1 /*
   2  * Copyright (c) 1999, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "jvm.h"
  27 #include "classfile/classLoader.hpp"
  28 #include "classfile/systemDictionary.hpp"
  29 #include "classfile/vmSymbols.hpp"
  30 #include "code/icBuffer.hpp"
  31 #include "code/vtableStubs.hpp"
  32 #include "compiler/compileBroker.hpp"
  33 #include "compiler/disassembler.hpp"
  34 #include "interpreter/interpreter.hpp"
  35 #include "logging/log.hpp"
  36 #include "memory/allocation.inline.hpp"
  37 #include "memory/filemap.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_linux.inline.hpp"
  40 #include "os_share_linux.hpp"
  41 #include "osContainer_linux.hpp"
  42 #include "prims/jniFastGetField.hpp"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.inline.hpp"
  49 #include "runtime/init.hpp"
  50 #include "runtime/java.hpp"
  51 #include "runtime/javaCalls.hpp"
  52 #include "runtime/mutexLocker.hpp"
  53 #include "runtime/objectMonitor.hpp"
  54 #include "runtime/orderAccess.inline.hpp"
  55 #include "runtime/osThread.hpp"
  56 #include "runtime/perfMemory.hpp"
  57 #include "runtime/sharedRuntime.hpp"
  58 #include "runtime/statSampler.hpp"
  59 #include "runtime/stubRoutines.hpp"
  60 #include "runtime/thread.inline.hpp"
  61 #include "runtime/threadCritical.hpp"
  62 #include "runtime/threadSMR.hpp"
  63 #include "runtime/timer.hpp"
  64 #include "semaphore_posix.hpp"
  65 #include "services/attachListener.hpp"
  66 #include "services/memTracker.hpp"
  67 #include "services/runtimeService.hpp"
  68 #include "utilities/align.hpp"
  69 #include "utilities/decoder.hpp"
  70 #include "utilities/defaultStream.hpp"
  71 #include "utilities/events.hpp"
  72 #include "utilities/elfFile.hpp"
  73 #include "utilities/growableArray.hpp"
  74 #include "utilities/macros.hpp"
  75 #include "utilities/vmError.hpp"
  76 
  77 // put OS-includes here
  78 # include <sys/types.h>
  79 # include <sys/mman.h>
  80 # include <sys/stat.h>
  81 # include <sys/select.h>
  82 # include <pthread.h>
  83 # include <signal.h>
  84 # include <errno.h>
  85 # include <dlfcn.h>
  86 # include <stdio.h>
  87 # include <unistd.h>
  88 # include <sys/resource.h>
  89 # include <pthread.h>
  90 # include <sys/stat.h>
  91 # include <sys/time.h>
  92 # include <sys/times.h>
  93 # include <sys/utsname.h>
  94 # include <sys/socket.h>
  95 # include <sys/wait.h>
  96 # include <pwd.h>
  97 # include <poll.h>
  98 # include <fcntl.h>
  99 # include <string.h>
 100 # include <syscall.h>
 101 # include <sys/sysinfo.h>
 102 # include <gnu/libc-version.h>
 103 # include <sys/ipc.h>
 104 # include <sys/shm.h>
 105 # include <link.h>
 106 # include <stdint.h>
 107 # include <inttypes.h>
 108 # include <sys/ioctl.h>
 109 
 110 #ifndef _GNU_SOURCE
 111   #define _GNU_SOURCE
 112   #include <sched.h>
 113   #undef _GNU_SOURCE
 114 #else
 115   #include <sched.h>
 116 #endif
 117 
 118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 119 // getrusage() is prepared to handle the associated failure.
 120 #ifndef RUSAGE_THREAD
 121   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 122 #endif
 123 
 124 #define MAX_PATH    (2 * K)
 125 
 126 #define MAX_SECS 100000000
 127 
 128 // for timer info max values which include all bits
 129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 130 
 131 #define LARGEPAGES_BIT (1 << 6)
 132 #define DAX_SHARED_BIT (1 << 8)
 133 ////////////////////////////////////////////////////////////////////////////////
 134 // global variables
 135 julong os::Linux::_physical_memory = 0;
 136 
 137 address   os::Linux::_initial_thread_stack_bottom = NULL;
 138 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 139 
 140 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 141 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 142 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 143 Mutex* os::Linux::_createThread_lock = NULL;
 144 pthread_t os::Linux::_main_thread;
 145 int os::Linux::_page_size = -1;
 146 bool os::Linux::_supports_fast_thread_cpu_time = false;
 147 uint32_t os::Linux::_os_version = 0;
 148 const char * os::Linux::_glibc_version = NULL;
 149 const char * os::Linux::_libpthread_version = NULL;
 150 
 151 static jlong initial_time_count=0;
 152 
 153 static int clock_tics_per_sec = 100;
 154 
 155 // For diagnostics to print a message once. see run_periodic_checks
 156 static sigset_t check_signal_done;
 157 static bool check_signals = true;
 158 
 159 // Signal number used to suspend/resume a thread
 160 
 161 // do not use any signal number less than SIGSEGV, see 4355769
 162 static int SR_signum = SIGUSR2;
 163 sigset_t SR_sigset;
 164 
 165 // utility functions
 166 
 167 static int SR_initialize();
 168 
 169 julong os::available_memory() {
 170   return Linux::available_memory();
 171 }
 172 
 173 julong os::Linux::available_memory() {
 174   // values in struct sysinfo are "unsigned long"
 175   struct sysinfo si;
 176   julong avail_mem;
 177 
 178   if (OSContainer::is_containerized()) {
 179     jlong mem_limit, mem_usage;
 180     if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
 181       log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
 182                              mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
 183     }
 184     if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
 185       log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
 186     }
 187     if (mem_limit > 0 && mem_usage > 0 ) {
 188       avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
 189       log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
 190       return avail_mem;
 191     }
 192   }
 193 
 194   sysinfo(&si);
 195   avail_mem = (julong)si.freeram * si.mem_unit;
 196   log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
 197   return avail_mem;
 198 }
 199 
 200 julong os::physical_memory() {
 201   jlong phys_mem = 0;
 202   if (OSContainer::is_containerized()) {
 203     jlong mem_limit;
 204     if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
 205       log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
 206       return mem_limit;
 207     }
 208     log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
 209                             mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
 210   }
 211 
 212   phys_mem = Linux::physical_memory();
 213   log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
 214   return phys_mem;
 215 }
 216 
 217 // Return true if user is running as root.
 218 
 219 bool os::have_special_privileges() {
 220   static bool init = false;
 221   static bool privileges = false;
 222   if (!init) {
 223     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 224     init = true;
 225   }
 226   return privileges;
 227 }
 228 
 229 
 230 #ifndef SYS_gettid
 231 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 232   #ifdef __ia64__
 233     #define SYS_gettid 1105
 234   #else
 235     #ifdef __i386__
 236       #define SYS_gettid 224
 237     #else
 238       #ifdef __amd64__
 239         #define SYS_gettid 186
 240       #else
 241         #ifdef __sparc__
 242           #define SYS_gettid 143
 243         #else
 244           #error define gettid for the arch
 245         #endif
 246       #endif
 247     #endif
 248   #endif
 249 #endif
 250 
 251 
 252 // pid_t gettid()
 253 //
 254 // Returns the kernel thread id of the currently running thread. Kernel
 255 // thread id is used to access /proc.
 256 pid_t os::Linux::gettid() {
 257   int rslt = syscall(SYS_gettid);
 258   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 259   return (pid_t)rslt;
 260 }
 261 
 262 // Most versions of linux have a bug where the number of processors are
 263 // determined by looking at the /proc file system.  In a chroot environment,
 264 // the system call returns 1.  This causes the VM to act as if it is
 265 // a single processor and elide locking (see is_MP() call).
 266 static bool unsafe_chroot_detected = false;
 267 static const char *unstable_chroot_error = "/proc file system not found.\n"
 268                      "Java may be unstable running multithreaded in a chroot "
 269                      "environment on Linux when /proc filesystem is not mounted.";
 270 
 271 void os::Linux::initialize_system_info() {
 272   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 273   if (processor_count() == 1) {
 274     pid_t pid = os::Linux::gettid();
 275     char fname[32];
 276     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 277     FILE *fp = fopen(fname, "r");
 278     if (fp == NULL) {
 279       unsafe_chroot_detected = true;
 280     } else {
 281       fclose(fp);
 282     }
 283   }
 284   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 285   assert(processor_count() > 0, "linux error");
 286 }
 287 
 288 void os::init_system_properties_values() {
 289   // The next steps are taken in the product version:
 290   //
 291   // Obtain the JAVA_HOME value from the location of libjvm.so.
 292   // This library should be located at:
 293   // <JAVA_HOME>/lib/{client|server}/libjvm.so.
 294   //
 295   // If "/jre/lib/" appears at the right place in the path, then we
 296   // assume libjvm.so is installed in a JDK and we use this path.
 297   //
 298   // Otherwise exit with message: "Could not create the Java virtual machine."
 299   //
 300   // The following extra steps are taken in the debugging version:
 301   //
 302   // If "/jre/lib/" does NOT appear at the right place in the path
 303   // instead of exit check for $JAVA_HOME environment variable.
 304   //
 305   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 306   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 307   // it looks like libjvm.so is installed there
 308   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 309   //
 310   // Otherwise exit.
 311   //
 312   // Important note: if the location of libjvm.so changes this
 313   // code needs to be changed accordingly.
 314 
 315   // See ld(1):
 316   //      The linker uses the following search paths to locate required
 317   //      shared libraries:
 318   //        1: ...
 319   //        ...
 320   //        7: The default directories, normally /lib and /usr/lib.
 321 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
 322   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 323 #else
 324   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 325 #endif
 326 
 327 // Base path of extensions installed on the system.
 328 #define SYS_EXT_DIR     "/usr/java/packages"
 329 #define EXTENSIONS_DIR  "/lib/ext"
 330 
 331   // Buffer that fits several sprintfs.
 332   // Note that the space for the colon and the trailing null are provided
 333   // by the nulls included by the sizeof operator.
 334   const size_t bufsize =
 335     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 336          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 337   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 338 
 339   // sysclasspath, java_home, dll_dir
 340   {
 341     char *pslash;
 342     os::jvm_path(buf, bufsize);
 343 
 344     // Found the full path to libjvm.so.
 345     // Now cut the path to <java_home>/jre if we can.
 346     pslash = strrchr(buf, '/');
 347     if (pslash != NULL) {
 348       *pslash = '\0';            // Get rid of /libjvm.so.
 349     }
 350     pslash = strrchr(buf, '/');
 351     if (pslash != NULL) {
 352       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 353     }
 354     Arguments::set_dll_dir(buf);
 355 
 356     if (pslash != NULL) {
 357       pslash = strrchr(buf, '/');
 358       if (pslash != NULL) {
 359         *pslash = '\0';        // Get rid of /lib.
 360       }
 361     }
 362     Arguments::set_java_home(buf);
 363     set_boot_path('/', ':');
 364   }
 365 
 366   // Where to look for native libraries.
 367   //
 368   // Note: Due to a legacy implementation, most of the library path
 369   // is set in the launcher. This was to accomodate linking restrictions
 370   // on legacy Linux implementations (which are no longer supported).
 371   // Eventually, all the library path setting will be done here.
 372   //
 373   // However, to prevent the proliferation of improperly built native
 374   // libraries, the new path component /usr/java/packages is added here.
 375   // Eventually, all the library path setting will be done here.
 376   {
 377     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 378     // should always exist (until the legacy problem cited above is
 379     // addressed).
 380     const char *v = ::getenv("LD_LIBRARY_PATH");
 381     const char *v_colon = ":";
 382     if (v == NULL) { v = ""; v_colon = ""; }
 383     // That's +1 for the colon and +1 for the trailing '\0'.
 384     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 385                                                      strlen(v) + 1 +
 386                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
 387                                                      mtInternal);
 388     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
 389     Arguments::set_library_path(ld_library_path);
 390     FREE_C_HEAP_ARRAY(char, ld_library_path);
 391   }
 392 
 393   // Extensions directories.
 394   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 395   Arguments::set_ext_dirs(buf);
 396 
 397   FREE_C_HEAP_ARRAY(char, buf);
 398 
 399 #undef DEFAULT_LIBPATH
 400 #undef SYS_EXT_DIR
 401 #undef EXTENSIONS_DIR
 402 }
 403 
 404 ////////////////////////////////////////////////////////////////////////////////
 405 // breakpoint support
 406 
 407 void os::breakpoint() {
 408   BREAKPOINT;
 409 }
 410 
 411 extern "C" void breakpoint() {
 412   // use debugger to set breakpoint here
 413 }
 414 
 415 ////////////////////////////////////////////////////////////////////////////////
 416 // signal support
 417 
 418 debug_only(static bool signal_sets_initialized = false);
 419 static sigset_t unblocked_sigs, vm_sigs;
 420 
 421 bool os::Linux::is_sig_ignored(int sig) {
 422   struct sigaction oact;
 423   sigaction(sig, (struct sigaction*)NULL, &oact);
 424   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 425                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 426   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 427     return true;
 428   } else {
 429     return false;
 430   }
 431 }
 432 
 433 void os::Linux::signal_sets_init() {
 434   // Should also have an assertion stating we are still single-threaded.
 435   assert(!signal_sets_initialized, "Already initialized");
 436   // Fill in signals that are necessarily unblocked for all threads in
 437   // the VM. Currently, we unblock the following signals:
 438   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 439   //                         by -Xrs (=ReduceSignalUsage));
 440   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 441   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 442   // the dispositions or masks wrt these signals.
 443   // Programs embedding the VM that want to use the above signals for their
 444   // own purposes must, at this time, use the "-Xrs" option to prevent
 445   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 446   // (See bug 4345157, and other related bugs).
 447   // In reality, though, unblocking these signals is really a nop, since
 448   // these signals are not blocked by default.
 449   sigemptyset(&unblocked_sigs);
 450   sigaddset(&unblocked_sigs, SIGILL);
 451   sigaddset(&unblocked_sigs, SIGSEGV);
 452   sigaddset(&unblocked_sigs, SIGBUS);
 453   sigaddset(&unblocked_sigs, SIGFPE);
 454 #if defined(PPC64)
 455   sigaddset(&unblocked_sigs, SIGTRAP);
 456 #endif
 457   sigaddset(&unblocked_sigs, SR_signum);
 458 
 459   if (!ReduceSignalUsage) {
 460     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 461       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 462     }
 463     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 464       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 465     }
 466     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 467       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 468     }
 469   }
 470   // Fill in signals that are blocked by all but the VM thread.
 471   sigemptyset(&vm_sigs);
 472   if (!ReduceSignalUsage) {
 473     sigaddset(&vm_sigs, BREAK_SIGNAL);
 474   }
 475   debug_only(signal_sets_initialized = true);
 476 
 477 }
 478 
 479 // These are signals that are unblocked while a thread is running Java.
 480 // (For some reason, they get blocked by default.)
 481 sigset_t* os::Linux::unblocked_signals() {
 482   assert(signal_sets_initialized, "Not initialized");
 483   return &unblocked_sigs;
 484 }
 485 
 486 // These are the signals that are blocked while a (non-VM) thread is
 487 // running Java. Only the VM thread handles these signals.
 488 sigset_t* os::Linux::vm_signals() {
 489   assert(signal_sets_initialized, "Not initialized");
 490   return &vm_sigs;
 491 }
 492 
 493 void os::Linux::hotspot_sigmask(Thread* thread) {
 494 
 495   //Save caller's signal mask before setting VM signal mask
 496   sigset_t caller_sigmask;
 497   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 498 
 499   OSThread* osthread = thread->osthread();
 500   osthread->set_caller_sigmask(caller_sigmask);
 501 
 502   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 503 
 504   if (!ReduceSignalUsage) {
 505     if (thread->is_VM_thread()) {
 506       // Only the VM thread handles BREAK_SIGNAL ...
 507       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 508     } else {
 509       // ... all other threads block BREAK_SIGNAL
 510       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 511     }
 512   }
 513 }
 514 
 515 //////////////////////////////////////////////////////////////////////////////
 516 // detecting pthread library
 517 
 518 void os::Linux::libpthread_init() {
 519   // Save glibc and pthread version strings.
 520 #if !defined(_CS_GNU_LIBC_VERSION) || \
 521     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 522   #error "glibc too old (< 2.3.2)"
 523 #endif
 524 
 525   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 526   assert(n > 0, "cannot retrieve glibc version");
 527   char *str = (char *)malloc(n, mtInternal);
 528   confstr(_CS_GNU_LIBC_VERSION, str, n);
 529   os::Linux::set_glibc_version(str);
 530 
 531   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 532   assert(n > 0, "cannot retrieve pthread version");
 533   str = (char *)malloc(n, mtInternal);
 534   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 535   os::Linux::set_libpthread_version(str);
 536 }
 537 
 538 /////////////////////////////////////////////////////////////////////////////
 539 // thread stack expansion
 540 
 541 // os::Linux::manually_expand_stack() takes care of expanding the thread
 542 // stack. Note that this is normally not needed: pthread stacks allocate
 543 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 544 // committed. Therefore it is not necessary to expand the stack manually.
 545 //
 546 // Manually expanding the stack was historically needed on LinuxThreads
 547 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 548 // it is kept to deal with very rare corner cases:
 549 //
 550 // For one, user may run the VM on an own implementation of threads
 551 // whose stacks are - like the old LinuxThreads - implemented using
 552 // mmap(MAP_GROWSDOWN).
 553 //
 554 // Also, this coding may be needed if the VM is running on the primordial
 555 // thread. Normally we avoid running on the primordial thread; however,
 556 // user may still invoke the VM on the primordial thread.
 557 //
 558 // The following historical comment describes the details about running
 559 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 560 
 561 
 562 // Force Linux kernel to expand current thread stack. If "bottom" is close
 563 // to the stack guard, caller should block all signals.
 564 //
 565 // MAP_GROWSDOWN:
 566 //   A special mmap() flag that is used to implement thread stacks. It tells
 567 //   kernel that the memory region should extend downwards when needed. This
 568 //   allows early versions of LinuxThreads to only mmap the first few pages
 569 //   when creating a new thread. Linux kernel will automatically expand thread
 570 //   stack as needed (on page faults).
 571 //
 572 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 573 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 574 //   region, it's hard to tell if the fault is due to a legitimate stack
 575 //   access or because of reading/writing non-exist memory (e.g. buffer
 576 //   overrun). As a rule, if the fault happens below current stack pointer,
 577 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 578 //   application (see Linux kernel fault.c).
 579 //
 580 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 581 //   stack overflow detection.
 582 //
 583 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 584 //   not use MAP_GROWSDOWN.
 585 //
 586 // To get around the problem and allow stack banging on Linux, we need to
 587 // manually expand thread stack after receiving the SIGSEGV.
 588 //
 589 // There are two ways to expand thread stack to address "bottom", we used
 590 // both of them in JVM before 1.5:
 591 //   1. adjust stack pointer first so that it is below "bottom", and then
 592 //      touch "bottom"
 593 //   2. mmap() the page in question
 594 //
 595 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 596 // if current sp is already near the lower end of page 101, and we need to
 597 // call mmap() to map page 100, it is possible that part of the mmap() frame
 598 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 599 // That will destroy the mmap() frame and cause VM to crash.
 600 //
 601 // The following code works by adjusting sp first, then accessing the "bottom"
 602 // page to force a page fault. Linux kernel will then automatically expand the
 603 // stack mapping.
 604 //
 605 // _expand_stack_to() assumes its frame size is less than page size, which
 606 // should always be true if the function is not inlined.
 607 
 608 static void NOINLINE _expand_stack_to(address bottom) {
 609   address sp;
 610   size_t size;
 611   volatile char *p;
 612 
 613   // Adjust bottom to point to the largest address within the same page, it
 614   // gives us a one-page buffer if alloca() allocates slightly more memory.
 615   bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
 616   bottom += os::Linux::page_size() - 1;
 617 
 618   // sp might be slightly above current stack pointer; if that's the case, we
 619   // will alloca() a little more space than necessary, which is OK. Don't use
 620   // os::current_stack_pointer(), as its result can be slightly below current
 621   // stack pointer, causing us to not alloca enough to reach "bottom".
 622   sp = (address)&sp;
 623 
 624   if (sp > bottom) {
 625     size = sp - bottom;
 626     p = (volatile char *)alloca(size);
 627     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 628     p[0] = '\0';
 629   }
 630 }
 631 
 632 void os::Linux::expand_stack_to(address bottom) {
 633   _expand_stack_to(bottom);
 634 }
 635 
 636 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 637   assert(t!=NULL, "just checking");
 638   assert(t->osthread()->expanding_stack(), "expand should be set");
 639   assert(t->stack_base() != NULL, "stack_base was not initialized");
 640 
 641   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 642     sigset_t mask_all, old_sigset;
 643     sigfillset(&mask_all);
 644     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 645     _expand_stack_to(addr);
 646     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 647     return true;
 648   }
 649   return false;
 650 }
 651 
 652 //////////////////////////////////////////////////////////////////////////////
 653 // create new thread
 654 
 655 // Thread start routine for all newly created threads
 656 static void *thread_native_entry(Thread *thread) {
 657   // Try to randomize the cache line index of hot stack frames.
 658   // This helps when threads of the same stack traces evict each other's
 659   // cache lines. The threads can be either from the same JVM instance, or
 660   // from different JVM instances. The benefit is especially true for
 661   // processors with hyperthreading technology.
 662   static int counter = 0;
 663   int pid = os::current_process_id();
 664   alloca(((pid ^ counter++) & 7) * 128);
 665 
 666   thread->initialize_thread_current();
 667 
 668   OSThread* osthread = thread->osthread();
 669   Monitor* sync = osthread->startThread_lock();
 670 
 671   osthread->set_thread_id(os::current_thread_id());
 672 
 673   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 674     os::current_thread_id(), (uintx) pthread_self());
 675 
 676   if (UseNUMA) {
 677     int lgrp_id = os::numa_get_group_id();
 678     if (lgrp_id != -1) {
 679       thread->set_lgrp_id(lgrp_id);
 680     }
 681   }
 682   // initialize signal mask for this thread
 683   os::Linux::hotspot_sigmask(thread);
 684 
 685   // initialize floating point control register
 686   os::Linux::init_thread_fpu_state();
 687 
 688   // handshaking with parent thread
 689   {
 690     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 691 
 692     // notify parent thread
 693     osthread->set_state(INITIALIZED);
 694     sync->notify_all();
 695 
 696     // wait until os::start_thread()
 697     while (osthread->get_state() == INITIALIZED) {
 698       sync->wait(Mutex::_no_safepoint_check_flag);
 699     }
 700   }
 701 
 702   // call one more level start routine
 703   thread->run();
 704 
 705   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 706     os::current_thread_id(), (uintx) pthread_self());
 707 
 708   // If a thread has not deleted itself ("delete this") as part of its
 709   // termination sequence, we have to ensure thread-local-storage is
 710   // cleared before we actually terminate. No threads should ever be
 711   // deleted asynchronously with respect to their termination.
 712   if (Thread::current_or_null_safe() != NULL) {
 713     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 714     thread->clear_thread_current();
 715   }
 716 
 717   return 0;
 718 }
 719 
 720 bool os::create_thread(Thread* thread, ThreadType thr_type,
 721                        size_t req_stack_size) {
 722   assert(thread->osthread() == NULL, "caller responsible");
 723 
 724   // Allocate the OSThread object
 725   OSThread* osthread = new OSThread(NULL, NULL);
 726   if (osthread == NULL) {
 727     return false;
 728   }
 729 
 730   // set the correct thread state
 731   osthread->set_thread_type(thr_type);
 732 
 733   // Initial state is ALLOCATED but not INITIALIZED
 734   osthread->set_state(ALLOCATED);
 735 
 736   thread->set_osthread(osthread);
 737 
 738   // init thread attributes
 739   pthread_attr_t attr;
 740   pthread_attr_init(&attr);
 741   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 742 
 743   // Calculate stack size if it's not specified by caller.
 744   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 745   // In the Linux NPTL pthread implementation the guard size mechanism
 746   // is not implemented properly. The posix standard requires adding
 747   // the size of the guard pages to the stack size, instead Linux
 748   // takes the space out of 'stacksize'. Thus we adapt the requested
 749   // stack_size by the size of the guard pages to mimick proper
 750   // behaviour. However, be careful not to end up with a size
 751   // of zero due to overflow. Don't add the guard page in that case.
 752   size_t guard_size = os::Linux::default_guard_size(thr_type);
 753   if (stack_size <= SIZE_MAX - guard_size) {
 754     stack_size += guard_size;
 755   }
 756   assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
 757 
 758   int status = pthread_attr_setstacksize(&attr, stack_size);
 759   assert_status(status == 0, status, "pthread_attr_setstacksize");
 760 
 761   // Configure glibc guard page.
 762   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 763 
 764   ThreadState state;
 765 
 766   {
 767     pthread_t tid;
 768     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 769 
 770     char buf[64];
 771     if (ret == 0) {
 772       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 773         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 774     } else {
 775       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 776         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 777     }
 778 
 779     pthread_attr_destroy(&attr);
 780 
 781     if (ret != 0) {
 782       // Need to clean up stuff we've allocated so far
 783       thread->set_osthread(NULL);
 784       delete osthread;
 785       return false;
 786     }
 787 
 788     // Store pthread info into the OSThread
 789     osthread->set_pthread_id(tid);
 790 
 791     // Wait until child thread is either initialized or aborted
 792     {
 793       Monitor* sync_with_child = osthread->startThread_lock();
 794       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 795       while ((state = osthread->get_state()) == ALLOCATED) {
 796         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 797       }
 798     }
 799   }
 800 
 801   // Aborted due to thread limit being reached
 802   if (state == ZOMBIE) {
 803     thread->set_osthread(NULL);
 804     delete osthread;
 805     return false;
 806   }
 807 
 808   // The thread is returned suspended (in state INITIALIZED),
 809   // and is started higher up in the call chain
 810   assert(state == INITIALIZED, "race condition");
 811   return true;
 812 }
 813 
 814 /////////////////////////////////////////////////////////////////////////////
 815 // attach existing thread
 816 
 817 // bootstrap the main thread
 818 bool os::create_main_thread(JavaThread* thread) {
 819   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 820   return create_attached_thread(thread);
 821 }
 822 
 823 bool os::create_attached_thread(JavaThread* thread) {
 824 #ifdef ASSERT
 825   thread->verify_not_published();
 826 #endif
 827 
 828   // Allocate the OSThread object
 829   OSThread* osthread = new OSThread(NULL, NULL);
 830 
 831   if (osthread == NULL) {
 832     return false;
 833   }
 834 
 835   // Store pthread info into the OSThread
 836   osthread->set_thread_id(os::Linux::gettid());
 837   osthread->set_pthread_id(::pthread_self());
 838 
 839   // initialize floating point control register
 840   os::Linux::init_thread_fpu_state();
 841 
 842   // Initial thread state is RUNNABLE
 843   osthread->set_state(RUNNABLE);
 844 
 845   thread->set_osthread(osthread);
 846 
 847   if (UseNUMA) {
 848     int lgrp_id = os::numa_get_group_id();
 849     if (lgrp_id != -1) {
 850       thread->set_lgrp_id(lgrp_id);
 851     }
 852   }
 853 
 854   if (os::is_primordial_thread()) {
 855     // If current thread is primordial thread, its stack is mapped on demand,
 856     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 857     // the entire stack region to avoid SEGV in stack banging.
 858     // It is also useful to get around the heap-stack-gap problem on SuSE
 859     // kernel (see 4821821 for details). We first expand stack to the top
 860     // of yellow zone, then enable stack yellow zone (order is significant,
 861     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 862     // is no gap between the last two virtual memory regions.
 863 
 864     JavaThread *jt = (JavaThread *)thread;
 865     address addr = jt->stack_reserved_zone_base();
 866     assert(addr != NULL, "initialization problem?");
 867     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 868 
 869     osthread->set_expanding_stack();
 870     os::Linux::manually_expand_stack(jt, addr);
 871     osthread->clear_expanding_stack();
 872   }
 873 
 874   // initialize signal mask for this thread
 875   // and save the caller's signal mask
 876   os::Linux::hotspot_sigmask(thread);
 877 
 878   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 879     os::current_thread_id(), (uintx) pthread_self());
 880 
 881   return true;
 882 }
 883 
 884 void os::pd_start_thread(Thread* thread) {
 885   OSThread * osthread = thread->osthread();
 886   assert(osthread->get_state() != INITIALIZED, "just checking");
 887   Monitor* sync_with_child = osthread->startThread_lock();
 888   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 889   sync_with_child->notify();
 890 }
 891 
 892 // Free Linux resources related to the OSThread
 893 void os::free_thread(OSThread* osthread) {
 894   assert(osthread != NULL, "osthread not set");
 895 
 896   // We are told to free resources of the argument thread,
 897   // but we can only really operate on the current thread.
 898   assert(Thread::current()->osthread() == osthread,
 899          "os::free_thread but not current thread");
 900 
 901 #ifdef ASSERT
 902   sigset_t current;
 903   sigemptyset(&current);
 904   pthread_sigmask(SIG_SETMASK, NULL, &current);
 905   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 906 #endif
 907 
 908   // Restore caller's signal mask
 909   sigset_t sigmask = osthread->caller_sigmask();
 910   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 911 
 912   delete osthread;
 913 }
 914 
 915 //////////////////////////////////////////////////////////////////////////////
 916 // primordial thread
 917 
 918 // Check if current thread is the primordial thread, similar to Solaris thr_main.
 919 bool os::is_primordial_thread(void) {
 920   char dummy;
 921   // If called before init complete, thread stack bottom will be null.
 922   // Can be called if fatal error occurs before initialization.
 923   if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
 924   assert(os::Linux::initial_thread_stack_bottom() != NULL &&
 925          os::Linux::initial_thread_stack_size()   != 0,
 926          "os::init did not locate primordial thread's stack region");
 927   if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
 928       (address)&dummy < os::Linux::initial_thread_stack_bottom() +
 929                         os::Linux::initial_thread_stack_size()) {
 930     return true;
 931   } else {
 932     return false;
 933   }
 934 }
 935 
 936 // Find the virtual memory area that contains addr
 937 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 938   FILE *fp = fopen("/proc/self/maps", "r");
 939   if (fp) {
 940     address low, high;
 941     while (!feof(fp)) {
 942       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 943         if (low <= addr && addr < high) {
 944           if (vma_low)  *vma_low  = low;
 945           if (vma_high) *vma_high = high;
 946           fclose(fp);
 947           return true;
 948         }
 949       }
 950       for (;;) {
 951         int ch = fgetc(fp);
 952         if (ch == EOF || ch == (int)'\n') break;
 953       }
 954     }
 955     fclose(fp);
 956   }
 957   return false;
 958 }
 959 
 960 // Locate primordial thread stack. This special handling of primordial thread stack
 961 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 962 // bogus value for the primordial process thread. While the launcher has created
 963 // the VM in a new thread since JDK 6, we still have to allow for the use of the
 964 // JNI invocation API from a primordial thread.
 965 void os::Linux::capture_initial_stack(size_t max_size) {
 966 
 967   // max_size is either 0 (which means accept OS default for thread stacks) or
 968   // a user-specified value known to be at least the minimum needed. If we
 969   // are actually on the primordial thread we can make it appear that we have a
 970   // smaller max_size stack by inserting the guard pages at that location. But we
 971   // cannot do anything to emulate a larger stack than what has been provided by
 972   // the OS or threading library. In fact if we try to use a stack greater than
 973   // what is set by rlimit then we will crash the hosting process.
 974 
 975   // Maximum stack size is the easy part, get it from RLIMIT_STACK.
 976   // If this is "unlimited" then it will be a huge value.
 977   struct rlimit rlim;
 978   getrlimit(RLIMIT_STACK, &rlim);
 979   size_t stack_size = rlim.rlim_cur;
 980 
 981   // 6308388: a bug in ld.so will relocate its own .data section to the
 982   //   lower end of primordial stack; reduce ulimit -s value a little bit
 983   //   so we won't install guard page on ld.so's data section.
 984   //   But ensure we don't underflow the stack size - allow 1 page spare
 985   if (stack_size >= (size_t)(3 * page_size())) {
 986     stack_size -= 2 * page_size();
 987   }
 988 
 989   // Try to figure out where the stack base (top) is. This is harder.
 990   //
 991   // When an application is started, glibc saves the initial stack pointer in
 992   // a global variable "__libc_stack_end", which is then used by system
 993   // libraries. __libc_stack_end should be pretty close to stack top. The
 994   // variable is available since the very early days. However, because it is
 995   // a private interface, it could disappear in the future.
 996   //
 997   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 998   // to __libc_stack_end, it is very close to stack top, but isn't the real
 999   // stack top. Note that /proc may not exist if VM is running as a chroot
1000   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1001   // /proc/<pid>/stat could change in the future (though unlikely).
1002   //
1003   // We try __libc_stack_end first. If that doesn't work, look for
1004   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1005   // as a hint, which should work well in most cases.
1006 
1007   uintptr_t stack_start;
1008 
1009   // try __libc_stack_end first
1010   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1011   if (p && *p) {
1012     stack_start = *p;
1013   } else {
1014     // see if we can get the start_stack field from /proc/self/stat
1015     FILE *fp;
1016     int pid;
1017     char state;
1018     int ppid;
1019     int pgrp;
1020     int session;
1021     int nr;
1022     int tpgrp;
1023     unsigned long flags;
1024     unsigned long minflt;
1025     unsigned long cminflt;
1026     unsigned long majflt;
1027     unsigned long cmajflt;
1028     unsigned long utime;
1029     unsigned long stime;
1030     long cutime;
1031     long cstime;
1032     long prio;
1033     long nice;
1034     long junk;
1035     long it_real;
1036     uintptr_t start;
1037     uintptr_t vsize;
1038     intptr_t rss;
1039     uintptr_t rsslim;
1040     uintptr_t scodes;
1041     uintptr_t ecode;
1042     int i;
1043 
1044     // Figure what the primordial thread stack base is. Code is inspired
1045     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1046     // followed by command name surrounded by parentheses, state, etc.
1047     char stat[2048];
1048     int statlen;
1049 
1050     fp = fopen("/proc/self/stat", "r");
1051     if (fp) {
1052       statlen = fread(stat, 1, 2047, fp);
1053       stat[statlen] = '\0';
1054       fclose(fp);
1055 
1056       // Skip pid and the command string. Note that we could be dealing with
1057       // weird command names, e.g. user could decide to rename java launcher
1058       // to "java 1.4.2 :)", then the stat file would look like
1059       //                1234 (java 1.4.2 :)) R ... ...
1060       // We don't really need to know the command string, just find the last
1061       // occurrence of ")" and then start parsing from there. See bug 4726580.
1062       char * s = strrchr(stat, ')');
1063 
1064       i = 0;
1065       if (s) {
1066         // Skip blank chars
1067         do { s++; } while (s && isspace(*s));
1068 
1069 #define _UFM UINTX_FORMAT
1070 #define _DFM INTX_FORMAT
1071 
1072         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1073         //              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
1074         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,
1075                    &state,          // 3  %c
1076                    &ppid,           // 4  %d
1077                    &pgrp,           // 5  %d
1078                    &session,        // 6  %d
1079                    &nr,             // 7  %d
1080                    &tpgrp,          // 8  %d
1081                    &flags,          // 9  %lu
1082                    &minflt,         // 10 %lu
1083                    &cminflt,        // 11 %lu
1084                    &majflt,         // 12 %lu
1085                    &cmajflt,        // 13 %lu
1086                    &utime,          // 14 %lu
1087                    &stime,          // 15 %lu
1088                    &cutime,         // 16 %ld
1089                    &cstime,         // 17 %ld
1090                    &prio,           // 18 %ld
1091                    &nice,           // 19 %ld
1092                    &junk,           // 20 %ld
1093                    &it_real,        // 21 %ld
1094                    &start,          // 22 UINTX_FORMAT
1095                    &vsize,          // 23 UINTX_FORMAT
1096                    &rss,            // 24 INTX_FORMAT
1097                    &rsslim,         // 25 UINTX_FORMAT
1098                    &scodes,         // 26 UINTX_FORMAT
1099                    &ecode,          // 27 UINTX_FORMAT
1100                    &stack_start);   // 28 UINTX_FORMAT
1101       }
1102 
1103 #undef _UFM
1104 #undef _DFM
1105 
1106       if (i != 28 - 2) {
1107         assert(false, "Bad conversion from /proc/self/stat");
1108         // product mode - assume we are the primordial thread, good luck in the
1109         // embedded case.
1110         warning("Can't detect primordial thread stack location - bad conversion");
1111         stack_start = (uintptr_t) &rlim;
1112       }
1113     } else {
1114       // For some reason we can't open /proc/self/stat (for example, running on
1115       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1116       // most cases, so don't abort:
1117       warning("Can't detect primordial thread stack location - no /proc/self/stat");
1118       stack_start = (uintptr_t) &rlim;
1119     }
1120   }
1121 
1122   // Now we have a pointer (stack_start) very close to the stack top, the
1123   // next thing to do is to figure out the exact location of stack top. We
1124   // can find out the virtual memory area that contains stack_start by
1125   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1126   // and its upper limit is the real stack top. (again, this would fail if
1127   // running inside chroot, because /proc may not exist.)
1128 
1129   uintptr_t stack_top;
1130   address low, high;
1131   if (find_vma((address)stack_start, &low, &high)) {
1132     // success, "high" is the true stack top. (ignore "low", because initial
1133     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1134     stack_top = (uintptr_t)high;
1135   } else {
1136     // failed, likely because /proc/self/maps does not exist
1137     warning("Can't detect primordial thread stack location - find_vma failed");
1138     // best effort: stack_start is normally within a few pages below the real
1139     // stack top, use it as stack top, and reduce stack size so we won't put
1140     // guard page outside stack.
1141     stack_top = stack_start;
1142     stack_size -= 16 * page_size();
1143   }
1144 
1145   // stack_top could be partially down the page so align it
1146   stack_top = align_up(stack_top, page_size());
1147 
1148   // Allowed stack value is minimum of max_size and what we derived from rlimit
1149   if (max_size > 0) {
1150     _initial_thread_stack_size = MIN2(max_size, stack_size);
1151   } else {
1152     // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1153     // clamp it at 8MB as we do on Solaris
1154     _initial_thread_stack_size = MIN2(stack_size, 8*M);
1155   }
1156   _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1157   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1158 
1159   assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1160 
1161   if (log_is_enabled(Info, os, thread)) {
1162     // See if we seem to be on primordial process thread
1163     bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1164                       uintptr_t(&rlim) < stack_top;
1165 
1166     log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1167                          SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1168                          primordial ? "primordial" : "user", max_size / K,  _initial_thread_stack_size / K,
1169                          stack_top, intptr_t(_initial_thread_stack_bottom));
1170   }
1171 }
1172 
1173 ////////////////////////////////////////////////////////////////////////////////
1174 // time support
1175 
1176 // Time since start-up in seconds to a fine granularity.
1177 // Used by VMSelfDestructTimer and the MemProfiler.
1178 double os::elapsedTime() {
1179 
1180   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1181 }
1182 
1183 jlong os::elapsed_counter() {
1184   return javaTimeNanos() - initial_time_count;
1185 }
1186 
1187 jlong os::elapsed_frequency() {
1188   return NANOSECS_PER_SEC; // nanosecond resolution
1189 }
1190 
1191 bool os::supports_vtime() { return true; }
1192 bool os::enable_vtime()   { return false; }
1193 bool os::vtime_enabled()  { return false; }
1194 
1195 double os::elapsedVTime() {
1196   struct rusage usage;
1197   int retval = getrusage(RUSAGE_THREAD, &usage);
1198   if (retval == 0) {
1199     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);
1200   } else {
1201     // better than nothing, but not much
1202     return elapsedTime();
1203   }
1204 }
1205 
1206 jlong os::javaTimeMillis() {
1207   timeval time;
1208   int status = gettimeofday(&time, NULL);
1209   assert(status != -1, "linux error");
1210   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1211 }
1212 
1213 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1214   timeval time;
1215   int status = gettimeofday(&time, NULL);
1216   assert(status != -1, "linux error");
1217   seconds = jlong(time.tv_sec);
1218   nanos = jlong(time.tv_usec) * 1000;
1219 }
1220 
1221 
1222 #ifndef CLOCK_MONOTONIC
1223   #define CLOCK_MONOTONIC (1)
1224 #endif
1225 
1226 void os::Linux::clock_init() {
1227   // we do dlopen's in this particular order due to bug in linux
1228   // dynamical loader (see 6348968) leading to crash on exit
1229   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1230   if (handle == NULL) {
1231     handle = dlopen("librt.so", RTLD_LAZY);
1232   }
1233 
1234   if (handle) {
1235     int (*clock_getres_func)(clockid_t, struct timespec*) =
1236            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1237     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1238            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1239     if (clock_getres_func && clock_gettime_func) {
1240       // See if monotonic clock is supported by the kernel. Note that some
1241       // early implementations simply return kernel jiffies (updated every
1242       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1243       // for nano time (though the monotonic property is still nice to have).
1244       // It's fixed in newer kernels, however clock_getres() still returns
1245       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1246       // resolution for now. Hopefully as people move to new kernels, this
1247       // won't be a problem.
1248       struct timespec res;
1249       struct timespec tp;
1250       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1251           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1252         // yes, monotonic clock is supported
1253         _clock_gettime = clock_gettime_func;
1254         return;
1255       } else {
1256         // close librt if there is no monotonic clock
1257         dlclose(handle);
1258       }
1259     }
1260   }
1261   warning("No monotonic clock was available - timed services may " \
1262           "be adversely affected if the time-of-day clock changes");
1263 }
1264 
1265 #ifndef SYS_clock_getres
1266   #if defined(X86) || defined(PPC64) || defined(S390)
1267     #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1268     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1269   #else
1270     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1271     #define sys_clock_getres(x,y)  -1
1272   #endif
1273 #else
1274   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1275 #endif
1276 
1277 void os::Linux::fast_thread_clock_init() {
1278   if (!UseLinuxPosixThreadCPUClocks) {
1279     return;
1280   }
1281   clockid_t clockid;
1282   struct timespec tp;
1283   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1284       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1285 
1286   // Switch to using fast clocks for thread cpu time if
1287   // the sys_clock_getres() returns 0 error code.
1288   // Note, that some kernels may support the current thread
1289   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1290   // returned by the pthread_getcpuclockid().
1291   // If the fast Posix clocks are supported then the sys_clock_getres()
1292   // must return at least tp.tv_sec == 0 which means a resolution
1293   // better than 1 sec. This is extra check for reliability.
1294 
1295   if (pthread_getcpuclockid_func &&
1296       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1297       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1298     _supports_fast_thread_cpu_time = true;
1299     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1300   }
1301 }
1302 
1303 jlong os::javaTimeNanos() {
1304   if (os::supports_monotonic_clock()) {
1305     struct timespec tp;
1306     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1307     assert(status == 0, "gettime error");
1308     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1309     return result;
1310   } else {
1311     timeval time;
1312     int status = gettimeofday(&time, NULL);
1313     assert(status != -1, "linux error");
1314     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1315     return 1000 * usecs;
1316   }
1317 }
1318 
1319 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1320   if (os::supports_monotonic_clock()) {
1321     info_ptr->max_value = ALL_64_BITS;
1322 
1323     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1324     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1325     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1326   } else {
1327     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1328     info_ptr->max_value = ALL_64_BITS;
1329 
1330     // gettimeofday is a real time clock so it skips
1331     info_ptr->may_skip_backward = true;
1332     info_ptr->may_skip_forward = true;
1333   }
1334 
1335   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1336 }
1337 
1338 // Return the real, user, and system times in seconds from an
1339 // arbitrary fixed point in the past.
1340 bool os::getTimesSecs(double* process_real_time,
1341                       double* process_user_time,
1342                       double* process_system_time) {
1343   struct tms ticks;
1344   clock_t real_ticks = times(&ticks);
1345 
1346   if (real_ticks == (clock_t) (-1)) {
1347     return false;
1348   } else {
1349     double ticks_per_second = (double) clock_tics_per_sec;
1350     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1351     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1352     *process_real_time = ((double) real_ticks) / ticks_per_second;
1353 
1354     return true;
1355   }
1356 }
1357 
1358 
1359 char * os::local_time_string(char *buf, size_t buflen) {
1360   struct tm t;
1361   time_t long_time;
1362   time(&long_time);
1363   localtime_r(&long_time, &t);
1364   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1365                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1366                t.tm_hour, t.tm_min, t.tm_sec);
1367   return buf;
1368 }
1369 
1370 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1371   return localtime_r(clock, res);
1372 }
1373 
1374 ////////////////////////////////////////////////////////////////////////////////
1375 // runtime exit support
1376 
1377 // Note: os::shutdown() might be called very early during initialization, or
1378 // called from signal handler. Before adding something to os::shutdown(), make
1379 // sure it is async-safe and can handle partially initialized VM.
1380 void os::shutdown() {
1381 
1382   // allow PerfMemory to attempt cleanup of any persistent resources
1383   perfMemory_exit();
1384 
1385   // needs to remove object in file system
1386   AttachListener::abort();
1387 
1388   // flush buffered output, finish log files
1389   ostream_abort();
1390 
1391   // Check for abort hook
1392   abort_hook_t abort_hook = Arguments::abort_hook();
1393   if (abort_hook != NULL) {
1394     abort_hook();
1395   }
1396 
1397 }
1398 
1399 // Note: os::abort() might be called very early during initialization, or
1400 // called from signal handler. Before adding something to os::abort(), make
1401 // sure it is async-safe and can handle partially initialized VM.
1402 void os::abort(bool dump_core, void* siginfo, const void* context) {
1403   os::shutdown();
1404   if (dump_core) {
1405 #ifndef PRODUCT
1406     fdStream out(defaultStream::output_fd());
1407     out.print_raw("Current thread is ");
1408     char buf[16];
1409     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1410     out.print_raw_cr(buf);
1411     out.print_raw_cr("Dumping core ...");
1412 #endif
1413     ::abort(); // dump core
1414   }
1415 
1416   ::exit(1);
1417 }
1418 
1419 // Die immediately, no exit hook, no abort hook, no cleanup.
1420 void os::die() {
1421   ::abort();
1422 }
1423 
1424 
1425 // This method is a copy of JDK's sysGetLastErrorString
1426 // from src/solaris/hpi/src/system_md.c
1427 
1428 size_t os::lasterror(char *buf, size_t len) {
1429   if (errno == 0)  return 0;
1430 
1431   const char *s = os::strerror(errno);
1432   size_t n = ::strlen(s);
1433   if (n >= len) {
1434     n = len - 1;
1435   }
1436   ::strncpy(buf, s, n);
1437   buf[n] = '\0';
1438   return n;
1439 }
1440 
1441 // thread_id is kernel thread id (similar to Solaris LWP id)
1442 intx os::current_thread_id() { return os::Linux::gettid(); }
1443 int os::current_process_id() {
1444   return ::getpid();
1445 }
1446 
1447 // DLL functions
1448 
1449 const char* os::dll_file_extension() { return ".so"; }
1450 
1451 // This must be hard coded because it's the system's temporary
1452 // directory not the java application's temp directory, ala java.io.tmpdir.
1453 const char* os::get_temp_directory() { return "/tmp"; }
1454 
1455 static bool file_exists(const char* filename) {
1456   struct stat statbuf;
1457   if (filename == NULL || strlen(filename) == 0) {
1458     return false;
1459   }
1460   return os::stat(filename, &statbuf) == 0;
1461 }
1462 
1463 // check if addr is inside libjvm.so
1464 bool os::address_is_in_vm(address addr) {
1465   static address libjvm_base_addr;
1466   Dl_info dlinfo;
1467 
1468   if (libjvm_base_addr == NULL) {
1469     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1470       libjvm_base_addr = (address)dlinfo.dli_fbase;
1471     }
1472     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1473   }
1474 
1475   if (dladdr((void *)addr, &dlinfo) != 0) {
1476     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1477   }
1478 
1479   return false;
1480 }
1481 
1482 bool os::dll_address_to_function_name(address addr, char *buf,
1483                                       int buflen, int *offset,
1484                                       bool demangle) {
1485   // buf is not optional, but offset is optional
1486   assert(buf != NULL, "sanity check");
1487 
1488   Dl_info dlinfo;
1489 
1490   if (dladdr((void*)addr, &dlinfo) != 0) {
1491     // see if we have a matching symbol
1492     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1493       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1494         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1495       }
1496       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1497       return true;
1498     }
1499     // no matching symbol so try for just file info
1500     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1501       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1502                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1503         return true;
1504       }
1505     }
1506   }
1507 
1508   buf[0] = '\0';
1509   if (offset != NULL) *offset = -1;
1510   return false;
1511 }
1512 
1513 struct _address_to_library_name {
1514   address addr;          // input : memory address
1515   size_t  buflen;        //         size of fname
1516   char*   fname;         // output: library name
1517   address base;          //         library base addr
1518 };
1519 
1520 static int address_to_library_name_callback(struct dl_phdr_info *info,
1521                                             size_t size, void *data) {
1522   int i;
1523   bool found = false;
1524   address libbase = NULL;
1525   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1526 
1527   // iterate through all loadable segments
1528   for (i = 0; i < info->dlpi_phnum; i++) {
1529     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1530     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1531       // base address of a library is the lowest address of its loaded
1532       // segments.
1533       if (libbase == NULL || libbase > segbase) {
1534         libbase = segbase;
1535       }
1536       // see if 'addr' is within current segment
1537       if (segbase <= d->addr &&
1538           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1539         found = true;
1540       }
1541     }
1542   }
1543 
1544   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1545   // so dll_address_to_library_name() can fall through to use dladdr() which
1546   // can figure out executable name from argv[0].
1547   if (found && info->dlpi_name && info->dlpi_name[0]) {
1548     d->base = libbase;
1549     if (d->fname) {
1550       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1551     }
1552     return 1;
1553   }
1554   return 0;
1555 }
1556 
1557 bool os::dll_address_to_library_name(address addr, char* buf,
1558                                      int buflen, int* offset) {
1559   // buf is not optional, but offset is optional
1560   assert(buf != NULL, "sanity check");
1561 
1562   Dl_info dlinfo;
1563   struct _address_to_library_name data;
1564 
1565   // There is a bug in old glibc dladdr() implementation that it could resolve
1566   // to wrong library name if the .so file has a base address != NULL. Here
1567   // we iterate through the program headers of all loaded libraries to find
1568   // out which library 'addr' really belongs to. This workaround can be
1569   // removed once the minimum requirement for glibc is moved to 2.3.x.
1570   data.addr = addr;
1571   data.fname = buf;
1572   data.buflen = buflen;
1573   data.base = NULL;
1574   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1575 
1576   if (rslt) {
1577     // buf already contains library name
1578     if (offset) *offset = addr - data.base;
1579     return true;
1580   }
1581   if (dladdr((void*)addr, &dlinfo) != 0) {
1582     if (dlinfo.dli_fname != NULL) {
1583       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1584     }
1585     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1586       *offset = addr - (address)dlinfo.dli_fbase;
1587     }
1588     return true;
1589   }
1590 
1591   buf[0] = '\0';
1592   if (offset) *offset = -1;
1593   return false;
1594 }
1595 
1596 // Loads .dll/.so and
1597 // in case of error it checks if .dll/.so was built for the
1598 // same architecture as Hotspot is running on
1599 
1600 
1601 // Remember the stack's state. The Linux dynamic linker will change
1602 // the stack to 'executable' at most once, so we must safepoint only once.
1603 bool os::Linux::_stack_is_executable = false;
1604 
1605 // VM operation that loads a library.  This is necessary if stack protection
1606 // of the Java stacks can be lost during loading the library.  If we
1607 // do not stop the Java threads, they can stack overflow before the stacks
1608 // are protected again.
1609 class VM_LinuxDllLoad: public VM_Operation {
1610  private:
1611   const char *_filename;
1612   char *_ebuf;
1613   int _ebuflen;
1614   void *_lib;
1615  public:
1616   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1617     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1618   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1619   void doit() {
1620     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1621     os::Linux::_stack_is_executable = true;
1622   }
1623   void* loaded_library() { return _lib; }
1624 };
1625 
1626 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1627   void * result = NULL;
1628   bool load_attempted = false;
1629 
1630   // Check whether the library to load might change execution rights
1631   // of the stack. If they are changed, the protection of the stack
1632   // guard pages will be lost. We need a safepoint to fix this.
1633   //
1634   // See Linux man page execstack(8) for more info.
1635   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1636     if (!ElfFile::specifies_noexecstack(filename)) {
1637       if (!is_init_completed()) {
1638         os::Linux::_stack_is_executable = true;
1639         // This is OK - No Java threads have been created yet, and hence no
1640         // stack guard pages to fix.
1641         //
1642         // This should happen only when you are building JDK7 using a very
1643         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1644         //
1645         // Dynamic loader will make all stacks executable after
1646         // this function returns, and will not do that again.
1647         assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1648       } else {
1649         warning("You have loaded library %s which might have disabled stack guard. "
1650                 "The VM will try to fix the stack guard now.\n"
1651                 "It's highly recommended that you fix the library with "
1652                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1653                 filename);
1654 
1655         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1656         JavaThread *jt = JavaThread::current();
1657         if (jt->thread_state() != _thread_in_native) {
1658           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1659           // that requires ExecStack. Cannot enter safe point. Let's give up.
1660           warning("Unable to fix stack guard. Giving up.");
1661         } else {
1662           if (!LoadExecStackDllInVMThread) {
1663             // This is for the case where the DLL has an static
1664             // constructor function that executes JNI code. We cannot
1665             // load such DLLs in the VMThread.
1666             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1667           }
1668 
1669           ThreadInVMfromNative tiv(jt);
1670           debug_only(VMNativeEntryWrapper vew;)
1671 
1672           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1673           VMThread::execute(&op);
1674           if (LoadExecStackDllInVMThread) {
1675             result = op.loaded_library();
1676           }
1677           load_attempted = true;
1678         }
1679       }
1680     }
1681   }
1682 
1683   if (!load_attempted) {
1684     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1685   }
1686 
1687   if (result != NULL) {
1688     // Successful loading
1689     return result;
1690   }
1691 
1692   Elf32_Ehdr elf_head;
1693   int diag_msg_max_length=ebuflen-strlen(ebuf);
1694   char* diag_msg_buf=ebuf+strlen(ebuf);
1695 
1696   if (diag_msg_max_length==0) {
1697     // No more space in ebuf for additional diagnostics message
1698     return NULL;
1699   }
1700 
1701 
1702   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1703 
1704   if (file_descriptor < 0) {
1705     // Can't open library, report dlerror() message
1706     return NULL;
1707   }
1708 
1709   bool failed_to_read_elf_head=
1710     (sizeof(elf_head)!=
1711      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1712 
1713   ::close(file_descriptor);
1714   if (failed_to_read_elf_head) {
1715     // file i/o error - report dlerror() msg
1716     return NULL;
1717   }
1718 
1719   typedef struct {
1720     Elf32_Half    code;         // Actual value as defined in elf.h
1721     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1722     unsigned char elf_class;    // 32 or 64 bit
1723     unsigned char endianess;    // MSB or LSB
1724     char*         name;         // String representation
1725   } arch_t;
1726 
1727 #ifndef EM_486
1728   #define EM_486          6               /* Intel 80486 */
1729 #endif
1730 #ifndef EM_AARCH64
1731   #define EM_AARCH64    183               /* ARM AARCH64 */
1732 #endif
1733 
1734   static const arch_t arch_array[]={
1735     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1736     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1737     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1738     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1739     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1740     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1741     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1742     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1743 #if defined(VM_LITTLE_ENDIAN)
1744     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1745     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1746 #else
1747     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1748     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1749 #endif
1750     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1751     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1752     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1753     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1754     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1755     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1756     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1757     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1758   };
1759 
1760 #if  (defined IA32)
1761   static  Elf32_Half running_arch_code=EM_386;
1762 #elif   (defined AMD64)
1763   static  Elf32_Half running_arch_code=EM_X86_64;
1764 #elif  (defined IA64)
1765   static  Elf32_Half running_arch_code=EM_IA_64;
1766 #elif  (defined __sparc) && (defined _LP64)
1767   static  Elf32_Half running_arch_code=EM_SPARCV9;
1768 #elif  (defined __sparc) && (!defined _LP64)
1769   static  Elf32_Half running_arch_code=EM_SPARC;
1770 #elif  (defined __powerpc64__)
1771   static  Elf32_Half running_arch_code=EM_PPC64;
1772 #elif  (defined __powerpc__)
1773   static  Elf32_Half running_arch_code=EM_PPC;
1774 #elif  (defined AARCH64)
1775   static  Elf32_Half running_arch_code=EM_AARCH64;
1776 #elif  (defined ARM)
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_container_info(st);
1994 }
1995 
1996 // Try to identify popular distros.
1997 // Most Linux distributions have a /etc/XXX-release file, which contains
1998 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1999 // file that also contains the OS version string. Some have more than one
2000 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2001 // /etc/redhat-release.), so the order is important.
2002 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2003 // their own specific XXX-release file as well as a redhat-release file.
2004 // Because of this the XXX-release file needs to be searched for before the
2005 // redhat-release file.
2006 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2007 // search for redhat-release / SuSE-release needs to be before lsb-release.
2008 // Since the lsb-release file is the new standard it needs to be searched
2009 // before the older style release files.
2010 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2011 // next to last resort.  The os-release file is a new standard that contains
2012 // distribution information and the system-release file seems to be an old
2013 // standard that has been replaced by the lsb-release and os-release files.
2014 // Searching for the debian_version file is the last resort.  It contains
2015 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2016 // "Debian " is printed before the contents of the debian_version file.
2017 
2018 const char* distro_files[] = {
2019   "/etc/oracle-release",
2020   "/etc/mandriva-release",
2021   "/etc/mandrake-release",
2022   "/etc/sun-release",
2023   "/etc/redhat-release",
2024   "/etc/SuSE-release",
2025   "/etc/lsb-release",
2026   "/etc/turbolinux-release",
2027   "/etc/gentoo-release",
2028   "/etc/ltib-release",
2029   "/etc/angstrom-version",
2030   "/etc/system-release",
2031   "/etc/os-release",
2032   NULL };
2033 
2034 void os::Linux::print_distro_info(outputStream* st) {
2035   for (int i = 0;; i++) {
2036     const char* file = distro_files[i];
2037     if (file == NULL) {
2038       break;  // done
2039     }
2040     // If file prints, we found it.
2041     if (_print_ascii_file(file, st)) {
2042       return;
2043     }
2044   }
2045 
2046   if (file_exists("/etc/debian_version")) {
2047     st->print("Debian ");
2048     _print_ascii_file("/etc/debian_version", st);
2049   } else {
2050     st->print("Linux");
2051   }
2052   st->cr();
2053 }
2054 
2055 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2056   char buf[256];
2057   while (fgets(buf, sizeof(buf), fp)) {
2058     // Edit out extra stuff in expected format
2059     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2060       char* ptr = strstr(buf, "\"");  // the name is in quotes
2061       if (ptr != NULL) {
2062         ptr++; // go beyond first quote
2063         char* nl = strchr(ptr, '\"');
2064         if (nl != NULL) *nl = '\0';
2065         strncpy(distro, ptr, length);
2066       } else {
2067         ptr = strstr(buf, "=");
2068         ptr++; // go beyond equals then
2069         char* nl = strchr(ptr, '\n');
2070         if (nl != NULL) *nl = '\0';
2071         strncpy(distro, ptr, length);
2072       }
2073       return;
2074     } else if (get_first_line) {
2075       char* nl = strchr(buf, '\n');
2076       if (nl != NULL) *nl = '\0';
2077       strncpy(distro, buf, length);
2078       return;
2079     }
2080   }
2081   // print last line and close
2082   char* nl = strchr(buf, '\n');
2083   if (nl != NULL) *nl = '\0';
2084   strncpy(distro, buf, length);
2085 }
2086 
2087 static void parse_os_info(char* distro, size_t length, const char* file) {
2088   FILE* fp = fopen(file, "r");
2089   if (fp != NULL) {
2090     // if suse format, print out first line
2091     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2092     parse_os_info_helper(fp, distro, length, get_first_line);
2093     fclose(fp);
2094   }
2095 }
2096 
2097 void os::get_summary_os_info(char* buf, size_t buflen) {
2098   for (int i = 0;; i++) {
2099     const char* file = distro_files[i];
2100     if (file == NULL) {
2101       break; // ran out of distro_files
2102     }
2103     if (file_exists(file)) {
2104       parse_os_info(buf, buflen, file);
2105       return;
2106     }
2107   }
2108   // special case for debian
2109   if (file_exists("/etc/debian_version")) {
2110     strncpy(buf, "Debian ", buflen);
2111     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2112   } else {
2113     strncpy(buf, "Linux", buflen);
2114   }
2115 }
2116 
2117 void os::Linux::print_libversion_info(outputStream* st) {
2118   // libc, pthread
2119   st->print("libc:");
2120   st->print("%s ", os::Linux::glibc_version());
2121   st->print("%s ", os::Linux::libpthread_version());
2122   st->cr();
2123 }
2124 
2125 void os::Linux::print_full_memory_info(outputStream* st) {
2126   st->print("\n/proc/meminfo:\n");
2127   _print_ascii_file("/proc/meminfo", st);
2128   st->cr();
2129 }
2130 
2131 void os::Linux::print_container_info(outputStream* st) {
2132   if (!OSContainer::is_containerized()) {
2133     return;
2134   }
2135 
2136   st->print("container (cgroup) information:\n");
2137 
2138   const char *p_ct = OSContainer::container_type();
2139   st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2140 
2141   char *p = OSContainer::cpu_cpuset_cpus();
2142   st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2143   free(p);
2144 
2145   p = OSContainer::cpu_cpuset_memory_nodes();
2146   st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2147   free(p);
2148 
2149   int i = OSContainer::active_processor_count();
2150   if (i > 0) {
2151     st->print("active_processor_count: %d\n", i);
2152   } else {
2153     st->print("active_processor_count: failed\n");
2154   }
2155 
2156   i = OSContainer::cpu_quota();
2157   st->print("cpu_quota: %d\n", i);
2158 
2159   i = OSContainer::cpu_period();
2160   st->print("cpu_period: %d\n", i);
2161 
2162   i = OSContainer::cpu_shares();
2163   st->print("cpu_shares: %d\n", i);
2164 
2165   jlong j = OSContainer::memory_limit_in_bytes();
2166   st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2167 
2168   j = OSContainer::memory_and_swap_limit_in_bytes();
2169   st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2170 
2171   j = OSContainer::memory_soft_limit_in_bytes();
2172   st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2173 
2174   j = OSContainer::OSContainer::memory_usage_in_bytes();
2175   st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2176 
2177   j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2178   st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2179   st->cr();
2180 }
2181 
2182 void os::print_memory_info(outputStream* st) {
2183 
2184   st->print("Memory:");
2185   st->print(" %dk page", os::vm_page_size()>>10);
2186 
2187   // values in struct sysinfo are "unsigned long"
2188   struct sysinfo si;
2189   sysinfo(&si);
2190 
2191   st->print(", physical " UINT64_FORMAT "k",
2192             os::physical_memory() >> 10);
2193   st->print("(" UINT64_FORMAT "k free)",
2194             os::available_memory() >> 10);
2195   st->print(", swap " UINT64_FORMAT "k",
2196             ((jlong)si.totalswap * si.mem_unit) >> 10);
2197   st->print("(" UINT64_FORMAT "k free)",
2198             ((jlong)si.freeswap * si.mem_unit) >> 10);
2199   st->cr();
2200 }
2201 
2202 // Print the first "model name" line and the first "flags" line
2203 // that we find and nothing more. We assume "model name" comes
2204 // before "flags" so if we find a second "model name", then the
2205 // "flags" field is considered missing.
2206 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2207 #if defined(IA32) || defined(AMD64)
2208   // Other platforms have less repetitive cpuinfo files
2209   FILE *fp = fopen("/proc/cpuinfo", "r");
2210   if (fp) {
2211     while (!feof(fp)) {
2212       if (fgets(buf, buflen, fp)) {
2213         // Assume model name comes before flags
2214         bool model_name_printed = false;
2215         if (strstr(buf, "model name") != NULL) {
2216           if (!model_name_printed) {
2217             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2218             st->print_raw(buf);
2219             model_name_printed = true;
2220           } else {
2221             // model name printed but not flags?  Odd, just return
2222             fclose(fp);
2223             return true;
2224           }
2225         }
2226         // print the flags line too
2227         if (strstr(buf, "flags") != NULL) {
2228           st->print_raw(buf);
2229           fclose(fp);
2230           return true;
2231         }
2232       }
2233     }
2234     fclose(fp);
2235   }
2236 #endif // x86 platforms
2237   return false;
2238 }
2239 
2240 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2241   // Only print the model name if the platform provides this as a summary
2242   if (!print_model_name_and_flags(st, buf, buflen)) {
2243     st->print("\n/proc/cpuinfo:\n");
2244     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2245       st->print_cr("  <Not Available>");
2246     }
2247   }
2248 }
2249 
2250 #if defined(AMD64) || defined(IA32) || defined(X32)
2251 const char* search_string = "model name";
2252 #elif defined(M68K)
2253 const char* search_string = "CPU";
2254 #elif defined(PPC64)
2255 const char* search_string = "cpu";
2256 #elif defined(S390)
2257 const char* search_string = "processor";
2258 #elif defined(SPARC)
2259 const char* search_string = "cpu";
2260 #else
2261 const char* search_string = "Processor";
2262 #endif
2263 
2264 // Parses the cpuinfo file for string representing the model name.
2265 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2266   FILE* fp = fopen("/proc/cpuinfo", "r");
2267   if (fp != NULL) {
2268     while (!feof(fp)) {
2269       char buf[256];
2270       if (fgets(buf, sizeof(buf), fp)) {
2271         char* start = strstr(buf, search_string);
2272         if (start != NULL) {
2273           char *ptr = start + strlen(search_string);
2274           char *end = buf + strlen(buf);
2275           while (ptr != end) {
2276              // skip whitespace and colon for the rest of the name.
2277              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2278                break;
2279              }
2280              ptr++;
2281           }
2282           if (ptr != end) {
2283             // reasonable string, get rid of newline and keep the rest
2284             char* nl = strchr(buf, '\n');
2285             if (nl != NULL) *nl = '\0';
2286             strncpy(cpuinfo, ptr, length);
2287             fclose(fp);
2288             return;
2289           }
2290         }
2291       }
2292     }
2293     fclose(fp);
2294   }
2295   // cpuinfo not found or parsing failed, just print generic string.  The entire
2296   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2297 #if   defined(AARCH64)
2298   strncpy(cpuinfo, "AArch64", length);
2299 #elif defined(AMD64)
2300   strncpy(cpuinfo, "x86_64", length);
2301 #elif defined(ARM)  // Order wrt. AARCH64 is relevant!
2302   strncpy(cpuinfo, "ARM", length);
2303 #elif defined(IA32)
2304   strncpy(cpuinfo, "x86_32", length);
2305 #elif defined(IA64)
2306   strncpy(cpuinfo, "IA64", length);
2307 #elif defined(PPC)
2308   strncpy(cpuinfo, "PPC64", length);
2309 #elif defined(S390)
2310   strncpy(cpuinfo, "S390", length);
2311 #elif defined(SPARC)
2312   strncpy(cpuinfo, "sparcv9", length);
2313 #elif defined(ZERO_LIBARCH)
2314   strncpy(cpuinfo, ZERO_LIBARCH, length);
2315 #else
2316   strncpy(cpuinfo, "unknown", length);
2317 #endif
2318 }
2319 
2320 static void print_signal_handler(outputStream* st, int sig,
2321                                  char* buf, size_t buflen);
2322 
2323 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2324   st->print_cr("Signal Handlers:");
2325   print_signal_handler(st, SIGSEGV, buf, buflen);
2326   print_signal_handler(st, SIGBUS , buf, buflen);
2327   print_signal_handler(st, SIGFPE , buf, buflen);
2328   print_signal_handler(st, SIGPIPE, buf, buflen);
2329   print_signal_handler(st, SIGXFSZ, buf, buflen);
2330   print_signal_handler(st, SIGILL , buf, buflen);
2331   print_signal_handler(st, SR_signum, buf, buflen);
2332   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2333   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2334   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2335   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2336 #if defined(PPC64)
2337   print_signal_handler(st, SIGTRAP, buf, buflen);
2338 #endif
2339 }
2340 
2341 static char saved_jvm_path[MAXPATHLEN] = {0};
2342 
2343 // Find the full path to the current module, libjvm.so
2344 void os::jvm_path(char *buf, jint buflen) {
2345   // Error checking.
2346   if (buflen < MAXPATHLEN) {
2347     assert(false, "must use a large-enough buffer");
2348     buf[0] = '\0';
2349     return;
2350   }
2351   // Lazy resolve the path to current module.
2352   if (saved_jvm_path[0] != 0) {
2353     strcpy(buf, saved_jvm_path);
2354     return;
2355   }
2356 
2357   char dli_fname[MAXPATHLEN];
2358   bool ret = dll_address_to_library_name(
2359                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2360                                          dli_fname, sizeof(dli_fname), NULL);
2361   assert(ret, "cannot locate libjvm");
2362   char *rp = NULL;
2363   if (ret && dli_fname[0] != '\0') {
2364     rp = os::Posix::realpath(dli_fname, buf, buflen);
2365   }
2366   if (rp == NULL) {
2367     return;
2368   }
2369 
2370   if (Arguments::sun_java_launcher_is_altjvm()) {
2371     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2372     // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2373     // If "/jre/lib/" appears at the right place in the string, then
2374     // assume we are installed in a JDK and we're done. Otherwise, check
2375     // for a JAVA_HOME environment variable and fix up the path so it
2376     // looks like libjvm.so is installed there (append a fake suffix
2377     // hotspot/libjvm.so).
2378     const char *p = buf + strlen(buf) - 1;
2379     for (int count = 0; p > buf && count < 5; ++count) {
2380       for (--p; p > buf && *p != '/'; --p)
2381         /* empty */ ;
2382     }
2383 
2384     if (strncmp(p, "/jre/lib/", 9) != 0) {
2385       // Look for JAVA_HOME in the environment.
2386       char* java_home_var = ::getenv("JAVA_HOME");
2387       if (java_home_var != NULL && java_home_var[0] != 0) {
2388         char* jrelib_p;
2389         int len;
2390 
2391         // Check the current module name "libjvm.so".
2392         p = strrchr(buf, '/');
2393         if (p == NULL) {
2394           return;
2395         }
2396         assert(strstr(p, "/libjvm") == p, "invalid library name");
2397 
2398         rp = os::Posix::realpath(java_home_var, buf, buflen);
2399         if (rp == NULL) {
2400           return;
2401         }
2402 
2403         // determine if this is a legacy image or modules image
2404         // modules image doesn't have "jre" subdirectory
2405         len = strlen(buf);
2406         assert(len < buflen, "Ran out of buffer room");
2407         jrelib_p = buf + len;
2408         snprintf(jrelib_p, buflen-len, "/jre/lib");
2409         if (0 != access(buf, F_OK)) {
2410           snprintf(jrelib_p, buflen-len, "/lib");
2411         }
2412 
2413         if (0 == access(buf, F_OK)) {
2414           // Use current module name "libjvm.so"
2415           len = strlen(buf);
2416           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2417         } else {
2418           // Go back to path of .so
2419           rp = os::Posix::realpath(dli_fname, buf, buflen);
2420           if (rp == NULL) {
2421             return;
2422           }
2423         }
2424       }
2425     }
2426   }
2427 
2428   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2429   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2430 }
2431 
2432 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2433   // no prefix required, not even "_"
2434 }
2435 
2436 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2437   // no suffix required
2438 }
2439 
2440 ////////////////////////////////////////////////////////////////////////////////
2441 // sun.misc.Signal support
2442 
2443 static volatile jint sigint_count = 0;
2444 
2445 static void UserHandler(int sig, void *siginfo, void *context) {
2446   // 4511530 - sem_post is serialized and handled by the manager thread. When
2447   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2448   // don't want to flood the manager thread with sem_post requests.
2449   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2450     return;
2451   }
2452 
2453   // Ctrl-C is pressed during error reporting, likely because the error
2454   // handler fails to abort. Let VM die immediately.
2455   if (sig == SIGINT && VMError::is_error_reported()) {
2456     os::die();
2457   }
2458 
2459   os::signal_notify(sig);
2460 }
2461 
2462 void* os::user_handler() {
2463   return CAST_FROM_FN_PTR(void*, UserHandler);
2464 }
2465 
2466 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2467   struct timespec ts;
2468   // Semaphore's are always associated with CLOCK_REALTIME
2469   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2470   // see os_posix.cpp for discussion on overflow checking
2471   if (sec >= MAX_SECS) {
2472     ts.tv_sec += MAX_SECS;
2473     ts.tv_nsec = 0;
2474   } else {
2475     ts.tv_sec += sec;
2476     ts.tv_nsec += nsec;
2477     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2478       ts.tv_nsec -= NANOSECS_PER_SEC;
2479       ++ts.tv_sec; // note: this must be <= max_secs
2480     }
2481   }
2482 
2483   return ts;
2484 }
2485 
2486 extern "C" {
2487   typedef void (*sa_handler_t)(int);
2488   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2489 }
2490 
2491 void* os::signal(int signal_number, void* handler) {
2492   struct sigaction sigAct, oldSigAct;
2493 
2494   sigfillset(&(sigAct.sa_mask));
2495   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2496   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2497 
2498   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2499     // -1 means registration failed
2500     return (void *)-1;
2501   }
2502 
2503   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2504 }
2505 
2506 void os::signal_raise(int signal_number) {
2507   ::raise(signal_number);
2508 }
2509 
2510 // The following code is moved from os.cpp for making this
2511 // code platform specific, which it is by its very nature.
2512 
2513 // Will be modified when max signal is changed to be dynamic
2514 int os::sigexitnum_pd() {
2515   return NSIG;
2516 }
2517 
2518 // a counter for each possible signal value
2519 static volatile jint pending_signals[NSIG+1] = { 0 };
2520 
2521 // Linux(POSIX) specific hand shaking semaphore.
2522 static Semaphore* sig_sem = NULL;
2523 static PosixSemaphore sr_semaphore;
2524 
2525 void os::signal_init_pd() {
2526   // Initialize signal structures
2527   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2528 
2529   // Initialize signal semaphore
2530   sig_sem = new Semaphore();
2531 }
2532 
2533 void os::signal_notify(int sig) {
2534   if (sig_sem != NULL) {
2535     Atomic::inc(&pending_signals[sig]);
2536     sig_sem->signal();
2537   } else {
2538     // Signal thread is not created with ReduceSignalUsage and signal_init_pd
2539     // initialization isn't called.
2540     assert(ReduceSignalUsage, "signal semaphore should be created");
2541   }
2542 }
2543 
2544 static int check_pending_signals() {
2545   Atomic::store(0, &sigint_count);
2546   for (;;) {
2547     for (int i = 0; i < NSIG + 1; i++) {
2548       jint n = pending_signals[i];
2549       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2550         return i;
2551       }
2552     }
2553     JavaThread *thread = JavaThread::current();
2554     ThreadBlockInVM tbivm(thread);
2555 
2556     bool threadIsSuspended;
2557     do {
2558       thread->set_suspend_equivalent();
2559       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2560       sig_sem->wait();
2561 
2562       // were we externally suspended while we were waiting?
2563       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2564       if (threadIsSuspended) {
2565         // The semaphore has been incremented, but while we were waiting
2566         // another thread suspended us. We don't want to continue running
2567         // while suspended because that would surprise the thread that
2568         // suspended us.
2569         sig_sem->signal();
2570 
2571         thread->java_suspend_self();
2572       }
2573     } while (threadIsSuspended);
2574   }
2575 }
2576 
2577 int os::signal_wait() {
2578   return check_pending_signals();
2579 }
2580 
2581 ////////////////////////////////////////////////////////////////////////////////
2582 // Virtual Memory
2583 
2584 int os::vm_page_size() {
2585   // Seems redundant as all get out
2586   assert(os::Linux::page_size() != -1, "must call os::init");
2587   return os::Linux::page_size();
2588 }
2589 
2590 // Solaris allocates memory by pages.
2591 int os::vm_allocation_granularity() {
2592   assert(os::Linux::page_size() != -1, "must call os::init");
2593   return os::Linux::page_size();
2594 }
2595 
2596 // Rationale behind this function:
2597 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2598 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2599 //  samples for JITted code. Here we create private executable mapping over the code cache
2600 //  and then we can use standard (well, almost, as mapping can change) way to provide
2601 //  info for the reporting script by storing timestamp and location of symbol
2602 void linux_wrap_code(char* base, size_t size) {
2603   static volatile jint cnt = 0;
2604 
2605   if (!UseOprofile) {
2606     return;
2607   }
2608 
2609   char buf[PATH_MAX+1];
2610   int num = Atomic::add(1, &cnt);
2611 
2612   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2613            os::get_temp_directory(), os::current_process_id(), num);
2614   unlink(buf);
2615 
2616   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2617 
2618   if (fd != -1) {
2619     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2620     if (rv != (off_t)-1) {
2621       if (::write(fd, "", 1) == 1) {
2622         mmap(base, size,
2623              PROT_READ|PROT_WRITE|PROT_EXEC,
2624              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2625       }
2626     }
2627     ::close(fd);
2628     unlink(buf);
2629   }
2630 }
2631 
2632 static bool recoverable_mmap_error(int err) {
2633   // See if the error is one we can let the caller handle. This
2634   // list of errno values comes from JBS-6843484. I can't find a
2635   // Linux man page that documents this specific set of errno
2636   // values so while this list currently matches Solaris, it may
2637   // change as we gain experience with this failure mode.
2638   switch (err) {
2639   case EBADF:
2640   case EINVAL:
2641   case ENOTSUP:
2642     // let the caller deal with these errors
2643     return true;
2644 
2645   default:
2646     // Any remaining errors on this OS can cause our reserved mapping
2647     // to be lost. That can cause confusion where different data
2648     // structures think they have the same memory mapped. The worst
2649     // scenario is if both the VM and a library think they have the
2650     // same memory mapped.
2651     return false;
2652   }
2653 }
2654 
2655 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2656                                     int err) {
2657   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2658           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2659           os::strerror(err), err);
2660 }
2661 
2662 static void warn_fail_commit_memory(char* addr, size_t size,
2663                                     size_t alignment_hint, bool exec,
2664                                     int err) {
2665   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2666           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2667           alignment_hint, exec, os::strerror(err), err);
2668 }
2669 
2670 // NOTE: Linux kernel does not really reserve the pages for us.
2671 //       All it does is to check if there are enough free pages
2672 //       left at the time of mmap(). This could be a potential
2673 //       problem.
2674 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2675   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2676   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2677                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2678   if (res != (uintptr_t) MAP_FAILED) {
2679     if (UseNUMAInterleaving) {
2680       numa_make_global(addr, size);
2681     }
2682     return 0;
2683   }
2684 
2685   int err = errno;  // save errno from mmap() call above
2686 
2687   if (!recoverable_mmap_error(err)) {
2688     warn_fail_commit_memory(addr, size, exec, err);
2689     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2690   }
2691 
2692   return err;
2693 }
2694 
2695 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2696   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2697 }
2698 
2699 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2700                                   const char* mesg) {
2701   assert(mesg != NULL, "mesg must be specified");
2702   int err = os::Linux::commit_memory_impl(addr, size, exec);
2703   if (err != 0) {
2704     // the caller wants all commit errors to exit with the specified mesg:
2705     warn_fail_commit_memory(addr, size, exec, err);
2706     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2707   }
2708 }
2709 
2710 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2711 #ifndef MAP_HUGETLB
2712   #define MAP_HUGETLB 0x40000
2713 #endif
2714 
2715 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2716 #ifndef MADV_HUGEPAGE
2717   #define MADV_HUGEPAGE 14
2718 #endif
2719 
2720 int os::Linux::commit_memory_impl(char* addr, size_t size,
2721                                   size_t alignment_hint, bool exec) {
2722   int err = os::Linux::commit_memory_impl(addr, size, exec);
2723   if (err == 0) {
2724     realign_memory(addr, size, alignment_hint);
2725   }
2726   return err;
2727 }
2728 
2729 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2730                           bool exec) {
2731   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2732 }
2733 
2734 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2735                                   size_t alignment_hint, bool exec,
2736                                   const char* mesg) {
2737   assert(mesg != NULL, "mesg must be specified");
2738   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2739   if (err != 0) {
2740     // the caller wants all commit errors to exit with the specified mesg:
2741     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2742     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2743   }
2744 }
2745 
2746 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2747   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2748     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2749     // be supported or the memory may already be backed by huge pages.
2750     ::madvise(addr, bytes, MADV_HUGEPAGE);
2751   }
2752 }
2753 
2754 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2755   // This method works by doing an mmap over an existing mmaping and effectively discarding
2756   // the existing pages. However it won't work for SHM-based large pages that cannot be
2757   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2758   // small pages on top of the SHM segment. This method always works for small pages, so we
2759   // allow that in any case.
2760   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2761     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2762   }
2763 }
2764 
2765 void os::numa_make_global(char *addr, size_t bytes) {
2766   Linux::numa_interleave_memory(addr, bytes);
2767 }
2768 
2769 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2770 // bind policy to MPOL_PREFERRED for the current thread.
2771 #define USE_MPOL_PREFERRED 0
2772 
2773 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2774   // To make NUMA and large pages more robust when both enabled, we need to ease
2775   // the requirements on where the memory should be allocated. MPOL_BIND is the
2776   // default policy and it will force memory to be allocated on the specified
2777   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2778   // the specified node, but will not force it. Using this policy will prevent
2779   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2780   // free large pages.
2781   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2782   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2783 }
2784 
2785 bool os::numa_topology_changed() { return false; }
2786 
2787 size_t os::numa_get_groups_num() {
2788   // Return just the number of nodes in which it's possible to allocate memory
2789   // (in numa terminology, configured nodes).
2790   return Linux::numa_num_configured_nodes();
2791 }
2792 
2793 int os::numa_get_group_id() {
2794   int cpu_id = Linux::sched_getcpu();
2795   if (cpu_id != -1) {
2796     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2797     if (lgrp_id != -1) {
2798       return lgrp_id;
2799     }
2800   }
2801   return 0;
2802 }
2803 
2804 int os::Linux::get_existing_num_nodes() {
2805   size_t node;
2806   size_t highest_node_number = Linux::numa_max_node();
2807   int num_nodes = 0;
2808 
2809   // Get the total number of nodes in the system including nodes without memory.
2810   for (node = 0; node <= highest_node_number; node++) {
2811     if (isnode_in_existing_nodes(node)) {
2812       num_nodes++;
2813     }
2814   }
2815   return num_nodes;
2816 }
2817 
2818 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2819   size_t highest_node_number = Linux::numa_max_node();
2820   size_t i = 0;
2821 
2822   // Map all node ids in which is possible to allocate memory. Also nodes are
2823   // not always consecutively available, i.e. available from 0 to the highest
2824   // node number.
2825   for (size_t node = 0; node <= highest_node_number; node++) {
2826     if (Linux::isnode_in_configured_nodes(node)) {
2827       ids[i++] = node;
2828     }
2829   }
2830   return i;
2831 }
2832 
2833 bool os::get_page_info(char *start, page_info* info) {
2834   return false;
2835 }
2836 
2837 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2838                      page_info* page_found) {
2839   return end;
2840 }
2841 
2842 
2843 int os::Linux::sched_getcpu_syscall(void) {
2844   unsigned int cpu = 0;
2845   int retval = -1;
2846 
2847 #if defined(IA32)
2848   #ifndef SYS_getcpu
2849     #define SYS_getcpu 318
2850   #endif
2851   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2852 #elif defined(AMD64)
2853 // Unfortunately we have to bring all these macros here from vsyscall.h
2854 // to be able to compile on old linuxes.
2855   #define __NR_vgetcpu 2
2856   #define VSYSCALL_START (-10UL << 20)
2857   #define VSYSCALL_SIZE 1024
2858   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2859   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2860   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2861   retval = vgetcpu(&cpu, NULL, NULL);
2862 #endif
2863 
2864   return (retval == -1) ? retval : cpu;
2865 }
2866 
2867 void os::Linux::sched_getcpu_init() {
2868   // sched_getcpu() should be in libc.
2869   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2870                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2871 
2872   // If it's not, try a direct syscall.
2873   if (sched_getcpu() == -1) {
2874     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2875                                     (void*)&sched_getcpu_syscall));
2876   }
2877 }
2878 
2879 // Something to do with the numa-aware allocator needs these symbols
2880 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2881 extern "C" JNIEXPORT void numa_error(char *where) { }
2882 
2883 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2884 // load symbol from base version instead.
2885 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2886   void *f = dlvsym(handle, name, "libnuma_1.1");
2887   if (f == NULL) {
2888     f = dlsym(handle, name);
2889   }
2890   return f;
2891 }
2892 
2893 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2894 // Return NULL if the symbol is not defined in this particular version.
2895 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2896   return dlvsym(handle, name, "libnuma_1.2");
2897 }
2898 
2899 bool os::Linux::libnuma_init() {
2900   if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2901     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2902     if (handle != NULL) {
2903       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2904                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2905       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2906                                        libnuma_dlsym(handle, "numa_max_node")));
2907       set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2908                                                    libnuma_dlsym(handle, "numa_num_configured_nodes")));
2909       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2910                                         libnuma_dlsym(handle, "numa_available")));
2911       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2912                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2913       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2914                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2915       set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2916                                                 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2917       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2918                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2919       set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2920                                                libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2921       set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2922                                        libnuma_dlsym(handle, "numa_distance")));
2923 
2924       if (numa_available() != -1) {
2925         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2926         set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2927         set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2928         // Create an index -> node mapping, since nodes are not always consecutive
2929         _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2930         rebuild_nindex_to_node_map();
2931         // Create a cpu -> node mapping
2932         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2933         rebuild_cpu_to_node_map();
2934         return true;
2935       }
2936     }
2937   }
2938   return false;
2939 }
2940 
2941 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2942   // Creating guard page is very expensive. Java thread has HotSpot
2943   // guard pages, only enable glibc guard page for non-Java threads.
2944   // (Remember: compiler thread is a Java thread, too!)
2945   return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2946 }
2947 
2948 void os::Linux::rebuild_nindex_to_node_map() {
2949   int highest_node_number = Linux::numa_max_node();
2950 
2951   nindex_to_node()->clear();
2952   for (int node = 0; node <= highest_node_number; node++) {
2953     if (Linux::isnode_in_existing_nodes(node)) {
2954       nindex_to_node()->append(node);
2955     }
2956   }
2957 }
2958 
2959 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2960 // The table is later used in get_node_by_cpu().
2961 void os::Linux::rebuild_cpu_to_node_map() {
2962   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2963                               // in libnuma (possible values are starting from 16,
2964                               // and continuing up with every other power of 2, but less
2965                               // than the maximum number of CPUs supported by kernel), and
2966                               // is a subject to change (in libnuma version 2 the requirements
2967                               // are more reasonable) we'll just hardcode the number they use
2968                               // in the library.
2969   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2970 
2971   size_t cpu_num = processor_count();
2972   size_t cpu_map_size = NCPUS / BitsPerCLong;
2973   size_t cpu_map_valid_size =
2974     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2975 
2976   cpu_to_node()->clear();
2977   cpu_to_node()->at_grow(cpu_num - 1);
2978 
2979   size_t node_num = get_existing_num_nodes();
2980 
2981   int distance = 0;
2982   int closest_distance = INT_MAX;
2983   int closest_node = 0;
2984   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2985   for (size_t i = 0; i < node_num; i++) {
2986     // Check if node is configured (not a memory-less node). If it is not, find
2987     // the closest configured node.
2988     if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
2989       closest_distance = INT_MAX;
2990       // Check distance from all remaining nodes in the system. Ignore distance
2991       // from itself and from another non-configured node.
2992       for (size_t m = 0; m < node_num; m++) {
2993         if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
2994           distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
2995           // If a closest node is found, update. There is always at least one
2996           // configured node in the system so there is always at least one node
2997           // close.
2998           if (distance != 0 && distance < closest_distance) {
2999             closest_distance = distance;
3000             closest_node = nindex_to_node()->at(m);
3001           }
3002         }
3003       }
3004      } else {
3005        // Current node is already a configured node.
3006        closest_node = nindex_to_node()->at(i);
3007      }
3008 
3009     // Get cpus from the original node and map them to the closest node. If node
3010     // is a configured node (not a memory-less node), then original node and
3011     // closest node are the same.
3012     if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3013       for (size_t j = 0; j < cpu_map_valid_size; j++) {
3014         if (cpu_map[j] != 0) {
3015           for (size_t k = 0; k < BitsPerCLong; k++) {
3016             if (cpu_map[j] & (1UL << k)) {
3017               cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3018             }
3019           }
3020         }
3021       }
3022     }
3023   }
3024   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3025 }
3026 
3027 int os::Linux::get_node_by_cpu(int cpu_id) {
3028   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3029     return cpu_to_node()->at(cpu_id);
3030   }
3031   return -1;
3032 }
3033 
3034 GrowableArray<int>* os::Linux::_cpu_to_node;
3035 GrowableArray<int>* os::Linux::_nindex_to_node;
3036 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3037 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3038 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3039 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3040 os::Linux::numa_available_func_t os::Linux::_numa_available;
3041 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3042 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3043 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3044 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3045 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3046 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3047 unsigned long* os::Linux::_numa_all_nodes;
3048 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3049 struct bitmask* os::Linux::_numa_nodes_ptr;
3050 
3051 bool os::pd_uncommit_memory(char* addr, size_t size) {
3052   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3053                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3054   return res  != (uintptr_t) MAP_FAILED;
3055 }
3056 
3057 static address get_stack_commited_bottom(address bottom, size_t size) {
3058   address nbot = bottom;
3059   address ntop = bottom + size;
3060 
3061   size_t page_sz = os::vm_page_size();
3062   unsigned pages = size / page_sz;
3063 
3064   unsigned char vec[1];
3065   unsigned imin = 1, imax = pages + 1, imid;
3066   int mincore_return_value = 0;
3067 
3068   assert(imin <= imax, "Unexpected page size");
3069 
3070   while (imin < imax) {
3071     imid = (imax + imin) / 2;
3072     nbot = ntop - (imid * page_sz);
3073 
3074     // Use a trick with mincore to check whether the page is mapped or not.
3075     // mincore sets vec to 1 if page resides in memory and to 0 if page
3076     // is swapped output but if page we are asking for is unmapped
3077     // it returns -1,ENOMEM
3078     mincore_return_value = mincore(nbot, page_sz, vec);
3079 
3080     if (mincore_return_value == -1) {
3081       // Page is not mapped go up
3082       // to find first mapped page
3083       if (errno != EAGAIN) {
3084         assert(errno == ENOMEM, "Unexpected mincore errno");
3085         imax = imid;
3086       }
3087     } else {
3088       // Page is mapped go down
3089       // to find first not mapped page
3090       imin = imid + 1;
3091     }
3092   }
3093 
3094   nbot = nbot + page_sz;
3095 
3096   // Adjust stack bottom one page up if last checked page is not mapped
3097   if (mincore_return_value == -1) {
3098     nbot = nbot + page_sz;
3099   }
3100 
3101   return nbot;
3102 }
3103 
3104 
3105 // Linux uses a growable mapping for the stack, and if the mapping for
3106 // the stack guard pages is not removed when we detach a thread the
3107 // stack cannot grow beyond the pages where the stack guard was
3108 // mapped.  If at some point later in the process the stack expands to
3109 // that point, the Linux kernel cannot expand the stack any further
3110 // because the guard pages are in the way, and a segfault occurs.
3111 //
3112 // However, it's essential not to split the stack region by unmapping
3113 // a region (leaving a hole) that's already part of the stack mapping,
3114 // so if the stack mapping has already grown beyond the guard pages at
3115 // the time we create them, we have to truncate the stack mapping.
3116 // So, we need to know the extent of the stack mapping when
3117 // create_stack_guard_pages() is called.
3118 
3119 // We only need this for stacks that are growable: at the time of
3120 // writing thread stacks don't use growable mappings (i.e. those
3121 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3122 // only applies to the main thread.
3123 
3124 // If the (growable) stack mapping already extends beyond the point
3125 // where we're going to put our guard pages, truncate the mapping at
3126 // that point by munmap()ping it.  This ensures that when we later
3127 // munmap() the guard pages we don't leave a hole in the stack
3128 // mapping. This only affects the main/primordial thread
3129 
3130 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3131   if (os::is_primordial_thread()) {
3132     // As we manually grow stack up to bottom inside create_attached_thread(),
3133     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3134     // we don't need to do anything special.
3135     // Check it first, before calling heavy function.
3136     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3137     unsigned char vec[1];
3138 
3139     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3140       // Fallback to slow path on all errors, including EAGAIN
3141       stack_extent = (uintptr_t) get_stack_commited_bottom(
3142                                                            os::Linux::initial_thread_stack_bottom(),
3143                                                            (size_t)addr - stack_extent);
3144     }
3145 
3146     if (stack_extent < (uintptr_t)addr) {
3147       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3148     }
3149   }
3150 
3151   return os::commit_memory(addr, size, !ExecMem);
3152 }
3153 
3154 // If this is a growable mapping, remove the guard pages entirely by
3155 // munmap()ping them.  If not, just call uncommit_memory(). This only
3156 // affects the main/primordial thread, but guard against future OS changes.
3157 // It's safe to always unmap guard pages for primordial thread because we
3158 // always place it right after end of the mapped region.
3159 
3160 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3161   uintptr_t stack_extent, stack_base;
3162 
3163   if (os::is_primordial_thread()) {
3164     return ::munmap(addr, size) == 0;
3165   }
3166 
3167   return os::uncommit_memory(addr, size);
3168 }
3169 
3170 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3171 // at 'requested_addr'. If there are existing memory mappings at the same
3172 // location, however, they will be overwritten. If 'fixed' is false,
3173 // 'requested_addr' is only treated as a hint, the return value may or
3174 // may not start from the requested address. Unlike Linux mmap(), this
3175 // function returns NULL to indicate failure.
3176 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3177   char * addr;
3178   int flags;
3179 
3180   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3181   if (fixed) {
3182     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3183     flags |= MAP_FIXED;
3184   }
3185 
3186   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3187   // touch an uncommitted page. Otherwise, the read/write might
3188   // succeed if we have enough swap space to back the physical page.
3189   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3190                        flags, -1, 0);
3191 
3192   return addr == MAP_FAILED ? NULL : addr;
3193 }
3194 
3195 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3196 //   (req_addr != NULL) or with a given alignment.
3197 //  - bytes shall be a multiple of alignment.
3198 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3199 //  - alignment sets the alignment at which memory shall be allocated.
3200 //     It must be a multiple of allocation granularity.
3201 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3202 //  req_addr or NULL.
3203 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3204 
3205   size_t extra_size = bytes;
3206   if (req_addr == NULL && alignment > 0) {
3207     extra_size += alignment;
3208   }
3209 
3210   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3211     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3212     -1, 0);
3213   if (start == MAP_FAILED) {
3214     start = NULL;
3215   } else {
3216     if (req_addr != NULL) {
3217       if (start != req_addr) {
3218         ::munmap(start, extra_size);
3219         start = NULL;
3220       }
3221     } else {
3222       char* const start_aligned = align_up(start, alignment);
3223       char* const end_aligned = start_aligned + bytes;
3224       char* const end = start + extra_size;
3225       if (start_aligned > start) {
3226         ::munmap(start, start_aligned - start);
3227       }
3228       if (end_aligned < end) {
3229         ::munmap(end_aligned, end - end_aligned);
3230       }
3231       start = start_aligned;
3232     }
3233   }
3234   return start;
3235 }
3236 
3237 static int anon_munmap(char * addr, size_t size) {
3238   return ::munmap(addr, size) == 0;
3239 }
3240 
3241 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3242                             size_t alignment_hint) {
3243   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3244 }
3245 
3246 bool os::pd_release_memory(char* addr, size_t size) {
3247   return anon_munmap(addr, size);
3248 }
3249 
3250 static bool linux_mprotect(char* addr, size_t size, int prot) {
3251   // Linux wants the mprotect address argument to be page aligned.
3252   char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3253 
3254   // According to SUSv3, mprotect() should only be used with mappings
3255   // established by mmap(), and mmap() always maps whole pages. Unaligned
3256   // 'addr' likely indicates problem in the VM (e.g. trying to change
3257   // protection of malloc'ed or statically allocated memory). Check the
3258   // caller if you hit this assert.
3259   assert(addr == bottom, "sanity check");
3260 
3261   size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3262   return ::mprotect(bottom, size, prot) == 0;
3263 }
3264 
3265 // Set protections specified
3266 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3267                         bool is_committed) {
3268   unsigned int p = 0;
3269   switch (prot) {
3270   case MEM_PROT_NONE: p = PROT_NONE; break;
3271   case MEM_PROT_READ: p = PROT_READ; break;
3272   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3273   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3274   default:
3275     ShouldNotReachHere();
3276   }
3277   // is_committed is unused.
3278   return linux_mprotect(addr, bytes, p);
3279 }
3280 
3281 bool os::guard_memory(char* addr, size_t size) {
3282   return linux_mprotect(addr, size, PROT_NONE);
3283 }
3284 
3285 bool os::unguard_memory(char* addr, size_t size) {
3286   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3287 }
3288 
3289 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3290                                                     size_t page_size) {
3291   bool result = false;
3292   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3293                  MAP_ANONYMOUS|MAP_PRIVATE,
3294                  -1, 0);
3295   if (p != MAP_FAILED) {
3296     void *aligned_p = align_up(p, page_size);
3297 
3298     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3299 
3300     munmap(p, page_size * 2);
3301   }
3302 
3303   if (warn && !result) {
3304     warning("TransparentHugePages is not supported by the operating system.");
3305   }
3306 
3307   return result;
3308 }
3309 
3310 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3311   bool result = false;
3312   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3313                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3314                  -1, 0);
3315 
3316   if (p != MAP_FAILED) {
3317     // We don't know if this really is a huge page or not.
3318     FILE *fp = fopen("/proc/self/maps", "r");
3319     if (fp) {
3320       while (!feof(fp)) {
3321         char chars[257];
3322         long x = 0;
3323         if (fgets(chars, sizeof(chars), fp)) {
3324           if (sscanf(chars, "%lx-%*x", &x) == 1
3325               && x == (long)p) {
3326             if (strstr (chars, "hugepage")) {
3327               result = true;
3328               break;
3329             }
3330           }
3331         }
3332       }
3333       fclose(fp);
3334     }
3335     munmap(p, page_size);
3336   }
3337 
3338   if (warn && !result) {
3339     warning("HugeTLBFS is not supported by the operating system.");
3340   }
3341 
3342   return result;
3343 }
3344 
3345 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3346 //
3347 // From the coredump_filter documentation:
3348 //
3349 // - (bit 0) anonymous private memory
3350 // - (bit 1) anonymous shared memory
3351 // - (bit 2) file-backed private memory
3352 // - (bit 3) file-backed shared memory
3353 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3354 //           effective only if the bit 2 is cleared)
3355 // - (bit 5) hugetlb private memory
3356 // - (bit 6) hugetlb shared memory
3357 // - (bit 7) dax private memory
3358 // - (bit 8) dax shared memory
3359 //
3360 static void set_coredump_filter(bool largepages, bool dax_shared) {
3361   FILE *f;
3362   long cdm;
3363   bool filter_changed = false;
3364 
3365   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3366     return;
3367   }
3368 
3369   if (fscanf(f, "%lx", &cdm) != 1) {
3370     fclose(f);
3371     return;
3372   }
3373 
3374   rewind(f);
3375 
3376   if (largepages && (cdm & LARGEPAGES_BIT) == 0) {
3377     cdm |= LARGEPAGES_BIT;
3378     filter_changed = true;
3379   }
3380   if (dax_shared && (cdm & DAX_SHARED_BIT) == 0) {
3381     cdm |= DAX_SHARED_BIT;
3382     filter_changed = true;
3383   }
3384   if (filter_changed) {
3385     fprintf(f, "%#lx", cdm);
3386   }
3387 
3388   fclose(f);
3389 }
3390 
3391 // Large page support
3392 
3393 static size_t _large_page_size = 0;
3394 
3395 size_t os::Linux::find_large_page_size() {
3396   size_t large_page_size = 0;
3397 
3398   // large_page_size on Linux is used to round up heap size. x86 uses either
3399   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3400   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3401   // page as large as 256M.
3402   //
3403   // Here we try to figure out page size by parsing /proc/meminfo and looking
3404   // for a line with the following format:
3405   //    Hugepagesize:     2048 kB
3406   //
3407   // If we can't determine the value (e.g. /proc is not mounted, or the text
3408   // format has been changed), we'll use the largest page size supported by
3409   // the processor.
3410 
3411 #ifndef ZERO
3412   large_page_size =
3413     AARCH64_ONLY(2 * M)
3414     AMD64_ONLY(2 * M)
3415     ARM32_ONLY(2 * M)
3416     IA32_ONLY(4 * M)
3417     IA64_ONLY(256 * M)
3418     PPC_ONLY(4 * M)
3419     S390_ONLY(1 * M)
3420     SPARC_ONLY(4 * M);
3421 #endif // ZERO
3422 
3423   FILE *fp = fopen("/proc/meminfo", "r");
3424   if (fp) {
3425     while (!feof(fp)) {
3426       int x = 0;
3427       char buf[16];
3428       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3429         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3430           large_page_size = x * K;
3431           break;
3432         }
3433       } else {
3434         // skip to next line
3435         for (;;) {
3436           int ch = fgetc(fp);
3437           if (ch == EOF || ch == (int)'\n') break;
3438         }
3439       }
3440     }
3441     fclose(fp);
3442   }
3443 
3444   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3445     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3446             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3447             proper_unit_for_byte_size(large_page_size));
3448   }
3449 
3450   return large_page_size;
3451 }
3452 
3453 size_t os::Linux::setup_large_page_size() {
3454   _large_page_size = Linux::find_large_page_size();
3455   const size_t default_page_size = (size_t)Linux::page_size();
3456   if (_large_page_size > default_page_size) {
3457     _page_sizes[0] = _large_page_size;
3458     _page_sizes[1] = default_page_size;
3459     _page_sizes[2] = 0;
3460   }
3461 
3462   return _large_page_size;
3463 }
3464 
3465 bool os::Linux::setup_large_page_type(size_t page_size) {
3466   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3467       FLAG_IS_DEFAULT(UseSHM) &&
3468       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3469 
3470     // The type of large pages has not been specified by the user.
3471 
3472     // Try UseHugeTLBFS and then UseSHM.
3473     UseHugeTLBFS = UseSHM = true;
3474 
3475     // Don't try UseTransparentHugePages since there are known
3476     // performance issues with it turned on. This might change in the future.
3477     UseTransparentHugePages = false;
3478   }
3479 
3480   if (UseTransparentHugePages) {
3481     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3482     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3483       UseHugeTLBFS = false;
3484       UseSHM = false;
3485       return true;
3486     }
3487     UseTransparentHugePages = false;
3488   }
3489 
3490   if (UseHugeTLBFS) {
3491     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3492     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3493       UseSHM = false;
3494       return true;
3495     }
3496     UseHugeTLBFS = false;
3497   }
3498 
3499   return UseSHM;
3500 }
3501 
3502 void os::large_page_init() {
3503   if (!UseLargePages &&
3504       !UseTransparentHugePages &&
3505       !UseHugeTLBFS &&
3506       !UseSHM) {
3507     // Not using large pages.
3508     return;
3509   }
3510 
3511   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3512     // The user explicitly turned off large pages.
3513     // Ignore the rest of the large pages flags.
3514     UseTransparentHugePages = false;
3515     UseHugeTLBFS = false;
3516     UseSHM = false;
3517     return;
3518   }
3519 
3520   size_t large_page_size = Linux::setup_large_page_size();
3521   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3522 
3523   set_coredump_filter(true /*largepages*/, false /*dax_shared*/);
3524 }
3525 
3526 #ifndef SHM_HUGETLB
3527   #define SHM_HUGETLB 04000
3528 #endif
3529 
3530 #define shm_warning_format(format, ...)              \
3531   do {                                               \
3532     if (UseLargePages &&                             \
3533         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3534          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3535          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3536       warning(format, __VA_ARGS__);                  \
3537     }                                                \
3538   } while (0)
3539 
3540 #define shm_warning(str) shm_warning_format("%s", str)
3541 
3542 #define shm_warning_with_errno(str)                \
3543   do {                                             \
3544     int err = errno;                               \
3545     shm_warning_format(str " (error = %d)", err);  \
3546   } while (0)
3547 
3548 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3549   assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3550 
3551   if (!is_aligned(alignment, SHMLBA)) {
3552     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3553     return NULL;
3554   }
3555 
3556   // To ensure that we get 'alignment' aligned memory from shmat,
3557   // we pre-reserve aligned virtual memory and then attach to that.
3558 
3559   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3560   if (pre_reserved_addr == NULL) {
3561     // Couldn't pre-reserve aligned memory.
3562     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3563     return NULL;
3564   }
3565 
3566   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3567   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3568 
3569   if ((intptr_t)addr == -1) {
3570     int err = errno;
3571     shm_warning_with_errno("Failed to attach shared memory.");
3572 
3573     assert(err != EACCES, "Unexpected error");
3574     assert(err != EIDRM,  "Unexpected error");
3575     assert(err != EINVAL, "Unexpected error");
3576 
3577     // Since we don't know if the kernel unmapped the pre-reserved memory area
3578     // we can't unmap it, since that would potentially unmap memory that was
3579     // mapped from other threads.
3580     return NULL;
3581   }
3582 
3583   return addr;
3584 }
3585 
3586 static char* shmat_at_address(int shmid, char* req_addr) {
3587   if (!is_aligned(req_addr, SHMLBA)) {
3588     assert(false, "Requested address needs to be SHMLBA aligned");
3589     return NULL;
3590   }
3591 
3592   char* addr = (char*)shmat(shmid, req_addr, 0);
3593 
3594   if ((intptr_t)addr == -1) {
3595     shm_warning_with_errno("Failed to attach shared memory.");
3596     return NULL;
3597   }
3598 
3599   return addr;
3600 }
3601 
3602 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3603   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3604   if (req_addr != NULL) {
3605     assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3606     assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3607     return shmat_at_address(shmid, req_addr);
3608   }
3609 
3610   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3611   // return large page size aligned memory addresses when req_addr == NULL.
3612   // However, if the alignment is larger than the large page size, we have
3613   // to manually ensure that the memory returned is 'alignment' aligned.
3614   if (alignment > os::large_page_size()) {
3615     assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3616     return shmat_with_alignment(shmid, bytes, alignment);
3617   } else {
3618     return shmat_at_address(shmid, NULL);
3619   }
3620 }
3621 
3622 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3623                                             char* req_addr, bool exec) {
3624   // "exec" is passed in but not used.  Creating the shared image for
3625   // the code cache doesn't have an SHM_X executable permission to check.
3626   assert(UseLargePages && UseSHM, "only for SHM large pages");
3627   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3628   assert(is_aligned(req_addr, alignment), "Unaligned address");
3629 
3630   if (!is_aligned(bytes, os::large_page_size())) {
3631     return NULL; // Fallback to small pages.
3632   }
3633 
3634   // Create a large shared memory region to attach to based on size.
3635   // Currently, size is the total size of the heap.
3636   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3637   if (shmid == -1) {
3638     // Possible reasons for shmget failure:
3639     // 1. shmmax is too small for Java heap.
3640     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3641     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3642     // 2. not enough large page memory.
3643     //    > check available large pages: cat /proc/meminfo
3644     //    > increase amount of large pages:
3645     //          echo new_value > /proc/sys/vm/nr_hugepages
3646     //      Note 1: different Linux may use different name for this property,
3647     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3648     //      Note 2: it's possible there's enough physical memory available but
3649     //            they are so fragmented after a long run that they can't
3650     //            coalesce into large pages. Try to reserve large pages when
3651     //            the system is still "fresh".
3652     shm_warning_with_errno("Failed to reserve shared memory.");
3653     return NULL;
3654   }
3655 
3656   // Attach to the region.
3657   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3658 
3659   // Remove shmid. If shmat() is successful, the actual shared memory segment
3660   // will be deleted when it's detached by shmdt() or when the process
3661   // terminates. If shmat() is not successful this will remove the shared
3662   // segment immediately.
3663   shmctl(shmid, IPC_RMID, NULL);
3664 
3665   return addr;
3666 }
3667 
3668 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3669                                         int error) {
3670   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3671 
3672   bool warn_on_failure = UseLargePages &&
3673       (!FLAG_IS_DEFAULT(UseLargePages) ||
3674        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3675        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3676 
3677   if (warn_on_failure) {
3678     char msg[128];
3679     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3680                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3681     warning("%s", msg);
3682   }
3683 }
3684 
3685 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3686                                                         char* req_addr,
3687                                                         bool exec) {
3688   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3689   assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3690   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3691 
3692   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3693   char* addr = (char*)::mmap(req_addr, bytes, prot,
3694                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3695                              -1, 0);
3696 
3697   if (addr == MAP_FAILED) {
3698     warn_on_large_pages_failure(req_addr, bytes, errno);
3699     return NULL;
3700   }
3701 
3702   assert(is_aligned(addr, os::large_page_size()), "Must be");
3703 
3704   return addr;
3705 }
3706 
3707 // Reserve memory using mmap(MAP_HUGETLB).
3708 //  - bytes shall be a multiple of alignment.
3709 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3710 //  - alignment sets the alignment at which memory shall be allocated.
3711 //     It must be a multiple of allocation granularity.
3712 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3713 //  req_addr or NULL.
3714 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3715                                                          size_t alignment,
3716                                                          char* req_addr,
3717                                                          bool exec) {
3718   size_t large_page_size = os::large_page_size();
3719   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3720 
3721   assert(is_aligned(req_addr, alignment), "Must be");
3722   assert(is_aligned(bytes, alignment), "Must be");
3723 
3724   // First reserve - but not commit - the address range in small pages.
3725   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3726 
3727   if (start == NULL) {
3728     return NULL;
3729   }
3730 
3731   assert(is_aligned(start, alignment), "Must be");
3732 
3733   char* end = start + bytes;
3734 
3735   // Find the regions of the allocated chunk that can be promoted to large pages.
3736   char* lp_start = align_up(start, large_page_size);
3737   char* lp_end   = align_down(end, large_page_size);
3738 
3739   size_t lp_bytes = lp_end - lp_start;
3740 
3741   assert(is_aligned(lp_bytes, large_page_size), "Must be");
3742 
3743   if (lp_bytes == 0) {
3744     // The mapped region doesn't even span the start and the end of a large page.
3745     // Fall back to allocate a non-special area.
3746     ::munmap(start, end - start);
3747     return NULL;
3748   }
3749 
3750   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3751 
3752   void* result;
3753 
3754   // Commit small-paged leading area.
3755   if (start != lp_start) {
3756     result = ::mmap(start, lp_start - start, prot,
3757                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3758                     -1, 0);
3759     if (result == MAP_FAILED) {
3760       ::munmap(lp_start, end - lp_start);
3761       return NULL;
3762     }
3763   }
3764 
3765   // Commit large-paged area.
3766   result = ::mmap(lp_start, lp_bytes, prot,
3767                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3768                   -1, 0);
3769   if (result == MAP_FAILED) {
3770     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3771     // If the mmap above fails, the large pages region will be unmapped and we
3772     // have regions before and after with small pages. Release these regions.
3773     //
3774     // |  mapped  |  unmapped  |  mapped  |
3775     // ^          ^            ^          ^
3776     // start      lp_start     lp_end     end
3777     //
3778     ::munmap(start, lp_start - start);
3779     ::munmap(lp_end, end - lp_end);
3780     return NULL;
3781   }
3782 
3783   // Commit small-paged trailing area.
3784   if (lp_end != end) {
3785     result = ::mmap(lp_end, end - lp_end, prot,
3786                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3787                     -1, 0);
3788     if (result == MAP_FAILED) {
3789       ::munmap(start, lp_end - start);
3790       return NULL;
3791     }
3792   }
3793 
3794   return start;
3795 }
3796 
3797 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3798                                                    size_t alignment,
3799                                                    char* req_addr,
3800                                                    bool exec) {
3801   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3802   assert(is_aligned(req_addr, alignment), "Must be");
3803   assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3804   assert(is_power_of_2(os::large_page_size()), "Must be");
3805   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3806 
3807   if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3808     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3809   } else {
3810     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3811   }
3812 }
3813 
3814 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3815                                  char* req_addr, bool exec) {
3816   assert(UseLargePages, "only for large pages");
3817 
3818   char* addr;
3819   if (UseSHM) {
3820     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3821   } else {
3822     assert(UseHugeTLBFS, "must be");
3823     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3824   }
3825 
3826   if (addr != NULL) {
3827     if (UseNUMAInterleaving) {
3828       numa_make_global(addr, bytes);
3829     }
3830 
3831     // The memory is committed
3832     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3833   }
3834 
3835   return addr;
3836 }
3837 
3838 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3839   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3840   return shmdt(base) == 0;
3841 }
3842 
3843 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3844   return pd_release_memory(base, bytes);
3845 }
3846 
3847 bool os::release_memory_special(char* base, size_t bytes) {
3848   bool res;
3849   if (MemTracker::tracking_level() > NMT_minimal) {
3850     Tracker tkr(Tracker::release);
3851     res = os::Linux::release_memory_special_impl(base, bytes);
3852     if (res) {
3853       tkr.record((address)base, bytes);
3854     }
3855 
3856   } else {
3857     res = os::Linux::release_memory_special_impl(base, bytes);
3858   }
3859   return res;
3860 }
3861 
3862 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3863   assert(UseLargePages, "only for large pages");
3864   bool res;
3865 
3866   if (UseSHM) {
3867     res = os::Linux::release_memory_special_shm(base, bytes);
3868   } else {
3869     assert(UseHugeTLBFS, "must be");
3870     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3871   }
3872   return res;
3873 }
3874 
3875 size_t os::large_page_size() {
3876   return _large_page_size;
3877 }
3878 
3879 // With SysV SHM the entire memory region must be allocated as shared
3880 // memory.
3881 // HugeTLBFS allows application to commit large page memory on demand.
3882 // However, when committing memory with HugeTLBFS fails, the region
3883 // that was supposed to be committed will lose the old reservation
3884 // and allow other threads to steal that memory region. Because of this
3885 // behavior we can't commit HugeTLBFS memory.
3886 bool os::can_commit_large_page_memory() {
3887   return UseTransparentHugePages;
3888 }
3889 
3890 bool os::can_execute_large_page_memory() {
3891   return UseTransparentHugePages || UseHugeTLBFS;
3892 }
3893 
3894 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3895   assert(file_desc >= 0, "file_desc is not valid");
3896   char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3897   if (result != NULL) {
3898     if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3899       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3900     }
3901   }
3902   return result;
3903 }
3904 
3905 // Reserve memory at an arbitrary address, only if that area is
3906 // available (and not reserved for something else).
3907 
3908 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3909   const int max_tries = 10;
3910   char* base[max_tries];
3911   size_t size[max_tries];
3912   const size_t gap = 0x000000;
3913 
3914   // Assert only that the size is a multiple of the page size, since
3915   // that's all that mmap requires, and since that's all we really know
3916   // about at this low abstraction level.  If we need higher alignment,
3917   // we can either pass an alignment to this method or verify alignment
3918   // in one of the methods further up the call chain.  See bug 5044738.
3919   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3920 
3921   // Repeatedly allocate blocks until the block is allocated at the
3922   // right spot.
3923 
3924   // Linux mmap allows caller to pass an address as hint; give it a try first,
3925   // if kernel honors the hint then we can return immediately.
3926   char * addr = anon_mmap(requested_addr, bytes, false);
3927   if (addr == requested_addr) {
3928     return requested_addr;
3929   }
3930 
3931   if (addr != NULL) {
3932     // mmap() is successful but it fails to reserve at the requested address
3933     anon_munmap(addr, bytes);
3934   }
3935 
3936   int i;
3937   for (i = 0; i < max_tries; ++i) {
3938     base[i] = reserve_memory(bytes);
3939 
3940     if (base[i] != NULL) {
3941       // Is this the block we wanted?
3942       if (base[i] == requested_addr) {
3943         size[i] = bytes;
3944         break;
3945       }
3946 
3947       // Does this overlap the block we wanted? Give back the overlapped
3948       // parts and try again.
3949 
3950       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3951       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3952         unmap_memory(base[i], top_overlap);
3953         base[i] += top_overlap;
3954         size[i] = bytes - top_overlap;
3955       } else {
3956         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3957         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3958           unmap_memory(requested_addr, bottom_overlap);
3959           size[i] = bytes - bottom_overlap;
3960         } else {
3961           size[i] = bytes;
3962         }
3963       }
3964     }
3965   }
3966 
3967   // Give back the unused reserved pieces.
3968 
3969   for (int j = 0; j < i; ++j) {
3970     if (base[j] != NULL) {
3971       unmap_memory(base[j], size[j]);
3972     }
3973   }
3974 
3975   if (i < max_tries) {
3976     return requested_addr;
3977   } else {
3978     return NULL;
3979   }
3980 }
3981 
3982 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3983   return ::read(fd, buf, nBytes);
3984 }
3985 
3986 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3987   return ::pread(fd, buf, nBytes, offset);
3988 }
3989 
3990 // Short sleep, direct OS call.
3991 //
3992 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3993 // sched_yield(2) will actually give up the CPU:
3994 //
3995 //   * Alone on this pariticular CPU, keeps running.
3996 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3997 //     (pre 2.6.39).
3998 //
3999 // So calling this with 0 is an alternative.
4000 //
4001 void os::naked_short_sleep(jlong ms) {
4002   struct timespec req;
4003 
4004   assert(ms < 1000, "Un-interruptable sleep, short time use only");
4005   req.tv_sec = 0;
4006   if (ms > 0) {
4007     req.tv_nsec = (ms % 1000) * 1000000;
4008   } else {
4009     req.tv_nsec = 1;
4010   }
4011 
4012   nanosleep(&req, NULL);
4013 
4014   return;
4015 }
4016 
4017 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4018 void os::infinite_sleep() {
4019   while (true) {    // sleep forever ...
4020     ::sleep(100);   // ... 100 seconds at a time
4021   }
4022 }
4023 
4024 // Used to convert frequent JVM_Yield() to nops
4025 bool os::dont_yield() {
4026   return DontYieldALot;
4027 }
4028 
4029 void os::naked_yield() {
4030   sched_yield();
4031 }
4032 
4033 ////////////////////////////////////////////////////////////////////////////////
4034 // thread priority support
4035 
4036 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4037 // only supports dynamic priority, static priority must be zero. For real-time
4038 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4039 // However, for large multi-threaded applications, SCHED_RR is not only slower
4040 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4041 // of 5 runs - Sep 2005).
4042 //
4043 // The following code actually changes the niceness of kernel-thread/LWP. It
4044 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4045 // not the entire user process, and user level threads are 1:1 mapped to kernel
4046 // threads. It has always been the case, but could change in the future. For
4047 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4048 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4049 
4050 int os::java_to_os_priority[CriticalPriority + 1] = {
4051   19,              // 0 Entry should never be used
4052 
4053    4,              // 1 MinPriority
4054    3,              // 2
4055    2,              // 3
4056 
4057    1,              // 4
4058    0,              // 5 NormPriority
4059   -1,              // 6
4060 
4061   -2,              // 7
4062   -3,              // 8
4063   -4,              // 9 NearMaxPriority
4064 
4065   -5,              // 10 MaxPriority
4066 
4067   -5               // 11 CriticalPriority
4068 };
4069 
4070 static int prio_init() {
4071   if (ThreadPriorityPolicy == 1) {
4072     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4073     // if effective uid is not root. Perhaps, a more elegant way of doing
4074     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4075     if (geteuid() != 0) {
4076       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4077         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4078       }
4079       ThreadPriorityPolicy = 0;
4080     }
4081   }
4082   if (UseCriticalJavaThreadPriority) {
4083     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4084   }
4085   return 0;
4086 }
4087 
4088 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4089   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4090 
4091   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4092   return (ret == 0) ? OS_OK : OS_ERR;
4093 }
4094 
4095 OSReturn os::get_native_priority(const Thread* const thread,
4096                                  int *priority_ptr) {
4097   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4098     *priority_ptr = java_to_os_priority[NormPriority];
4099     return OS_OK;
4100   }
4101 
4102   errno = 0;
4103   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4104   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4105 }
4106 
4107 // Hint to the underlying OS that a task switch would not be good.
4108 // Void return because it's a hint and can fail.
4109 void os::hint_no_preempt() {}
4110 
4111 ////////////////////////////////////////////////////////////////////////////////
4112 // suspend/resume support
4113 
4114 //  The low-level signal-based suspend/resume support is a remnant from the
4115 //  old VM-suspension that used to be for java-suspension, safepoints etc,
4116 //  within hotspot. Currently used by JFR's OSThreadSampler
4117 //
4118 //  The remaining code is greatly simplified from the more general suspension
4119 //  code that used to be used.
4120 //
4121 //  The protocol is quite simple:
4122 //  - suspend:
4123 //      - sends a signal to the target thread
4124 //      - polls the suspend state of the osthread using a yield loop
4125 //      - target thread signal handler (SR_handler) sets suspend state
4126 //        and blocks in sigsuspend until continued
4127 //  - resume:
4128 //      - sets target osthread state to continue
4129 //      - sends signal to end the sigsuspend loop in the SR_handler
4130 //
4131 //  Note that the SR_lock plays no role in this suspend/resume protocol,
4132 //  but is checked for NULL in SR_handler as a thread termination indicator.
4133 //  The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4134 //
4135 //  Note that resume_clear_context() and suspend_save_context() are needed
4136 //  by SR_handler(), so that fetch_frame_from_ucontext() works,
4137 //  which in part is used by:
4138 //    - Forte Analyzer: AsyncGetCallTrace()
4139 //    - StackBanging: get_frame_at_stack_banging_point()
4140 
4141 static void resume_clear_context(OSThread *osthread) {
4142   osthread->set_ucontext(NULL);
4143   osthread->set_siginfo(NULL);
4144 }
4145 
4146 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4147                                  ucontext_t* context) {
4148   osthread->set_ucontext(context);
4149   osthread->set_siginfo(siginfo);
4150 }
4151 
4152 // Handler function invoked when a thread's execution is suspended or
4153 // resumed. We have to be careful that only async-safe functions are
4154 // called here (Note: most pthread functions are not async safe and
4155 // should be avoided.)
4156 //
4157 // Note: sigwait() is a more natural fit than sigsuspend() from an
4158 // interface point of view, but sigwait() prevents the signal hander
4159 // from being run. libpthread would get very confused by not having
4160 // its signal handlers run and prevents sigwait()'s use with the
4161 // mutex granting granting signal.
4162 //
4163 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4164 //
4165 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4166   // Save and restore errno to avoid confusing native code with EINTR
4167   // after sigsuspend.
4168   int old_errno = errno;
4169 
4170   Thread* thread = Thread::current_or_null_safe();
4171   assert(thread != NULL, "Missing current thread in SR_handler");
4172 
4173   // On some systems we have seen signal delivery get "stuck" until the signal
4174   // mask is changed as part of thread termination. Check that the current thread
4175   // has not already terminated (via SR_lock()) - else the following assertion
4176   // will fail because the thread is no longer a JavaThread as the ~JavaThread
4177   // destructor has completed.
4178 
4179   if (thread->SR_lock() == NULL) {
4180     return;
4181   }
4182 
4183   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4184 
4185   OSThread* osthread = thread->osthread();
4186 
4187   os::SuspendResume::State current = osthread->sr.state();
4188   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4189     suspend_save_context(osthread, siginfo, context);
4190 
4191     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4192     os::SuspendResume::State state = osthread->sr.suspended();
4193     if (state == os::SuspendResume::SR_SUSPENDED) {
4194       sigset_t suspend_set;  // signals for sigsuspend()
4195       sigemptyset(&suspend_set);
4196       // get current set of blocked signals and unblock resume signal
4197       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4198       sigdelset(&suspend_set, SR_signum);
4199 
4200       sr_semaphore.signal();
4201       // wait here until we are resumed
4202       while (1) {
4203         sigsuspend(&suspend_set);
4204 
4205         os::SuspendResume::State result = osthread->sr.running();
4206         if (result == os::SuspendResume::SR_RUNNING) {
4207           sr_semaphore.signal();
4208           break;
4209         }
4210       }
4211 
4212     } else if (state == os::SuspendResume::SR_RUNNING) {
4213       // request was cancelled, continue
4214     } else {
4215       ShouldNotReachHere();
4216     }
4217 
4218     resume_clear_context(osthread);
4219   } else if (current == os::SuspendResume::SR_RUNNING) {
4220     // request was cancelled, continue
4221   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4222     // ignore
4223   } else {
4224     // ignore
4225   }
4226 
4227   errno = old_errno;
4228 }
4229 
4230 static int SR_initialize() {
4231   struct sigaction act;
4232   char *s;
4233 
4234   // Get signal number to use for suspend/resume
4235   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4236     int sig = ::strtol(s, 0, 10);
4237     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4238         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4239       SR_signum = sig;
4240     } else {
4241       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4242               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4243     }
4244   }
4245 
4246   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4247          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4248 
4249   sigemptyset(&SR_sigset);
4250   sigaddset(&SR_sigset, SR_signum);
4251 
4252   // Set up signal handler for suspend/resume
4253   act.sa_flags = SA_RESTART|SA_SIGINFO;
4254   act.sa_handler = (void (*)(int)) SR_handler;
4255 
4256   // SR_signum is blocked by default.
4257   // 4528190 - We also need to block pthread restart signal (32 on all
4258   // supported Linux platforms). Note that LinuxThreads need to block
4259   // this signal for all threads to work properly. So we don't have
4260   // to use hard-coded signal number when setting up the mask.
4261   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4262 
4263   if (sigaction(SR_signum, &act, 0) == -1) {
4264     return -1;
4265   }
4266 
4267   // Save signal flag
4268   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4269   return 0;
4270 }
4271 
4272 static int sr_notify(OSThread* osthread) {
4273   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4274   assert_status(status == 0, status, "pthread_kill");
4275   return status;
4276 }
4277 
4278 // "Randomly" selected value for how long we want to spin
4279 // before bailing out on suspending a thread, also how often
4280 // we send a signal to a thread we want to resume
4281 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4282 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4283 
4284 // returns true on success and false on error - really an error is fatal
4285 // but this seems the normal response to library errors
4286 static bool do_suspend(OSThread* osthread) {
4287   assert(osthread->sr.is_running(), "thread should be running");
4288   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4289 
4290   // mark as suspended and send signal
4291   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4292     // failed to switch, state wasn't running?
4293     ShouldNotReachHere();
4294     return false;
4295   }
4296 
4297   if (sr_notify(osthread) != 0) {
4298     ShouldNotReachHere();
4299   }
4300 
4301   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4302   while (true) {
4303     if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4304       break;
4305     } else {
4306       // timeout
4307       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4308       if (cancelled == os::SuspendResume::SR_RUNNING) {
4309         return false;
4310       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4311         // make sure that we consume the signal on the semaphore as well
4312         sr_semaphore.wait();
4313         break;
4314       } else {
4315         ShouldNotReachHere();
4316         return false;
4317       }
4318     }
4319   }
4320 
4321   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4322   return true;
4323 }
4324 
4325 static void do_resume(OSThread* osthread) {
4326   assert(osthread->sr.is_suspended(), "thread should be suspended");
4327   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4328 
4329   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4330     // failed to switch to WAKEUP_REQUEST
4331     ShouldNotReachHere();
4332     return;
4333   }
4334 
4335   while (true) {
4336     if (sr_notify(osthread) == 0) {
4337       if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4338         if (osthread->sr.is_running()) {
4339           return;
4340         }
4341       }
4342     } else {
4343       ShouldNotReachHere();
4344     }
4345   }
4346 
4347   guarantee(osthread->sr.is_running(), "Must be running!");
4348 }
4349 
4350 ///////////////////////////////////////////////////////////////////////////////////
4351 // signal handling (except suspend/resume)
4352 
4353 // This routine may be used by user applications as a "hook" to catch signals.
4354 // The user-defined signal handler must pass unrecognized signals to this
4355 // routine, and if it returns true (non-zero), then the signal handler must
4356 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4357 // routine will never retun false (zero), but instead will execute a VM panic
4358 // routine kill the process.
4359 //
4360 // If this routine returns false, it is OK to call it again.  This allows
4361 // the user-defined signal handler to perform checks either before or after
4362 // the VM performs its own checks.  Naturally, the user code would be making
4363 // a serious error if it tried to handle an exception (such as a null check
4364 // or breakpoint) that the VM was generating for its own correct operation.
4365 //
4366 // This routine may recognize any of the following kinds of signals:
4367 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4368 // It should be consulted by handlers for any of those signals.
4369 //
4370 // The caller of this routine must pass in the three arguments supplied
4371 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4372 // field of the structure passed to sigaction().  This routine assumes that
4373 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4374 //
4375 // Note that the VM will print warnings if it detects conflicting signal
4376 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4377 //
4378 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4379                                                  siginfo_t* siginfo,
4380                                                  void* ucontext,
4381                                                  int abort_if_unrecognized);
4382 
4383 void signalHandler(int sig, siginfo_t* info, void* uc) {
4384   assert(info != NULL && uc != NULL, "it must be old kernel");
4385   int orig_errno = errno;  // Preserve errno value over signal handler.
4386   JVM_handle_linux_signal(sig, info, uc, true);
4387   errno = orig_errno;
4388 }
4389 
4390 
4391 // This boolean allows users to forward their own non-matching signals
4392 // to JVM_handle_linux_signal, harmlessly.
4393 bool os::Linux::signal_handlers_are_installed = false;
4394 
4395 // For signal-chaining
4396 struct sigaction sigact[NSIG];
4397 uint64_t sigs = 0;
4398 #if (64 < NSIG-1)
4399 #error "Not all signals can be encoded in sigs. Adapt its type!"
4400 #endif
4401 bool os::Linux::libjsig_is_loaded = false;
4402 typedef struct sigaction *(*get_signal_t)(int);
4403 get_signal_t os::Linux::get_signal_action = NULL;
4404 
4405 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4406   struct sigaction *actp = NULL;
4407 
4408   if (libjsig_is_loaded) {
4409     // Retrieve the old signal handler from libjsig
4410     actp = (*get_signal_action)(sig);
4411   }
4412   if (actp == NULL) {
4413     // Retrieve the preinstalled signal handler from jvm
4414     actp = get_preinstalled_handler(sig);
4415   }
4416 
4417   return actp;
4418 }
4419 
4420 static bool call_chained_handler(struct sigaction *actp, int sig,
4421                                  siginfo_t *siginfo, void *context) {
4422   // Call the old signal handler
4423   if (actp->sa_handler == SIG_DFL) {
4424     // It's more reasonable to let jvm treat it as an unexpected exception
4425     // instead of taking the default action.
4426     return false;
4427   } else if (actp->sa_handler != SIG_IGN) {
4428     if ((actp->sa_flags & SA_NODEFER) == 0) {
4429       // automaticlly block the signal
4430       sigaddset(&(actp->sa_mask), sig);
4431     }
4432 
4433     sa_handler_t hand = NULL;
4434     sa_sigaction_t sa = NULL;
4435     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4436     // retrieve the chained handler
4437     if (siginfo_flag_set) {
4438       sa = actp->sa_sigaction;
4439     } else {
4440       hand = actp->sa_handler;
4441     }
4442 
4443     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4444       actp->sa_handler = SIG_DFL;
4445     }
4446 
4447     // try to honor the signal mask
4448     sigset_t oset;
4449     sigemptyset(&oset);
4450     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4451 
4452     // call into the chained handler
4453     if (siginfo_flag_set) {
4454       (*sa)(sig, siginfo, context);
4455     } else {
4456       (*hand)(sig);
4457     }
4458 
4459     // restore the signal mask
4460     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4461   }
4462   // Tell jvm's signal handler the signal is taken care of.
4463   return true;
4464 }
4465 
4466 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4467   bool chained = false;
4468   // signal-chaining
4469   if (UseSignalChaining) {
4470     struct sigaction *actp = get_chained_signal_action(sig);
4471     if (actp != NULL) {
4472       chained = call_chained_handler(actp, sig, siginfo, context);
4473     }
4474   }
4475   return chained;
4476 }
4477 
4478 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4479   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4480     return &sigact[sig];
4481   }
4482   return NULL;
4483 }
4484 
4485 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4486   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4487   sigact[sig] = oldAct;
4488   sigs |= (uint64_t)1 << (sig-1);
4489 }
4490 
4491 // for diagnostic
4492 int sigflags[NSIG];
4493 
4494 int os::Linux::get_our_sigflags(int sig) {
4495   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4496   return sigflags[sig];
4497 }
4498 
4499 void os::Linux::set_our_sigflags(int sig, int flags) {
4500   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4501   if (sig > 0 && sig < NSIG) {
4502     sigflags[sig] = flags;
4503   }
4504 }
4505 
4506 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4507   // Check for overwrite.
4508   struct sigaction oldAct;
4509   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4510 
4511   void* oldhand = oldAct.sa_sigaction
4512                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4513                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4514   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4515       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4516       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4517     if (AllowUserSignalHandlers || !set_installed) {
4518       // Do not overwrite; user takes responsibility to forward to us.
4519       return;
4520     } else if (UseSignalChaining) {
4521       // save the old handler in jvm
4522       save_preinstalled_handler(sig, oldAct);
4523       // libjsig also interposes the sigaction() call below and saves the
4524       // old sigaction on it own.
4525     } else {
4526       fatal("Encountered unexpected pre-existing sigaction handler "
4527             "%#lx for signal %d.", (long)oldhand, sig);
4528     }
4529   }
4530 
4531   struct sigaction sigAct;
4532   sigfillset(&(sigAct.sa_mask));
4533   sigAct.sa_handler = SIG_DFL;
4534   if (!set_installed) {
4535     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4536   } else {
4537     sigAct.sa_sigaction = signalHandler;
4538     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4539   }
4540   // Save flags, which are set by ours
4541   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4542   sigflags[sig] = sigAct.sa_flags;
4543 
4544   int ret = sigaction(sig, &sigAct, &oldAct);
4545   assert(ret == 0, "check");
4546 
4547   void* oldhand2  = oldAct.sa_sigaction
4548                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4549                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4550   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4551 }
4552 
4553 // install signal handlers for signals that HotSpot needs to
4554 // handle in order to support Java-level exception handling.
4555 
4556 void os::Linux::install_signal_handlers() {
4557   if (!signal_handlers_are_installed) {
4558     signal_handlers_are_installed = true;
4559 
4560     // signal-chaining
4561     typedef void (*signal_setting_t)();
4562     signal_setting_t begin_signal_setting = NULL;
4563     signal_setting_t end_signal_setting = NULL;
4564     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4565                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4566     if (begin_signal_setting != NULL) {
4567       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4568                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4569       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4570                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4571       libjsig_is_loaded = true;
4572       assert(UseSignalChaining, "should enable signal-chaining");
4573     }
4574     if (libjsig_is_loaded) {
4575       // Tell libjsig jvm is setting signal handlers
4576       (*begin_signal_setting)();
4577     }
4578 
4579     set_signal_handler(SIGSEGV, true);
4580     set_signal_handler(SIGPIPE, true);
4581     set_signal_handler(SIGBUS, true);
4582     set_signal_handler(SIGILL, true);
4583     set_signal_handler(SIGFPE, true);
4584 #if defined(PPC64)
4585     set_signal_handler(SIGTRAP, true);
4586 #endif
4587     set_signal_handler(SIGXFSZ, true);
4588 
4589     if (libjsig_is_loaded) {
4590       // Tell libjsig jvm finishes setting signal handlers
4591       (*end_signal_setting)();
4592     }
4593 
4594     // We don't activate signal checker if libjsig is in place, we trust ourselves
4595     // and if UserSignalHandler is installed all bets are off.
4596     // Log that signal checking is off only if -verbose:jni is specified.
4597     if (CheckJNICalls) {
4598       if (libjsig_is_loaded) {
4599         if (PrintJNIResolving) {
4600           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4601         }
4602         check_signals = false;
4603       }
4604       if (AllowUserSignalHandlers) {
4605         if (PrintJNIResolving) {
4606           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4607         }
4608         check_signals = false;
4609       }
4610     }
4611   }
4612 }
4613 
4614 // This is the fastest way to get thread cpu time on Linux.
4615 // Returns cpu time (user+sys) for any thread, not only for current.
4616 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4617 // It might work on 2.6.10+ with a special kernel/glibc patch.
4618 // For reference, please, see IEEE Std 1003.1-2004:
4619 //   http://www.unix.org/single_unix_specification
4620 
4621 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4622   struct timespec tp;
4623   int rc = os::Linux::clock_gettime(clockid, &tp);
4624   assert(rc == 0, "clock_gettime is expected to return 0 code");
4625 
4626   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4627 }
4628 
4629 void os::Linux::initialize_os_info() {
4630   assert(_os_version == 0, "OS info already initialized");
4631 
4632   struct utsname _uname;
4633 
4634   uint32_t major;
4635   uint32_t minor;
4636   uint32_t fix;
4637 
4638   int rc;
4639 
4640   // Kernel version is unknown if
4641   // verification below fails.
4642   _os_version = 0x01000000;
4643 
4644   rc = uname(&_uname);
4645   if (rc != -1) {
4646 
4647     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4648     if (rc == 3) {
4649 
4650       if (major < 256 && minor < 256 && fix < 256) {
4651         // Kernel version format is as expected,
4652         // set it overriding unknown state.
4653         _os_version = (major << 16) |
4654                       (minor << 8 ) |
4655                       (fix   << 0 ) ;
4656       }
4657     }
4658   }
4659 }
4660 
4661 uint32_t os::Linux::os_version() {
4662   assert(_os_version != 0, "not initialized");
4663   return _os_version & 0x00FFFFFF;
4664 }
4665 
4666 bool os::Linux::os_version_is_known() {
4667   assert(_os_version != 0, "not initialized");
4668   return _os_version & 0x01000000 ? false : true;
4669 }
4670 
4671 /////
4672 // glibc on Linux platform uses non-documented flag
4673 // to indicate, that some special sort of signal
4674 // trampoline is used.
4675 // We will never set this flag, and we should
4676 // ignore this flag in our diagnostic
4677 #ifdef SIGNIFICANT_SIGNAL_MASK
4678   #undef SIGNIFICANT_SIGNAL_MASK
4679 #endif
4680 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4681 
4682 static const char* get_signal_handler_name(address handler,
4683                                            char* buf, int buflen) {
4684   int offset = 0;
4685   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4686   if (found) {
4687     // skip directory names
4688     const char *p1, *p2;
4689     p1 = buf;
4690     size_t len = strlen(os::file_separator());
4691     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4692     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4693   } else {
4694     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4695   }
4696   return buf;
4697 }
4698 
4699 static void print_signal_handler(outputStream* st, int sig,
4700                                  char* buf, size_t buflen) {
4701   struct sigaction sa;
4702 
4703   sigaction(sig, NULL, &sa);
4704 
4705   // See comment for SIGNIFICANT_SIGNAL_MASK define
4706   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4707 
4708   st->print("%s: ", os::exception_name(sig, buf, buflen));
4709 
4710   address handler = (sa.sa_flags & SA_SIGINFO)
4711     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4712     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4713 
4714   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4715     st->print("SIG_DFL");
4716   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4717     st->print("SIG_IGN");
4718   } else {
4719     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4720   }
4721 
4722   st->print(", sa_mask[0]=");
4723   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4724 
4725   address rh = VMError::get_resetted_sighandler(sig);
4726   // May be, handler was resetted by VMError?
4727   if (rh != NULL) {
4728     handler = rh;
4729     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4730   }
4731 
4732   st->print(", sa_flags=");
4733   os::Posix::print_sa_flags(st, sa.sa_flags);
4734 
4735   // Check: is it our handler?
4736   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4737       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4738     // It is our signal handler
4739     // check for flags, reset system-used one!
4740     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4741       st->print(
4742                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4743                 os::Linux::get_our_sigflags(sig));
4744     }
4745   }
4746   st->cr();
4747 }
4748 
4749 
4750 #define DO_SIGNAL_CHECK(sig)                      \
4751   do {                                            \
4752     if (!sigismember(&check_signal_done, sig)) {  \
4753       os::Linux::check_signal_handler(sig);       \
4754     }                                             \
4755   } while (0)
4756 
4757 // This method is a periodic task to check for misbehaving JNI applications
4758 // under CheckJNI, we can add any periodic checks here
4759 
4760 void os::run_periodic_checks() {
4761   if (check_signals == false) return;
4762 
4763   // SEGV and BUS if overridden could potentially prevent
4764   // generation of hs*.log in the event of a crash, debugging
4765   // such a case can be very challenging, so we absolutely
4766   // check the following for a good measure:
4767   DO_SIGNAL_CHECK(SIGSEGV);
4768   DO_SIGNAL_CHECK(SIGILL);
4769   DO_SIGNAL_CHECK(SIGFPE);
4770   DO_SIGNAL_CHECK(SIGBUS);
4771   DO_SIGNAL_CHECK(SIGPIPE);
4772   DO_SIGNAL_CHECK(SIGXFSZ);
4773 #if defined(PPC64)
4774   DO_SIGNAL_CHECK(SIGTRAP);
4775 #endif
4776 
4777   // ReduceSignalUsage allows the user to override these handlers
4778   // see comments at the very top and jvm_md.h
4779   if (!ReduceSignalUsage) {
4780     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4781     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4782     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4783     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4784   }
4785 
4786   DO_SIGNAL_CHECK(SR_signum);
4787 }
4788 
4789 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4790 
4791 static os_sigaction_t os_sigaction = NULL;
4792 
4793 void os::Linux::check_signal_handler(int sig) {
4794   char buf[O_BUFLEN];
4795   address jvmHandler = NULL;
4796 
4797 
4798   struct sigaction act;
4799   if (os_sigaction == NULL) {
4800     // only trust the default sigaction, in case it has been interposed
4801     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4802     if (os_sigaction == NULL) return;
4803   }
4804 
4805   os_sigaction(sig, (struct sigaction*)NULL, &act);
4806 
4807 
4808   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4809 
4810   address thisHandler = (act.sa_flags & SA_SIGINFO)
4811     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4812     : CAST_FROM_FN_PTR(address, act.sa_handler);
4813 
4814 
4815   switch (sig) {
4816   case SIGSEGV:
4817   case SIGBUS:
4818   case SIGFPE:
4819   case SIGPIPE:
4820   case SIGILL:
4821   case SIGXFSZ:
4822     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4823     break;
4824 
4825   case SHUTDOWN1_SIGNAL:
4826   case SHUTDOWN2_SIGNAL:
4827   case SHUTDOWN3_SIGNAL:
4828   case BREAK_SIGNAL:
4829     jvmHandler = (address)user_handler();
4830     break;
4831 
4832   default:
4833     if (sig == SR_signum) {
4834       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4835     } else {
4836       return;
4837     }
4838     break;
4839   }
4840 
4841   if (thisHandler != jvmHandler) {
4842     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4843     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4844     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4845     // No need to check this sig any longer
4846     sigaddset(&check_signal_done, sig);
4847     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4848     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4849       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4850                     exception_name(sig, buf, O_BUFLEN));
4851     }
4852   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4853     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4854     tty->print("expected:");
4855     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4856     tty->cr();
4857     tty->print("  found:");
4858     os::Posix::print_sa_flags(tty, act.sa_flags);
4859     tty->cr();
4860     // No need to check this sig any longer
4861     sigaddset(&check_signal_done, sig);
4862   }
4863 
4864   // Dump all the signal
4865   if (sigismember(&check_signal_done, sig)) {
4866     print_signal_handlers(tty, buf, O_BUFLEN);
4867   }
4868 }
4869 
4870 extern void report_error(char* file_name, int line_no, char* title,
4871                          char* format, ...);
4872 
4873 // this is called _before_ most of the global arguments have been parsed
4874 void os::init(void) {
4875   char dummy;   // used to get a guess on initial stack address
4876 
4877   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4878 
4879   init_random(1234567);
4880 
4881   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4882   if (Linux::page_size() == -1) {
4883     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4884           os::strerror(errno));
4885   }
4886   init_page_sizes((size_t) Linux::page_size());
4887 
4888   Linux::initialize_system_info();
4889 
4890   Linux::initialize_os_info();
4891 
4892   // _main_thread points to the thread that created/loaded the JVM.
4893   Linux::_main_thread = pthread_self();
4894 
4895   Linux::clock_init();
4896   initial_time_count = javaTimeNanos();
4897 
4898   // retrieve entry point for pthread_setname_np
4899   Linux::_pthread_setname_np =
4900     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4901 
4902   os::Posix::init();
4903 }
4904 
4905 // To install functions for atexit system call
4906 extern "C" {
4907   static void perfMemory_exit_helper() {
4908     perfMemory_exit();
4909   }
4910 }
4911 
4912 void os::pd_init_container_support() {
4913   OSContainer::init();
4914 }
4915 
4916 // this is called _after_ the global arguments have been parsed
4917 jint os::init_2(void) {
4918 
4919   os::Posix::init_2();
4920 
4921   Linux::fast_thread_clock_init();
4922 
4923   // initialize suspend/resume support - must do this before signal_sets_init()
4924   if (SR_initialize() != 0) {
4925     perror("SR_initialize failed");
4926     return JNI_ERR;
4927   }
4928 
4929   Linux::signal_sets_init();
4930   Linux::install_signal_handlers();
4931 
4932   // Check and sets minimum stack sizes against command line options
4933   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4934     return JNI_ERR;
4935   }
4936   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4937 
4938 #if defined(IA32)
4939   workaround_expand_exec_shield_cs_limit();
4940 #endif
4941 
4942   Linux::libpthread_init();
4943   Linux::sched_getcpu_init();
4944   log_info(os)("HotSpot is running with %s, %s",
4945                Linux::glibc_version(), Linux::libpthread_version());
4946 
4947   if (UseNUMA) {
4948     if (!Linux::libnuma_init()) {
4949       UseNUMA = false;
4950     } else {
4951       if ((Linux::numa_max_node() < 1)) {
4952         // There's only one node(they start from 0), disable NUMA.
4953         UseNUMA = false;
4954       }
4955     }
4956 
4957     if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4958       // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4959       // we can make the adaptive lgrp chunk resizing work. If the user specified both
4960       // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
4961       // and disable adaptive resizing.
4962       if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4963         warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
4964                 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4965         UseAdaptiveSizePolicy = false;
4966         UseAdaptiveNUMAChunkSizing = false;
4967       }
4968     }
4969 
4970     if (!UseNUMA && ForceNUMA) {
4971       UseNUMA = true;
4972     }
4973   }
4974 
4975   if (MaxFDLimit) {
4976     // set the number of file descriptors to max. print out error
4977     // if getrlimit/setrlimit fails but continue regardless.
4978     struct rlimit nbr_files;
4979     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4980     if (status != 0) {
4981       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4982     } else {
4983       nbr_files.rlim_cur = nbr_files.rlim_max;
4984       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4985       if (status != 0) {
4986         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4987       }
4988     }
4989   }
4990 
4991   // Initialize lock used to serialize thread creation (see os::create_thread)
4992   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4993 
4994   // at-exit methods are called in the reverse order of their registration.
4995   // atexit functions are called on return from main or as a result of a
4996   // call to exit(3C). There can be only 32 of these functions registered
4997   // and atexit() does not set errno.
4998 
4999   if (PerfAllowAtExitRegistration) {
5000     // only register atexit functions if PerfAllowAtExitRegistration is set.
5001     // atexit functions can be delayed until process exit time, which
5002     // can be problematic for embedded VM situations. Embedded VMs should
5003     // call DestroyJavaVM() to assure that VM resources are released.
5004 
5005     // note: perfMemory_exit_helper atexit function may be removed in
5006     // the future if the appropriate cleanup code can be added to the
5007     // VM_Exit VMOperation's doit method.
5008     if (atexit(perfMemory_exit_helper) != 0) {
5009       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5010     }
5011   }
5012 
5013   // initialize thread priority policy
5014   prio_init();
5015 
5016   if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5017     set_coredump_filter(false /*largepages*/, true /*dax_shared*/);
5018   }
5019   return JNI_OK;
5020 }
5021 
5022 // Mark the polling page as unreadable
5023 void os::make_polling_page_unreadable(void) {
5024   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5025     fatal("Could not disable polling page");
5026   }
5027 }
5028 
5029 // Mark the polling page as readable
5030 void os::make_polling_page_readable(void) {
5031   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5032     fatal("Could not enable polling page");
5033   }
5034 }
5035 
5036 // older glibc versions don't have this macro (which expands to
5037 // an optimized bit-counting function) so we have to roll our own
5038 #ifndef CPU_COUNT
5039 
5040 static int _cpu_count(const cpu_set_t* cpus) {
5041   int count = 0;
5042   // only look up to the number of configured processors
5043   for (int i = 0; i < os::processor_count(); i++) {
5044     if (CPU_ISSET(i, cpus)) {
5045       count++;
5046     }
5047   }
5048   return count;
5049 }
5050 
5051 #define CPU_COUNT(cpus) _cpu_count(cpus)
5052 
5053 #endif // CPU_COUNT
5054 
5055 // Get the current number of available processors for this process.
5056 // This value can change at any time during a process's lifetime.
5057 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5058 // If it appears there may be more than 1024 processors then we do a
5059 // dynamic check - see 6515172 for details.
5060 // If anything goes wrong we fallback to returning the number of online
5061 // processors - which can be greater than the number available to the process.
5062 int os::Linux::active_processor_count() {
5063   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
5064   cpu_set_t* cpus_p = &cpus;
5065   int cpus_size = sizeof(cpu_set_t);
5066 
5067   int configured_cpus = os::processor_count();  // upper bound on available cpus
5068   int cpu_count = 0;
5069 
5070 // old build platforms may not support dynamic cpu sets
5071 #ifdef CPU_ALLOC
5072 
5073   // To enable easy testing of the dynamic path on different platforms we
5074   // introduce a diagnostic flag: UseCpuAllocPath
5075   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5076     // kernel may use a mask bigger than cpu_set_t
5077     log_trace(os)("active_processor_count: using dynamic path %s"
5078                   "- configured processors: %d",
5079                   UseCpuAllocPath ? "(forced) " : "",
5080                   configured_cpus);
5081     cpus_p = CPU_ALLOC(configured_cpus);
5082     if (cpus_p != NULL) {
5083       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5084       // zero it just to be safe
5085       CPU_ZERO_S(cpus_size, cpus_p);
5086     }
5087     else {
5088        // failed to allocate so fallback to online cpus
5089        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5090        log_trace(os)("active_processor_count: "
5091                      "CPU_ALLOC failed (%s) - using "
5092                      "online processor count: %d",
5093                      os::strerror(errno), online_cpus);
5094        return online_cpus;
5095     }
5096   }
5097   else {
5098     log_trace(os)("active_processor_count: using static path - configured processors: %d",
5099                   configured_cpus);
5100   }
5101 #else // CPU_ALLOC
5102 // these stubs won't be executed
5103 #define CPU_COUNT_S(size, cpus) -1
5104 #define CPU_FREE(cpus)
5105 
5106   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5107                 configured_cpus);
5108 #endif // CPU_ALLOC
5109 
5110   // pid 0 means the current thread - which we have to assume represents the process
5111   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5112     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5113       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5114     }
5115     else {
5116       cpu_count = CPU_COUNT(cpus_p);
5117     }
5118     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5119   }
5120   else {
5121     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5122     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5123             "which may exceed available processors", os::strerror(errno), cpu_count);
5124   }
5125 
5126   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5127     CPU_FREE(cpus_p);
5128   }
5129 
5130   assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5131   return cpu_count;
5132 }
5133 
5134 // Determine the active processor count from one of
5135 // three different sources:
5136 //
5137 // 1. User option -XX:ActiveProcessorCount
5138 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5139 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5140 //
5141 // Option 1, if specified, will always override.
5142 // If the cgroup subsystem is active and configured, we
5143 // will return the min of the cgroup and option 2 results.
5144 // This is required since tools, such as numactl, that
5145 // alter cpu affinity do not update cgroup subsystem
5146 // cpuset configuration files.
5147 int os::active_processor_count() {
5148   // User has overridden the number of active processors
5149   if (ActiveProcessorCount > 0) {
5150     log_trace(os)("active_processor_count: "
5151                   "active processor count set by user : %d",
5152                   ActiveProcessorCount);
5153     return ActiveProcessorCount;
5154   }
5155 
5156   int active_cpus;
5157   if (OSContainer::is_containerized()) {
5158     active_cpus = OSContainer::active_processor_count();
5159     log_trace(os)("active_processor_count: determined by OSContainer: %d",
5160                    active_cpus);
5161   } else {
5162     active_cpus = os::Linux::active_processor_count();
5163   }
5164 
5165   return active_cpus;
5166 }
5167 
5168 void os::set_native_thread_name(const char *name) {
5169   if (Linux::_pthread_setname_np) {
5170     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5171     snprintf(buf, sizeof(buf), "%s", name);
5172     buf[sizeof(buf) - 1] = '\0';
5173     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5174     // ERANGE should not happen; all other errors should just be ignored.
5175     assert(rc != ERANGE, "pthread_setname_np failed");
5176   }
5177 }
5178 
5179 bool os::distribute_processes(uint length, uint* distribution) {
5180   // Not yet implemented.
5181   return false;
5182 }
5183 
5184 bool os::bind_to_processor(uint processor_id) {
5185   // Not yet implemented.
5186   return false;
5187 }
5188 
5189 ///
5190 
5191 void os::SuspendedThreadTask::internal_do_task() {
5192   if (do_suspend(_thread->osthread())) {
5193     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5194     do_task(context);
5195     do_resume(_thread->osthread());
5196   }
5197 }
5198 
5199 ////////////////////////////////////////////////////////////////////////////////
5200 // debug support
5201 
5202 bool os::find(address addr, outputStream* st) {
5203   Dl_info dlinfo;
5204   memset(&dlinfo, 0, sizeof(dlinfo));
5205   if (dladdr(addr, &dlinfo) != 0) {
5206     st->print(PTR_FORMAT ": ", p2i(addr));
5207     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5208       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5209                 p2i(addr) - p2i(dlinfo.dli_saddr));
5210     } else if (dlinfo.dli_fbase != NULL) {
5211       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5212     } else {
5213       st->print("<absolute address>");
5214     }
5215     if (dlinfo.dli_fname != NULL) {
5216       st->print(" in %s", dlinfo.dli_fname);
5217     }
5218     if (dlinfo.dli_fbase != NULL) {
5219       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5220     }
5221     st->cr();
5222 
5223     if (Verbose) {
5224       // decode some bytes around the PC
5225       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5226       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5227       address       lowest = (address) dlinfo.dli_sname;
5228       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5229       if (begin < lowest)  begin = lowest;
5230       Dl_info dlinfo2;
5231       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5232           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5233         end = (address) dlinfo2.dli_saddr;
5234       }
5235       Disassembler::decode(begin, end, st);
5236     }
5237     return true;
5238   }
5239   return false;
5240 }
5241 
5242 ////////////////////////////////////////////////////////////////////////////////
5243 // misc
5244 
5245 // This does not do anything on Linux. This is basically a hook for being
5246 // able to use structured exception handling (thread-local exception filters)
5247 // on, e.g., Win32.
5248 void
5249 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5250                          JavaCallArguments* args, Thread* thread) {
5251   f(value, method, args, thread);
5252 }
5253 
5254 void os::print_statistics() {
5255 }
5256 
5257 bool os::message_box(const char* title, const char* message) {
5258   int i;
5259   fdStream err(defaultStream::error_fd());
5260   for (i = 0; i < 78; i++) err.print_raw("=");
5261   err.cr();
5262   err.print_raw_cr(title);
5263   for (i = 0; i < 78; i++) err.print_raw("-");
5264   err.cr();
5265   err.print_raw_cr(message);
5266   for (i = 0; i < 78; i++) err.print_raw("=");
5267   err.cr();
5268 
5269   char buf[16];
5270   // Prevent process from exiting upon "read error" without consuming all CPU
5271   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5272 
5273   return buf[0] == 'y' || buf[0] == 'Y';
5274 }
5275 
5276 int os::stat(const char *path, struct stat *sbuf) {
5277   char pathbuf[MAX_PATH];
5278   if (strlen(path) > MAX_PATH - 1) {
5279     errno = ENAMETOOLONG;
5280     return -1;
5281   }
5282   os::native_path(strcpy(pathbuf, path));
5283   return ::stat(pathbuf, sbuf);
5284 }
5285 
5286 // Is a (classpath) directory empty?
5287 bool os::dir_is_empty(const char* path) {
5288   DIR *dir = NULL;
5289   struct dirent *ptr;
5290 
5291   dir = opendir(path);
5292   if (dir == NULL) return true;
5293 
5294   // Scan the directory
5295   bool result = true;
5296   char buf[sizeof(struct dirent) + MAX_PATH];
5297   while (result && (ptr = ::readdir(dir)) != NULL) {
5298     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5299       result = false;
5300     }
5301   }
5302   closedir(dir);
5303   return result;
5304 }
5305 
5306 // This code originates from JDK's sysOpen and open64_w
5307 // from src/solaris/hpi/src/system_md.c
5308 
5309 int os::open(const char *path, int oflag, int mode) {
5310   if (strlen(path) > MAX_PATH - 1) {
5311     errno = ENAMETOOLONG;
5312     return -1;
5313   }
5314 
5315   // All file descriptors that are opened in the Java process and not
5316   // specifically destined for a subprocess should have the close-on-exec
5317   // flag set.  If we don't set it, then careless 3rd party native code
5318   // might fork and exec without closing all appropriate file descriptors
5319   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5320   // turn might:
5321   //
5322   // - cause end-of-file to fail to be detected on some file
5323   //   descriptors, resulting in mysterious hangs, or
5324   //
5325   // - might cause an fopen in the subprocess to fail on a system
5326   //   suffering from bug 1085341.
5327   //
5328   // (Yes, the default setting of the close-on-exec flag is a Unix
5329   // design flaw)
5330   //
5331   // See:
5332   // 1085341: 32-bit stdio routines should support file descriptors >255
5333   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5334   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5335   //
5336   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5337   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5338   // because it saves a system call and removes a small window where the flag
5339   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5340   // and we fall back to using FD_CLOEXEC (see below).
5341 #ifdef O_CLOEXEC
5342   oflag |= O_CLOEXEC;
5343 #endif
5344 
5345   int fd = ::open64(path, oflag, mode);
5346   if (fd == -1) return -1;
5347 
5348   //If the open succeeded, the file might still be a directory
5349   {
5350     struct stat64 buf64;
5351     int ret = ::fstat64(fd, &buf64);
5352     int st_mode = buf64.st_mode;
5353 
5354     if (ret != -1) {
5355       if ((st_mode & S_IFMT) == S_IFDIR) {
5356         errno = EISDIR;
5357         ::close(fd);
5358         return -1;
5359       }
5360     } else {
5361       ::close(fd);
5362       return -1;
5363     }
5364   }
5365 
5366 #ifdef FD_CLOEXEC
5367   // Validate that the use of the O_CLOEXEC flag on open above worked.
5368   // With recent kernels, we will perform this check exactly once.
5369   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5370   if (!O_CLOEXEC_is_known_to_work) {
5371     int flags = ::fcntl(fd, F_GETFD);
5372     if (flags != -1) {
5373       if ((flags & FD_CLOEXEC) != 0)
5374         O_CLOEXEC_is_known_to_work = 1;
5375       else
5376         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5377     }
5378   }
5379 #endif
5380 
5381   return fd;
5382 }
5383 
5384 
5385 // create binary file, rewriting existing file if required
5386 int os::create_binary_file(const char* path, bool rewrite_existing) {
5387   int oflags = O_WRONLY | O_CREAT;
5388   if (!rewrite_existing) {
5389     oflags |= O_EXCL;
5390   }
5391   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5392 }
5393 
5394 // return current position of file pointer
5395 jlong os::current_file_offset(int fd) {
5396   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5397 }
5398 
5399 // move file pointer to the specified offset
5400 jlong os::seek_to_file_offset(int fd, jlong offset) {
5401   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5402 }
5403 
5404 // This code originates from JDK's sysAvailable
5405 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5406 
5407 int os::available(int fd, jlong *bytes) {
5408   jlong cur, end;
5409   int mode;
5410   struct stat64 buf64;
5411 
5412   if (::fstat64(fd, &buf64) >= 0) {
5413     mode = buf64.st_mode;
5414     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5415       int n;
5416       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5417         *bytes = n;
5418         return 1;
5419       }
5420     }
5421   }
5422   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5423     return 0;
5424   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5425     return 0;
5426   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5427     return 0;
5428   }
5429   *bytes = end - cur;
5430   return 1;
5431 }
5432 
5433 // Map a block of memory.
5434 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5435                         char *addr, size_t bytes, bool read_only,
5436                         bool allow_exec) {
5437   int prot;
5438   int flags = MAP_PRIVATE;
5439 
5440   if (read_only) {
5441     prot = PROT_READ;
5442   } else {
5443     prot = PROT_READ | PROT_WRITE;
5444   }
5445 
5446   if (allow_exec) {
5447     prot |= PROT_EXEC;
5448   }
5449 
5450   if (addr != NULL) {
5451     flags |= MAP_FIXED;
5452   }
5453 
5454   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5455                                      fd, file_offset);
5456   if (mapped_address == MAP_FAILED) {
5457     return NULL;
5458   }
5459   return mapped_address;
5460 }
5461 
5462 
5463 // Remap a block of memory.
5464 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5465                           char *addr, size_t bytes, bool read_only,
5466                           bool allow_exec) {
5467   // same as map_memory() on this OS
5468   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5469                         allow_exec);
5470 }
5471 
5472 
5473 // Unmap a block of memory.
5474 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5475   return munmap(addr, bytes) == 0;
5476 }
5477 
5478 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5479 
5480 static clockid_t thread_cpu_clockid(Thread* thread) {
5481   pthread_t tid = thread->osthread()->pthread_id();
5482   clockid_t clockid;
5483 
5484   // Get thread clockid
5485   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5486   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5487   return clockid;
5488 }
5489 
5490 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5491 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5492 // of a thread.
5493 //
5494 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5495 // the fast estimate available on the platform.
5496 
5497 jlong os::current_thread_cpu_time() {
5498   if (os::Linux::supports_fast_thread_cpu_time()) {
5499     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5500   } else {
5501     // return user + sys since the cost is the same
5502     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5503   }
5504 }
5505 
5506 jlong os::thread_cpu_time(Thread* thread) {
5507   // consistent with what current_thread_cpu_time() returns
5508   if (os::Linux::supports_fast_thread_cpu_time()) {
5509     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5510   } else {
5511     return slow_thread_cpu_time(thread, true /* user + sys */);
5512   }
5513 }
5514 
5515 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5516   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5517     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5518   } else {
5519     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5520   }
5521 }
5522 
5523 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5524   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5525     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5526   } else {
5527     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5528   }
5529 }
5530 
5531 //  -1 on error.
5532 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5533   pid_t  tid = thread->osthread()->thread_id();
5534   char *s;
5535   char stat[2048];
5536   int statlen;
5537   char proc_name[64];
5538   int count;
5539   long sys_time, user_time;
5540   char cdummy;
5541   int idummy;
5542   long ldummy;
5543   FILE *fp;
5544 
5545   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5546   fp = fopen(proc_name, "r");
5547   if (fp == NULL) return -1;
5548   statlen = fread(stat, 1, 2047, fp);
5549   stat[statlen] = '\0';
5550   fclose(fp);
5551 
5552   // Skip pid and the command string. Note that we could be dealing with
5553   // weird command names, e.g. user could decide to rename java launcher
5554   // to "java 1.4.2 :)", then the stat file would look like
5555   //                1234 (java 1.4.2 :)) R ... ...
5556   // We don't really need to know the command string, just find the last
5557   // occurrence of ")" and then start parsing from there. See bug 4726580.
5558   s = strrchr(stat, ')');
5559   if (s == NULL) return -1;
5560 
5561   // Skip blank chars
5562   do { s++; } while (s && isspace(*s));
5563 
5564   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5565                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5566                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5567                  &user_time, &sys_time);
5568   if (count != 13) return -1;
5569   if (user_sys_cpu_time) {
5570     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5571   } else {
5572     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5573   }
5574 }
5575 
5576 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5577   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5578   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5579   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5580   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5581 }
5582 
5583 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5584   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5585   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5586   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5587   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5588 }
5589 
5590 bool os::is_thread_cpu_time_supported() {
5591   return true;
5592 }
5593 
5594 // System loadavg support.  Returns -1 if load average cannot be obtained.
5595 // Linux doesn't yet have a (official) notion of processor sets,
5596 // so just return the system wide load average.
5597 int os::loadavg(double loadavg[], int nelem) {
5598   return ::getloadavg(loadavg, nelem);
5599 }
5600 
5601 void os::pause() {
5602   char filename[MAX_PATH];
5603   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5604     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5605   } else {
5606     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5607   }
5608 
5609   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5610   if (fd != -1) {
5611     struct stat buf;
5612     ::close(fd);
5613     while (::stat(filename, &buf) == 0) {
5614       (void)::poll(NULL, 0, 100);
5615     }
5616   } else {
5617     jio_fprintf(stderr,
5618                 "Could not open pause file '%s', continuing immediately.\n", filename);
5619   }
5620 }
5621 
5622 extern char** environ;
5623 
5624 // Run the specified command in a separate process. Return its exit value,
5625 // or -1 on failure (e.g. can't fork a new process).
5626 // Unlike system(), this function can be called from signal handler. It
5627 // doesn't block SIGINT et al.
5628 int os::fork_and_exec(char* cmd) {
5629   const char * argv[4] = {"sh", "-c", cmd, NULL};
5630 
5631   pid_t pid = fork();
5632 
5633   if (pid < 0) {
5634     // fork failed
5635     return -1;
5636 
5637   } else if (pid == 0) {
5638     // child process
5639 
5640     execve("/bin/sh", (char* const*)argv, environ);
5641 
5642     // execve failed
5643     _exit(-1);
5644 
5645   } else  {
5646     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5647     // care about the actual exit code, for now.
5648 
5649     int status;
5650 
5651     // Wait for the child process to exit.  This returns immediately if
5652     // the child has already exited. */
5653     while (waitpid(pid, &status, 0) < 0) {
5654       switch (errno) {
5655       case ECHILD: return 0;
5656       case EINTR: break;
5657       default: return -1;
5658       }
5659     }
5660 
5661     if (WIFEXITED(status)) {
5662       // The child exited normally; get its exit code.
5663       return WEXITSTATUS(status);
5664     } else if (WIFSIGNALED(status)) {
5665       // The child exited because of a signal
5666       // The best value to return is 0x80 + signal number,
5667       // because that is what all Unix shells do, and because
5668       // it allows callers to distinguish between process exit and
5669       // process death by signal.
5670       return 0x80 + WTERMSIG(status);
5671     } else {
5672       // Unknown exit code; pass it through
5673       return status;
5674     }
5675   }
5676 }
5677 
5678 // Get the default path to the core file
5679 // Returns the length of the string
5680 int os::get_core_path(char* buffer, size_t bufferSize) {
5681   /*
5682    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5683    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5684    */
5685   const int core_pattern_len = 129;
5686   char core_pattern[core_pattern_len] = {0};
5687 
5688   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5689   if (core_pattern_file == -1) {
5690     return -1;
5691   }
5692 
5693   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5694   ::close(core_pattern_file);
5695   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5696     return -1;
5697   }
5698   if (core_pattern[ret-1] == '\n') {
5699     core_pattern[ret-1] = '\0';
5700   } else {
5701     core_pattern[ret] = '\0';
5702   }
5703 
5704   char *pid_pos = strstr(core_pattern, "%p");
5705   int written;
5706 
5707   if (core_pattern[0] == '/') {
5708     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5709   } else {
5710     char cwd[PATH_MAX];
5711 
5712     const char* p = get_current_directory(cwd, PATH_MAX);
5713     if (p == NULL) {
5714       return -1;
5715     }
5716 
5717     if (core_pattern[0] == '|') {
5718       written = jio_snprintf(buffer, bufferSize,
5719                              "\"%s\" (or dumping to %s/core.%d)",
5720                              &core_pattern[1], p, current_process_id());
5721     } else {
5722       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5723     }
5724   }
5725 
5726   if (written < 0) {
5727     return -1;
5728   }
5729 
5730   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5731     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5732 
5733     if (core_uses_pid_file != -1) {
5734       char core_uses_pid = 0;
5735       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5736       ::close(core_uses_pid_file);
5737 
5738       if (core_uses_pid == '1') {
5739         jio_snprintf(buffer + written, bufferSize - written,
5740                                           ".%d", current_process_id());
5741       }
5742     }
5743   }
5744 
5745   return strlen(buffer);
5746 }
5747 
5748 bool os::start_debugging(char *buf, int buflen) {
5749   int len = (int)strlen(buf);
5750   char *p = &buf[len];
5751 
5752   jio_snprintf(p, buflen-len,
5753                "\n\n"
5754                "Do you want to debug the problem?\n\n"
5755                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5756                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5757                "Otherwise, press RETURN to abort...",
5758                os::current_process_id(), os::current_process_id(),
5759                os::current_thread_id(), os::current_thread_id());
5760 
5761   bool yes = os::message_box("Unexpected Error", buf);
5762 
5763   if (yes) {
5764     // yes, user asked VM to launch debugger
5765     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5766                  os::current_process_id(), os::current_process_id());
5767 
5768     os::fork_and_exec(buf);
5769     yes = false;
5770   }
5771   return yes;
5772 }
5773 
5774 
5775 // Java/Compiler thread:
5776 //
5777 //   Low memory addresses
5778 // P0 +------------------------+
5779 //    |                        |\  Java thread created by VM does not have glibc
5780 //    |    glibc guard page    | - guard page, attached Java thread usually has
5781 //    |                        |/  1 glibc guard page.
5782 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5783 //    |                        |\
5784 //    |  HotSpot Guard Pages   | - red, yellow and reserved pages
5785 //    |                        |/
5786 //    +------------------------+ JavaThread::stack_reserved_zone_base()
5787 //    |                        |\
5788 //    |      Normal Stack      | -
5789 //    |                        |/
5790 // P2 +------------------------+ Thread::stack_base()
5791 //
5792 // Non-Java thread:
5793 //
5794 //   Low memory addresses
5795 // P0 +------------------------+
5796 //    |                        |\
5797 //    |  glibc guard page      | - usually 1 page
5798 //    |                        |/
5799 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5800 //    |                        |\
5801 //    |      Normal Stack      | -
5802 //    |                        |/
5803 // P2 +------------------------+ Thread::stack_base()
5804 //
5805 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5806 //    returned from pthread_attr_getstack().
5807 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5808 //    of the stack size given in pthread_attr. We work around this for
5809 //    threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5810 //
5811 #ifndef ZERO
5812 static void current_stack_region(address * bottom, size_t * size) {
5813   if (os::is_primordial_thread()) {
5814     // primordial thread needs special handling because pthread_getattr_np()
5815     // may return bogus value.
5816     *bottom = os::Linux::initial_thread_stack_bottom();
5817     *size   = os::Linux::initial_thread_stack_size();
5818   } else {
5819     pthread_attr_t attr;
5820 
5821     int rslt = pthread_getattr_np(pthread_self(), &attr);
5822 
5823     // JVM needs to know exact stack location, abort if it fails
5824     if (rslt != 0) {
5825       if (rslt == ENOMEM) {
5826         vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5827       } else {
5828         fatal("pthread_getattr_np failed with error = %d", rslt);
5829       }
5830     }
5831 
5832     if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5833       fatal("Cannot locate current stack attributes!");
5834     }
5835 
5836     // Work around NPTL stack guard error.
5837     size_t guard_size = 0;
5838     rslt = pthread_attr_getguardsize(&attr, &guard_size);
5839     if (rslt != 0) {
5840       fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5841     }
5842     *bottom += guard_size;
5843     *size   -= guard_size;
5844 
5845     pthread_attr_destroy(&attr);
5846 
5847   }
5848   assert(os::current_stack_pointer() >= *bottom &&
5849          os::current_stack_pointer() < *bottom + *size, "just checking");
5850 }
5851 
5852 address os::current_stack_base() {
5853   address bottom;
5854   size_t size;
5855   current_stack_region(&bottom, &size);
5856   return (bottom + size);
5857 }
5858 
5859 size_t os::current_stack_size() {
5860   // This stack size includes the usable stack and HotSpot guard pages
5861   // (for the threads that have Hotspot guard pages).
5862   address bottom;
5863   size_t size;
5864   current_stack_region(&bottom, &size);
5865   return size;
5866 }
5867 #endif
5868 
5869 static inline struct timespec get_mtime(const char* filename) {
5870   struct stat st;
5871   int ret = os::stat(filename, &st);
5872   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5873   return st.st_mtim;
5874 }
5875 
5876 int os::compare_file_modified_times(const char* file1, const char* file2) {
5877   struct timespec filetime1 = get_mtime(file1);
5878   struct timespec filetime2 = get_mtime(file2);
5879   int diff = filetime1.tv_sec - filetime2.tv_sec;
5880   if (diff == 0) {
5881     return filetime1.tv_nsec - filetime2.tv_nsec;
5882   }
5883   return diff;
5884 }
5885 
5886 /////////////// Unit tests ///////////////
5887 
5888 #ifndef PRODUCT
5889 
5890 #define test_log(...)              \
5891   do {                             \
5892     if (VerboseInternalVMTests) {  \
5893       tty->print_cr(__VA_ARGS__);  \
5894       tty->flush();                \
5895     }                              \
5896   } while (false)
5897 
5898 class TestReserveMemorySpecial : AllStatic {
5899  public:
5900   static void small_page_write(void* addr, size_t size) {
5901     size_t page_size = os::vm_page_size();
5902 
5903     char* end = (char*)addr + size;
5904     for (char* p = (char*)addr; p < end; p += page_size) {
5905       *p = 1;
5906     }
5907   }
5908 
5909   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5910     if (!UseHugeTLBFS) {
5911       return;
5912     }
5913 
5914     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5915 
5916     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5917 
5918     if (addr != NULL) {
5919       small_page_write(addr, size);
5920 
5921       os::Linux::release_memory_special_huge_tlbfs(addr, size);
5922     }
5923   }
5924 
5925   static void test_reserve_memory_special_huge_tlbfs_only() {
5926     if (!UseHugeTLBFS) {
5927       return;
5928     }
5929 
5930     size_t lp = os::large_page_size();
5931 
5932     for (size_t size = lp; size <= lp * 10; size += lp) {
5933       test_reserve_memory_special_huge_tlbfs_only(size);
5934     }
5935   }
5936 
5937   static void test_reserve_memory_special_huge_tlbfs_mixed() {
5938     size_t lp = os::large_page_size();
5939     size_t ag = os::vm_allocation_granularity();
5940 
5941     // sizes to test
5942     const size_t sizes[] = {
5943       lp, lp + ag, lp + lp / 2, lp * 2,
5944       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
5945       lp * 10, lp * 10 + lp / 2
5946     };
5947     const int num_sizes = sizeof(sizes) / sizeof(size_t);
5948 
5949     // For each size/alignment combination, we test three scenarios:
5950     // 1) with req_addr == NULL
5951     // 2) with a non-null req_addr at which we expect to successfully allocate
5952     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
5953     //    expect the allocation to either fail or to ignore req_addr
5954 
5955     // Pre-allocate two areas; they shall be as large as the largest allocation
5956     //  and aligned to the largest alignment we will be testing.
5957     const size_t mapping_size = sizes[num_sizes - 1] * 2;
5958     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
5959       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5960       -1, 0);
5961     assert(mapping1 != MAP_FAILED, "should work");
5962 
5963     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
5964       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
5965       -1, 0);
5966     assert(mapping2 != MAP_FAILED, "should work");
5967 
5968     // Unmap the first mapping, but leave the second mapping intact: the first
5969     // mapping will serve as a value for a "good" req_addr (case 2). The second
5970     // mapping, still intact, as "bad" req_addr (case 3).
5971     ::munmap(mapping1, mapping_size);
5972 
5973     // Case 1
5974     test_log("%s, req_addr NULL:", __FUNCTION__);
5975     test_log("size            align           result");
5976 
5977     for (int i = 0; i < num_sizes; i++) {
5978       const size_t size = sizes[i];
5979       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
5980         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
5981         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
5982                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
5983         if (p != NULL) {
5984           assert(is_aligned(p, alignment), "must be");
5985           small_page_write(p, size);
5986           os::Linux::release_memory_special_huge_tlbfs(p, size);
5987         }
5988       }
5989     }
5990 
5991     // Case 2
5992     test_log("%s, req_addr non-NULL:", __FUNCTION__);
5993     test_log("size            align           req_addr         result");
5994 
5995     for (int i = 0; i < num_sizes; i++) {
5996       const size_t size = sizes[i];
5997       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
5998         char* const req_addr = align_up(mapping1, alignment);
5999         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6000         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6001                  size, alignment, p2i(req_addr), p2i(p),
6002                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6003         if (p != NULL) {
6004           assert(p == req_addr, "must be");
6005           small_page_write(p, size);
6006           os::Linux::release_memory_special_huge_tlbfs(p, size);
6007         }
6008       }
6009     }
6010 
6011     // Case 3
6012     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6013     test_log("size            align           req_addr         result");
6014 
6015     for (int i = 0; i < num_sizes; i++) {
6016       const size_t size = sizes[i];
6017       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6018         char* const req_addr = align_up(mapping2, alignment);
6019         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6020         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6021                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6022         // as the area around req_addr contains already existing mappings, the API should always
6023         // return NULL (as per contract, it cannot return another address)
6024         assert(p == NULL, "must be");
6025       }
6026     }
6027 
6028     ::munmap(mapping2, mapping_size);
6029 
6030   }
6031 
6032   static void test_reserve_memory_special_huge_tlbfs() {
6033     if (!UseHugeTLBFS) {
6034       return;
6035     }
6036 
6037     test_reserve_memory_special_huge_tlbfs_only();
6038     test_reserve_memory_special_huge_tlbfs_mixed();
6039   }
6040 
6041   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6042     if (!UseSHM) {
6043       return;
6044     }
6045 
6046     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6047 
6048     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6049 
6050     if (addr != NULL) {
6051       assert(is_aligned(addr, alignment), "Check");
6052       assert(is_aligned(addr, os::large_page_size()), "Check");
6053 
6054       small_page_write(addr, size);
6055 
6056       os::Linux::release_memory_special_shm(addr, size);
6057     }
6058   }
6059 
6060   static void test_reserve_memory_special_shm() {
6061     size_t lp = os::large_page_size();
6062     size_t ag = os::vm_allocation_granularity();
6063 
6064     for (size_t size = ag; size < lp * 3; size += ag) {
6065       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6066         test_reserve_memory_special_shm(size, alignment);
6067       }
6068     }
6069   }
6070 
6071   static void test() {
6072     test_reserve_memory_special_huge_tlbfs();
6073     test_reserve_memory_special_shm();
6074   }
6075 };
6076 
6077 void TestReserveMemorySpecial_test() {
6078   TestReserveMemorySpecial::test();
6079 }
6080 
6081 #endif