContainer::memory_usage_in_bytes();
2197   st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2198 
2199   j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2200   st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2201   st->cr();
2202 }
2203 
2204 void os::print_memory_info(outputStream* st) {
2205 
2206   st->print("Memory:");
2207   st->print(" %dk page", os::vm_page_size()>>10);
2208 
2209   // values in struct sysinfo are "unsigned long"
2210   struct sysinfo si;
2211   sysinfo(&si);
2212 
2213   st->print(", physical " UINT64_FORMAT "k",
2214             os::physical_memory() >> 10);
2215   st->print("(" UINT64_FORMAT "k free)",
2216             os::available_memory() >> 10);
2217   st->print(", swap " UINT64_FORMAT "k",
2218             ((jlong)si.totalswap * si.mem_unit) >> 10);
2219   st->print("(" UINT64_FORMAT "k free)",
2220             ((jlong)si.freeswap * si.mem_unit) >> 10);
2221   st->cr();
2222 }
2223 
2224 // Print the first "model name" line and the first "flags" line
2225 // that we find and nothing more. We assume "model name" comes
2226 // before "flags" so if we find a second "model name", then the
2227 // "flags" field is considered missing.
2228 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2229 #if defined(IA32) || defined(AMD64)
2230   // Other platforms have less repetitive cpuinfo files
2231   FILE *fp = fopen("/proc/cpuinfo", "r");
2232   if (fp) {
2233     while (!feof(fp)) {
2234       if (fgets(buf, buflen, fp)) {
2235         // Assume model name comes before flags
2236         bool model_name_printed = false;
2237         if (strstr(buf, "model name") != NULL) {
2238           if (!model_name_printed) {
2239             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2240             st->print_raw(buf);
2241             model_name_printed = true;
2242           } else {
2243             // model name printed but not flags?  Odd, just return
2244             fclose(fp);
2245             return true;
2246           }
2247         }
2248         // print the flags line too
2249         if (strstr(buf, "flags") != NULL) {
2250           st->print_raw(buf);
2251           fclose(fp);
2252           return true;
2253         }
2254       }
2255     }
2256     fclose(fp);
2257   }
2258 #endif // x86 platforms
2259   return false;
2260 }
2261 
2262 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2263   // Only print the model name if the platform provides this as a summary
2264   if (!print_model_name_and_flags(st, buf, buflen)) {
2265     st->print("\n/proc/cpuinfo:\n");
2266     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2267       st->print_cr("  <Not Available>");
2268     }
2269   }
2270 }
2271 
2272 #if defined(AMD64) || defined(IA32) || defined(X32)
2273 const char* search_string = "model name";
2274 #elif defined(M68K)
2275 const char* search_string = "CPU";
2276 #elif defined(PPC64)
2277 const char* search_string = "cpu";
2278 #elif defined(S390)
2279 const char* search_string = "processor";
2280 #elif defined(SPARC)
2281 const char* search_string = "cpu";
2282 #else
2283 const char* search_string = "Processor";
2284 #endif
2285 
2286 // Parses the cpuinfo file for string representing the model name.
2287 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2288   FILE* fp = fopen("/proc/cpuinfo", "r");
2289   if (fp != NULL) {
2290     while (!feof(fp)) {
2291       char buf[256];
2292       if (fgets(buf, sizeof(buf), fp)) {
2293         char* start = strstr(buf, search_string);
2294         if (start != NULL) {
2295           char *ptr = start + strlen(search_string);
2296           char *end = buf + strlen(buf);
2297           while (ptr != end) {
2298              // skip whitespace and colon for the rest of the name.
2299              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2300                break;
2301              }
2302              ptr++;
2303           }
2304           if (ptr != end) {
2305             // reasonable string, get rid of newline and keep the rest
2306             char* nl = strchr(buf, '\n');
2307             if (nl != NULL) *nl = '\0';
2308             strncpy(cpuinfo, ptr, length);
2309             fclose(fp);
2310             return;
2311           }
2312         }
2313       }
2314     }
2315     fclose(fp);
2316   }
2317   // cpuinfo not found or parsing failed, just print generic string.  The entire
2318   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2319 #if   defined(AARCH64)
2320   strncpy(cpuinfo, "AArch64", length);
2321 #elif defined(AMD64)
2322   strncpy(cpuinfo, "x86_64", length);
2323 #elif defined(ARM)  // Order wrt. AARCH64 is relevant!
2324   strncpy(cpuinfo, "ARM", length);
2325 #elif defined(IA32)
2326   strncpy(cpuinfo, "x86_32", length);
2327 #elif defined(IA64)
2328   strncpy(cpuinfo, "IA64", length);
2329 #elif defined(PPC)
2330   strncpy(cpuinfo, "PPC64", length);
2331 #elif defined(S390)
2332   strncpy(cpuinfo, "S390", length);
2333 #elif defined(SPARC)
2334   strncpy(cpuinfo, "sparcv9", length);
2335 #elif defined(ZERO_LIBARCH)
2336   strncpy(cpuinfo, ZERO_LIBARCH, length);
2337 #else
2338   strncpy(cpuinfo, "unknown", length);
2339 #endif
2340 }
2341 
2342 static void print_signal_handler(outputStream* st, int sig,
2343                                  char* buf, size_t buflen);
2344 
2345 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2346   st->print_cr("Signal Handlers:");
2347   print_signal_handler(st, SIGSEGV, buf, buflen);
2348   print_signal_handler(st, SIGBUS , buf, buflen);
2349   print_signal_handler(st, SIGFPE , buf, buflen);
2350   print_signal_handler(st, SIGPIPE, buf, buflen);
2351   print_signal_handler(st, SIGXFSZ, buf, buflen);
2352   print_signal_handler(st, SIGILL , buf, buflen);
2353   print_signal_handler(st, SR_signum, buf, buflen);
2354   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2355   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2356   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2357   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2358 #if defined(PPC64)
2359   print_signal_handler(st, SIGTRAP, buf, buflen);
2360 #endif
2361 }
2362 
2363 static char saved_jvm_path[MAXPATHLEN] = {0};
2364 
2365 // Find the full path to the current module, libjvm.so
2366 void os::jvm_path(char *buf, jint buflen) {
2367   // Error checking.
2368   if (buflen < MAXPATHLEN) {
2369     assert(false, "must use a large-enough buffer");
2370     buf[0] = '\0';
2371     return;
2372   }
2373   // Lazy resolve the path to current module.
2374   if (saved_jvm_path[0] != 0) {
2375     strcpy(buf, saved_jvm_path);
2376     return;
2377   }
2378 
2379   char dli_fname[MAXPATHLEN];
2380   bool ret = dll_address_to_library_name(
2381                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2382                                          dli_fname, sizeof(dli_fname), NULL);
2383   assert(ret, "cannot locate libjvm");
2384   char *rp = NULL;
2385   if (ret && dli_fname[0] != '\0') {
2386     rp = os::Posix::realpath(dli_fname, buf, buflen);
2387   }
2388   if (rp == NULL) {
2389     return;
2390   }
2391 
2392   if (Arguments::sun_java_launcher_is_altjvm()) {
2393     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2394     // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2395     // If "/jre/lib/" appears at the right place in the string, then
2396     // assume we are installed in a JDK and we're done. Otherwise, check
2397     // for a JAVA_HOME environment variable and fix up the path so it
2398     // looks like libjvm.so is installed there (append a fake suffix
2399     // hotspot/libjvm.so).
2400     const char *p = buf + strlen(buf) - 1;
2401     for (int count = 0; p > buf && count < 5; ++count) {
2402       for (--p; p > buf && *p != '/'; --p)
2403         /* empty */ ;
2404     }
2405 
2406     if (strncmp(p, "/jre/lib/", 9) != 0) {
2407       // Look for JAVA_HOME in the environment.
2408       char* java_home_var = ::getenv("JAVA_HOME");
2409       if (java_home_var != NULL && java_home_var[0] != 0) {
2410         char* jrelib_p;
2411         int len;
2412 
2413         // Check the current module name "libjvm.so".
2414         p = strrchr(buf, '/');
2415         if (p == NULL) {
2416           return;
2417         }
2418         assert(strstr(p, "/libjvm") == p, "invalid library name");
2419 
2420         rp = os::Posix::realpath(java_home_var, buf, buflen);
2421         if (rp == NULL) {
2422           return;
2423         }
2424 
2425         // determine if this is a legacy image or modules image
2426         // modules image doesn't have "jre" subdirectory
2427         len = strlen(buf);
2428         assert(len < buflen, "Ran out of buffer room");
2429         jrelib_p = buf + len;
2430         snprintf(jrelib_p, buflen-len, "/jre/lib");
2431         if (0 != access(buf, F_OK)) {
2432           snprintf(jrelib_p, buflen-len, "/lib");
2433         }
2434 
2435         if (0 == access(buf, F_OK)) {
2436           // Use current module name "libjvm.so"
2437           len = strlen(buf);
2438           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2439         } else {
2440           // Go back to path of .so
2441           rp = os::Posix::realpath(dli_fname, buf, buflen);
2442           if (rp == NULL) {
2443             return;
2444           }
2445         }
2446       }
2447     }
2448   }
2449 
2450   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2451   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2452 }
2453 
2454 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2455   // no prefix required, not even "_"
2456 }
2457 
2458 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2459   // no suffix required
2460 }
2461 
2462 ////////////////////////////////////////////////////////////////////////////////
2463 // sun.misc.Signal support
2464 
2465 static volatile jint sigint_count = 0;
2466 
2467 static void UserHandler(int sig, void *siginfo, void *context) {
2468   // 4511530 - sem_post is serialized and handled by the manager thread. When
2469   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2470   // don't want to flood the manager thread with sem_post requests.
2471   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2472     return;
2473   }
2474 
2475   // Ctrl-C is pressed during error reporting, likely because the error
2476   // handler fails to abort. Let VM die immediately.
2477   if (sig == SIGINT && VMError::is_error_reported()) {
2478     os::die();
2479   }
2480 
2481   os::signal_notify(sig);
2482 }
2483 
2484 void* os::user_handler() {
2485   return CAST_FROM_FN_PTR(void*, UserHandler);
2486 }
2487 
2488 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2489   struct timespec ts;
2490   // Semaphore's are always associated with CLOCK_REALTIME
2491   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2492   // see os_posix.cpp for discussion on overflow checking
2493   if (sec >= MAX_SECS) {
2494     ts.tv_sec += MAX_SECS;
2495     ts.tv_nsec = 0;
2496   } else {
2497     ts.tv_sec += sec;
2498     ts.tv_nsec += nsec;
2499     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2500       ts.tv_nsec -= NANOSECS_PER_SEC;
2501       ++ts.tv_sec; // note: this must be <= max_secs
2502     }
2503   }
2504 
2505   return ts;
2506 }
2507 
2508 extern "C" {
2509   typedef void (*sa_handler_t)(int);
2510   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2511 }
2512 
2513 void* os::signal(int signal_number, void* handler) {
2514   struct sigaction sigAct, oldSigAct;
2515 
2516   sigfillset(&(sigAct.sa_mask));
2517   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2518   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2519 
2520   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2521     // -1 means registration failed
2522     return (void *)-1;
2523   }
2524 
2525   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2526 }
2527 
2528 void os::signal_raise(int signal_number) {
2529   ::raise(signal_number);
2530 }
2531 
2532 // The following code is moved from os.cpp for making this
2533 // code platform specific, which it is by its very nature.
2534 
2535 // Will be modified when max signal is changed to be dynamic
2536 int os::sigexitnum_pd() {
2537   return NSIG;
2538 }
2539 
2540 // a counter for each possible signal value
2541 static volatile jint pending_signals[NSIG+1] = { 0 };
2542 
2543 // Linux(POSIX) specific hand shaking semaphore.
2544 static Semaphore* sig_sem = NULL;
2545 static PosixSemaphore sr_semaphore;
2546 
2547 static void jdk_misc_signal_init() {
2548   // Initialize signal structures
2549   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2550 
2551   // Initialize signal semaphore
2552   sig_sem = new Semaphore();
2553 }
2554 
2555 void os::signal_notify(int sig) {
2556   if (sig_sem != NULL) {
2557     Atomic::inc(&pending_signals[sig]);
2558     sig_sem->signal();
2559   } else {
2560     // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2561     // initialization isn't called.
2562     assert(ReduceSignalUsage, "signal semaphore should be created");
2563   }
2564 }
2565 
2566 static int check_pending_signals() {
2567   Atomic::store(0, &sigint_count);
2568   for (;;) {
2569     for (int i = 0; i < NSIG + 1; i++) {
2570       jint n = pending_signals[i];
2571       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2572         return i;
2573       }
2574     }
2575     JavaThread *thread = JavaThread::current();
2576     ThreadBlockInVM tbivm(thread);
2577 
2578     bool threadIsSuspended;
2579     do {
2580       thread->set_suspend_equivalent();
2581       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2582       sig_sem->wait();
2583 
2584       // were we externally suspended while we were waiting?
2585       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2586       if (threadIsSuspended) {
2587         // The semaphore has been incremented, but while we were waiting
2588         // another thread suspended us. We don't want to continue running
2589         // while suspended because that would surprise the thread that
2590         // suspended us.
2591         sig_sem->signal();
2592 
2593         thread->java_suspend_self();
2594       }
2595     } while (threadIsSuspended);
2596   }
2597 }
2598 
2599 int os::signal_wait() {
2600   return check_pending_signals();
2601 }
2602 
2603 ////////////////////////////////////////////////////////////////////////////////
2604 // Virtual Memory
2605 
2606 int os::vm_page_size() {
2607   // Seems redundant as all get out
2608   assert(os::Linux::page_size() != -1, "must call os::init");
2609   return os::Linux::page_size();
2610 }
2611 
2612 // Solaris allocates memory by pages.
2613 int os::vm_allocation_granularity() {
2614   assert(os::Linux::page_size() != -1, "must call os::init");
2615   return os::Linux::page_size();
2616 }
2617 
2618 // Rationale behind this function:
2619 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2620 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2621 //  samples for JITted code. Here we create private executable mapping over the code cache
2622 //  and then we can use standard (well, almost, as mapping can change) way to provide
2623 //  info for the reporting script by storing timestamp and location of symbol
2624 void linux_wrap_code(char* base, size_t size) {
2625   static volatile jint cnt = 0;
2626 
2627   if (!UseOprofile) {
2628     return;
2629   }
2630 
2631   char buf[PATH_MAX+1];
2632   int num = Atomic::add(1, &cnt);
2633 
2634   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2635            os::get_temp_directory(), os::current_process_id(), num);
2636   unlink(buf);
2637 
2638   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2639 
2640   if (fd != -1) {
2641     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2642     if (rv != (off_t)-1) {
2643       if (::write(fd, "", 1) == 1) {
2644         mmap(base, size,
2645              PROT_READ|PROT_WRITE|PROT_EXEC,
2646              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2647       }
2648     }
2649     ::close(fd);
2650     unlink(buf);
2651   }
2652 }
2653 
2654 static bool recoverable_mmap_error(int err) {
2655   // See if the error is one we can let the caller handle. This
2656   // list of errno values comes from JBS-6843484. I can't find a
2657   // Linux man page that documents this specific set of errno
2658   // values so while this list currently matches Solaris, it may
2659   // change as we gain experience with this failure mode.
2660   switch (err) {
2661   case EBADF:
2662   case EINVAL:
2663   case ENOTSUP:
2664     // let the caller deal with these errors
2665     return true;
2666 
2667   default:
2668     // Any remaining errors on this OS can cause our reserved mapping
2669     // to be lost. That can cause confusion where different data
2670     // structures think they have the same memory mapped. The worst
2671     // scenario is if both the VM and a library think they have the
2672     // same memory mapped.
2673     return false;
2674   }
2675 }
2676 
2677 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2678                                     int err) {
2679   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2680           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2681           os::strerror(err), err);
2682 }
2683 
2684 static void warn_fail_commit_memory(char* addr, size_t size,
2685                                     size_t alignment_hint, bool exec,
2686                                     int err) {
2687   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2688           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2689           alignment_hint, exec, os::strerror(err), err);
2690 }
2691 
2692 // NOTE: Linux kernel does not really reserve the pages for us.
2693 //       All it does is to check if there are enough free pages
2694 //       left at the time of mmap(). This could be a potential
2695 //       problem.
2696 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2697   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2698   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2699                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2700   if (res != (uintptr_t) MAP_FAILED) {
2701     if (UseNUMAInterleaving) {
2702       numa_make_global(addr, size);
2703     }
2704     return 0;
2705   }
2706 
2707   int err = errno;  // save errno from mmap() call above
2708 
2709   if (!recoverable_mmap_error(err)) {
2710     warn_fail_commit_memory(addr, size, exec, err);
2711     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2712   }
2713 
2714   return err;
2715 }
2716 
2717 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2718   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2719 }
2720 
2721 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2722                                   const char* mesg) {
2723   assert(mesg != NULL, "mesg must be specified");
2724   int err = os::Linux::commit_memory_impl(addr, size, exec);
2725   if (err != 0) {
2726     // the caller wants all commit errors to exit with the specified mesg:
2727     warn_fail_commit_memory(addr, size, exec, err);
2728     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2729   }
2730 }
2731 
2732 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2733 #ifndef MAP_HUGETLB
2734   #define MAP_HUGETLB 0x40000
2735 #endif
2736 
2737 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2738 #ifndef MADV_HUGEPAGE
2739   #define MADV_HUGEPAGE 14
2740 #endif
2741 
2742 int os::Linux::commit_memory_impl(char* addr, size_t size,
2743                                   size_t alignment_hint, bool exec) {
2744   int err = os::Linux::commit_memory_impl(addr, size, exec);
2745   if (err == 0) {
2746     realign_memory(addr, size, alignment_hint);
2747   }
2748   return err;
2749 }
2750 
2751 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2752                           bool exec) {
2753   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2754 }
2755 
2756 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2757                                   size_t alignment_hint, bool exec,
2758                                   const char* mesg) {
2759   assert(mesg != NULL, "mesg must be specified");
2760   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2761   if (err != 0) {
2762     // the caller wants all commit errors to exit with the specified mesg:
2763     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2764     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2765   }
2766 }
2767 
2768 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2769   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2770     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2771     // be supported or the memory may already be backed by huge pages.
2772     ::madvise(addr, bytes, MADV_HUGEPAGE);
2773   }
2774 }
2775 
2776 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2777   // This method works by doing an mmap over an existing mmaping and effectively discarding
2778   // the existing pages. However it won't work for SHM-based large pages that cannot be
2779   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2780   // small pages on top of the SHM segment. This method always works for small pages, so we
2781   // allow that in any case.
2782   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2783     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2784   }
2785 }
2786 
2787 void os::numa_make_global(char *addr, size_t bytes) {
2788   Linux::numa_interleave_memory(addr, bytes);
2789 }
2790 
2791 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2792 // bind policy to MPOL_PREFERRED for the current thread.
2793 #define USE_MPOL_PREFERRED 0
2794 
2795 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2796   // To make NUMA and large pages more robust when both enabled, we need to ease
2797   // the requirements on where the memory should be allocated. MPOL_BIND is the
2798   // default policy and it will force memory to be allocated on the specified
2799   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2800   // the specified node, but will not force it. Using this policy will prevent
2801   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2802   // free large pages.
2803   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2804   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2805 }
2806 
2807 bool os::numa_topology_changed() { return false; }
2808 
2809 size_t os::numa_get_groups_num() {
2810   // Return just the number of nodes in which it's possible to allocate memory
2811   // (in numa terminology, configured nodes).
2812   return Linux::numa_num_configured_nodes();
2813 }
2814 
2815 int os::numa_get_group_id() {
2816   int cpu_id = Linux::sched_getcpu();
2817   if (cpu_id != -1) {
2818     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2819     if (lgrp_id != -1) {
2820       return lgrp_id;
2821     }
2822   }
2823   return 0;
2824 }
2825 
2826 int os::Linux::get_existing_num_nodes() {
2827   int node;
2828   int highest_node_number = Linux::numa_max_node();
2829   int num_nodes = 0;
2830 
2831   // Get the total number of nodes in the system including nodes without memory.
2832   for (node = 0; node <= highest_node_number; node++) {
2833     if (isnode_in_existing_nodes(node)) {
2834       num_nodes++;
2835     }
2836   }
2837   return num_nodes;
2838 }
2839 
2840 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2841   int highest_node_number = Linux::numa_max_node();
2842   size_t i = 0;
2843 
2844   // Map all node ids in which it is possible to allocate memory. Also nodes are
2845   // not always consecutively available, i.e. available from 0 to the highest
2846   // node number. If the nodes have been bound explicitly using numactl membind,
2847   // then allocate memory from those nodes only.
2848   for (int node = 0; node <= highest_node_number; node++) {
2849     if (Linux::isnode_in_bound_nodes((unsigned int)node)) {
2850       ids[i++] = node;
2851     }
2852   }
2853   return i;
2854 }
2855 
2856 bool os::get_page_info(char *start, page_info* info) {
2857   return false;
2858 }
2859 
2860 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2861                      page_info* page_found) {
2862   return end;
2863 }
2864 
2865 
2866 int os::Linux::sched_getcpu_syscall(void) {
2867   unsigned int cpu = 0;
2868   int retval = -1;
2869 
2870 #if defined(IA32)
2871   #ifndef SYS_getcpu
2872     #define SYS_getcpu 318
2873   #endif
2874   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2875 #elif defined(AMD64)
2876 // Unfortunately we have to bring all these macros here from vsyscall.h
2877 // to be able to compile on old linuxes.
2878   #define __NR_vgetcpu 2
2879   #define VSYSCALL_START (-10UL << 20)
2880   #define VSYSCALL_SIZE 1024
2881   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2882   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2883   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2884   retval = vgetcpu(&cpu, NULL, NULL);
2885 #endif
2886 
2887   return (retval == -1) ? retval : cpu;
2888 }
2889 
2890 void os::Linux::sched_getcpu_init() {
2891   // sched_getcpu() should be in libc.
2892   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2893                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2894 
2895   // If it's not, try a direct syscall.
2896   if (sched_getcpu() == -1) {
2897     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2898                                     (void*)&sched_getcpu_syscall));
2899   }
2900 
2901   if (sched_getcpu() == -1) {
2902     vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
2903   }
2904 }
2905 
2906 // Something to do with the numa-aware allocator needs these symbols
2907 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2908 extern "C" JNIEXPORT void numa_error(char *where) { }
2909 
2910 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2911 // load symbol from base version instead.
2912 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2913   void *f = dlvsym(handle, name, "libnuma_1.1");
2914   if (f == NULL) {
2915     f = dlsym(handle, name);
2916   }
2917   return f;
2918 }
2919 
2920 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2921 // Return NULL if the symbol is not defined in this particular version.
2922 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2923   return dlvsym(handle, name, "libnuma_1.2");
2924 }
2925 
2926 bool os::Linux::libnuma_init() {
2927   if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2928     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2929     if (handle != NULL) {
2930       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2931                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2932       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2933                                        libnuma_dlsym(handle, "numa_max_node")));
2934       set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2935                                                    libnuma_dlsym(handle, "numa_num_configured_nodes")));
2936       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2937                                         libnuma_dlsym(handle, "numa_available")));
2938       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2939                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2940       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2941                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2942       set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2943                                                 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2944       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2945                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2946       set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2947                                                libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2948       set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2949                                        libnuma_dlsym(handle, "numa_distance")));
2950       set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
2951                                           libnuma_v2_dlsym(handle, "numa_get_membind")));
2952 
2953       if (numa_available() != -1) {
2954         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2955         set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2956         set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2957         // Create an index -> node mapping, since nodes are not always consecutive
2958         _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2959         rebuild_nindex_to_node_map();
2960         // Create a cpu -> node mapping
2961         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2962         rebuild_cpu_to_node_map();
2963         return true;
2964       }
2965     }
2966   }
2967   return false;
2968 }
2969 
2970 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2971   // Creating guard page is very expensive. Java thread has HotSpot
2972   // guard pages, only enable glibc guard page for non-Java threads.
2973   // (Remember: compiler thread is a Java thread, too!)
2974   return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2975 }
2976 
2977 void os::Linux::rebuild_nindex_to_node_map() {
2978   int highest_node_number = Linux::numa_max_node();
2979 
2980   nindex_to_node()->clear();
2981   for (int node = 0; node <= highest_node_number; node++) {
2982     if (Linux::isnode_in_existing_nodes(node)) {
2983       nindex_to_node()->append(node);
2984     }
2985   }
2986 }
2987 
2988 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2989 // The table is later used in get_node_by_cpu().
2990 void os::Linux::rebuild_cpu_to_node_map() {
2991   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2992                               // in libnuma (possible values are starting from 16,
2993                               // and continuing up with every other power of 2, but less
2994                               // than the maximum number of CPUs supported by kernel), and
2995                               // is a subject to change (in libnuma version 2 the requirements
2996                               // are more reasonable) we'll just hardcode the number they use
2997                               // in the library.
2998   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2999 
3000   size_t cpu_num = processor_count();
3001   size_t cpu_map_size = NCPUS / BitsPerCLong;
3002   size_t cpu_map_valid_size =
3003     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3004 
3005   cpu_to_node()->clear();
3006   cpu_to_node()->at_grow(cpu_num - 1);
3007 
3008   size_t node_num = get_existing_num_nodes();
3009 
3010   int distance = 0;
3011   int closest_distance = INT_MAX;
3012   int closest_node = 0;
3013   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3014   for (size_t i = 0; i < node_num; i++) {
3015     // Check if node is configured (not a memory-less node). If it is not, find
3016     // the closest configured node. Check also if node is bound, i.e. it's allowed
3017     // to allocate memory from the node. If it's not allowed, map cpus in that node
3018     // to the closest node from which memory allocation is allowed.
3019     if (!isnode_in_configured_nodes(nindex_to_node()->at(i)) ||
3020         !isnode_in_bound_nodes(nindex_to_node()->at(i))) {
3021       closest_distance = INT_MAX;
3022       // Check distance from all remaining nodes in the system. Ignore distance
3023       // from itself, from another non-configured node, and from another non-bound
3024       // node.
3025       for (size_t m = 0; m < node_num; m++) {
3026         if (m != i &&
3027             isnode_in_configured_nodes(nindex_to_node()->at(m)) &&
3028             isnode_in_bound_nodes(nindex_to_node()->at(m))) {
3029           distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3030           // If a closest node is found, update. There is always at least one
3031           // configured and bound node in the system so there is always at least
3032           // one node close.
3033           if (distance != 0 && distance < closest_distance) {
3034             closest_distance = distance;
3035             closest_node = nindex_to_node()->at(m);
3036           }
3037         }
3038       }
3039      } else {
3040        // Current node is already a configured node.
3041        closest_node = nindex_to_node()->at(i);
3042      }
3043 
3044     // Get cpus from the original node and map them to the closest node. If node
3045     // is a configured node (not a memory-less node), then original node and
3046     // closest node are the same.
3047     if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3048       for (size_t j = 0; j < cpu_map_valid_size; j++) {
3049         if (cpu_map[j] != 0) {
3050           for (size_t k = 0; k < BitsPerCLong; k++) {
3051             if (cpu_map[j] & (1UL << k)) {
3052               cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3053             }
3054           }
3055         }
3056       }
3057     }
3058   }
3059   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3060 }
3061 
3062 int os::Linux::get_node_by_cpu(int cpu_id) {
3063   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3064     return cpu_to_node()->at(cpu_id);
3065   }
3066   return -1;
3067 }
3068 
3069 GrowableArray<int>* os::Linux::_cpu_to_node;
3070 GrowableArray<int>* os::Linux::_nindex_to_node;
3071 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3072 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3073 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3074 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3075 os::Linux::numa_available_func_t os::Linux::_numa_available;
3076 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3077 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3078 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3079 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3080 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3081 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3082 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3083 unsigned long* os::Linux::_numa_all_nodes;
3084 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3085 struct bitmask* os::Linux::_numa_nodes_ptr;
3086 
3087 bool os::pd_uncommit_memory(char* addr, size_t size) {
3088   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3089                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3090   return res  != (uintptr_t) MAP_FAILED;
3091 }
3092 
3093 static address get_stack_commited_bottom(address bottom, size_t size) {
3094   address nbot = bottom;
3095   address ntop = bottom + size;
3096 
3097   size_t page_sz = os::vm_page_size();
3098   unsigned pages = size / page_sz;
3099 
3100   unsigned char vec[1];
3101   unsigned imin = 1, imax = pages + 1, imid;
3102   int mincore_return_value = 0;
3103 
3104   assert(imin <= imax, "Unexpected page size");
3105 
3106   while (imin < imax) {
3107     imid = (imax + imin) / 2;
3108     nbot = ntop - (imid * page_sz);
3109 
3110     // Use a trick with mincore to check whether the page is mapped or not.
3111     // mincore sets vec to 1 if page resides in memory and to 0 if page
3112     // is swapped output but if page we are asking for is unmapped
3113     // it returns -1,ENOMEM
3114     mincore_return_value = mincore(nbot, page_sz, vec);
3115 
3116     if (mincore_return_value == -1) {
3117       // Page is not mapped go up
3118       // to find first mapped page
3119       if (errno != EAGAIN) {
3120         assert(errno == ENOMEM, "Unexpected mincore errno");
3121         imax = imid;
3122       }
3123     } else {
3124       // Page is mapped go down
3125       // to find first not mapped page
3126       imin = imid + 1;
3127     }
3128   }
3129 
3130   nbot = nbot + page_sz;
3131 
3132   // Adjust stack bottom one page up if last checked page is not mapped
3133   if (mincore_return_value == -1) {
3134     nbot = nbot + page_sz;
3135   }
3136 
3137   return nbot;
3138 }
3139 
3140 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3141   int mincore_return_value;
3142   const size_t stripe = 1024;  // query this many pages each time
3143   unsigned char vec[stripe + 1];
3144   // set a guard
3145   vec[stripe] = 'X';
3146 
3147   const size_t page_sz = os::vm_page_size();
3148   size_t pages = size / page_sz;
3149 
3150   assert(is_aligned(start, page_sz), "Start address must be page aligned");
3151   assert(is_aligned(size, page_sz), "Size must be page aligned");
3152 
3153   committed_start = NULL;
3154 
3155   int loops = (pages + stripe - 1) / stripe;
3156   int committed_pages = 0;
3157   address loop_base = start;
3158   bool found_range = false;
3159 
3160   for (int index = 0; index < loops && !found_range; index ++) {
3161     assert(pages > 0, "Nothing to do");
3162     int pages_to_query = (pages >= stripe) ? stripe : pages;
3163     pages -= pages_to_query;
3164 
3165     // Get stable read
3166     while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3167 
3168     // During shutdown, some memory goes away without properly notifying NMT,
3169     // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3170     // Bailout and return as not committed for now.
3171     if (mincore_return_value == -1 && errno == ENOMEM) {
3172       return false;
3173     }
3174 
3175     assert(vec[stripe] == 'X', "overflow guard");
3176     assert(mincore_return_value == 0, "Range must be valid");
3177     // Process this stripe
3178     for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3179       if ((vec[vecIdx] & 0x01) == 0) { // not committed
3180         // End of current contiguous region
3181         if (committed_start != NULL) {
3182           found_range = true;
3183           break;
3184         }
3185       } else { // committed
3186         // Start of region
3187         if (committed_start == NULL) {
3188           committed_start = loop_base + page_sz * vecIdx;
3189         }
3190         committed_pages ++;
3191       }
3192     }
3193 
3194     loop_base += pages_to_query * page_sz;
3195   }
3196 
3197   if (committed_start != NULL) {
3198     assert(committed_pages > 0, "Must have committed region");
3199     assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3200     assert(committed_start >= start && committed_start < start + size, "Out of range");
3201     committed_size = page_sz * committed_pages;
3202     return true;
3203   } else {
3204     assert(committed_pages == 0, "Should not have committed region");
3205     return false;
3206   }
3207 }
3208 
3209 
3210 // Linux uses a growable mapping for the stack, and if the mapping for
3211 // the stack guard pages is not removed when we detach a thread the
3212 // stack cannot grow beyond the pages where the stack guard was
3213 // mapped.  If at some point later in the process the stack expands to
3214 // that point, the Linux kernel cannot expand the stack any further
3215 // because the guard pages are in the way, and a segfault occurs.
3216 //
3217 // However, it's essential not to split the stack region by unmapping
3218 // a region (leaving a hole) that's already part of the stack mapping,
3219 // so if the stack mapping has already grown beyond the guard pages at
3220 // the time we create them, we have to truncate the stack mapping.
3221 // So, we need to know the extent of the stack mapping when
3222 // create_stack_guard_pages() is called.
3223 
3224 // We only need this for stacks that are growable: at the time of
3225 // writing thread stacks don't use growable mappings (i.e. those
3226 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3227 // only applies to the main thread.
3228 
3229 // If the (growable) stack mapping already extends beyond the point
3230 // where we're going to put our guard pages, truncate the mapping at
3231 // that point by munmap()ping it.  This ensures that when we later
3232 // munmap() the guard pages we don't leave a hole in the stack
3233 // mapping. This only affects the main/primordial thread
3234 
3235 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3236   if (os::is_primordial_thread()) {
3237     // As we manually grow stack up to bottom inside create_attached_thread(),
3238     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3239     // we don't need to do anything special.
3240     // Check it first, before calling heavy function.
3241     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3242     unsigned char vec[1];
3243 
3244     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3245       // Fallback to slow path on all errors, including EAGAIN
3246       stack_extent = (uintptr_t) get_stack_commited_bottom(
3247                                                            os::Linux::initial_thread_stack_bottom(),
3248                                                            (size_t)addr - stack_extent);
3249     }
3250 
3251     if (stack_extent < (uintptr_t)addr) {
3252       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3253     }
3254   }
3255 
3256   return os::commit_memory(addr, size, !ExecMem);
3257 }
3258 
3259 // If this is a growable mapping, remove the guard pages entirely by
3260 // munmap()ping them.  If not, just call uncommit_memory(). This only
3261 // affects the main/primordial thread, but guard against future OS changes.
3262 // It's safe to always unmap guard pages for primordial thread because we
3263 // always place it right after end of the mapped region.
3264 
3265 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3266   uintptr_t stack_extent, stack_base;
3267 
3268   if (os::is_primordial_thread()) {
3269     return ::munmap(addr, size) == 0;
3270   }
3271 
3272   return os::uncommit_memory(addr, size);
3273 }
3274 
3275 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3276 // at 'requested_addr'. If there are existing memory mappings at the same
3277 // location, however, they will be overwritten. If 'fixed' is false,
3278 // 'requested_addr' is only treated as a hint, the return value may or
3279 // may not start from the requested address. Unlike Linux mmap(), this
3280 // function returns NULL to indicate failure.
3281 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3282   char * addr;
3283   int flags;
3284 
3285   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3286   if (fixed) {
3287     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3288     flags |= MAP_FIXED;
3289   }
3290 
3291   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3292   // touch an uncommitted page. Otherwise, the read/write might
3293   // succeed if we have enough swap space to back the physical page.
3294   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3295                        flags, -1, 0);
3296 
3297   return addr == MAP_FAILED ? NULL : addr;
3298 }
3299 
3300 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3301 //   (req_addr != NULL) or with a given alignment.
3302 //  - bytes shall be a multiple of alignment.
3303 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3304 //  - alignment sets the alignment at which memory shall be allocated.
3305 //     It must be a multiple of allocation granularity.
3306 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3307 //  req_addr or NULL.
3308 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3309 
3310   size_t extra_size = bytes;
3311   if (req_addr == NULL && alignment > 0) {
3312     extra_size += alignment;
3313   }
3314 
3315   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3316     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3317     -1, 0);
3318   if (start == MAP_FAILED) {
3319     start = NULL;
3320   } else {
3321     if (req_addr != NULL) {
3322       if (start != req_addr) {
3323         ::munmap(start, extra_size);
3324         start = NULL;
3325       }
3326     } else {
3327       char* const start_aligned = align_up(start, alignment);
3328       char* const end_aligned = start_aligned + bytes;
3329       char* const end = start + extra_size;
3330       if (start_aligned > start) {
3331         ::munmap(start, start_aligned - start);
3332       }
3333       if (end_aligned < end) {
3334         ::munmap(end_aligned, end - end_aligned);
3335       }
3336       start = start_aligned;
3337     }
3338   }
3339   return start;
3340 }
3341 
3342 static int anon_munmap(char * addr, size_t size) {
3343   return ::munmap(addr, size) == 0;
3344 }
3345 
3346 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3347                             size_t alignment_hint) {
3348   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3349 }
3350 
3351 bool os::pd_release_memory(char* addr, size_t size) {
3352   return anon_munmap(addr, size);
3353 }
3354 
3355 static bool linux_mprotect(char* addr, size_t size, int prot) {
3356   // Linux wants the mprotect address argument to be page aligned.
3357   char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3358 
3359   // According to SUSv3, mprotect() should only be used with mappings
3360   // established by mmap(), and mmap() always maps whole pages. Unaligned
3361   // 'addr' likely indicates problem in the VM (e.g. trying to change
3362   // protection of malloc'ed or statically allocated memory). Check the
3363   // caller if you hit this assert.
3364   assert(addr == bottom, "sanity check");
3365 
3366   size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3367   return ::mprotect(bottom, size, prot) == 0;
3368 }
3369 
3370 // Set protections specified
3371 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3372                         bool is_committed) {
3373   unsigned int p = 0;
3374   switch (prot) {
3375   case MEM_PROT_NONE: p = PROT_NONE; break;
3376   case MEM_PROT_READ: p = PROT_READ; break;
3377   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3378   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3379   default:
3380     ShouldNotReachHere();
3381   }
3382   // is_committed is unused.
3383   return linux_mprotect(addr, bytes, p);
3384 }
3385 
3386 bool os::guard_memory(char* addr, size_t size) {
3387   return linux_mprotect(addr, size, PROT_NONE);
3388 }
3389 
3390 bool os::unguard_memory(char* addr, size_t size) {
3391   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3392 }
3393 
3394 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3395                                                     size_t page_size) {
3396   bool result = false;
3397   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3398                  MAP_ANONYMOUS|MAP_PRIVATE,
3399                  -1, 0);
3400   if (p != MAP_FAILED) {
3401     void *aligned_p = align_up(p, page_size);
3402 
3403     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3404 
3405     munmap(p, page_size * 2);
3406   }
3407 
3408   if (warn && !result) {
3409     warning("TransparentHugePages is not supported by the operating system.");
3410   }
3411 
3412   return result;
3413 }
3414 
3415 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3416   bool result = false;
3417   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3418                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3419                  -1, 0);
3420 
3421   if (p != MAP_FAILED) {
3422     // We don't know if this really is a huge page or not.
3423     FILE *fp = fopen("/proc/self/maps", "r");
3424     if (fp) {
3425       while (!feof(fp)) {
3426         char chars[257];
3427         long x = 0;
3428         if (fgets(chars, sizeof(chars), fp)) {
3429           if (sscanf(chars, "%lx-%*x", &x) == 1
3430               && x == (long)p) {
3431             if (strstr (chars, "hugepage")) {
3432               result = true;
3433               break;
3434             }
3435           }
3436         }
3437       }
3438       fclose(fp);
3439     }
3440     munmap(p, page_size);
3441   }
3442 
3443   if (warn && !result) {
3444     warning("HugeTLBFS is not supported by the operating system.");
3445   }
3446 
3447   return result;
3448 }
3449 
3450 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3451 //
3452 // From the coredump_filter documentation:
3453 //
3454 // - (bit 0) anonymous private memory
3455 // - (bit 1) anonymous shared memory
3456 // - (bit 2) file-backed private memory
3457 // - (bit 3) file-backed shared memory
3458 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3459 //           effective only if the bit 2 is cleared)
3460 // - (bit 5) hugetlb private memory
3461 // - (bit 6) hugetlb shared memory
3462 // - (bit 7) dax private memory
3463 // - (bit 8) dax shared memory
3464 //
3465 static void set_coredump_filter(bool largepages, bool dax_shared) {
3466   FILE *f;
3467   long cdm;
3468   bool filter_changed = false;
3469 
3470   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3471     return;
3472   }
3473 
3474   if (fscanf(f, "%lx", &cdm) != 1) {
3475     fclose(f);
3476     return;
3477   }
3478 
3479   rewind(f);
3480 
3481   if (largepages && (cdm & LARGEPAGES_BIT) == 0) {
3482     cdm |= LARGEPAGES_BIT;
3483     filter_changed = true;
3484   }
3485   if (dax_shared && (cdm & DAX_SHARED_BIT) == 0) {
3486     cdm |= DAX_SHARED_BIT;
3487     filter_changed = true;
3488   }
3489   if (filter_changed) {
3490     fprintf(f, "%#lx", cdm);
3491   }
3492 
3493   fclose(f);
3494 }
3495 
3496 // Large page support
3497 
3498 static size_t _large_page_size = 0;
3499 
3500 size_t os::Linux::find_large_page_size() {
3501   size_t large_page_size = 0;
3502 
3503   // large_page_size on Linux is used to round up heap size. x86 uses either
3504   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3505   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3506   // page as large as 256M.
3507   //
3508   // Here we try to figure out page size by parsing /proc/meminfo and looking
3509   // for a line with the following format:
3510   //    Hugepagesize:     2048 kB
3511   //
3512   // If we can't determine the value (e.g. /proc is not mounted, or the text
3513   // format has been changed), we'll use the largest page size supported by
3514   // the processor.
3515 
3516 #ifndef ZERO
3517   large_page_size =
3518     AARCH64_ONLY(2 * M)
3519     AMD64_ONLY(2 * M)
3520     ARM32_ONLY(2 * M)
3521     AARCH32_ONLY(2 * M)
3522     IA32_ONLY(4 * M)
3523     IA64_ONLY(256 * M)
3524     PPC_ONLY(4 * M)
3525     S390_ONLY(1 * M)
3526     SPARC_ONLY(4 * M);
3527 #endif // ZERO
3528 
3529   FILE *fp = fopen("/proc/meminfo", "r");
3530   if (fp) {
3531     while (!feof(fp)) {
3532       int x = 0;
3533       char buf[16];
3534       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3535         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3536           large_page_size = x * K;
3537           break;
3538         }
3539       } else {
3540         // skip to next line
3541         for (;;) {
3542           int ch = fgetc(fp);
3543           if (ch == EOF || ch == (int)'\n') break;
3544         }
3545       }
3546     }
3547     fclose(fp);
3548   }
3549 
3550   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3551     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3552             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3553             proper_unit_for_byte_size(large_page_size));
3554   }
3555 
3556   return large_page_size;
3557 }
3558 
3559 size_t os::Linux::setup_large_page_size() {
3560   _large_page_size = Linux::find_large_page_size();
3561   const size_t default_page_size = (size_t)Linux::page_size();
3562   if (_large_page_size > default_page_size) {
3563     _page_sizes[0] = _large_page_size;
3564     _page_sizes[1] = default_page_size;
3565     _page_sizes[2] = 0;
3566   }
3567 
3568   return _large_page_size;
3569 }
3570 
3571 bool os::Linux::setup_large_page_type(size_t page_size) {
3572   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3573       FLAG_IS_DEFAULT(UseSHM) &&
3574       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3575 
3576     // The type of large pages has not been specified by the user.
3577 
3578     // Try UseHugeTLBFS and then UseSHM.
3579     UseHugeTLBFS = UseSHM = true;
3580 
3581     // Don't try UseTransparentHugePages since there are known
3582     // performance issues with it turned on. This might change in the future.
3583     UseTransparentHugePages = false;
3584   }
3585 
3586   if (UseTransparentHugePages) {
3587     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3588     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3589       UseHugeTLBFS = false;
3590       UseSHM = false;
3591       return true;
3592     }
3593     UseTransparentHugePages = false;
3594   }
3595 
3596   if (UseHugeTLBFS) {
3597     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3598     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3599       UseSHM = false;
3600       return true;
3601     }
3602     UseHugeTLBFS = false;
3603   }
3604 
3605   return UseSHM;
3606 }
3607 
3608 void os::large_page_init() {
3609   if (!UseLargePages &&
3610       !UseTransparentHugePages &&
3611       !UseHugeTLBFS &&
3612       !UseSHM) {
3613     // Not using large pages.
3614     return;
3615   }
3616 
3617   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3618     // The user explicitly turned off large pages.
3619     // Ignore the rest of the large pages flags.
3620     UseTransparentHugePages = false;
3621     UseHugeTLBFS = false;
3622     UseSHM = false;
3623     return;
3624   }
3625 
3626   size_t large_page_size = Linux::setup_large_page_size();
3627   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3628 
3629   set_coredump_filter(true /*largepages*/, false /*dax_shared*/);
3630 }
3631 
3632 #ifndef SHM_HUGETLB
3633   #define SHM_HUGETLB 04000
3634 #endif
3635 
3636 #define shm_warning_format(format, ...)              \
3637   do {                                               \
3638     if (UseLargePages &&                             \
3639         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3640          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3641          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3642       warning(format, __VA_ARGS__);                  \
3643     }                                                \
3644   } while (0)
3645 
3646 #define shm_warning(str) shm_warning_format("%s", str)
3647 
3648 #define shm_warning_with_errno(str)                \
3649   do {                                             \
3650     int err = errno;                               \
3651     shm_warning_format(str " (error = %d)", err);  \
3652   } while (0)
3653 
3654 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3655   assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3656 
3657   if (!is_aligned(alignment, SHMLBA)) {
3658     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3659     return NULL;
3660   }
3661 
3662   // To ensure that we get 'alignment' aligned memory from shmat,
3663   // we pre-reserve aligned virtual memory and then attach to that.
3664 
3665   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3666   if (pre_reserved_addr == NULL) {
3667     // Couldn't pre-reserve aligned memory.
3668     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3669     return NULL;
3670   }
3671 
3672   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3673   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3674 
3675   if ((intptr_t)addr == -1) {
3676     int err = errno;
3677     shm_warning_with_errno("Failed to attach shared memory.");
3678 
3679     assert(err != EACCES, "Unexpected error");
3680     assert(err != EIDRM,  "Unexpected error");
3681     assert(err != EINVAL, "Unexpected error");
3682 
3683     // Since we don't know if the kernel unmapped the pre-reserved memory area
3684     // we can't unmap it, since that would potentially unmap memory that was
3685     // mapped from other threads.
3686     return NULL;
3687   }
3688 
3689   return addr;
3690 }
3691 
3692 static char* shmat_at_address(int shmid, char* req_addr) {
3693   if (!is_aligned(req_addr, SHMLBA)) {
3694     assert(false, "Requested address needs to be SHMLBA aligned");
3695     return NULL;
3696   }
3697 
3698   char* addr = (char*)shmat(shmid, req_addr, 0);
3699 
3700   if ((intptr_t)addr == -1) {
3701     shm_warning_with_errno("Failed to attach shared memory.");
3702     return NULL;
3703   }
3704 
3705   return addr;
3706 }
3707 
3708 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3709   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3710   if (req_addr != NULL) {
3711     assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3712     assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3713     return shmat_at_address(shmid, req_addr);
3714   }
3715 
3716   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3717   // return large page size aligned memory addresses when req_addr == NULL.
3718   // However, if the alignment is larger than the large page size, we have
3719   // to manually ensure that the memory returned is 'alignment' aligned.
3720   if (alignment > os::large_page_size()) {
3721     assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3722     return shmat_with_alignment(shmid, bytes, alignment);
3723   } else {
3724     return shmat_at_address(shmid, NULL);
3725   }
3726 }
3727 
3728 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3729                                             char* req_addr, bool exec) {
3730   // "exec" is passed in but not used.  Creating the shared image for
3731   // the code cache doesn't have an SHM_X executable permission to check.
3732   assert(UseLargePages && UseSHM, "only for SHM large pages");
3733   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3734   assert(is_aligned(req_addr, alignment), "Unaligned address");
3735 
3736   if (!is_aligned(bytes, os::large_page_size())) {
3737     return NULL; // Fallback to small pages.
3738   }
3739 
3740   // Create a large shared memory region to attach to based on size.
3741   // Currently, size is the total size of the heap.
3742   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3743   if (shmid == -1) {
3744     // Possible reasons for shmget failure:
3745     // 1. shmmax is too small for Java heap.
3746     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3747     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3748     // 2. not enough large page memory.
3749     //    > check available large pages: cat /proc/meminfo
3750     //    > increase amount of large pages:
3751     //          echo new_value > /proc/sys/vm/nr_hugepages
3752     //      Note 1: different Linux may use different name for this property,
3753     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3754     //      Note 2: it's possible there's enough physical memory available but
3755     //            they are so fragmented after a long run that they can't
3756     //            coalesce into large pages. Try to reserve large pages when
3757     //            the system is still "fresh".
3758     shm_warning_with_errno("Failed to reserve shared memory.");
3759     return NULL;
3760   }
3761 
3762   // Attach to the region.
3763   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3764 
3765   // Remove shmid. If shmat() is successful, the actual shared memory segment
3766   // will be deleted when it's detached by shmdt() or when the process
3767   // terminates. If shmat() is not successful this will remove the shared
3768   // segment immediately.
3769   shmctl(shmid, IPC_RMID, NULL);
3770 
3771   return addr;
3772 }
3773 
3774 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3775                                         int error) {
3776   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3777 
3778   bool warn_on_failure = UseLargePages &&
3779       (!FLAG_IS_DEFAULT(UseLargePages) ||
3780        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3781        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3782 
3783   if (warn_on_failure) {
3784     char msg[128];
3785     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3786                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3787     warning("%s", msg);
3788   }
3789 }
3790 
3791 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3792                                                         char* req_addr,
3793                                                         bool exec) {
3794   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3795   assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3796   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3797 
3798   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3799   char* addr = (char*)::mmap(req_addr, bytes, prot,
3800                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3801                              -1, 0);
3802 
3803   if (addr == MAP_FAILED) {
3804     warn_on_large_pages_failure(req_addr, bytes, errno);
3805     return NULL;
3806   }
3807 
3808   assert(is_aligned(addr, os::large_page_size()), "Must be");
3809 
3810   return addr;
3811 }
3812 
3813 // Reserve memory using mmap(MAP_HUGETLB).
3814 //  - bytes shall be a multiple of alignment.
3815 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3816 //  - alignment sets the alignment at which memory shall be allocated.
3817 //     It must be a multiple of allocation granularity.
3818 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3819 //  req_addr or NULL.
3820 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3821                                                          size_t alignment,
3822                                                          char* req_addr,
3823                                                          bool exec) {
3824   size_t large_page_size = os::large_page_size();
3825   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3826 
3827   assert(is_aligned(req_addr, alignment), "Must be");
3828   assert(is_aligned(bytes, alignment), "Must be");
3829 
3830   // First reserve - but not commit - the address range in small pages.
3831   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3832 
3833   if (start == NULL) {
3834     return NULL;
3835   }
3836 
3837   assert(is_aligned(start, alignment), "Must be");
3838 
3839   char* end = start + bytes;
3840 
3841   // Find the regions of the allocated chunk that can be promoted to large pages.
3842   char* lp_start = align_up(start, large_page_size);
3843   char* lp_end   = align_down(end, large_page_size);
3844 
3845   size_t lp_bytes = lp_end - lp_start;
3846 
3847   assert(is_aligned(lp_bytes, large_page_size), "Must be");
3848 
3849   if (lp_bytes == 0) {
3850     // The mapped region doesn't even span the start and the end of a large page.
3851     // Fall back to allocate a non-special area.
3852     ::munmap(start, end - start);
3853     return NULL;
3854   }
3855 
3856   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3857 
3858   void* result;
3859 
3860   // Commit small-paged leading area.
3861   if (start != lp_start) {
3862     result = ::mmap(start, lp_start - start, prot,
3863                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3864                     -1, 0);
3865     if (result == MAP_FAILED) {
3866       ::munmap(lp_start, end - lp_start);
3867       return NULL;
3868     }
3869   }
3870 
3871   // Commit large-paged area.
3872   result = ::mmap(lp_start, lp_bytes, prot,
3873                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3874                   -1, 0);
3875   if (result == MAP_FAILED) {
3876     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3877     // If the mmap above fails, the large pages region will be unmapped and we
3878     // have regions before and after with small pages. Release these regions.
3879     //
3880     // |  mapped  |  unmapped  |  mapped  |
3881     // ^          ^            ^          ^
3882     // start      lp_start     lp_end     end
3883     //
3884     ::munmap(start, lp_start - start);
3885     ::munmap(lp_end, end - lp_end);
3886     return NULL;
3887   }
3888 
3889   // Commit small-paged trailing area.
3890   if (lp_end != end) {
3891     result = ::mmap(lp_end, end - lp_end, prot,
3892                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3893                     -1, 0);
3894     if (result == MAP_FAILED) {
3895       ::munmap(start, lp_end - start);
3896       return NULL;
3897     }
3898   }
3899 
3900   return start;
3901 }
3902 
3903 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3904                                                    size_t alignment,
3905                                                    char* req_addr,
3906                                                    bool exec) {
3907   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3908   assert(is_aligned(req_addr, alignment), "Must be");
3909   assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3910   assert(is_power_of_2(os::large_page_size()), "Must be");
3911   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3912 
3913   if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3914     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3915   } else {
3916     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3917   }
3918 }
3919 
3920 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3921                                  char* req_addr, bool exec) {
3922   assert(UseLargePages, "only for large pages");
3923 
3924   char* addr;
3925   if (UseSHM) {
3926     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3927   } else {
3928     assert(UseHugeTLBFS, "must be");
3929     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3930   }
3931 
3932   if (addr != NULL) {
3933     if (UseNUMAInterleaving) {
3934       numa_make_global(addr, bytes);
3935     }
3936 
3937     // The memory is committed
3938     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3939   }
3940 
3941   return addr;
3942 }
3943 
3944 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3945   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3946   return shmdt(base) == 0;
3947 }
3948 
3949 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3950   return pd_release_memory(base, bytes);
3951 }
3952 
3953 bool os::release_memory_special(char* base, size_t bytes) {
3954   bool res;
3955   if (MemTracker::tracking_level() > NMT_minimal) {
3956     Tracker tkr(Tracker::release);
3957     res = os::Linux::release_memory_special_impl(base, bytes);
3958     if (res) {
3959       tkr.record((address)base, bytes);
3960     }
3961 
3962   } else {
3963     res = os::Linux::release_memory_special_impl(base, bytes);
3964   }
3965   return res;
3966 }
3967 
3968 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3969   assert(UseLargePages, "only for large pages");
3970   bool res;
3971 
3972   if (UseSHM) {
3973     res = os::Linux::release_memory_special_shm(base, bytes);
3974   } else {
3975     assert(UseHugeTLBFS, "must be");
3976     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3977   }
3978   return res;
3979 }
3980 
3981 size_t os::large_page_size() {
3982   return _large_page_size;
3983 }
3984 
3985 // With SysV SHM the entire memory region must be allocated as shared
3986 // memory.
3987 // HugeTLBFS allows application to commit large page memory on demand.
3988 // However, when committing memory with HugeTLBFS fails, the region
3989 // that was supposed to be committed will lose the old reservation
3990 // and allow other threads to steal that memory region. Because of this
3991 // behavior we can't commit HugeTLBFS memory.
3992 bool os::can_commit_large_page_memory() {
3993   return UseTransparentHugePages;
3994 }
3995 
3996 bool os::can_execute_large_page_memory() {
3997   return UseTransparentHugePages || UseHugeTLBFS;
3998 }
3999 
4000 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
4001   assert(file_desc >= 0, "file_desc is not valid");
4002   char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
4003   if (result != NULL) {
4004     if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4005       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4006     }
4007   }
4008   return result;
4009 }
4010 
4011 // Reserve memory at an arbitrary address, only if that area is
4012 // available (and not reserved for something else).
4013 
4014 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4015   const int max_tries = 10;
4016   char* base[max_tries];
4017   size_t size[max_tries];
4018   const size_t gap = 0x000000;
4019 
4020   // Assert only that the size is a multiple of the page size, since
4021   // that's all that mmap requires, and since that's all we really know
4022   // about at this low abstraction level.  If we need higher alignment,
4023   // we can either pass an alignment to this method or verify alignment
4024   // in one of the methods further up the call chain.  See bug 5044738.
4025   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4026 
4027   // Repeatedly allocate blocks until the block is allocated at the
4028   // right spot.
4029 
4030   // Linux mmap allows caller to pass an address as hint; give it a try first,
4031   // if kernel honors the hint then we can return immediately.
4032   char * addr = anon_mmap(requested_addr, bytes, false);
4033   if (addr == requested_addr) {
4034     return requested_addr;
4035   }
4036 
4037   if (addr != NULL) {
4038     // mmap() is successful but it fails to reserve at the requested address
4039     anon_munmap(addr, bytes);
4040   }
4041 
4042   int i;
4043   for (i = 0; i < max_tries; ++i) {
4044     base[i] = reserve_memory(bytes);
4045 
4046     if (base[i] != NULL) {
4047       // Is this the block we wanted?
4048       if (base[i] == requested_addr) {
4049         size[i] = bytes;
4050         break;
4051       }
4052 
4053       // Does this overlap the block we wanted? Give back the overlapped
4054       // parts and try again.
4055 
4056       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
4057       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
4058         unmap_memory(base[i], top_overlap);
4059         base[i] += top_overlap;
4060         size[i] = bytes - top_overlap;
4061       } else {
4062         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
4063         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
4064           unmap_memory(requested_addr, bottom_overlap);
4065           size[i] = bytes - bottom_overlap;
4066         } else {
4067           size[i] = bytes;
4068         }
4069       }
4070     }
4071   }
4072 
4073   // Give back the unused reserved pieces.
4074 
4075   for (int j = 0; j < i; ++j) {
4076     if (base[j] != NULL) {
4077       unmap_memory(base[j], size[j]);
4078     }
4079   }
4080 
4081   if (i < max_tries) {
4082     return requested_addr;
4083   } else {
4084     return NULL;
4085   }
4086 }
4087 
4088 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4089   return ::read(fd, buf, nBytes);
4090 }
4091 
4092 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4093   return ::pread(fd, buf, nBytes, offset);
4094 }
4095 
4096 // Short sleep, direct OS call.
4097 //
4098 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4099 // sched_yield(2) will actually give up the CPU:
4100 //
4101 //   * Alone on this pariticular CPU, keeps running.
4102 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
4103 //     (pre 2.6.39).
4104 //
4105 // So calling this with 0 is an alternative.
4106 //
4107 void os::naked_short_sleep(jlong ms) {
4108   struct timespec req;
4109 
4110   assert(ms < 1000, "Un-interruptable sleep, short time use only");
4111   req.tv_sec = 0;
4112   if (ms > 0) {
4113     req.tv_nsec = (ms % 1000) * 1000000;
4114   } else {
4115     req.tv_nsec = 1;
4116   }
4117 
4118   nanosleep(&req, NULL);
4119 
4120   return;
4121 }
4122 
4123 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4124 void os::infinite_sleep() {
4125   while (true) {    // sleep forever ...
4126     ::sleep(100);   // ... 100 seconds at a time
4127   }
4128 }
4129 
4130 // Used to convert frequent JVM_Yield() to nops
4131 bool os::dont_yield() {
4132   return DontYieldALot;
4133 }
4134 
4135 void os::naked_yield() {
4136   sched_yield();
4137 }
4138 
4139 ////////////////////////////////////////////////////////////////////////////////
4140 // thread priority support
4141 
4142 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4143 // only supports dynamic priority, static priority must be zero. For real-time
4144 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4145 // However, for large multi-threaded applications, SCHED_RR is not only slower
4146 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4147 // of 5 runs - Sep 2005).
4148 //
4149 // The following code actually changes the niceness of kernel-thread/LWP. It
4150 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4151 // not the entire user process, and user level threads are 1:1 mapped to kernel
4152 // threads. It has always been the case, but could change in the future. For
4153 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4154 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4155 
4156 int os::java_to_os_priority[CriticalPriority + 1] = {
4157   19,              // 0 Entry should never be used
4158 
4159    4,              // 1 MinPriority
4160    3,              // 2
4161    2,              // 3
4162 
4163    1,              // 4
4164    0,              // 5 NormPriority
4165   -1,              // 6
4166 
4167   -2,              // 7
4168   -3,              // 8
4169   -4,              // 9 NearMaxPriority
4170 
4171   -5,              // 10 MaxPriority
4172 
4173   -5               // 11 CriticalPriority
4174 };
4175 
4176 static int prio_init() {
4177   if (ThreadPriorityPolicy == 1) {
4178     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4179     // if effective uid is not root. Perhaps, a more elegant way of doing
4180     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4181     if (geteuid() != 0) {
4182       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4183         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4184       }
4185       ThreadPriorityPolicy = 0;
4186     }
4187   }
4188   if (UseCriticalJavaThreadPriority) {
4189     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4190   }
4191   return 0;
4192 }
4193 
4194 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4195   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4196 
4197   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4198   return (ret == 0) ? OS_OK : OS_ERR;
4199 }
4200 
4201 OSReturn os::get_native_priority(const Thread* const thread,
4202                                  int *priority_ptr) {
4203   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4204     *priority_ptr = java_to_os_priority[NormPriority];
4205     return OS_OK;
4206   }
4207 
4208   errno = 0;
4209   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4210   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4211 }
4212 
4213 // Hint to the underlying OS that a task switch would not be good.
4214 // Void return because it's a hint and can fail.
4215 void os::hint_no_preempt() {}
4216 
4217 ////////////////////////////////////////////////////////////////////////////////
4218 // suspend/resume support
4219 
4220 //  The low-level signal-based suspend/resume support is a remnant from the
4221 //  old VM-suspension that used to be for java-suspension, safepoints etc,
4222 //  within hotspot. Currently used by JFR's OSThreadSampler
4223 //
4224 //  The remaining code is greatly simplified from the more general suspension
4225 //  code that used to be used.
4226 //
4227 //  The protocol is quite simple:
4228 //  - suspend:
4229 //      - sends a signal to the target thread
4230 //      - polls the suspend state of the osthread using a yield loop
4231 //      - target thread signal handler (SR_handler) sets suspend state
4232 //        and blocks in sigsuspend until continued
4233 //  - resume:
4234 //      - sets target osthread state to continue
4235 //      - sends signal to end the sigsuspend loop in the SR_handler
4236 //
4237 //  Note that the SR_lock plays no role in this suspend/resume protocol,
4238 //  but is checked for NULL in SR_handler as a thread termination indicator.
4239 //  The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4240 //
4241 //  Note that resume_clear_context() and suspend_save_context() are needed
4242 //  by SR_handler(), so that fetch_frame_from_ucontext() works,
4243 //  which in part is used by:
4244 //    - Forte Analyzer: AsyncGetCallTrace()
4245 //    - StackBanging: get_frame_at_stack_banging_point()
4246 
4247 static void resume_clear_context(OSThread *osthread) {
4248   osthread->set_ucontext(NULL);
4249   osthread->set_siginfo(NULL);
4250 }
4251 
4252 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4253                                  ucontext_t* context) {
4254   osthread->set_ucontext(context);
4255   osthread->set_siginfo(siginfo);
4256 }
4257 
4258 // Handler function invoked when a thread's execution is suspended or
4259 // resumed. We have to be careful that only async-safe functions are
4260 // called here (Note: most pthread functions are not async safe and
4261 // should be avoided.)
4262 //
4263 // Note: sigwait() is a more natural fit than sigsuspend() from an
4264 // interface point of view, but sigwait() prevents the signal hander
4265 // from being run. libpthread would get very confused by not having
4266 // its signal handlers run and prevents sigwait()'s use with the
4267 // mutex granting granting signal.
4268 //
4269 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4270 //
4271 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4272   // Save and restore errno to avoid confusing native code with EINTR
4273   // after sigsuspend.
4274   int old_errno = errno;
4275 
4276   Thread* thread = Thread::current_or_null_safe();
4277   assert(thread != NULL, "Missing current thread in SR_handler");
4278 
4279   // On some systems we have seen signal delivery get "stuck" until the signal
4280   // mask is changed as part of thread termination. Check that the current thread
4281   // has not already terminated (via SR_lock()) - else the following assertion
4282   // will fail because the thread is no longer a JavaThread as the ~JavaThread
4283   // destructor has completed.
4284 
4285   if (thread->SR_lock() == NULL) {
4286     return;
4287   }
4288 
4289   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4290 
4291   OSThread* osthread = thread->osthread();
4292 
4293   os::SuspendResume::State current = osthread->sr.state();
4294   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4295     suspend_save_context(osthread, siginfo, context);
4296 
4297     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4298     os::SuspendResume::State state = osthread->sr.suspended();
4299     if (state == os::SuspendResume::SR_SUSPENDED) {
4300       sigset_t suspend_set;  // signals for sigsuspend()
4301       sigemptyset(&suspend_set);
4302       // get current set of blocked signals and unblock resume signal
4303       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4304       sigdelset(&suspend_set, SR_signum);
4305 
4306       sr_semaphore.signal();
4307       // wait here until we are resumed
4308       while (1) {
4309         sigsuspend(&suspend_set);
4310 
4311         os::SuspendResume::State result = osthread->sr.running();
4312         if (result == os::SuspendResume::SR_RUNNING) {
4313           sr_semaphore.signal();
4314           break;
4315         }
4316       }
4317 
4318     } else if (state == os::SuspendResume::SR_RUNNING) {
4319       // request was cancelled, continue
4320     } else {
4321       ShouldNotReachHere();
4322     }
4323 
4324     resume_clear_context(osthread);
4325   } else if (current == os::SuspendResume::SR_RUNNING) {
4326     // request was cancelled, continue
4327   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4328     // ignore
4329   } else {
4330     // ignore
4331   }
4332 
4333   errno = old_errno;
4334 }
4335 
4336 static int SR_initialize() {
4337   struct sigaction act;
4338   char *s;
4339 
4340   // Get signal number to use for suspend/resume
4341   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4342     int sig = ::strtol(s, 0, 10);
4343     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4344         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4345       SR_signum = sig;
4346     } else {
4347       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4348               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4349     }
4350   }
4351 
4352   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4353          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4354 
4355   sigemptyset(&SR_sigset);
4356   sigaddset(&SR_sigset, SR_signum);
4357 
4358   // Set up signal handler for suspend/resume
4359   act.sa_flags = SA_RESTART|SA_SIGINFO;
4360   act.sa_handler = (void (*)(int)) SR_handler;
4361 
4362   // SR_signum is blocked by default.
4363   // 4528190 - We also need to block pthread restart signal (32 on all
4364   // supported Linux platforms). Note that LinuxThreads need to block
4365   // this signal for all threads to work properly. So we don't have
4366   // to use hard-coded signal number when setting up the mask.
4367   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4368 
4369   if (sigaction(SR_signum, &act, 0) == -1) {
4370     return -1;
4371   }
4372 
4373   // Save signal flag
4374   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4375   return 0;
4376 }
4377 
4378 static int sr_notify(OSThread* osthread) {
4379   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4380   assert_status(status == 0, status, "pthread_kill");
4381   return status;
4382 }
4383 
4384 // "Randomly" selected value for how long we want to spin
4385 // before bailing out on suspending a thread, also how often
4386 // we send a signal to a thread we want to resume
4387 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4388 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4389 
4390 // returns true on success and false on error - really an error is fatal
4391 // but this seems the normal response to library errors
4392 static bool do_suspend(OSThread* osthread) {
4393   assert(osthread->sr.is_running(), "thread should be running");
4394   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4395 
4396   // mark as suspended and send signal
4397   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4398     // failed to switch, state wasn't running?
4399     ShouldNotReachHere();
4400     return false;
4401   }
4402 
4403   if (sr_notify(osthread) != 0) {
4404     ShouldNotReachHere();
4405   }
4406 
4407   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4408   while (true) {
4409     if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4410       break;
4411     } else {
4412       // timeout
4413       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4414       if (cancelled == os::SuspendResume::SR_RUNNING) {
4415         return false;
4416       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4417         // make sure that we consume the signal on the semaphore as well
4418         sr_semaphore.wait();
4419         break;
4420       } else {
4421         ShouldNotReachHere();
4422         return false;
4423       }
4424     }
4425   }
4426 
4427   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4428   return true;
4429 }
4430 
4431 static void do_resume(OSThread* osthread) {
4432   assert(osthread->sr.is_suspended(), "thread should be suspended");
4433   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4434 
4435   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4436     // failed to switch to WAKEUP_REQUEST
4437     ShouldNotReachHere();
4438     return;
4439   }
4440 
4441   while (true) {
4442     if (sr_notify(osthread) == 0) {
4443       if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4444         if (osthread->sr.is_running()) {
4445           return;
4446         }
4447       }
4448     } else {
4449       ShouldNotReachHere();
4450     }
4451   }
4452 
4453   guarantee(osthread->sr.is_running(), "Must be running!");
4454 }
4455 
4456 ///////////////////////////////////////////////////////////////////////////////////
4457 // signal handling (except suspend/resume)
4458 
4459 // This routine may be used by user applications as a "hook" to catch signals.
4460 // The user-defined signal handler must pass unrecognized signals to this
4461 // routine, and if it returns true (non-zero), then the signal handler must
4462 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4463 // routine will never retun false (zero), but instead will execute a VM panic
4464 // routine kill the process.
4465 //
4466 // If this routine returns false, it is OK to call it again.  This allows
4467 // the user-defined signal handler to perform checks either before or after
4468 // the VM performs its own checks.  Naturally, the user code would be making
4469 // a serious error if it tried to handle an exception (such as a null check
4470 // or breakpoint) that the VM was generating for its own correct operation.
4471 //
4472 // This routine may recognize any of the following kinds of signals:
4473 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4474 // It should be consulted by handlers for any of those signals.
4475 //
4476 // The caller of this routine must pass in the three arguments supplied
4477 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4478 // field of the structure passed to sigaction().  This routine assumes that
4479 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4480 //
4481 // Note that the VM will print warnings if it detects conflicting signal
4482 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4483 //
4484 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4485                                                  siginfo_t* siginfo,
4486                                                  void* ucontext,
4487                                                  int abort_if_unrecognized);
4488 
4489 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4490   assert(info != NULL && uc != NULL, "it must be old kernel");
4491   int orig_errno = errno;  // Preserve errno value over signal handler.
4492   JVM_handle_linux_signal(sig, info, uc, true);
4493   errno = orig_errno;
4494 }
4495 
4496 
4497 // This boolean allows users to forward their own non-matching signals
4498 // to JVM_handle_linux_signal, harmlessly.
4499 bool os::Linux::signal_handlers_are_installed = false;
4500 
4501 // For signal-chaining
4502 struct sigaction sigact[NSIG];
4503 uint64_t sigs = 0;
4504 #if (64 < NSIG-1)
4505 #error "Not all signals can be encoded in sigs. Adapt its type!"
4506 #endif
4507 bool os::Linux::libjsig_is_loaded = false;
4508 typedef struct sigaction *(*get_signal_t)(int);
4509 get_signal_t os::Linux::get_signal_action = NULL;
4510 
4511 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4512   struct sigaction *actp = NULL;
4513 
4514   if (libjsig_is_loaded) {
4515     // Retrieve the old signal handler from libjsig
4516     actp = (*get_signal_action)(sig);
4517   }
4518   if (actp == NULL) {
4519     // Retrieve the preinstalled signal handler from jvm
4520     actp = get_preinstalled_handler(sig);
4521   }
4522 
4523   return actp;
4524 }
4525 
4526 static bool call_chained_handler(struct sigaction *actp, int sig,
4527                                  siginfo_t *siginfo, void *context) {
4528   // Call the old signal handler
4529   if (actp->sa_handler == SIG_DFL) {
4530     // It's more reasonable to let jvm treat it as an unexpected exception
4531     // instead of taking the default action.
4532     return false;
4533   } else if (actp->sa_handler != SIG_IGN) {
4534     if ((actp->sa_flags & SA_NODEFER) == 0) {
4535       // automaticlly block the signal
4536       sigaddset(&(actp->sa_mask), sig);
4537     }
4538 
4539     sa_handler_t hand = NULL;
4540     sa_sigaction_t sa = NULL;
4541     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4542     // retrieve the chained handler
4543     if (siginfo_flag_set) {
4544       sa = actp->sa_sigaction;
4545     } else {
4546       hand = actp->sa_handler;
4547     }
4548 
4549     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4550       actp->sa_handler = SIG_DFL;
4551     }
4552 
4553     // try to honor the signal mask
4554     sigset_t oset;
4555     sigemptyset(&oset);
4556     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4557 
4558     // call into the chained handler
4559     if (siginfo_flag_set) {
4560       (*sa)(sig, siginfo, context);
4561     } else {
4562       (*hand)(sig);
4563     }
4564 
4565     // restore the signal mask
4566     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4567   }
4568   // Tell jvm's signal handler the signal is taken care of.
4569   return true;
4570 }
4571 
4572 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4573   bool chained = false;
4574   // signal-chaining
4575   if (UseSignalChaining) {
4576     struct sigaction *actp = get_chained_signal_action(sig);
4577     if (actp != NULL) {
4578       chained = call_chained_handler(actp, sig, siginfo, context);
4579     }
4580   }
4581   return chained;
4582 }
4583 
4584 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4585   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4586     return &sigact[sig];
4587   }
4588   return NULL;
4589 }
4590 
4591 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4592   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4593   sigact[sig] = oldAct;
4594   sigs |= (uint64_t)1 << (sig-1);
4595 }
4596 
4597 // for diagnostic
4598 int sigflags[NSIG];
4599 
4600 int os::Linux::get_our_sigflags(int sig) {
4601   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4602   return sigflags[sig];
4603 }
4604 
4605 void os::Linux::set_our_sigflags(int sig, int flags) {
4606   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4607   if (sig > 0 && sig < NSIG) {
4608     sigflags[sig] = flags;
4609   }
4610 }
4611 
4612 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4613   // Check for overwrite.
4614   struct sigaction oldAct;
4615   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4616 
4617   void* oldhand = oldAct.sa_sigaction
4618                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4619                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4620   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4621       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4622       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4623     if (AllowUserSignalHandlers || !set_installed) {
4624       // Do not overwrite; user takes responsibility to forward to us.
4625       return;
4626     } else if (UseSignalChaining) {
4627       // save the old handler in jvm
4628       save_preinstalled_handler(sig, oldAct);
4629       // libjsig also interposes the sigaction() call below and saves the
4630       // old sigaction on it own.
4631     } else {
4632       fatal("Encountered unexpected pre-existing sigaction handler "
4633             "%#lx for signal %d.", (long)oldhand, sig);
4634     }
4635   }
4636 
4637   struct sigaction sigAct;
4638   sigfillset(&(sigAct.sa_mask));
4639   sigAct.sa_handler = SIG_DFL;
4640   if (!set_installed) {
4641     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4642   } else {
4643     sigAct.sa_sigaction = signalHandler;
4644     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4645   }
4646   // Save flags, which are set by ours
4647   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4648   sigflags[sig] = sigAct.sa_flags;
4649 
4650   int ret = sigaction(sig, &sigAct, &oldAct);
4651   assert(ret == 0, "check");
4652 
4653   void* oldhand2  = oldAct.sa_sigaction
4654                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4655                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4656   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4657 }
4658 
4659 // install signal handlers for signals that HotSpot needs to
4660 // handle in order to support Java-level exception handling.
4661 
4662 void os::Linux::install_signal_handlers() {
4663   if (!signal_handlers_are_installed) {
4664     signal_handlers_are_installed = true;
4665 
4666     // signal-chaining
4667     typedef void (*signal_setting_t)();
4668     signal_setting_t begin_signal_setting = NULL;
4669     signal_setting_t end_signal_setting = NULL;
4670     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4671                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4672     if (begin_signal_setting != NULL) {
4673       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4674                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4675       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4676                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4677       libjsig_is_loaded = true;
4678       assert(UseSignalChaining, "should enable signal-chaining");
4679     }
4680     if (libjsig_is_loaded) {
4681       // Tell libjsig jvm is setting signal handlers
4682       (*begin_signal_setting)();
4683     }
4684 
4685     set_signal_handler(SIGSEGV, true);
4686     set_signal_handler(SIGPIPE, true);
4687     set_signal_handler(SIGBUS, true);
4688     set_signal_handler(SIGILL, true);
4689     set_signal_handler(SIGFPE, true);
4690 #if defined(PPC64)
4691     set_signal_handler(SIGTRAP, true);
4692 #endif
4693     set_signal_handler(SIGXFSZ, true);
4694 
4695     if (libjsig_is_loaded) {
4696       // Tell libjsig jvm finishes setting signal handlers
4697       (*end_signal_setting)();
4698     }
4699 
4700     // We don't activate signal checker if libjsig is in place, we trust ourselves
4701     // and if UserSignalHandler is installed all bets are off.
4702     // Log that signal checking is off only if -verbose:jni is specified.
4703     if (CheckJNICalls) {
4704       if (libjsig_is_loaded) {
4705         if (PrintJNIResolving) {
4706           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4707         }
4708         check_signals = false;
4709       }
4710       if (AllowUserSignalHandlers) {
4711         if (PrintJNIResolving) {
4712           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4713         }
4714         check_signals = false;
4715       }
4716     }
4717   }
4718 }
4719 
4720 // This is the fastest way to get thread cpu time on Linux.
4721 // Returns cpu time (user+sys) for any thread, not only for current.
4722 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4723 // It might work on 2.6.10+ with a special kernel/glibc patch.
4724 // For reference, please, see IEEE Std 1003.1-2004:
4725 //   http://www.unix.org/single_unix_specification
4726 
4727 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4728   struct timespec tp;
4729   int rc = os::Linux::clock_gettime(clockid, &tp);
4730   assert(rc == 0, "clock_gettime is expected to return 0 code");
4731 
4732   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4733 }
4734 
4735 void os::Linux::initialize_os_info() {
4736   assert(_os_version == 0, "OS info already initialized");
4737 
4738   struct utsname _uname;
4739 
4740   uint32_t major;
4741   uint32_t minor;
4742   uint32_t fix;
4743 
4744   int rc;
4745 
4746   // Kernel version is unknown if
4747   // verification below fails.
4748   _os_version = 0x01000000;
4749 
4750   rc = uname(&_uname);
4751   if (rc != -1) {
4752 
4753     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4754     if (rc == 3) {
4755 
4756       if (major < 256 && minor < 256 && fix < 256) {
4757         // Kernel version format is as expected,
4758         // set it overriding unknown state.
4759         _os_version = (major << 16) |
4760                       (minor << 8 ) |
4761                       (fix   << 0 ) ;
4762       }
4763     }
4764   }
4765 }
4766 
4767 uint32_t os::Linux::os_version() {
4768   assert(_os_version != 0, "not initialized");
4769   return _os_version & 0x00FFFFFF;
4770 }
4771 
4772 bool os::Linux::os_version_is_known() {
4773   assert(_os_version != 0, "not initialized");
4774   return _os_version & 0x01000000 ? false : true;
4775 }
4776 
4777 /////
4778 // glibc on Linux platform uses non-documented flag
4779 // to indicate, that some special sort of signal
4780 // trampoline is used.
4781 // We will never set this flag, and we should
4782 // ignore this flag in our diagnostic
4783 #ifdef SIGNIFICANT_SIGNAL_MASK
4784   #undef SIGNIFICANT_SIGNAL_MASK
4785 #endif
4786 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4787 
4788 static const char* get_signal_handler_name(address handler,
4789                                            char* buf, int buflen) {
4790   int offset = 0;
4791   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4792   if (found) {
4793     // skip directory names
4794     const char *p1, *p2;
4795     p1 = buf;
4796     size_t len = strlen(os::file_separator());
4797     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4798     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4799   } else {
4800     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4801   }
4802   return buf;
4803 }
4804 
4805 static void print_signal_handler(outputStream* st, int sig,
4806                                  char* buf, size_t buflen) {
4807   struct sigaction sa;
4808 
4809   sigaction(sig, NULL, &sa);
4810 
4811   // See comment for SIGNIFICANT_SIGNAL_MASK define
4812   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4813 
4814   st->print("%s: ", os::exception_name(sig, buf, buflen));
4815 
4816   address handler = (sa.sa_flags & SA_SIGINFO)
4817     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4818     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4819 
4820   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4821     st->print("SIG_DFL");
4822   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4823     st->print("SIG_IGN");
4824   } else {
4825     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4826   }
4827 
4828   st->print(", sa_mask[0]=");
4829   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4830 
4831   address rh = VMError::get_resetted_sighandler(sig);
4832   // May be, handler was resetted by VMError?
4833   if (rh != NULL) {
4834     handler = rh;
4835     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4836   }
4837 
4838   st->print(", sa_flags=");
4839   os::Posix::print_sa_flags(st, sa.sa_flags);
4840 
4841   // Check: is it our handler?
4842   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4843       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4844     // It is our signal handler
4845     // check for flags, reset system-used one!
4846     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4847       st->print(
4848                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4849                 os::Linux::get_our_sigflags(sig));
4850     }
4851   }
4852   st->cr();
4853 }
4854 
4855 
4856 #define DO_SIGNAL_CHECK(sig)                      \
4857   do {                                            \
4858     if (!sigismember(&check_signal_done, sig)) {  \
4859       os::Linux::check_signal_handler(sig);       \
4860     }                                             \
4861   } while (0)
4862 
4863 // This method is a periodic task to check for misbehaving JNI applications
4864 // under CheckJNI, we can add any periodic checks here
4865 
4866 void os::run_periodic_checks() {
4867   if (check_signals == false) return;
4868 
4869   // SEGV and BUS if overridden could potentially prevent
4870   // generation of hs*.log in the event of a crash, debugging
4871   // such a case can be very challenging, so we absolutely
4872   // check the following for a good measure:
4873   DO_SIGNAL_CHECK(SIGSEGV);
4874   DO_SIGNAL_CHECK(SIGILL);
4875   DO_SIGNAL_CHECK(SIGFPE);
4876   DO_SIGNAL_CHECK(SIGBUS);
4877   DO_SIGNAL_CHECK(SIGPIPE);
4878   DO_SIGNAL_CHECK(SIGXFSZ);
4879 #if defined(PPC64)
4880   DO_SIGNAL_CHECK(SIGTRAP);
4881 #endif
4882 
4883   // ReduceSignalUsage allows the user to override these handlers
4884   // see comments at the very top and jvm_md.h
4885   if (!ReduceSignalUsage) {
4886     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4887     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4888     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4889     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4890   }
4891 
4892   DO_SIGNAL_CHECK(SR_signum);
4893 }
4894 
4895 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4896 
4897 static os_sigaction_t os_sigaction = NULL;
4898 
4899 void os::Linux::check_signal_handler(int sig) {
4900   char buf[O_BUFLEN];
4901   address jvmHandler = NULL;
4902 
4903 
4904   struct sigaction act;
4905   if (os_sigaction == NULL) {
4906     // only trust the default sigaction, in case it has been interposed
4907     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4908     if (os_sigaction == NULL) return;
4909   }
4910 
4911   os_sigaction(sig, (struct sigaction*)NULL, &act);
4912 
4913 
4914   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4915 
4916   address thisHandler = (act.sa_flags & SA_SIGINFO)
4917     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4918     : CAST_FROM_FN_PTR(address, act.sa_handler);
4919 
4920 
4921   switch (sig) {
4922   case SIGSEGV:
4923   case SIGBUS:
4924   case SIGFPE:
4925   case SIGPIPE:
4926   case SIGILL:
4927   case SIGXFSZ:
4928     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4929     break;
4930 
4931   case SHUTDOWN1_SIGNAL:
4932   case SHUTDOWN2_SIGNAL:
4933   case SHUTDOWN3_SIGNAL:
4934   case BREAK_SIGNAL:
4935     jvmHandler = (address)user_handler();
4936     break;
4937 
4938   default:
4939     if (sig == SR_signum) {
4940       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4941     } else {
4942       return;
4943     }
4944     break;
4945   }
4946 
4947   if (thisHandler != jvmHandler) {
4948     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4949     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4950     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4951     // No need to check this sig any longer
4952     sigaddset(&check_signal_done, sig);
4953     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4954     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4955       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4956                     exception_name(sig, buf, O_BUFLEN));
4957     }
4958   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4959     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4960     tty->print("expected:");
4961     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4962     tty->cr();
4963     tty->print("  found:");
4964     os::Posix::print_sa_flags(tty, act.sa_flags);
4965     tty->cr();
4966     // No need to check this sig any longer
4967     sigaddset(&check_signal_done, sig);
4968   }
4969 
4970   // Dump all the signal
4971   if (sigismember(&check_signal_done, sig)) {
4972     print_signal_handlers(tty, buf, O_BUFLEN);
4973   }
4974 }
4975 
4976 extern void report_error(char* file_name, int line_no, char* title,
4977                          char* format, ...);
4978 
4979 // this is called _before_ most of the global arguments have been parsed
4980 void os::init(void) {
4981   char dummy;   // used to get a guess on initial stack address
4982 
4983   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4984 
4985   init_random(1234567);
4986 
4987   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4988   if (Linux::page_size() == -1) {
4989     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4990           os::strerror(errno));
4991   }
4992   init_page_sizes((size_t) Linux::page_size());
4993 
4994   Linux::initialize_system_info();
4995 
4996   Linux::initialize_os_info();
4997 
4998   // _main_thread points to the thread that created/loaded the JVM.
4999   Linux::_main_thread = pthread_self();
5000 
5001   Linux::clock_init();
5002   initial_time_count = javaTimeNanos();
5003 
5004   // retrieve entry point for pthread_setname_np
5005   Linux::_pthread_setname_np =
5006     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5007 
5008   os::Posix::init();
5009 }
5010 
5011 // To install functions for atexit system call
5012 extern "C" {
5013   static void perfMemory_exit_helper() {
5014     perfMemory_exit();
5015   }
5016 }
5017 
5018 void os::pd_init_container_support() {
5019   OSContainer::init();
5020 }
5021 
5022 // this is called _after_ the global arguments have been parsed
5023 jint os::init_2(void) {
5024 
5025   os::Posix::init_2();
5026 
5027   Linux::fast_thread_clock_init();
5028 
5029   // initialize suspend/resume support - must do this before signal_sets_init()
5030   if (SR_initialize() != 0) {
5031     perror("SR_initialize failed");
5032     return JNI_ERR;
5033   }
5034 
5035   Linux::signal_sets_init();
5036   Linux::install_signal_handlers();
5037   // Initialize data for jdk.internal.misc.Signal
5038   if (!ReduceSignalUsage) {
5039     jdk_misc_signal_init();
5040   }
5041 
5042   // Check and sets minimum stack sizes against command line options
5043   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5044     return JNI_ERR;
5045   }
5046 
5047   suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5048   if (!suppress_primordial_thread_resolution) {
5049     Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5050   }
5051 
5052 #if defined(IA32)
5053   workaround_expand_exec_shield_cs_limit();
5054 #endif
5055 
5056   Linux::libpthread_init();
5057   Linux::sched_getcpu_init();
5058   log_info(os)("HotSpot is running with %s, %s",
5059                Linux::glibc_version(), Linux::libpthread_version());
5060 
5061   if (UseNUMA) {
5062     if (!Linux::libnuma_init()) {
5063       UseNUMA = false;
5064     } else {
5065       if ((Linux::numa_max_node() < 1) || Linux::isbound_to_single_node()) {
5066         // If there's only one node (they start from 0) or if the process
5067         // is bound explicitly to a single node using membind, disable NUMA.
5068         UseNUMA = false;
5069       }
5070     }
5071 
5072     if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5073       // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5074       // we can make the adaptive lgrp chunk resizing work. If the user specified both
5075       // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5076       // and disable adaptive resizing.
5077       if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5078         warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5079                 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5080         UseAdaptiveSizePolicy = false;
5081         UseAdaptiveNUMAChunkSizing = false;
5082       }
5083     }
5084 
5085     if (!UseNUMA && ForceNUMA) {
5086       UseNUMA = true;
5087     }
5088   }
5089 
5090   if (MaxFDLimit) {
5091     // set the number of file descriptors to max. print out error
5092     // if getrlimit/setrlimit fails but continue regardless.
5093     struct rlimit nbr_files;
5094     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5095     if (status != 0) {
5096       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5097     } else {
5098       nbr_files.rlim_cur = nbr_files.rlim_max;
5099       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5100       if (status != 0) {
5101         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5102       }
5103     }
5104   }
5105 
5106   // Initialize lock used to serialize thread creation (see os::create_thread)
5107   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5108 
5109   // at-exit methods are called in the reverse order of their registration.
5110   // atexit functions are called on return from main or as a result of a
5111   // call to exit(3C). There can be only 32 of these functions registered
5112   // and atexit() does not set errno.
5113 
5114   if (PerfAllowAtExitRegistration) {
5115     // only register atexit functions if PerfAllowAtExitRegistration is set.
5116     // atexit functions can be delayed until process exit time, which
5117     // can be problematic for embedded VM situations. Embedded VMs should
5118     // call DestroyJavaVM() to assure that VM resources are released.
5119 
5120     // note: perfMemory_exit_helper atexit function may be removed in
5121     // the future if the appropriate cleanup code can be added to the
5122     // VM_Exit VMOperation's doit method.
5123     if (atexit(perfMemory_exit_helper) != 0) {
5124       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5125     }
5126   }
5127 
5128   // initialize thread priority policy
5129   prio_init();
5130 
5131   if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5132     set_coredump_filter(false /*largepages*/, true /*dax_shared*/);
5133   }
5134   return JNI_OK;
5135 }
5136 
5137 // Mark the polling page as unreadable
5138 void os::make_polling_page_unreadable(void) {
5139   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5140     fatal("Could not disable polling page");
5141   }
5142 }
5143 
5144 // Mark the polling page as readable
5145 void os::make_polling_page_readable(void) {
5146   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5147     fatal("Could not enable polling page");
5148   }
5149 }
5150 
5151 // older glibc versions don't have this macro (which expands to
5152 // an optimized bit-counting function) so we have to roll our own
5153 #ifndef CPU_COUNT
5154 
5155 static int _cpu_count(const cpu_set_t* cpus) {
5156   int count = 0;
5157   // only look up to the number of configured processors
5158   for (int i = 0; i < os::processor_count(); i++) {
5159     if (CPU_ISSET(i, cpus)) {
5160       count++;
5161     }
5162   }
5163   return count;
5164 }
5165 
5166 #define CPU_COUNT(cpus) _cpu_count(cpus)
5167 
5168 #endif // CPU_COUNT
5169 
5170 // Get the current number of available processors for this process.
5171 // This value can change at any time during a process's lifetime.
5172 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5173 // If it appears there may be more than 1024 processors then we do a
5174 // dynamic check - see 6515172 for details.
5175 // If anything goes wrong we fallback to returning the number of online
5176 // processors - which can be greater than the number available to the process.
5177 int os::Linux::active_processor_count() {
5178   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
5179   cpu_set_t* cpus_p = &cpus;
5180   int cpus_size = sizeof(cpu_set_t);
5181 
5182   int configured_cpus = os::processor_count();  // upper bound on available cpus
5183   int cpu_count = 0;
5184 
5185 // old build platforms may not support dynamic cpu sets
5186 #ifdef CPU_ALLOC
5187 
5188   // To enable easy testing of the dynamic path on different platforms we
5189   // introduce a diagnostic flag: UseCpuAllocPath
5190   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5191     // kernel may use a mask bigger than cpu_set_t
5192     log_trace(os)("active_processor_count: using dynamic path %s"
5193                   "- configured processors: %d",
5194                   UseCpuAllocPath ? "(forced) " : "",
5195                   configured_cpus);
5196     cpus_p = CPU_ALLOC(configured_cpus);
5197     if (cpus_p != NULL) {
5198       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5199       // zero it just to be safe
5200       CPU_ZERO_S(cpus_size, cpus_p);
5201     }
5202     else {
5203        // failed to allocate so fallback to online cpus
5204        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5205        log_trace(os)("active_processor_count: "
5206                      "CPU_ALLOC failed (%s) - using "
5207                      "online processor count: %d",
5208                      os::strerror(errno), online_cpus);
5209        return online_cpus;
5210     }
5211   }
5212   else {
5213     log_trace(os)("active_processor_count: using static path - configured processors: %d",
5214                   configured_cpus);
5215   }
5216 #else // CPU_ALLOC
5217 // these stubs won't be executed
5218 #define CPU_COUNT_S(size, cpus) -1
5219 #define CPU_FREE(cpus)
5220 
5221   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5222                 configured_cpus);
5223 #endif // CPU_ALLOC
5224 
5225   // pid 0 means the current thread - which we have to assume represents the process
5226   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5227     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5228       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5229     }
5230     else {
5231       cpu_count = CPU_COUNT(cpus_p);
5232     }
5233     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5234   }
5235   else {
5236     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5237     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5238             "which may exceed available processors", os::strerror(errno), cpu_count);
5239   }
5240 
5241   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5242     CPU_FREE(cpus_p);
5243   }
5244 
5245   assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5246   return cpu_count;
5247 }
5248 
5249 // Determine the active processor count from one of
5250 // three different sources:
5251 //
5252 // 1. User option -XX:ActiveProcessorCount
5253 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5254 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5255 //
5256 // Option 1, if specified, will always override.
5257 // If the cgroup subsystem is active and configured, we
5258 // will return the min of the cgroup and option 2 results.
5259 // This is required since tools, such as numactl, that
5260 // alter cpu affinity do not update cgroup subsystem
5261 // cpuset configuration files.
5262 int os::active_processor_count() {
5263   // User has overridden the number of active processors
5264   if (ActiveProcessorCount > 0) {
5265     log_trace(os)("active_processor_count: "
5266                   "active processor count set by user : %d",
5267                   ActiveProcessorCount);
5268     return ActiveProcessorCount;
5269   }
5270 
5271   int active_cpus;
5272   if (OSContainer::is_containerized()) {
5273     active_cpus = OSContainer::active_processor_count();
5274     log_trace(os)("active_processor_count: determined by OSContainer: %d",
5275                    active_cpus);
5276   } else {
5277     active_cpus = os::Linux::active_processor_count();
5278   }
5279 
5280   return active_cpus;
5281 }
5282 
5283 uint os::processor_id() {
5284   const int id = Linux::sched_getcpu();
5285   assert(id >= 0 && id < _processor_count, "Invalid processor id");
5286   return (uint)id;
5287 }
5288 
5289 void os::set_native_thread_name(const char *name) {
5290   if (Linux::_pthread_setname_np) {
5291     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5292     snprintf(buf, sizeof(buf), "%s", name);
5293     buf[sizeof(buf) - 1] = '\0';
5294     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5295     // ERANGE should not happen; all other errors should just be ignored.
5296     assert(rc != ERANGE, "pthread_setname_np failed");
5297   }
5298 }
5299 
5300 bool os::distribute_processes(uint length, uint* distribution) {
5301   // Not yet implemented.
5302   return false;
5303 }
5304 
5305 bool os::bind_to_processor(uint processor_id) {
5306   // Not yet implemented.
5307   return false;
5308 }
5309 
5310 ///
5311 
5312 void os::SuspendedThreadTask::internal_do_task() {
5313   if (do_suspend(_thread->osthread())) {
5314     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5315     do_task(context);
5316     do_resume(_thread->osthread());
5317   }
5318 }
5319 
5320 ////////////////////////////////////////////////////////////////////////////////
5321 // debug support
5322 
5323 bool os::find(address addr, outputStream* st) {
5324   Dl_info dlinfo;
5325   memset(&dlinfo, 0, sizeof(dlinfo));
5326   if (dladdr(addr, &dlinfo) != 0) {
5327     st->print(PTR_FORMAT ": ", p2i(addr));
5328     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5329       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5330                 p2i(addr) - p2i(dlinfo.dli_saddr));
5331     } else if (dlinfo.dli_fbase != NULL) {
5332       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5333     } else {
5334       st->print("<absolute address>");
5335     }
5336     if (dlinfo.dli_fname != NULL) {
5337       st->print(" in %s", dlinfo.dli_fname);
5338     }
5339     if (dlinfo.dli_fbase != NULL) {
5340       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5341     }
5342     st->cr();
5343 
5344     if (Verbose) {
5345       // decode some bytes around the PC
5346       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5347       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5348       address       lowest = (address) dlinfo.dli_sname;
5349       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5350       if (begin < lowest)  begin = lowest;
5351       Dl_info dlinfo2;
5352       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5353           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5354         end = (address) dlinfo2.dli_saddr;
5355       }
5356       Disassembler::decode(begin, end, st);
5357     }
5358     return true;
5359   }
5360   return false;
5361 }
5362 
5363 ////////////////////////////////////////////////////////////////////////////////
5364 // misc
5365 
5366 // This does not do anything on Linux. This is basically a hook for being
5367 // able to use structured exception handling (thread-local exception filters)
5368 // on, e.g., Win32.
5369 void
5370 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5371                          JavaCallArguments* args, Thread* thread) {
5372   f(value, method, args, thread);
5373 }
5374 
5375 void os::print_statistics() {
5376 }
5377 
5378 bool os::message_box(const char* title, const char* message) {
5379   int i;
5380   fdStream err(defaultStream::error_fd());
5381   for (i = 0; i < 78; i++) err.print_raw("=");
5382   err.cr();
5383   err.print_raw_cr(title);
5384   for (i = 0; i < 78; i++) err.print_raw("-");
5385   err.cr();
5386   err.print_raw_cr(message);
5387   for (i = 0; i < 78; i++) err.print_raw("=");
5388   err.cr();
5389 
5390   char buf[16];
5391   // Prevent process from exiting upon "read error" without consuming all CPU
5392   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5393 
5394   return buf[0] == 'y' || buf[0] == 'Y';
5395 }
5396 
5397 // Is a (classpath) directory empty?
5398 bool os::dir_is_empty(const char* path) {
5399   DIR *dir = NULL;
5400   struct dirent *ptr;
5401 
5402   dir = opendir(path);
5403   if (dir == NULL) return true;
5404 
5405   // Scan the directory
5406   bool result = true;
5407   char buf[sizeof(struct dirent) + MAX_PATH];
5408   while (result && (ptr = ::readdir(dir)) != NULL) {
5409     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5410       result = false;
5411     }
5412   }
5413   closedir(dir);
5414   return result;
5415 }
5416 
5417 // This code originates from JDK's sysOpen and open64_w
5418 // from src/solaris/hpi/src/system_md.c
5419 
5420 int os::open(const char *path, int oflag, int mode) {
5421   if (strlen(path) > MAX_PATH - 1) {
5422     errno = ENAMETOOLONG;
5423     return -1;
5424   }
5425 
5426   // All file descriptors that are opened in the Java process and not
5427   // specifically destined for a subprocess should have the close-on-exec
5428   // flag set.  If we don't set it, then careless 3rd party native code
5429   // might fork and exec without closing all appropriate file descriptors
5430   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5431   // turn might:
5432   //
5433   // - cause end-of-file to fail to be detected on some file
5434   //   descriptors, resulting in mysterious hangs, or
5435   //
5436   // - might cause an fopen in the subprocess to fail on a system
5437   //   suffering from bug 1085341.
5438   //
5439   // (Yes, the default setting of the close-on-exec flag is a Unix
5440   // design flaw)
5441   //
5442   // See:
5443   // 1085341: 32-bit stdio routines should support file descriptors >255
5444   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5445   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5446   //
5447   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5448   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5449   // because it saves a system call and removes a small window where the flag
5450   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5451   // and we fall back to using FD_CLOEXEC (see below).
5452 #ifdef O_CLOEXEC
5453   oflag |= O_CLOEXEC;
5454 #endif
5455 
5456   int fd = ::open64(path, oflag, mode);
5457   if (fd == -1) return -1;
5458 
5459   //If the open succeeded, the file might still be a directory
5460   {
5461     struct stat64 buf64;
5462     int ret = ::fstat64(fd, &buf64);
5463     int st_mode = buf64.st_mode;
5464 
5465     if (ret != -1) {
5466       if ((st_mode & S_IFMT) == S_IFDIR) {
5467         errno = EISDIR;
5468         ::close(fd);
5469         return -1;
5470       }
5471     } else {
5472       ::close(fd);
5473       return -1;
5474     }
5475   }
5476 
5477 #ifdef FD_CLOEXEC
5478   // Validate that the use of the O_CLOEXEC flag on open above worked.
5479   // With recent kernels, we will perform this check exactly once.
5480   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5481   if (!O_CLOEXEC_is_known_to_work) {
5482     int flags = ::fcntl(fd, F_GETFD);
5483     if (flags != -1) {
5484       if ((flags & FD_CLOEXEC) != 0)
5485         O_CLOEXEC_is_known_to_work = 1;
5486       else
5487         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5488     }
5489   }
5490 #endif
5491 
5492   return fd;
5493 }
5494 
5495 
5496 // create binary file, rewriting existing file if required
5497 int os::create_binary_file(const char* path, bool rewrite_existing) {
5498   int oflags = O_WRONLY | O_CREAT;
5499   if (!rewrite_existing) {
5500     oflags |= O_EXCL;
5501   }
5502   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5503 }
5504 
5505 // return current position of file pointer
5506 jlong os::current_file_offset(int fd) {
5507   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5508 }
5509 
5510 // move file pointer to the specified offset
5511 jlong os::seek_to_file_offset(int fd, jlong offset) {
5512   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5513 }
5514 
5515 // This code originates from JDK's sysAvailable
5516 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5517 
5518 int os::available(int fd, jlong *bytes) {
5519   jlong cur, end;
5520   int mode;
5521   struct stat64 buf64;
5522 
5523   if (::fstat64(fd, &buf64) >= 0) {
5524     mode = buf64.st_mode;
5525     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5526       int n;
5527       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5528         *bytes = n;
5529         return 1;
5530       }
5531     }
5532   }
5533   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5534     return 0;
5535   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5536     return 0;
5537   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5538     return 0;
5539   }
5540   *bytes = end - cur;
5541   return 1;
5542 }
5543 
5544 // Map a block of memory.
5545 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5546                         char *addr, size_t bytes, bool read_only,
5547                         bool allow_exec) {
5548   int prot;
5549   int flags = MAP_PRIVATE;
5550 
5551   if (read_only) {
5552     prot = PROT_READ;
5553   } else {
5554     prot = PROT_READ | PROT_WRITE;
5555   }
5556 
5557   if (allow_exec) {
5558     prot |= PROT_EXEC;
5559   }
5560 
5561   if (addr != NULL) {
5562     flags |= MAP_FIXED;
5563   }
5564 
5565   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5566                                      fd, file_offset);
5567   if (mapped_address == MAP_FAILED) {
5568     return NULL;
5569   }
5570   return mapped_address;
5571 }
5572 
5573 
5574 // Remap a block of memory.
5575 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5576                           char *addr, size_t bytes, bool read_only,
5577                           bool allow_exec) {
5578   // same as map_memory() on this OS
5579   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5580                         allow_exec);
5581 }
5582 
5583 
5584 // Unmap a block of memory.
5585 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5586   return munmap(addr, bytes) == 0;
5587 }
5588 
5589 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5590 
5591 static jlong fast_cpu_time(Thread *thread) {
5592     clockid_t clockid;
5593     int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5594                                               &clockid);
5595     if (rc == 0) {
5596       return os::Linux::fast_thread_cpu_time(clockid);
5597     } else {
5598       // It's possible to encounter a terminated native thread that failed
5599       // to detach itself from the VM - which should result in ESRCH.
5600       assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5601       return -1;
5602     }
5603 }
5604 
5605 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5606 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5607 // of a thread.
5608 //
5609 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5610 // the fast estimate available on the platform.
5611 
5612 jlong os::current_thread_cpu_time() {
5613   if (os::Linux::supports_fast_thread_cpu_time()) {
5614     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5615   } else {
5616     // return user + sys since the cost is the same
5617     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5618   }
5619 }
5620 
5621 jlong os::thread_cpu_time(Thread* thread) {
5622   // consistent with what current_thread_cpu_time() returns
5623   if (os::Linux::supports_fast_thread_cpu_time()) {
5624     return fast_cpu_time(thread);
5625   } else {
5626     return slow_thread_cpu_time(thread, true /* user + sys */);
5627   }
5628 }
5629 
5630 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5631   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5632     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5633   } else {
5634     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5635   }
5636 }
5637 
5638 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5639   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5640     return fast_cpu_time(thread);
5641   } else {
5642     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5643   }
5644 }
5645 
5646 //  -1 on error.
5647 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5648   pid_t  tid = thread->osthread()->thread_id();
5649   char *s;
5650   char stat[2048];
5651   int statlen;
5652   char proc_name[64];
5653   int count;
5654   long sys_time, user_time;
5655   char cdummy;
5656   int idummy;
5657   long ldummy;
5658   FILE *fp;
5659 
5660   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5661   fp = fopen(proc_name, "r");
5662   if (fp == NULL) return -1;
5663   statlen = fread(stat, 1, 2047, fp);
5664   stat[statlen] = '\0';
5665   fclose(fp);
5666 
5667   // Skip pid and the command string. Note that we could be dealing with
5668   // weird command names, e.g. user could decide to rename java launcher
5669   // to "java 1.4.2 :)", then the stat file would look like
5670   //                1234 (java 1.4.2 :)) R ... ...
5671   // We don't really need to know the command string, just find the last
5672   // occurrence of ")" and then start parsing from there. See bug 4726580.
5673   s = strrchr(stat, ')');
5674   if (s == NULL) return -1;
5675 
5676   // Skip blank chars
5677   do { s++; } while (s && isspace(*s));
5678 
5679   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5680                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5681                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5682                  &user_time, &sys_time);
5683   if (count != 13) return -1;
5684   if (user_sys_cpu_time) {
5685     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5686   } else {
5687     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5688   }
5689 }
5690 
5691 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5692   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5693   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5694   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5695   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5696 }
5697 
5698 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5699   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5700   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5701   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5702   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5703 }
5704 
5705 bool os::is_thread_cpu_time_supported() {
5706   return true;
5707 }
5708 
5709 // System loadavg support.  Returns -1 if load average cannot be obtained.
5710 // Linux doesn't yet have a (official) notion of processor sets,
5711 // so just return the system wide load average.
5712 int os::loadavg(double loadavg[], int nelem) {
5713   return ::getloadavg(loadavg, nelem);
5714 }
5715 
5716 void os::pause() {
5717   char filename[MAX_PATH];
5718   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5719     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5720   } else {
5721     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5722   }
5723 
5724   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5725   if (fd != -1) {
5726     struct stat buf;
5727     ::close(fd);
5728     while (::stat(filename, &buf) == 0) {
5729       (void)::poll(NULL, 0, 100);
5730     }
5731   } else {
5732     jio_fprintf(stderr,
5733                 "Could not open pause file '%s', continuing immediately.\n", filename);
5734   }
5735 }
5736 
5737 extern char** environ;
5738 
5739 // Run the specified command in a separate process. Return its exit value,
5740 // or -1 on failure (e.g. can't fork a new process).
5741 // Unlike system(), this function can be called from signal handler. It
5742 // doesn't block SIGINT et al.
5743 int os::fork_and_exec(char* cmd) {
5744   const char * argv[4] = {"sh", "-c", cmd, NULL};
5745 
5746   pid_t pid = fork();
5747 
5748   if (pid < 0) {
5749     // fork failed
5750     return -1;
5751 
5752   } else if (pid == 0) {
5753     // child process
5754 
5755     execve("/bin/sh", (char* const*)argv, environ);
5756 
5757     // execve failed
5758     _exit(-1);
5759 
5760   } else  {
5761     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5762     // care about the actual exit code, for now.
5763 
5764     int status;
5765 
5766     // Wait for the child process to exit.  This returns immediately if
5767     // the child has already exited. */
5768     while (waitpid(pid, &status, 0) < 0) {
5769       switch (errno) {
5770       case ECHILD: return 0;
5771       case EINTR: break;
5772       default: return -1;
5773       }
5774     }
5775 
5776     if (WIFEXITED(status)) {
5777       // The child exited normally; get its exit code.
5778       return WEXITSTATUS(status);
5779     } else if (WIFSIGNALED(status)) {
5780       // The child exited because of a signal
5781       // The best value to return is 0x80 + signal number,
5782       // because that is what all Unix shells do, and because
5783       // it allows callers to distinguish between process exit and
5784       // process death by signal.
5785       return 0x80 + WTERMSIG(status);
5786     } else {
5787       // Unknown exit code; pass it through
5788       return status;
5789     }
5790   }
5791 }
5792 
5793 // Get the default path to the core file
5794 // Returns the length of the string
5795 int os::get_core_path(char* buffer, size_t bufferSize) {
5796   /*
5797    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5798    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5799    */
5800   const int core_pattern_len = 129;
5801   char core_pattern[core_pattern_len] = {0};
5802 
5803   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5804   if (core_pattern_file == -1) {
5805     return -1;
5806   }
5807 
5808   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5809   ::close(core_pattern_file);
5810   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5811     return -1;
5812   }
5813   if (core_pattern[ret-1] == '\n') {
5814     core_pattern[ret-1] = '\0';
5815   } else {
5816     core_pattern[ret] = '\0';
5817   }
5818 
5819   char *pid_pos = strstr(core_pattern, "%p");
5820   int written;
5821 
5822   if (core_pattern[0] == '/') {
5823     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5824   } else {
5825     char cwd[PATH_MAX];
5826 
5827     const char* p = get_current_directory(cwd, PATH_MAX);
5828     if (p == NULL) {
5829       return -1;
5830     }
5831 
5832     if (core_pattern[0] == '|') {
5833       written = jio_snprintf(buffer, bufferSize,
5834                              "\"%s\" (or dumping to %s/core.%d)",
5835                              &core_pattern[1], p, current_process_id());
5836     } else {
5837       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5838     }
5839   }
5840 
5841   if (written < 0) {
5842     return -1;
5843   }
5844 
5845   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5846     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5847 
5848     if (core_uses_pid_file != -1) {
5849       char core_uses_pid = 0;
5850       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5851       ::close(core_uses_pid_file);
5852 
5853       if (core_uses_pid == '1') {
5854         jio_snprintf(buffer + written, bufferSize - written,
5855                                           ".%d", current_process_id());
5856       }
5857     }
5858   }
5859 
5860   return strlen(buffer);
5861 }
5862 
5863 bool os::start_debugging(char *buf, int buflen) {
5864   int len = (int)strlen(buf);
5865   char *p = &buf[len];
5866 
5867   jio_snprintf(p, buflen-len,
5868                "\n\n"
5869                "Do you want to debug the problem?\n\n"
5870                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5871                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5872                "Otherwise, press RETURN to abort...",
5873                os::current_process_id(), os::current_process_id(),
5874                os::current_thread_id(), os::current_thread_id());
5875 
5876   bool yes = os::message_box("Unexpected Error", buf);
5877 
5878   if (yes) {
5879     // yes, user asked VM to launch debugger
5880     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5881                  os::current_process_id(), os::current_process_id());
5882 
5883     os::fork_and_exec(buf);
5884     yes = false;
5885   }
5886   return yes;
5887 }
5888 
5889 
5890 // Java/Compiler thread:
5891 //
5892 //   Low memory addresses
5893 // P0 +------------------------+
5894 //    |                        |\  Java thread created by VM does not have glibc
5895 //    |    glibc guard page    | - guard page, attached Java thread usually has
5896 //    |                        |/  1 glibc guard page.
5897 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5898 //    |                        |\
5899 //    |  HotSpot Guard Pages   | - red, yellow and reserved pages
5900 //    |                        |/
5901 //    +------------------------+ JavaThread::stack_reserved_zone_base()
5902 //    |                        |\
5903 //    |      Normal Stack      | -
5904 //    |                        |/
5905 // P2 +------------------------+ Thread::stack_base()
5906 //
5907 // Non-Java thread:
5908 //
5909 //   Low memory addresses
5910 // P0 +------------------------+
5911 //    |                        |\
5912 //    |  glibc guard page      | - usually 1 page
5913 //    |                        |/
5914 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5915 //    |                        |\
5916 //    |      Normal Stack      | -
5917 //    |                        |/
5918 // P2 +------------------------+ Thread::stack_base()
5919 //
5920 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5921 //    returned from pthread_attr_getstack().
5922 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5923 //    of the stack size given in pthread_attr. We work around this for
5924 //    threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5925 //
5926 #ifndef ZERO
5927 static void current_stack_region(address * bottom, size_t * size) {
5928   if (os::is_primordial_thread()) {
5929     // primordial thread needs special handling because pthread_getattr_np()
5930     // may return bogus value.
5931     *bottom = os::Linux::initial_thread_stack_bottom();
5932     *size   = os::Linux::initial_thread_stack_size();
5933   } else {
5934     pthread_attr_t attr;
5935 
5936     int rslt = pthread_getattr_np(pthread_self(), &attr);
5937 
5938     // JVM needs to know exact stack location, abort if it fails
5939     if (rslt != 0) {
5940       if (rslt == ENOMEM) {
5941         vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5942       } else {
5943         fatal("pthread_getattr_np failed with error = %d", rslt);
5944       }
5945     }
5946 
5947     if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5948       fatal("Cannot locate current stack attributes!");
5949     }
5950 
5951     // Work around NPTL stack guard error.
5952     size_t guard_size = 0;
5953     rslt = pthread_attr_getguardsize(&attr, &guard_size);
5954     if (rslt != 0) {
5955       fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5956     }
5957     *bottom += guard_size;
5958     *size   -= guard_size;
5959 
5960     pthread_attr_destroy(&attr);
5961 
5962   }
5963   assert(os::current_stack_pointer() >= *bottom &&
5964          os::current_stack_pointer() < *bottom + *size, "just checking");
5965 }
5966 
5967 address os::current_stack_base() {
5968   address bottom;
5969   size_t size;
5970   current_stack_region(&bottom, &size);
5971   return (bottom + size);
5972 }
5973 
5974 size_t os::current_stack_size() {
5975   // This stack size includes the usable stack and HotSpot guard pages
5976   // (for the threads that have Hotspot guard pages).
5977   address bottom;
5978   size_t size;
5979   current_stack_region(&bottom, &size);
5980   return size;
5981 }
5982 #endif
5983 
5984 static inline struct timespec get_mtime(const char* filename) {
5985   struct stat st;
5986   int ret = os::stat(filename, &st);
5987   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5988   return st.st_mtim;
5989 }
5990 
5991 int os::compare_file_modified_times(const char* file1, const char* file2) {
5992   struct timespec filetime1 = get_mtime(file1);
5993   struct timespec filetime2 = get_mtime(file2);
5994   int diff = filetime1.tv_sec - filetime2.tv_sec;
5995   if (diff == 0) {
5996     return filetime1.tv_nsec - filetime2.tv_nsec;
5997   }
5998   return diff;
5999 }
6000 
6001 /////////////// Unit tests ///////////////
6002 
6003 #ifndef PRODUCT
6004 
6005 #define test_log(...)              \
6006   do {                             \
6007     if (VerboseInternalVMTests) {  \
6008       tty->print_cr(__VA_ARGS__);  \
6009       tty->flush();                \
6010     }                              \
6011   } while (false)
6012 
6013 class TestReserveMemorySpecial : AllStatic {
6014  public:
6015   static void small_page_write(void* addr, size_t size) {
6016     size_t page_size = os::vm_page_size();
6017 
6018     char* end = (char*)addr + size;
6019     for (char* p = (char*)addr; p < end; p += page_size) {
6020       *p = 1;
6021     }
6022   }
6023 
6024   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6025     if (!UseHugeTLBFS) {
6026       return;
6027     }
6028 
6029     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6030 
6031     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6032 
6033     if (addr != NULL) {
6034       small_page_write(addr, size);
6035 
6036       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6037     }
6038   }
6039 
6040   static void test_reserve_memory_special_huge_tlbfs_only() {
6041     if (!UseHugeTLBFS) {
6042       return;
6043     }
6044 
6045     size_t lp = os::large_page_size();
6046 
6047     for (size_t size = lp; size <= lp * 10; size += lp) {
6048       test_reserve_memory_special_huge_tlbfs_only(size);
6049     }
6050   }
6051 
6052   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6053     size_t lp = os::large_page_size();
6054     size_t ag = os::vm_allocation_granularity();
6055 
6056     // sizes to test
6057     const size_t sizes[] = {
6058       lp, lp + ag, lp + lp / 2, lp * 2,
6059       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6060       lp * 10, lp * 10 + lp / 2
6061     };
6062     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6063 
6064     // For each size/alignment combination, we test three scenarios:
6065     // 1) with req_addr == NULL
6066     // 2) with a non-null req_addr at which we expect to successfully allocate
6067     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6068     //    expect the allocation to either fail or to ignore req_addr
6069 
6070     // Pre-allocate two areas; they shall be as large as the largest allocation
6071     //  and aligned to the largest alignment we will be testing.
6072     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6073     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6074       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6075       -1, 0);
6076     assert(mapping1 != MAP_FAILED, "should work");
6077 
6078     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6079       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6080       -1, 0);
6081     assert(mapping2 != MAP_FAILED, "should work");
6082 
6083     // Unmap the first mapping, but leave the second mapping intact: the first
6084     // mapping will serve as a value for a "good" req_addr (case 2). The second
6085     // mapping, still intact, as "bad" req_addr (case 3).
6086     ::munmap(mapping1, mapping_size);
6087 
6088     // Case 1
6089     test_log("%s, req_addr NULL:", __FUNCTION__);
6090     test_log("size            align           result");
6091 
6092     for (int i = 0; i < num_sizes; i++) {
6093       const size_t size = sizes[i];
6094       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6095         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6096         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6097                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6098         if (p != NULL) {
6099           assert(is_aligned(p, alignment), "must be");
6100           small_page_write(p, size);
6101           os::Linux::release_memory_special_huge_tlbfs(p, size);
6102         }
6103       }
6104     }
6105 
6106     // Case 2
6107     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6108     test_log("size            align           req_addr         result");
6109 
6110     for (int i = 0; i < num_sizes; i++) {
6111       const size_t size = sizes[i];
6112       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6113         char* const req_addr = align_up(mapping1, alignment);
6114         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6115         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6116                  size, alignment, p2i(req_addr), p2i(p),
6117                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6118         if (p != NULL) {
6119           assert(p == req_addr, "must be");
6120           small_page_write(p, size);
6121           os::Linux::release_memory_special_huge_tlbfs(p, size);
6122         }
6123       }
6124     }
6125 
6126     // Case 3
6127     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6128     test_log("size            align           req_addr         result");
6129 
6130     for (int i = 0; i < num_sizes; i++) {
6131       const size_t size = sizes[i];
6132       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6133         char* const req_addr = align_up(mapping2, alignment);
6134         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6135         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6136                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6137         // as the area around req_addr contains already existing mappings, the API should always
6138         // return NULL (as per contract, it cannot return another address)
6139         assert(p == NULL, "must be");
6140       }
6141     }
6142 
6143     ::munmap(mapping2, mapping_size);
6144 
6145   }
6146 
6147   static void test_reserve_memory_special_huge_tlbfs() {
6148     if (!UseHugeTLBFS) {
6149       return;
6150     }
6151 
6152     test_reserve_memory_special_huge_tlbfs_only();
6153     test_reserve_memory_special_huge_tlbfs_mixed();
6154   }
6155 
6156   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6157     if (!UseSHM) {
6158       return;
6159     }
6160 
6161     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6162 
6163     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6164 
6165     if (addr != NULL) {
6166       assert(is_aligned(addr, alignment), "Check");
6167       assert(is_aligned(addr, os::large_page_size()), "Check");
6168 
6169       small_page_write(addr, size);
6170 
6171       os::Linux::release_memory_special_shm(addr, size);
6172     }
6173   }
6174 
6175   static void test_reserve_memory_special_shm() {
6176     size_t lp = os::large_page_size();
6177     size_t ag = os::vm_allocation_granularity();
6178 
6179     for (size_t size = ag; size < lp * 3; size += ag) {
6180       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6181         test_reserve_memory_special_shm(size, alignment);
6182       }
6183     }
6184   }
6185 
6186   static void test() {
6187     test_reserve_memory_special_huge_tlbfs();
6188     test_reserve_memory_special_shm();
6189   }
6190 };
6191 
6192 void TestReserveMemorySpecial_test() {
6193   TestReserveMemorySpecial::test();
6194 }
6195 
6196 #endif