1 /*
   2  * Copyright (c) 1999, 2019, 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 "logging/logStream.hpp"
  37 #include "memory/allocation.inline.hpp"
  38 #include "memory/filemap.hpp"
  39 #include "oops/oop.inline.hpp"
  40 #include "os_linux.inline.hpp"
  41 #include "os_share_linux.hpp"
  42 #include "osContainer_linux.hpp"
  43 #include "prims/jniFastGetField.hpp"
  44 #include "prims/jvm_misc.hpp"
  45 #include "runtime/arguments.hpp"
  46 #include "runtime/atomic.hpp"
  47 #include "runtime/extendedPC.hpp"
  48 #include "runtime/globals.hpp"
  49 #include "runtime/interfaceSupport.inline.hpp"
  50 #include "runtime/init.hpp"
  51 #include "runtime/java.hpp"
  52 #include "runtime/javaCalls.hpp"
  53 #include "runtime/mutexLocker.hpp"
  54 #include "runtime/objectMonitor.hpp"
  55 #include "runtime/orderAccess.hpp"
  56 #include "runtime/osThread.hpp"
  57 #include "runtime/perfMemory.hpp"
  58 #include "runtime/sharedRuntime.hpp"
  59 #include "runtime/statSampler.hpp"
  60 #include "runtime/stubRoutines.hpp"
  61 #include "runtime/thread.inline.hpp"
  62 #include "runtime/threadCritical.hpp"
  63 #include "runtime/threadSMR.hpp"
  64 #include "runtime/timer.hpp"
  65 #include "semaphore_posix.hpp"
  66 #include "services/attachListener.hpp"
  67 #include "services/memTracker.hpp"
  68 #include "services/runtimeService.hpp"
  69 #include "utilities/align.hpp"
  70 #include "utilities/decoder.hpp"
  71 #include "utilities/defaultStream.hpp"
  72 #include "utilities/events.hpp"
  73 #include "utilities/elfFile.hpp"
  74 #include "utilities/growableArray.hpp"
  75 #include "utilities/macros.hpp"
  76 #include "utilities/vmError.hpp"
  77 
  78 // put OS-includes here
  79 # include <sys/types.h>
  80 # include <sys/mman.h>
  81 # include <sys/stat.h>
  82 # include <sys/select.h>
  83 # include <pthread.h>
  84 # include <signal.h>
  85 # include <errno.h>
  86 # include <dlfcn.h>
  87 # include <stdio.h>
  88 # include <unistd.h>
  89 # include <sys/resource.h>
  90 # include <pthread.h>
  91 # include <sys/stat.h>
  92 # include <sys/time.h>
  93 # include <sys/times.h>
  94 # include <sys/utsname.h>
  95 # include <sys/socket.h>
  96 # include <sys/wait.h>
  97 # include <pwd.h>
  98 # include <poll.h>
  99 # include <fcntl.h>
 100 # include <string.h>
 101 # include <syscall.h>
 102 # include <sys/sysinfo.h>
 103 # include <gnu/libc-version.h>
 104 # include <sys/ipc.h>
 105 # include <sys/shm.h>
 106 # include <link.h>
 107 # include <stdint.h>
 108 # include <inttypes.h>
 109 # include <sys/ioctl.h>
 110 
 111 #ifndef _GNU_SOURCE
 112   #define _GNU_SOURCE
 113   #include <sched.h>
 114   #undef _GNU_SOURCE
 115 #else
 116   #include <sched.h>
 117 #endif
 118 
 119 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 120 // getrusage() is prepared to handle the associated failure.
 121 #ifndef RUSAGE_THREAD
 122   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 123 #endif
 124 
 125 #define MAX_PATH    (2 * K)
 126 
 127 #define MAX_SECS 100000000
 128 
 129 // for timer info max values which include all bits
 130 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 131 
 132 enum CoredumpFilterBit {
 133   FILE_BACKED_PVT_BIT = 1 << 2,
 134   FILE_BACKED_SHARED_BIT = 1 << 3,
 135   LARGEPAGES_BIT = 1 << 6,
 136   DAX_SHARED_BIT = 1 << 8
 137 };
 138 
 139 ////////////////////////////////////////////////////////////////////////////////
 140 // global variables
 141 julong os::Linux::_physical_memory = 0;
 142 
 143 address   os::Linux::_initial_thread_stack_bottom = NULL;
 144 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 145 
 146 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 147 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 148 Mutex* os::Linux::_createThread_lock = NULL;
 149 pthread_t os::Linux::_main_thread;
 150 int os::Linux::_page_size = -1;
 151 bool os::Linux::_supports_fast_thread_cpu_time = false;
 152 uint32_t os::Linux::_os_version = 0;
 153 const char * os::Linux::_glibc_version = NULL;
 154 const char * os::Linux::_libpthread_version = NULL;
 155 
 156 static jlong initial_time_count=0;
 157 
 158 static int clock_tics_per_sec = 100;
 159 
 160 // If the VM might have been created on the primordial thread, we need to resolve the
 161 // primordial thread stack bounds and check if the current thread might be the
 162 // primordial thread in places. If we know that the primordial thread is never used,
 163 // such as when the VM was created by one of the standard java launchers, we can
 164 // avoid this
 165 static bool suppress_primordial_thread_resolution = false;
 166 
 167 // For diagnostics to print a message once. see run_periodic_checks
 168 static sigset_t check_signal_done;
 169 static bool check_signals = true;
 170 
 171 // Signal number used to suspend/resume a thread
 172 
 173 // do not use any signal number less than SIGSEGV, see 4355769
 174 static int SR_signum = SIGUSR2;
 175 sigset_t SR_sigset;
 176 
 177 // utility functions
 178 
 179 static int SR_initialize();
 180 
 181 julong os::available_memory() {
 182   return Linux::available_memory();
 183 }
 184 
 185 julong os::Linux::available_memory() {
 186   // values in struct sysinfo are "unsigned long"
 187   struct sysinfo si;
 188   julong avail_mem;
 189 
 190   if (OSContainer::is_containerized()) {
 191     jlong mem_limit, mem_usage;
 192     if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
 193       log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
 194                              mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
 195     }
 196     if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
 197       log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
 198     }
 199     if (mem_limit > 0 && mem_usage > 0 ) {
 200       avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
 201       log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
 202       return avail_mem;
 203     }
 204   }
 205 
 206   sysinfo(&si);
 207   avail_mem = (julong)si.freeram * si.mem_unit;
 208   log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
 209   return avail_mem;
 210 }
 211 
 212 julong os::physical_memory() {
 213   jlong phys_mem = 0;
 214   if (OSContainer::is_containerized()) {
 215     jlong mem_limit;
 216     if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
 217       log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
 218       return mem_limit;
 219     }
 220     log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
 221                             mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
 222   }
 223 
 224   phys_mem = Linux::physical_memory();
 225   log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
 226   return phys_mem;
 227 }
 228 
 229 // Return true if user is running as root.
 230 
 231 bool os::have_special_privileges() {
 232   static bool init = false;
 233   static bool privileges = false;
 234   if (!init) {
 235     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 236     init = true;
 237   }
 238   return privileges;
 239 }
 240 
 241 
 242 #ifndef SYS_gettid
 243 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 244   #ifdef __ia64__
 245     #define SYS_gettid 1105
 246   #else
 247     #ifdef __i386__
 248       #define SYS_gettid 224
 249     #else
 250       #ifdef __amd64__
 251         #define SYS_gettid 186
 252       #else
 253         #ifdef __sparc__
 254           #define SYS_gettid 143
 255         #else
 256           #error define gettid for the arch
 257         #endif
 258       #endif
 259     #endif
 260   #endif
 261 #endif
 262 
 263 
 264 // pid_t gettid()
 265 //
 266 // Returns the kernel thread id of the currently running thread. Kernel
 267 // thread id is used to access /proc.
 268 pid_t os::Linux::gettid() {
 269   int rslt = syscall(SYS_gettid);
 270   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 271   return (pid_t)rslt;
 272 }
 273 
 274 // Most versions of linux have a bug where the number of processors are
 275 // determined by looking at the /proc file system.  In a chroot environment,
 276 // the system call returns 1.
 277 static bool unsafe_chroot_detected = false;
 278 static const char *unstable_chroot_error = "/proc file system not found.\n"
 279                      "Java may be unstable running multithreaded in a chroot "
 280                      "environment on Linux when /proc filesystem is not mounted.";
 281 
 282 void os::Linux::initialize_system_info() {
 283   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 284   if (processor_count() == 1) {
 285     pid_t pid = os::Linux::gettid();
 286     char fname[32];
 287     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 288     FILE *fp = fopen(fname, "r");
 289     if (fp == NULL) {
 290       unsafe_chroot_detected = true;
 291     } else {
 292       fclose(fp);
 293     }
 294   }
 295   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 296   assert(processor_count() > 0, "linux error");
 297 }
 298 
 299 void os::init_system_properties_values() {
 300   // The next steps are taken in the product version:
 301   //
 302   // Obtain the JAVA_HOME value from the location of libjvm.so.
 303   // This library should be located at:
 304   // <JAVA_HOME>/lib/{client|server}/libjvm.so.
 305   //
 306   // If "/jre/lib/" appears at the right place in the path, then we
 307   // assume libjvm.so is installed in a JDK and we use this path.
 308   //
 309   // Otherwise exit with message: "Could not create the Java virtual machine."
 310   //
 311   // The following extra steps are taken in the debugging version:
 312   //
 313   // If "/jre/lib/" does NOT appear at the right place in the path
 314   // instead of exit check for $JAVA_HOME environment variable.
 315   //
 316   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 317   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 318   // it looks like libjvm.so is installed there
 319   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 320   //
 321   // Otherwise exit.
 322   //
 323   // Important note: if the location of libjvm.so changes this
 324   // code needs to be changed accordingly.
 325 
 326   // See ld(1):
 327   //      The linker uses the following search paths to locate required
 328   //      shared libraries:
 329   //        1: ...
 330   //        ...
 331   //        7: The default directories, normally /lib and /usr/lib.
 332 #ifndef OVERRIDE_LIBPATH
 333   #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
 334     #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 335   #else
 336     #define DEFAULT_LIBPATH "/lib:/usr/lib"
 337   #endif
 338 #else
 339   #define DEFAULT_LIBPATH OVERRIDE_LIBPATH
 340 #endif
 341 
 342 // Base path of extensions installed on the system.
 343 #define SYS_EXT_DIR     "/usr/java/packages"
 344 #define EXTENSIONS_DIR  "/lib/ext"
 345 
 346   // Buffer that fits several sprintfs.
 347   // Note that the space for the colon and the trailing null are provided
 348   // by the nulls included by the sizeof operator.
 349   const size_t bufsize =
 350     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 351          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 352   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 353 
 354   // sysclasspath, java_home, dll_dir
 355   {
 356     char *pslash;
 357     os::jvm_path(buf, bufsize);
 358 
 359     // Found the full path to libjvm.so.
 360     // Now cut the path to <java_home>/jre if we can.
 361     pslash = strrchr(buf, '/');
 362     if (pslash != NULL) {
 363       *pslash = '\0';            // Get rid of /libjvm.so.
 364     }
 365     pslash = strrchr(buf, '/');
 366     if (pslash != NULL) {
 367       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 368     }
 369     Arguments::set_dll_dir(buf);
 370 
 371     if (pslash != NULL) {
 372       pslash = strrchr(buf, '/');
 373       if (pslash != NULL) {
 374         *pslash = '\0';        // Get rid of /lib.
 375       }
 376     }
 377     Arguments::set_java_home(buf);
 378     if (!set_boot_path('/', ':')) {
 379       vm_exit_during_initialization("Failed setting boot class path.", NULL);
 380     }
 381   }
 382 
 383   // Where to look for native libraries.
 384   //
 385   // Note: Due to a legacy implementation, most of the library path
 386   // is set in the launcher. This was to accomodate linking restrictions
 387   // on legacy Linux implementations (which are no longer supported).
 388   // Eventually, all the library path setting will be done here.
 389   //
 390   // However, to prevent the proliferation of improperly built native
 391   // libraries, the new path component /usr/java/packages is added here.
 392   // Eventually, all the library path setting will be done here.
 393   {
 394     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 395     // should always exist (until the legacy problem cited above is
 396     // addressed).
 397     const char *v = ::getenv("LD_LIBRARY_PATH");
 398     const char *v_colon = ":";
 399     if (v == NULL) { v = ""; v_colon = ""; }
 400     // That's +1 for the colon and +1 for the trailing '\0'.
 401     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 402                                                      strlen(v) + 1 +
 403                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
 404                                                      mtInternal);
 405     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
 406     Arguments::set_library_path(ld_library_path);
 407     FREE_C_HEAP_ARRAY(char, ld_library_path);
 408   }
 409 
 410   // Extensions directories.
 411   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 412   Arguments::set_ext_dirs(buf);
 413 
 414   FREE_C_HEAP_ARRAY(char, buf);
 415 
 416 #undef DEFAULT_LIBPATH
 417 #undef SYS_EXT_DIR
 418 #undef EXTENSIONS_DIR
 419 }
 420 
 421 ////////////////////////////////////////////////////////////////////////////////
 422 // breakpoint support
 423 
 424 void os::breakpoint() {
 425   BREAKPOINT;
 426 }
 427 
 428 extern "C" void breakpoint() {
 429   // use debugger to set breakpoint here
 430 }
 431 
 432 ////////////////////////////////////////////////////////////////////////////////
 433 // signal support
 434 
 435 debug_only(static bool signal_sets_initialized = false);
 436 static sigset_t unblocked_sigs, vm_sigs;
 437 
 438 void os::Linux::signal_sets_init() {
 439   // Should also have an assertion stating we are still single-threaded.
 440   assert(!signal_sets_initialized, "Already initialized");
 441   // Fill in signals that are necessarily unblocked for all threads in
 442   // the VM. Currently, we unblock the following signals:
 443   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 444   //                         by -Xrs (=ReduceSignalUsage));
 445   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 446   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 447   // the dispositions or masks wrt these signals.
 448   // Programs embedding the VM that want to use the above signals for their
 449   // own purposes must, at this time, use the "-Xrs" option to prevent
 450   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 451   // (See bug 4345157, and other related bugs).
 452   // In reality, though, unblocking these signals is really a nop, since
 453   // these signals are not blocked by default.
 454   sigemptyset(&unblocked_sigs);
 455   sigaddset(&unblocked_sigs, SIGILL);
 456   sigaddset(&unblocked_sigs, SIGSEGV);
 457   sigaddset(&unblocked_sigs, SIGBUS);
 458   sigaddset(&unblocked_sigs, SIGFPE);
 459 #if defined(PPC64)
 460   sigaddset(&unblocked_sigs, SIGTRAP);
 461 #endif
 462   sigaddset(&unblocked_sigs, SR_signum);
 463 
 464   if (!ReduceSignalUsage) {
 465     if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 466       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 467     }
 468     if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 469       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 470     }
 471     if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 472       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 473     }
 474   }
 475   // Fill in signals that are blocked by all but the VM thread.
 476   sigemptyset(&vm_sigs);
 477   if (!ReduceSignalUsage) {
 478     sigaddset(&vm_sigs, BREAK_SIGNAL);
 479   }
 480   debug_only(signal_sets_initialized = true);
 481 
 482 }
 483 
 484 // These are signals that are unblocked while a thread is running Java.
 485 // (For some reason, they get blocked by default.)
 486 sigset_t* os::Linux::unblocked_signals() {
 487   assert(signal_sets_initialized, "Not initialized");
 488   return &unblocked_sigs;
 489 }
 490 
 491 // These are the signals that are blocked while a (non-VM) thread is
 492 // running Java. Only the VM thread handles these signals.
 493 sigset_t* os::Linux::vm_signals() {
 494   assert(signal_sets_initialized, "Not initialized");
 495   return &vm_sigs;
 496 }
 497 
 498 void os::Linux::hotspot_sigmask(Thread* thread) {
 499 
 500   //Save caller's signal mask before setting VM signal mask
 501   sigset_t caller_sigmask;
 502   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 503 
 504   OSThread* osthread = thread->osthread();
 505   osthread->set_caller_sigmask(caller_sigmask);
 506 
 507   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 508 
 509   if (!ReduceSignalUsage) {
 510     if (thread->is_VM_thread()) {
 511       // Only the VM thread handles BREAK_SIGNAL ...
 512       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 513     } else {
 514       // ... all other threads block BREAK_SIGNAL
 515       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 516     }
 517   }
 518 }
 519 
 520 //////////////////////////////////////////////////////////////////////////////
 521 // detecting pthread library
 522 
 523 void os::Linux::libpthread_init() {
 524   // Save glibc and pthread version strings.
 525 #if !defined(_CS_GNU_LIBC_VERSION) || \
 526     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 527   #error "glibc too old (< 2.3.2)"
 528 #endif
 529 
 530   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 531   assert(n > 0, "cannot retrieve glibc version");
 532   char *str = (char *)malloc(n, mtInternal);
 533   confstr(_CS_GNU_LIBC_VERSION, str, n);
 534   os::Linux::set_glibc_version(str);
 535 
 536   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 537   assert(n > 0, "cannot retrieve pthread version");
 538   str = (char *)malloc(n, mtInternal);
 539   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 540   os::Linux::set_libpthread_version(str);
 541 }
 542 
 543 /////////////////////////////////////////////////////////////////////////////
 544 // thread stack expansion
 545 
 546 // os::Linux::manually_expand_stack() takes care of expanding the thread
 547 // stack. Note that this is normally not needed: pthread stacks allocate
 548 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 549 // committed. Therefore it is not necessary to expand the stack manually.
 550 //
 551 // Manually expanding the stack was historically needed on LinuxThreads
 552 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 553 // it is kept to deal with very rare corner cases:
 554 //
 555 // For one, user may run the VM on an own implementation of threads
 556 // whose stacks are - like the old LinuxThreads - implemented using
 557 // mmap(MAP_GROWSDOWN).
 558 //
 559 // Also, this coding may be needed if the VM is running on the primordial
 560 // thread. Normally we avoid running on the primordial thread; however,
 561 // user may still invoke the VM on the primordial thread.
 562 //
 563 // The following historical comment describes the details about running
 564 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 565 
 566 
 567 // Force Linux kernel to expand current thread stack. If "bottom" is close
 568 // to the stack guard, caller should block all signals.
 569 //
 570 // MAP_GROWSDOWN:
 571 //   A special mmap() flag that is used to implement thread stacks. It tells
 572 //   kernel that the memory region should extend downwards when needed. This
 573 //   allows early versions of LinuxThreads to only mmap the first few pages
 574 //   when creating a new thread. Linux kernel will automatically expand thread
 575 //   stack as needed (on page faults).
 576 //
 577 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 578 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 579 //   region, it's hard to tell if the fault is due to a legitimate stack
 580 //   access or because of reading/writing non-exist memory (e.g. buffer
 581 //   overrun). As a rule, if the fault happens below current stack pointer,
 582 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 583 //   application (see Linux kernel fault.c).
 584 //
 585 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 586 //   stack overflow detection.
 587 //
 588 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 589 //   not use MAP_GROWSDOWN.
 590 //
 591 // To get around the problem and allow stack banging on Linux, we need to
 592 // manually expand thread stack after receiving the SIGSEGV.
 593 //
 594 // There are two ways to expand thread stack to address "bottom", we used
 595 // both of them in JVM before 1.5:
 596 //   1. adjust stack pointer first so that it is below "bottom", and then
 597 //      touch "bottom"
 598 //   2. mmap() the page in question
 599 //
 600 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 601 // if current sp is already near the lower end of page 101, and we need to
 602 // call mmap() to map page 100, it is possible that part of the mmap() frame
 603 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 604 // That will destroy the mmap() frame and cause VM to crash.
 605 //
 606 // The following code works by adjusting sp first, then accessing the "bottom"
 607 // page to force a page fault. Linux kernel will then automatically expand the
 608 // stack mapping.
 609 //
 610 // _expand_stack_to() assumes its frame size is less than page size, which
 611 // should always be true if the function is not inlined.
 612 
 613 static void NOINLINE _expand_stack_to(address bottom) {
 614   address sp;
 615   size_t size;
 616   volatile char *p;
 617 
 618   // Adjust bottom to point to the largest address within the same page, it
 619   // gives us a one-page buffer if alloca() allocates slightly more memory.
 620   bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
 621   bottom += os::Linux::page_size() - 1;
 622 
 623   // sp might be slightly above current stack pointer; if that's the case, we
 624   // will alloca() a little more space than necessary, which is OK. Don't use
 625   // os::current_stack_pointer(), as its result can be slightly below current
 626   // stack pointer, causing us to not alloca enough to reach "bottom".
 627   sp = (address)&sp;
 628 
 629   if (sp > bottom) {
 630     size = sp - bottom;
 631     p = (volatile char *)alloca(size);
 632     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 633     p[0] = '\0';
 634   }
 635 }
 636 
 637 void os::Linux::expand_stack_to(address bottom) {
 638   _expand_stack_to(bottom);
 639 }
 640 
 641 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 642   assert(t!=NULL, "just checking");
 643   assert(t->osthread()->expanding_stack(), "expand should be set");
 644   assert(t->stack_base() != NULL, "stack_base was not initialized");
 645 
 646   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 647     sigset_t mask_all, old_sigset;
 648     sigfillset(&mask_all);
 649     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 650     _expand_stack_to(addr);
 651     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 652     return true;
 653   }
 654   return false;
 655 }
 656 
 657 //////////////////////////////////////////////////////////////////////////////
 658 // create new thread
 659 
 660 // Thread start routine for all newly created threads
 661 static void *thread_native_entry(Thread *thread) {
 662 
 663   thread->record_stack_base_and_size();
 664 
 665   // Try to randomize the cache line index of hot stack frames.
 666   // This helps when threads of the same stack traces evict each other's
 667   // cache lines. The threads can be either from the same JVM instance, or
 668   // from different JVM instances. The benefit is especially true for
 669   // processors with hyperthreading technology.
 670   static int counter = 0;
 671   int pid = os::current_process_id();
 672   alloca(((pid ^ counter++) & 7) * 128);
 673 
 674   thread->initialize_thread_current();
 675 
 676   OSThread* osthread = thread->osthread();
 677   Monitor* sync = osthread->startThread_lock();
 678 
 679   osthread->set_thread_id(os::current_thread_id());
 680 
 681   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 682     os::current_thread_id(), (uintx) pthread_self());
 683 
 684   if (UseNUMA) {
 685     int lgrp_id = os::numa_get_group_id();
 686     if (lgrp_id != -1) {
 687       thread->set_lgrp_id(lgrp_id);
 688     }
 689   }
 690   // initialize signal mask for this thread
 691   os::Linux::hotspot_sigmask(thread);
 692 
 693   // initialize floating point control register
 694   os::Linux::init_thread_fpu_state();
 695 
 696   // handshaking with parent thread
 697   {
 698     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 699 
 700     // notify parent thread
 701     osthread->set_state(INITIALIZED);
 702     sync->notify_all();
 703 
 704     // wait until os::start_thread()
 705     while (osthread->get_state() == INITIALIZED) {
 706       sync->wait(Mutex::_no_safepoint_check_flag);
 707     }
 708   }
 709 
 710   assert(osthread->pthread_id() != 0, "pthread_id was not set as expected");
 711 
 712   // call one more level start routine
 713   thread->call_run();
 714 
 715   // Note: at this point the thread object may already have deleted itself.
 716   // Prevent dereferencing it from here on out.
 717   thread = NULL;
 718 
 719   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 720     os::current_thread_id(), (uintx) pthread_self());
 721 
 722   return 0;
 723 }
 724 
 725 bool os::create_thread(Thread* thread, ThreadType thr_type,
 726                        size_t req_stack_size) {
 727   assert(thread->osthread() == NULL, "caller responsible");
 728 
 729   // Allocate the OSThread object
 730   OSThread* osthread = new OSThread(NULL, NULL);
 731   if (osthread == NULL) {
 732     return false;
 733   }
 734 
 735   // set the correct thread state
 736   osthread->set_thread_type(thr_type);
 737 
 738   // Initial state is ALLOCATED but not INITIALIZED
 739   osthread->set_state(ALLOCATED);
 740 
 741   thread->set_osthread(osthread);
 742 
 743   // init thread attributes
 744   pthread_attr_t attr;
 745   pthread_attr_init(&attr);
 746   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 747 
 748   // Calculate stack size if it's not specified by caller.
 749   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 750   // In the Linux NPTL pthread implementation the guard size mechanism
 751   // is not implemented properly. The posix standard requires adding
 752   // the size of the guard pages to the stack size, instead Linux
 753   // takes the space out of 'stacksize'. Thus we adapt the requested
 754   // stack_size by the size of the guard pages to mimick proper
 755   // behaviour. However, be careful not to end up with a size
 756   // of zero due to overflow. Don't add the guard page in that case.
 757   size_t guard_size = os::Linux::default_guard_size(thr_type);
 758   if (stack_size <= SIZE_MAX - guard_size) {
 759     stack_size += guard_size;
 760   }
 761   assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
 762 
 763   int status = pthread_attr_setstacksize(&attr, stack_size);
 764   assert_status(status == 0, status, "pthread_attr_setstacksize");
 765 
 766   // Configure glibc guard page.
 767   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 768 
 769   ThreadState state;
 770 
 771   {
 772     pthread_t tid;
 773     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 774 
 775     char buf[64];
 776     if (ret == 0) {
 777       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 778         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 779     } else {
 780       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 781         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 782     }
 783 
 784     pthread_attr_destroy(&attr);
 785 
 786     if (ret != 0) {
 787       // Need to clean up stuff we've allocated so far
 788       thread->set_osthread(NULL);
 789       delete osthread;
 790       return false;
 791     }
 792 
 793     // Store pthread info into the OSThread
 794     osthread->set_pthread_id(tid);
 795 
 796     // Wait until child thread is either initialized or aborted
 797     {
 798       Monitor* sync_with_child = osthread->startThread_lock();
 799       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 800       while ((state = osthread->get_state()) == ALLOCATED) {
 801         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 802       }
 803     }
 804   }
 805 
 806   // Aborted due to thread limit being reached
 807   if (state == ZOMBIE) {
 808     thread->set_osthread(NULL);
 809     delete osthread;
 810     return false;
 811   }
 812 
 813   // The thread is returned suspended (in state INITIALIZED),
 814   // and is started higher up in the call chain
 815   assert(state == INITIALIZED, "race condition");
 816   return true;
 817 }
 818 
 819 /////////////////////////////////////////////////////////////////////////////
 820 // attach existing thread
 821 
 822 // bootstrap the main thread
 823 bool os::create_main_thread(JavaThread* thread) {
 824   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 825   return create_attached_thread(thread);
 826 }
 827 
 828 bool os::create_attached_thread(JavaThread* thread) {
 829 #ifdef ASSERT
 830   thread->verify_not_published();
 831 #endif
 832 
 833   // Allocate the OSThread object
 834   OSThread* osthread = new OSThread(NULL, NULL);
 835 
 836   if (osthread == NULL) {
 837     return false;
 838   }
 839 
 840   // Store pthread info into the OSThread
 841   osthread->set_thread_id(os::Linux::gettid());
 842   osthread->set_pthread_id(::pthread_self());
 843 
 844   // initialize floating point control register
 845   os::Linux::init_thread_fpu_state();
 846 
 847   // Initial thread state is RUNNABLE
 848   osthread->set_state(RUNNABLE);
 849 
 850   thread->set_osthread(osthread);
 851 
 852   if (UseNUMA) {
 853     int lgrp_id = os::numa_get_group_id();
 854     if (lgrp_id != -1) {
 855       thread->set_lgrp_id(lgrp_id);
 856     }
 857   }
 858 
 859   if (os::is_primordial_thread()) {
 860     // If current thread is primordial thread, its stack is mapped on demand,
 861     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 862     // the entire stack region to avoid SEGV in stack banging.
 863     // It is also useful to get around the heap-stack-gap problem on SuSE
 864     // kernel (see 4821821 for details). We first expand stack to the top
 865     // of yellow zone, then enable stack yellow zone (order is significant,
 866     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 867     // is no gap between the last two virtual memory regions.
 868 
 869     JavaThread *jt = (JavaThread *)thread;
 870     address addr = jt->stack_reserved_zone_base();
 871     assert(addr != NULL, "initialization problem?");
 872     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 873 
 874     osthread->set_expanding_stack();
 875     os::Linux::manually_expand_stack(jt, addr);
 876     osthread->clear_expanding_stack();
 877   }
 878 
 879   // initialize signal mask for this thread
 880   // and save the caller's signal mask
 881   os::Linux::hotspot_sigmask(thread);
 882 
 883   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 884     os::current_thread_id(), (uintx) pthread_self());
 885 
 886   return true;
 887 }
 888 
 889 void os::pd_start_thread(Thread* thread) {
 890   OSThread * osthread = thread->osthread();
 891   assert(osthread->get_state() != INITIALIZED, "just checking");
 892   Monitor* sync_with_child = osthread->startThread_lock();
 893   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 894   sync_with_child->notify();
 895 }
 896 
 897 // Free Linux resources related to the OSThread
 898 void os::free_thread(OSThread* osthread) {
 899   assert(osthread != NULL, "osthread not set");
 900 
 901   // We are told to free resources of the argument thread,
 902   // but we can only really operate on the current thread.
 903   assert(Thread::current()->osthread() == osthread,
 904          "os::free_thread but not current thread");
 905 
 906 #ifdef ASSERT
 907   sigset_t current;
 908   sigemptyset(&current);
 909   pthread_sigmask(SIG_SETMASK, NULL, &current);
 910   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 911 #endif
 912 
 913   // Restore caller's signal mask
 914   sigset_t sigmask = osthread->caller_sigmask();
 915   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 916 
 917   delete osthread;
 918 }
 919 
 920 //////////////////////////////////////////////////////////////////////////////
 921 // primordial thread
 922 
 923 // Check if current thread is the primordial thread, similar to Solaris thr_main.
 924 bool os::is_primordial_thread(void) {
 925   if (suppress_primordial_thread_resolution) {
 926     return false;
 927   }
 928   char dummy;
 929   // If called before init complete, thread stack bottom will be null.
 930   // Can be called if fatal error occurs before initialization.
 931   if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
 932   assert(os::Linux::initial_thread_stack_bottom() != NULL &&
 933          os::Linux::initial_thread_stack_size()   != 0,
 934          "os::init did not locate primordial thread's stack region");
 935   if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
 936       (address)&dummy < os::Linux::initial_thread_stack_bottom() +
 937                         os::Linux::initial_thread_stack_size()) {
 938     return true;
 939   } else {
 940     return false;
 941   }
 942 }
 943 
 944 // Find the virtual memory area that contains addr
 945 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 946   FILE *fp = fopen("/proc/self/maps", "r");
 947   if (fp) {
 948     address low, high;
 949     while (!feof(fp)) {
 950       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 951         if (low <= addr && addr < high) {
 952           if (vma_low)  *vma_low  = low;
 953           if (vma_high) *vma_high = high;
 954           fclose(fp);
 955           return true;
 956         }
 957       }
 958       for (;;) {
 959         int ch = fgetc(fp);
 960         if (ch == EOF || ch == (int)'\n') break;
 961       }
 962     }
 963     fclose(fp);
 964   }
 965   return false;
 966 }
 967 
 968 // Locate primordial thread stack. This special handling of primordial thread stack
 969 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 970 // bogus value for the primordial process thread. While the launcher has created
 971 // the VM in a new thread since JDK 6, we still have to allow for the use of the
 972 // JNI invocation API from a primordial thread.
 973 void os::Linux::capture_initial_stack(size_t max_size) {
 974 
 975   // max_size is either 0 (which means accept OS default for thread stacks) or
 976   // a user-specified value known to be at least the minimum needed. If we
 977   // are actually on the primordial thread we can make it appear that we have a
 978   // smaller max_size stack by inserting the guard pages at that location. But we
 979   // cannot do anything to emulate a larger stack than what has been provided by
 980   // the OS or threading library. In fact if we try to use a stack greater than
 981   // what is set by rlimit then we will crash the hosting process.
 982 
 983   // Maximum stack size is the easy part, get it from RLIMIT_STACK.
 984   // If this is "unlimited" then it will be a huge value.
 985   struct rlimit rlim;
 986   getrlimit(RLIMIT_STACK, &rlim);
 987   size_t stack_size = rlim.rlim_cur;
 988 
 989   // 6308388: a bug in ld.so will relocate its own .data section to the
 990   //   lower end of primordial stack; reduce ulimit -s value a little bit
 991   //   so we won't install guard page on ld.so's data section.
 992   //   But ensure we don't underflow the stack size - allow 1 page spare
 993   if (stack_size >= (size_t)(3 * page_size())) {
 994     stack_size -= 2 * page_size();
 995   }
 996 
 997   // Try to figure out where the stack base (top) is. This is harder.
 998   //
 999   // When an application is started, glibc saves the initial stack pointer in
1000   // a global variable "__libc_stack_end", which is then used by system
1001   // libraries. __libc_stack_end should be pretty close to stack top. The
1002   // variable is available since the very early days. However, because it is
1003   // a private interface, it could disappear in the future.
1004   //
1005   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1006   // to __libc_stack_end, it is very close to stack top, but isn't the real
1007   // stack top. Note that /proc may not exist if VM is running as a chroot
1008   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1009   // /proc/<pid>/stat could change in the future (though unlikely).
1010   //
1011   // We try __libc_stack_end first. If that doesn't work, look for
1012   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1013   // as a hint, which should work well in most cases.
1014 
1015   uintptr_t stack_start;
1016 
1017   // try __libc_stack_end first
1018   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1019   if (p && *p) {
1020     stack_start = *p;
1021   } else {
1022     // see if we can get the start_stack field from /proc/self/stat
1023     FILE *fp;
1024     int pid;
1025     char state;
1026     int ppid;
1027     int pgrp;
1028     int session;
1029     int nr;
1030     int tpgrp;
1031     unsigned long flags;
1032     unsigned long minflt;
1033     unsigned long cminflt;
1034     unsigned long majflt;
1035     unsigned long cmajflt;
1036     unsigned long utime;
1037     unsigned long stime;
1038     long cutime;
1039     long cstime;
1040     long prio;
1041     long nice;
1042     long junk;
1043     long it_real;
1044     uintptr_t start;
1045     uintptr_t vsize;
1046     intptr_t rss;
1047     uintptr_t rsslim;
1048     uintptr_t scodes;
1049     uintptr_t ecode;
1050     int i;
1051 
1052     // Figure what the primordial thread stack base is. Code is inspired
1053     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1054     // followed by command name surrounded by parentheses, state, etc.
1055     char stat[2048];
1056     int statlen;
1057 
1058     fp = fopen("/proc/self/stat", "r");
1059     if (fp) {
1060       statlen = fread(stat, 1, 2047, fp);
1061       stat[statlen] = '\0';
1062       fclose(fp);
1063 
1064       // Skip pid and the command string. Note that we could be dealing with
1065       // weird command names, e.g. user could decide to rename java launcher
1066       // to "java 1.4.2 :)", then the stat file would look like
1067       //                1234 (java 1.4.2 :)) R ... ...
1068       // We don't really need to know the command string, just find the last
1069       // occurrence of ")" and then start parsing from there. See bug 4726580.
1070       char * s = strrchr(stat, ')');
1071 
1072       i = 0;
1073       if (s) {
1074         // Skip blank chars
1075         do { s++; } while (s && isspace(*s));
1076 
1077 #define _UFM UINTX_FORMAT
1078 #define _DFM INTX_FORMAT
1079 
1080         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1081         //              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
1082         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,
1083                    &state,          // 3  %c
1084                    &ppid,           // 4  %d
1085                    &pgrp,           // 5  %d
1086                    &session,        // 6  %d
1087                    &nr,             // 7  %d
1088                    &tpgrp,          // 8  %d
1089                    &flags,          // 9  %lu
1090                    &minflt,         // 10 %lu
1091                    &cminflt,        // 11 %lu
1092                    &majflt,         // 12 %lu
1093                    &cmajflt,        // 13 %lu
1094                    &utime,          // 14 %lu
1095                    &stime,          // 15 %lu
1096                    &cutime,         // 16 %ld
1097                    &cstime,         // 17 %ld
1098                    &prio,           // 18 %ld
1099                    &nice,           // 19 %ld
1100                    &junk,           // 20 %ld
1101                    &it_real,        // 21 %ld
1102                    &start,          // 22 UINTX_FORMAT
1103                    &vsize,          // 23 UINTX_FORMAT
1104                    &rss,            // 24 INTX_FORMAT
1105                    &rsslim,         // 25 UINTX_FORMAT
1106                    &scodes,         // 26 UINTX_FORMAT
1107                    &ecode,          // 27 UINTX_FORMAT
1108                    &stack_start);   // 28 UINTX_FORMAT
1109       }
1110 
1111 #undef _UFM
1112 #undef _DFM
1113 
1114       if (i != 28 - 2) {
1115         assert(false, "Bad conversion from /proc/self/stat");
1116         // product mode - assume we are the primordial thread, good luck in the
1117         // embedded case.
1118         warning("Can't detect primordial thread stack location - bad conversion");
1119         stack_start = (uintptr_t) &rlim;
1120       }
1121     } else {
1122       // For some reason we can't open /proc/self/stat (for example, running on
1123       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1124       // most cases, so don't abort:
1125       warning("Can't detect primordial thread stack location - no /proc/self/stat");
1126       stack_start = (uintptr_t) &rlim;
1127     }
1128   }
1129 
1130   // Now we have a pointer (stack_start) very close to the stack top, the
1131   // next thing to do is to figure out the exact location of stack top. We
1132   // can find out the virtual memory area that contains stack_start by
1133   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1134   // and its upper limit is the real stack top. (again, this would fail if
1135   // running inside chroot, because /proc may not exist.)
1136 
1137   uintptr_t stack_top;
1138   address low, high;
1139   if (find_vma((address)stack_start, &low, &high)) {
1140     // success, "high" is the true stack top. (ignore "low", because initial
1141     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1142     stack_top = (uintptr_t)high;
1143   } else {
1144     // failed, likely because /proc/self/maps does not exist
1145     warning("Can't detect primordial thread stack location - find_vma failed");
1146     // best effort: stack_start is normally within a few pages below the real
1147     // stack top, use it as stack top, and reduce stack size so we won't put
1148     // guard page outside stack.
1149     stack_top = stack_start;
1150     stack_size -= 16 * page_size();
1151   }
1152 
1153   // stack_top could be partially down the page so align it
1154   stack_top = align_up(stack_top, page_size());
1155 
1156   // Allowed stack value is minimum of max_size and what we derived from rlimit
1157   if (max_size > 0) {
1158     _initial_thread_stack_size = MIN2(max_size, stack_size);
1159   } else {
1160     // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1161     // clamp it at 8MB as we do on Solaris
1162     _initial_thread_stack_size = MIN2(stack_size, 8*M);
1163   }
1164   _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1165   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1166 
1167   assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1168 
1169   if (log_is_enabled(Info, os, thread)) {
1170     // See if we seem to be on primordial process thread
1171     bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1172                       uintptr_t(&rlim) < stack_top;
1173 
1174     log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1175                          SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1176                          primordial ? "primordial" : "user", max_size / K,  _initial_thread_stack_size / K,
1177                          stack_top, intptr_t(_initial_thread_stack_bottom));
1178   }
1179 }
1180 
1181 ////////////////////////////////////////////////////////////////////////////////
1182 // time support
1183 
1184 #ifndef SUPPORTS_CLOCK_MONOTONIC
1185 #error "Build platform doesn't support clock_gettime and related functionality"
1186 #endif
1187 
1188 // Time since start-up in seconds to a fine granularity.
1189 // Used by VMSelfDestructTimer and the MemProfiler.
1190 double os::elapsedTime() {
1191 
1192   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1193 }
1194 
1195 jlong os::elapsed_counter() {
1196   return javaTimeNanos() - initial_time_count;
1197 }
1198 
1199 jlong os::elapsed_frequency() {
1200   return NANOSECS_PER_SEC; // nanosecond resolution
1201 }
1202 
1203 bool os::supports_vtime() { return true; }
1204 bool os::enable_vtime()   { return false; }
1205 bool os::vtime_enabled()  { return false; }
1206 
1207 double os::elapsedVTime() {
1208   struct rusage usage;
1209   int retval = getrusage(RUSAGE_THREAD, &usage);
1210   if (retval == 0) {
1211     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);
1212   } else {
1213     // better than nothing, but not much
1214     return elapsedTime();
1215   }
1216 }
1217 
1218 jlong os::javaTimeMillis() {
1219   timeval time;
1220   int status = gettimeofday(&time, NULL);
1221   assert(status != -1, "linux error");
1222   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1223 }
1224 
1225 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1226   timeval time;
1227   int status = gettimeofday(&time, NULL);
1228   assert(status != -1, "linux error");
1229   seconds = jlong(time.tv_sec);
1230   nanos = jlong(time.tv_usec) * 1000;
1231 }
1232 
1233 void os::Linux::fast_thread_clock_init() {
1234   if (!UseLinuxPosixThreadCPUClocks) {
1235     return;
1236   }
1237   clockid_t clockid;
1238   struct timespec tp;
1239   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1240       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1241 
1242   // Switch to using fast clocks for thread cpu time if
1243   // the clock_getres() returns 0 error code.
1244   // Note, that some kernels may support the current thread
1245   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1246   // returned by the pthread_getcpuclockid().
1247   // If the fast Posix clocks are supported then the clock_getres()
1248   // must return at least tp.tv_sec == 0 which means a resolution
1249   // better than 1 sec. This is extra check for reliability.
1250 
1251   if (pthread_getcpuclockid_func &&
1252       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1253       os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1254     _supports_fast_thread_cpu_time = true;
1255     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1256   }
1257 }
1258 
1259 jlong os::javaTimeNanos() {
1260   if (os::supports_monotonic_clock()) {
1261     struct timespec tp;
1262     int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1263     assert(status == 0, "gettime error");
1264     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1265     return result;
1266   } else {
1267     timeval time;
1268     int status = gettimeofday(&time, NULL);
1269     assert(status != -1, "linux error");
1270     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1271     return 1000 * usecs;
1272   }
1273 }
1274 
1275 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1276   if (os::supports_monotonic_clock()) {
1277     info_ptr->max_value = ALL_64_BITS;
1278 
1279     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1280     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1281     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1282   } else {
1283     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1284     info_ptr->max_value = ALL_64_BITS;
1285 
1286     // gettimeofday is a real time clock so it skips
1287     info_ptr->may_skip_backward = true;
1288     info_ptr->may_skip_forward = true;
1289   }
1290 
1291   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1292 }
1293 
1294 // Return the real, user, and system times in seconds from an
1295 // arbitrary fixed point in the past.
1296 bool os::getTimesSecs(double* process_real_time,
1297                       double* process_user_time,
1298                       double* process_system_time) {
1299   struct tms ticks;
1300   clock_t real_ticks = times(&ticks);
1301 
1302   if (real_ticks == (clock_t) (-1)) {
1303     return false;
1304   } else {
1305     double ticks_per_second = (double) clock_tics_per_sec;
1306     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1307     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1308     *process_real_time = ((double) real_ticks) / ticks_per_second;
1309 
1310     return true;
1311   }
1312 }
1313 
1314 
1315 char * os::local_time_string(char *buf, size_t buflen) {
1316   struct tm t;
1317   time_t long_time;
1318   time(&long_time);
1319   localtime_r(&long_time, &t);
1320   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1321                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1322                t.tm_hour, t.tm_min, t.tm_sec);
1323   return buf;
1324 }
1325 
1326 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1327   return localtime_r(clock, res);
1328 }
1329 
1330 ////////////////////////////////////////////////////////////////////////////////
1331 // runtime exit support
1332 
1333 // Note: os::shutdown() might be called very early during initialization, or
1334 // called from signal handler. Before adding something to os::shutdown(), make
1335 // sure it is async-safe and can handle partially initialized VM.
1336 void os::shutdown() {
1337 
1338   // allow PerfMemory to attempt cleanup of any persistent resources
1339   perfMemory_exit();
1340 
1341   // needs to remove object in file system
1342   AttachListener::abort();
1343 
1344   // flush buffered output, finish log files
1345   ostream_abort();
1346 
1347   // Check for abort hook
1348   abort_hook_t abort_hook = Arguments::abort_hook();
1349   if (abort_hook != NULL) {
1350     abort_hook();
1351   }
1352 
1353 }
1354 
1355 // Note: os::abort() might be called very early during initialization, or
1356 // called from signal handler. Before adding something to os::abort(), make
1357 // sure it is async-safe and can handle partially initialized VM.
1358 void os::abort(bool dump_core, void* siginfo, const void* context) {
1359   os::shutdown();
1360   if (dump_core) {
1361 #if INCLUDE_CDS
1362     if (UseSharedSpaces && DumpPrivateMappingsInCore) {
1363       ClassLoader::close_jrt_image();
1364     }
1365 #endif
1366 #ifndef PRODUCT
1367     fdStream out(defaultStream::output_fd());
1368     out.print_raw("Current thread is ");
1369     char buf[16];
1370     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1371     out.print_raw_cr(buf);
1372     out.print_raw_cr("Dumping core ...");
1373 #endif
1374     ::abort(); // dump core
1375   }
1376 
1377   ::exit(1);
1378 }
1379 
1380 // Die immediately, no exit hook, no abort hook, no cleanup.
1381 void os::die() {
1382   ::abort();
1383 }
1384 
1385 // thread_id is kernel thread id (similar to Solaris LWP id)
1386 intx os::current_thread_id() { return os::Linux::gettid(); }
1387 int os::current_process_id() {
1388   return ::getpid();
1389 }
1390 
1391 // DLL functions
1392 
1393 const char* os::dll_file_extension() { return ".so"; }
1394 
1395 // This must be hard coded because it's the system's temporary
1396 // directory not the java application's temp directory, ala java.io.tmpdir.
1397 const char* os::get_temp_directory() { return "/tmp"; }
1398 
1399 static bool file_exists(const char* filename) {
1400   struct stat statbuf;
1401   if (filename == NULL || strlen(filename) == 0) {
1402     return false;
1403   }
1404   return os::stat(filename, &statbuf) == 0;
1405 }
1406 
1407 // check if addr is inside libjvm.so
1408 bool os::address_is_in_vm(address addr) {
1409   static address libjvm_base_addr;
1410   Dl_info dlinfo;
1411 
1412   if (libjvm_base_addr == NULL) {
1413     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1414       libjvm_base_addr = (address)dlinfo.dli_fbase;
1415     }
1416     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1417   }
1418 
1419   if (dladdr((void *)addr, &dlinfo) != 0) {
1420     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1421   }
1422 
1423   return false;
1424 }
1425 
1426 bool os::dll_address_to_function_name(address addr, char *buf,
1427                                       int buflen, int *offset,
1428                                       bool demangle) {
1429   // buf is not optional, but offset is optional
1430   assert(buf != NULL, "sanity check");
1431 
1432   Dl_info dlinfo;
1433 
1434   if (dladdr((void*)addr, &dlinfo) != 0) {
1435     // see if we have a matching symbol
1436     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1437       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1438         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1439       }
1440       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1441       return true;
1442     }
1443     // no matching symbol so try for just file info
1444     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1445       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1446                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1447         return true;
1448       }
1449     }
1450   }
1451 
1452   buf[0] = '\0';
1453   if (offset != NULL) *offset = -1;
1454   return false;
1455 }
1456 
1457 struct _address_to_library_name {
1458   address addr;          // input : memory address
1459   size_t  buflen;        //         size of fname
1460   char*   fname;         // output: library name
1461   address base;          //         library base addr
1462 };
1463 
1464 static int address_to_library_name_callback(struct dl_phdr_info *info,
1465                                             size_t size, void *data) {
1466   int i;
1467   bool found = false;
1468   address libbase = NULL;
1469   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1470 
1471   // iterate through all loadable segments
1472   for (i = 0; i < info->dlpi_phnum; i++) {
1473     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1474     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1475       // base address of a library is the lowest address of its loaded
1476       // segments.
1477       if (libbase == NULL || libbase > segbase) {
1478         libbase = segbase;
1479       }
1480       // see if 'addr' is within current segment
1481       if (segbase <= d->addr &&
1482           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1483         found = true;
1484       }
1485     }
1486   }
1487 
1488   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1489   // so dll_address_to_library_name() can fall through to use dladdr() which
1490   // can figure out executable name from argv[0].
1491   if (found && info->dlpi_name && info->dlpi_name[0]) {
1492     d->base = libbase;
1493     if (d->fname) {
1494       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1495     }
1496     return 1;
1497   }
1498   return 0;
1499 }
1500 
1501 bool os::dll_address_to_library_name(address addr, char* buf,
1502                                      int buflen, int* offset) {
1503   // buf is not optional, but offset is optional
1504   assert(buf != NULL, "sanity check");
1505 
1506   Dl_info dlinfo;
1507   struct _address_to_library_name data;
1508 
1509   // There is a bug in old glibc dladdr() implementation that it could resolve
1510   // to wrong library name if the .so file has a base address != NULL. Here
1511   // we iterate through the program headers of all loaded libraries to find
1512   // out which library 'addr' really belongs to. This workaround can be
1513   // removed once the minimum requirement for glibc is moved to 2.3.x.
1514   data.addr = addr;
1515   data.fname = buf;
1516   data.buflen = buflen;
1517   data.base = NULL;
1518   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1519 
1520   if (rslt) {
1521     // buf already contains library name
1522     if (offset) *offset = addr - data.base;
1523     return true;
1524   }
1525   if (dladdr((void*)addr, &dlinfo) != 0) {
1526     if (dlinfo.dli_fname != NULL) {
1527       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1528     }
1529     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1530       *offset = addr - (address)dlinfo.dli_fbase;
1531     }
1532     return true;
1533   }
1534 
1535   buf[0] = '\0';
1536   if (offset) *offset = -1;
1537   return false;
1538 }
1539 
1540 // Loads .dll/.so and
1541 // in case of error it checks if .dll/.so was built for the
1542 // same architecture as Hotspot is running on
1543 
1544 
1545 // Remember the stack's state. The Linux dynamic linker will change
1546 // the stack to 'executable' at most once, so we must safepoint only once.
1547 bool os::Linux::_stack_is_executable = false;
1548 
1549 // VM operation that loads a library.  This is necessary if stack protection
1550 // of the Java stacks can be lost during loading the library.  If we
1551 // do not stop the Java threads, they can stack overflow before the stacks
1552 // are protected again.
1553 class VM_LinuxDllLoad: public VM_Operation {
1554  private:
1555   const char *_filename;
1556   char *_ebuf;
1557   int _ebuflen;
1558   void *_lib;
1559  public:
1560   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1561     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1562   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1563   void doit() {
1564     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1565     os::Linux::_stack_is_executable = true;
1566   }
1567   void* loaded_library() { return _lib; }
1568 };
1569 
1570 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1571   void * result = NULL;
1572   bool load_attempted = false;
1573 
1574   // Check whether the library to load might change execution rights
1575   // of the stack. If they are changed, the protection of the stack
1576   // guard pages will be lost. We need a safepoint to fix this.
1577   //
1578   // See Linux man page execstack(8) for more info.
1579   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1580     if (!ElfFile::specifies_noexecstack(filename)) {
1581       if (!is_init_completed()) {
1582         os::Linux::_stack_is_executable = true;
1583         // This is OK - No Java threads have been created yet, and hence no
1584         // stack guard pages to fix.
1585         //
1586         // Dynamic loader will make all stacks executable after
1587         // this function returns, and will not do that again.
1588         assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1589       } else {
1590         warning("You have loaded library %s which might have disabled stack guard. "
1591                 "The VM will try to fix the stack guard now.\n"
1592                 "It's highly recommended that you fix the library with "
1593                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1594                 filename);
1595 
1596         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1597         JavaThread *jt = JavaThread::current();
1598         if (jt->thread_state() != _thread_in_native) {
1599           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1600           // that requires ExecStack. Cannot enter safe point. Let's give up.
1601           warning("Unable to fix stack guard. Giving up.");
1602         } else {
1603           if (!LoadExecStackDllInVMThread) {
1604             // This is for the case where the DLL has an static
1605             // constructor function that executes JNI code. We cannot
1606             // load such DLLs in the VMThread.
1607             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1608           }
1609 
1610           ThreadInVMfromNative tiv(jt);
1611           debug_only(VMNativeEntryWrapper vew;)
1612 
1613           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1614           VMThread::execute(&op);
1615           if (LoadExecStackDllInVMThread) {
1616             result = op.loaded_library();
1617           }
1618           load_attempted = true;
1619         }
1620       }
1621     }
1622   }
1623 
1624   if (!load_attempted) {
1625     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1626   }
1627 
1628   if (result != NULL) {
1629     // Successful loading
1630     return result;
1631   }
1632 
1633   Elf32_Ehdr elf_head;
1634   int diag_msg_max_length=ebuflen-strlen(ebuf);
1635   char* diag_msg_buf=ebuf+strlen(ebuf);
1636 
1637   if (diag_msg_max_length==0) {
1638     // No more space in ebuf for additional diagnostics message
1639     return NULL;
1640   }
1641 
1642 
1643   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1644 
1645   if (file_descriptor < 0) {
1646     // Can't open library, report dlerror() message
1647     return NULL;
1648   }
1649 
1650   bool failed_to_read_elf_head=
1651     (sizeof(elf_head)!=
1652      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1653 
1654   ::close(file_descriptor);
1655   if (failed_to_read_elf_head) {
1656     // file i/o error - report dlerror() msg
1657     return NULL;
1658   }
1659 
1660   typedef struct {
1661     Elf32_Half    code;         // Actual value as defined in elf.h
1662     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1663     unsigned char elf_class;    // 32 or 64 bit
1664     unsigned char endianess;    // MSB or LSB
1665     char*         name;         // String representation
1666   } arch_t;
1667 
1668 #ifndef EM_486
1669   #define EM_486          6               /* Intel 80486 */
1670 #endif
1671 #ifndef EM_AARCH64
1672   #define EM_AARCH64    183               /* ARM AARCH64 */
1673 #endif
1674 
1675   static const arch_t arch_array[]={
1676     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1677     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1678     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1679     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1680     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1681     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1682     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1683     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1684 #if defined(VM_LITTLE_ENDIAN)
1685     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1686     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1687 #else
1688     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1689     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1690 #endif
1691     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1692     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1693     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1694     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1695     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1696     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1697     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1698     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1699   };
1700 
1701 #if  (defined IA32)
1702   static  Elf32_Half running_arch_code=EM_386;
1703 #elif   (defined AMD64) || (defined X32)
1704   static  Elf32_Half running_arch_code=EM_X86_64;
1705 #elif  (defined IA64)
1706   static  Elf32_Half running_arch_code=EM_IA_64;
1707 #elif  (defined __sparc) && (defined _LP64)
1708   static  Elf32_Half running_arch_code=EM_SPARCV9;
1709 #elif  (defined __sparc) && (!defined _LP64)
1710   static  Elf32_Half running_arch_code=EM_SPARC;
1711 #elif  (defined __powerpc64__)
1712   static  Elf32_Half running_arch_code=EM_PPC64;
1713 #elif  (defined __powerpc__)
1714   static  Elf32_Half running_arch_code=EM_PPC;
1715 #elif  (defined AARCH64)
1716   static  Elf32_Half running_arch_code=EM_AARCH64;
1717 #elif  (defined ARM)
1718   static  Elf32_Half running_arch_code=EM_ARM;
1719 #elif  (defined S390)
1720   static  Elf32_Half running_arch_code=EM_S390;
1721 #elif  (defined ALPHA)
1722   static  Elf32_Half running_arch_code=EM_ALPHA;
1723 #elif  (defined MIPSEL)
1724   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1725 #elif  (defined PARISC)
1726   static  Elf32_Half running_arch_code=EM_PARISC;
1727 #elif  (defined MIPS)
1728   static  Elf32_Half running_arch_code=EM_MIPS;
1729 #elif  (defined M68K)
1730   static  Elf32_Half running_arch_code=EM_68K;
1731 #elif  (defined SH)
1732   static  Elf32_Half running_arch_code=EM_SH;
1733 #else
1734     #error Method os::dll_load requires that one of following is defined:\
1735         AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1736 #endif
1737 
1738   // Identify compatability class for VM's architecture and library's architecture
1739   // Obtain string descriptions for architectures
1740 
1741   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1742   int running_arch_index=-1;
1743 
1744   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1745     if (running_arch_code == arch_array[i].code) {
1746       running_arch_index    = i;
1747     }
1748     if (lib_arch.code == arch_array[i].code) {
1749       lib_arch.compat_class = arch_array[i].compat_class;
1750       lib_arch.name         = arch_array[i].name;
1751     }
1752   }
1753 
1754   assert(running_arch_index != -1,
1755          "Didn't find running architecture code (running_arch_code) in arch_array");
1756   if (running_arch_index == -1) {
1757     // Even though running architecture detection failed
1758     // we may still continue with reporting dlerror() message
1759     return NULL;
1760   }
1761 
1762   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1763     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1764     return NULL;
1765   }
1766 
1767 #ifndef S390
1768   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1769     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1770     return NULL;
1771   }
1772 #endif // !S390
1773 
1774   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1775     if (lib_arch.name!=NULL) {
1776       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1777                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1778                  lib_arch.name, arch_array[running_arch_index].name);
1779     } else {
1780       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1781                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1782                  lib_arch.code,
1783                  arch_array[running_arch_index].name);
1784     }
1785   }
1786 
1787   return NULL;
1788 }
1789 
1790 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1791                                 int ebuflen) {
1792   void * result = ::dlopen(filename, RTLD_LAZY);
1793   if (result == NULL) {
1794     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1795     ebuf[ebuflen-1] = '\0';
1796   }
1797   return result;
1798 }
1799 
1800 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1801                                        int ebuflen) {
1802   void * result = NULL;
1803   if (LoadExecStackDllInVMThread) {
1804     result = dlopen_helper(filename, ebuf, ebuflen);
1805   }
1806 
1807   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1808   // library that requires an executable stack, or which does not have this
1809   // stack attribute set, dlopen changes the stack attribute to executable. The
1810   // read protection of the guard pages gets lost.
1811   //
1812   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1813   // may have been queued at the same time.
1814 
1815   if (!_stack_is_executable) {
1816     for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1817       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1818           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1819         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1820           warning("Attempt to reguard stack yellow zone failed.");
1821         }
1822       }
1823     }
1824   }
1825 
1826   return result;
1827 }
1828 
1829 void* os::dll_lookup(void* handle, const char* name) {
1830   void* res = dlsym(handle, name);
1831   return res;
1832 }
1833 
1834 void* os::get_default_process_handle() {
1835   return (void*)::dlopen(NULL, RTLD_LAZY);
1836 }
1837 
1838 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
1839   int fd = ::open(filename, O_RDONLY);
1840   if (fd == -1) {
1841     return false;
1842   }
1843 
1844   if (hdr != NULL) {
1845     st->print_cr("%s", hdr);
1846   }
1847 
1848   char buf[33];
1849   int bytes;
1850   buf[32] = '\0';
1851   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1852     st->print_raw(buf, bytes);
1853   }
1854 
1855   ::close(fd);
1856 
1857   return true;
1858 }
1859 
1860 #if defined(S390)
1861 // keywords_to_match - NULL terminated array of keywords
1862 static bool print_matching_lines_from_sysinfo_file(outputStream* st, const char* keywords_to_match[]) {
1863   const char* filename = "/proc/sysinfo";
1864   char* line = NULL;
1865   size_t length = 0;
1866   FILE* fp = fopen(filename, "r");
1867   if (fp == NULL) {
1868     return false;
1869   }
1870 
1871   st->print_cr("Virtualization information:");
1872   while (getline(&line, &length, fp) != -1) {
1873     int i = 0;
1874     while (keywords_to_match[i] != NULL) {
1875       if (strncmp(line, keywords_to_match[i], strlen(keywords_to_match[i])) == 0) {
1876         st->print("%s", line);
1877         break;
1878       }
1879       i++;
1880     }
1881   }
1882 
1883   free(line);
1884   fclose(fp);
1885 
1886   return true;
1887 }
1888 #endif
1889 
1890 void os::print_dll_info(outputStream *st) {
1891   st->print_cr("Dynamic libraries:");
1892 
1893   char fname[32];
1894   pid_t pid = os::Linux::gettid();
1895 
1896   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1897 
1898   if (!_print_ascii_file(fname, st)) {
1899     st->print("Can not get library information for pid = %d\n", pid);
1900   }
1901 }
1902 
1903 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1904   FILE *procmapsFile = NULL;
1905 
1906   // Open the procfs maps file for the current process
1907   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1908     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1909     char line[PATH_MAX + 100];
1910 
1911     // Read line by line from 'file'
1912     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1913       u8 base, top, offset, inode;
1914       char permissions[5];
1915       char device[6];
1916       char name[PATH_MAX + 1];
1917 
1918       // Parse fields from line
1919       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s",
1920              &base, &top, permissions, &offset, device, &inode, name);
1921 
1922       // Filter by device id '00:00' so that we only get file system mapped files.
1923       if (strcmp(device, "00:00") != 0) {
1924 
1925         // Call callback with the fields of interest
1926         if(callback(name, (address)base, (address)top, param)) {
1927           // Oops abort, callback aborted
1928           fclose(procmapsFile);
1929           return 1;
1930         }
1931       }
1932     }
1933     fclose(procmapsFile);
1934   }
1935   return 0;
1936 }
1937 
1938 void os::print_os_info_brief(outputStream* st) {
1939   os::Linux::print_distro_info(st);
1940 
1941   os::Posix::print_uname_info(st);
1942 
1943   os::Linux::print_libversion_info(st);
1944 
1945 }
1946 
1947 void os::print_os_info(outputStream* st) {
1948   st->print("OS:");
1949 
1950   os::Linux::print_distro_info(st);
1951 
1952   os::Posix::print_uname_info(st);
1953 
1954   // Print warning if unsafe chroot environment detected
1955   if (unsafe_chroot_detected) {
1956     st->print("WARNING!! ");
1957     st->print_cr("%s", unstable_chroot_error);
1958   }
1959 
1960   os::Linux::print_libversion_info(st);
1961 
1962   os::Posix::print_rlimit_info(st);
1963 
1964   os::Posix::print_load_average(st);
1965 
1966   os::Linux::print_full_memory_info(st);
1967 
1968   os::Linux::print_proc_sys_info(st);
1969 
1970   os::Linux::print_ld_preload_file(st);
1971 
1972   os::Linux::print_container_info(st);
1973 
1974   os::Linux::print_virtualization_info(st);
1975 }
1976 
1977 // Try to identify popular distros.
1978 // Most Linux distributions have a /etc/XXX-release file, which contains
1979 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1980 // file that also contains the OS version string. Some have more than one
1981 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1982 // /etc/redhat-release.), so the order is important.
1983 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1984 // their own specific XXX-release file as well as a redhat-release file.
1985 // Because of this the XXX-release file needs to be searched for before the
1986 // redhat-release file.
1987 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
1988 // search for redhat-release / SuSE-release needs to be before lsb-release.
1989 // Since the lsb-release file is the new standard it needs to be searched
1990 // before the older style release files.
1991 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
1992 // next to last resort.  The os-release file is a new standard that contains
1993 // distribution information and the system-release file seems to be an old
1994 // standard that has been replaced by the lsb-release and os-release files.
1995 // Searching for the debian_version file is the last resort.  It contains
1996 // an informative string like "6.0.6" or "wheezy/sid". Because of this
1997 // "Debian " is printed before the contents of the debian_version file.
1998 
1999 const char* distro_files[] = {
2000   "/etc/oracle-release",
2001   "/etc/mandriva-release",
2002   "/etc/mandrake-release",
2003   "/etc/sun-release",
2004   "/etc/redhat-release",
2005   "/etc/SuSE-release",
2006   "/etc/lsb-release",
2007   "/etc/turbolinux-release",
2008   "/etc/gentoo-release",
2009   "/etc/ltib-release",
2010   "/etc/angstrom-version",
2011   "/etc/system-release",
2012   "/etc/os-release",
2013   NULL };
2014 
2015 void os::Linux::print_distro_info(outputStream* st) {
2016   for (int i = 0;; i++) {
2017     const char* file = distro_files[i];
2018     if (file == NULL) {
2019       break;  // done
2020     }
2021     // If file prints, we found it.
2022     if (_print_ascii_file(file, st)) {
2023       return;
2024     }
2025   }
2026 
2027   if (file_exists("/etc/debian_version")) {
2028     st->print("Debian ");
2029     _print_ascii_file("/etc/debian_version", st);
2030   } else {
2031     st->print("Linux");
2032   }
2033   st->cr();
2034 }
2035 
2036 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2037   char buf[256];
2038   while (fgets(buf, sizeof(buf), fp)) {
2039     // Edit out extra stuff in expected format
2040     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2041       char* ptr = strstr(buf, "\"");  // the name is in quotes
2042       if (ptr != NULL) {
2043         ptr++; // go beyond first quote
2044         char* nl = strchr(ptr, '\"');
2045         if (nl != NULL) *nl = '\0';
2046         strncpy(distro, ptr, length);
2047       } else {
2048         ptr = strstr(buf, "=");
2049         ptr++; // go beyond equals then
2050         char* nl = strchr(ptr, '\n');
2051         if (nl != NULL) *nl = '\0';
2052         strncpy(distro, ptr, length);
2053       }
2054       return;
2055     } else if (get_first_line) {
2056       char* nl = strchr(buf, '\n');
2057       if (nl != NULL) *nl = '\0';
2058       strncpy(distro, buf, length);
2059       return;
2060     }
2061   }
2062   // print last line and close
2063   char* nl = strchr(buf, '\n');
2064   if (nl != NULL) *nl = '\0';
2065   strncpy(distro, buf, length);
2066 }
2067 
2068 static void parse_os_info(char* distro, size_t length, const char* file) {
2069   FILE* fp = fopen(file, "r");
2070   if (fp != NULL) {
2071     // if suse format, print out first line
2072     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2073     parse_os_info_helper(fp, distro, length, get_first_line);
2074     fclose(fp);
2075   }
2076 }
2077 
2078 void os::get_summary_os_info(char* buf, size_t buflen) {
2079   for (int i = 0;; i++) {
2080     const char* file = distro_files[i];
2081     if (file == NULL) {
2082       break; // ran out of distro_files
2083     }
2084     if (file_exists(file)) {
2085       parse_os_info(buf, buflen, file);
2086       return;
2087     }
2088   }
2089   // special case for debian
2090   if (file_exists("/etc/debian_version")) {
2091     strncpy(buf, "Debian ", buflen);
2092     if (buflen > 7) {
2093       parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2094     }
2095   } else {
2096     strncpy(buf, "Linux", buflen);
2097   }
2098 }
2099 
2100 void os::Linux::print_libversion_info(outputStream* st) {
2101   // libc, pthread
2102   st->print("libc:");
2103   st->print("%s ", os::Linux::glibc_version());
2104   st->print("%s ", os::Linux::libpthread_version());
2105   st->cr();
2106 }
2107 
2108 void os::Linux::print_proc_sys_info(outputStream* st) {
2109   st->cr();
2110   st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2111   _print_ascii_file("/proc/sys/kernel/threads-max", st);
2112   st->cr();
2113   st->cr();
2114 
2115   st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2116   _print_ascii_file("/proc/sys/vm/max_map_count", st);
2117   st->cr();
2118   st->cr();
2119 
2120   st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2121   _print_ascii_file("/proc/sys/kernel/pid_max", st);
2122   st->cr();
2123   st->cr();
2124 }
2125 
2126 void os::Linux::print_full_memory_info(outputStream* st) {
2127   st->print("\n/proc/meminfo:\n");
2128   _print_ascii_file("/proc/meminfo", st);
2129   st->cr();
2130 }
2131 
2132 void os::Linux::print_ld_preload_file(outputStream* st) {
2133   _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2134   st->cr();
2135 }
2136 
2137 void os::Linux::print_container_info(outputStream* st) {
2138   if (!OSContainer::is_containerized()) {
2139     return;
2140   }
2141 
2142   st->print("container (cgroup) information:\n");
2143 
2144   const char *p_ct = OSContainer::container_type();
2145   st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2146 
2147   char *p = OSContainer::cpu_cpuset_cpus();
2148   st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2149   free(p);
2150 
2151   p = OSContainer::cpu_cpuset_memory_nodes();
2152   st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2153   free(p);
2154 
2155   int i = OSContainer::active_processor_count();
2156   if (i > 0) {
2157     st->print("active_processor_count: %d\n", i);
2158   } else {
2159     st->print("active_processor_count: failed\n");
2160   }
2161 
2162   i = OSContainer::cpu_quota();
2163   st->print("cpu_quota: %d\n", i);
2164 
2165   i = OSContainer::cpu_period();
2166   st->print("cpu_period: %d\n", i);
2167 
2168   i = OSContainer::cpu_shares();
2169   st->print("cpu_shares: %d\n", i);
2170 
2171   jlong j = OSContainer::memory_limit_in_bytes();
2172   st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2173 
2174   j = OSContainer::memory_and_swap_limit_in_bytes();
2175   st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2176 
2177   j = OSContainer::memory_soft_limit_in_bytes();
2178   st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2179 
2180   j = OSContainer::OSContainer::memory_usage_in_bytes();
2181   st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2182 
2183   j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2184   st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2185   st->cr();
2186 }
2187 
2188 void os::Linux::print_virtualization_info(outputStream* st) {
2189 #if defined(S390)
2190   // /proc/sysinfo contains interesting information about
2191   // - LPAR
2192   // - whole "Box" (CPUs )
2193   // - z/VM / KVM (VM<nn>); this is not available in an LPAR-only setup
2194   const char* kw[] = { "LPAR", "CPUs", "VM", NULL };
2195 
2196   if (! print_matching_lines_from_sysinfo_file(st, kw)) {
2197     st->print_cr("  </proc/sysinfo Not Available>");
2198   }
2199 #endif
2200 }
2201 
2202 void os::print_memory_info(outputStream* st) {
2203 
2204   st->print("Memory:");
2205   st->print(" %dk page", os::vm_page_size()>>10);
2206 
2207   // values in struct sysinfo are "unsigned long"
2208   struct sysinfo si;
2209   sysinfo(&si);
2210 
2211   st->print(", physical " UINT64_FORMAT "k",
2212             os::physical_memory() >> 10);
2213   st->print("(" UINT64_FORMAT "k free)",
2214             os::available_memory() >> 10);
2215   st->print(", swap " UINT64_FORMAT "k",
2216             ((jlong)si.totalswap * si.mem_unit) >> 10);
2217   st->print("(" UINT64_FORMAT "k free)",
2218             ((jlong)si.freeswap * si.mem_unit) >> 10);
2219   st->cr();
2220 }
2221 
2222 // Print the first "model name" line and the first "flags" line
2223 // that we find and nothing more. We assume "model name" comes
2224 // before "flags" so if we find a second "model name", then the
2225 // "flags" field is considered missing.
2226 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2227 #if defined(IA32) || defined(AMD64)
2228   // Other platforms have less repetitive cpuinfo files
2229   FILE *fp = fopen("/proc/cpuinfo", "r");
2230   if (fp) {
2231     while (!feof(fp)) {
2232       if (fgets(buf, buflen, fp)) {
2233         // Assume model name comes before flags
2234         bool model_name_printed = false;
2235         if (strstr(buf, "model name") != NULL) {
2236           if (!model_name_printed) {
2237             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2238             st->print_raw(buf);
2239             model_name_printed = true;
2240           } else {
2241             // model name printed but not flags?  Odd, just return
2242             fclose(fp);
2243             return true;
2244           }
2245         }
2246         // print the flags line too
2247         if (strstr(buf, "flags") != NULL) {
2248           st->print_raw(buf);
2249           fclose(fp);
2250           return true;
2251         }
2252       }
2253     }
2254     fclose(fp);
2255   }
2256 #endif // x86 platforms
2257   return false;
2258 }
2259 
2260 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2261   // Only print the model name if the platform provides this as a summary
2262   if (!print_model_name_and_flags(st, buf, buflen)) {
2263     st->print("\n/proc/cpuinfo:\n");
2264     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2265       st->print_cr("  <Not Available>");
2266     }
2267   }
2268 }
2269 
2270 #if defined(AMD64) || defined(IA32) || defined(X32)
2271 const char* search_string = "model name";
2272 #elif defined(M68K)
2273 const char* search_string = "CPU";
2274 #elif defined(PPC64)
2275 const char* search_string = "cpu";
2276 #elif defined(S390)
2277 const char* search_string = "processor";
2278 #elif defined(SPARC)
2279 const char* search_string = "cpu";
2280 #else
2281 const char* search_string = "Processor";
2282 #endif
2283 
2284 // Parses the cpuinfo file for string representing the model name.
2285 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2286   FILE* fp = fopen("/proc/cpuinfo", "r");
2287   if (fp != NULL) {
2288     while (!feof(fp)) {
2289       char buf[256];
2290       if (fgets(buf, sizeof(buf), fp)) {
2291         char* start = strstr(buf, search_string);
2292         if (start != NULL) {
2293           char *ptr = start + strlen(search_string);
2294           char *end = buf + strlen(buf);
2295           while (ptr != end) {
2296              // skip whitespace and colon for the rest of the name.
2297              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2298                break;
2299              }
2300              ptr++;
2301           }
2302           if (ptr != end) {
2303             // reasonable string, get rid of newline and keep the rest
2304             char* nl = strchr(buf, '\n');
2305             if (nl != NULL) *nl = '\0';
2306             strncpy(cpuinfo, ptr, length);
2307             fclose(fp);
2308             return;
2309           }
2310         }
2311       }
2312     }
2313     fclose(fp);
2314   }
2315   // cpuinfo not found or parsing failed, just print generic string.  The entire
2316   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2317 #if   defined(AARCH64)
2318   strncpy(cpuinfo, "AArch64", length);
2319 #elif defined(AMD64)
2320   strncpy(cpuinfo, "x86_64", length);
2321 #elif defined(ARM)  // Order wrt. AARCH64 is relevant!
2322   strncpy(cpuinfo, "ARM", length);
2323 #elif defined(IA32)
2324   strncpy(cpuinfo, "x86_32", length);
2325 #elif defined(IA64)
2326   strncpy(cpuinfo, "IA64", length);
2327 #elif defined(PPC)
2328   strncpy(cpuinfo, "PPC64", length);
2329 #elif defined(S390)
2330   strncpy(cpuinfo, "S390", length);
2331 #elif defined(SPARC)
2332   strncpy(cpuinfo, "sparcv9", length);
2333 #elif defined(ZERO_LIBARCH)
2334   strncpy(cpuinfo, ZERO_LIBARCH, length);
2335 #else
2336   strncpy(cpuinfo, "unknown", length);
2337 #endif
2338 }
2339 
2340 static void print_signal_handler(outputStream* st, int sig,
2341                                  char* buf, size_t buflen);
2342 
2343 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2344   st->print_cr("Signal Handlers:");
2345   print_signal_handler(st, SIGSEGV, buf, buflen);
2346   print_signal_handler(st, SIGBUS , buf, buflen);
2347   print_signal_handler(st, SIGFPE , buf, buflen);
2348   print_signal_handler(st, SIGPIPE, buf, buflen);
2349   print_signal_handler(st, SIGXFSZ, buf, buflen);
2350   print_signal_handler(st, SIGILL , buf, buflen);
2351   print_signal_handler(st, SR_signum, buf, buflen);
2352   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2353   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2354   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2355   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2356 #if defined(PPC64)
2357   print_signal_handler(st, SIGTRAP, buf, buflen);
2358 #endif
2359 }
2360 
2361 static char saved_jvm_path[MAXPATHLEN] = {0};
2362 
2363 // Find the full path to the current module, libjvm.so
2364 void os::jvm_path(char *buf, jint buflen) {
2365   // Error checking.
2366   if (buflen < MAXPATHLEN) {
2367     assert(false, "must use a large-enough buffer");
2368     buf[0] = '\0';
2369     return;
2370   }
2371   // Lazy resolve the path to current module.
2372   if (saved_jvm_path[0] != 0) {
2373     strcpy(buf, saved_jvm_path);
2374     return;
2375   }
2376 
2377   char dli_fname[MAXPATHLEN];
2378   bool ret = dll_address_to_library_name(
2379                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2380                                          dli_fname, sizeof(dli_fname), NULL);
2381   assert(ret, "cannot locate libjvm");
2382   char *rp = NULL;
2383   if (ret && dli_fname[0] != '\0') {
2384     rp = os::Posix::realpath(dli_fname, buf, buflen);
2385   }
2386   if (rp == NULL) {
2387     return;
2388   }
2389 
2390   if (Arguments::sun_java_launcher_is_altjvm()) {
2391     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2392     // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2393     // If "/jre/lib/" appears at the right place in the string, then
2394     // assume we are installed in a JDK and we're done. Otherwise, check
2395     // for a JAVA_HOME environment variable and fix up the path so it
2396     // looks like libjvm.so is installed there (append a fake suffix
2397     // hotspot/libjvm.so).
2398     const char *p = buf + strlen(buf) - 1;
2399     for (int count = 0; p > buf && count < 5; ++count) {
2400       for (--p; p > buf && *p != '/'; --p)
2401         /* empty */ ;
2402     }
2403 
2404     if (strncmp(p, "/jre/lib/", 9) != 0) {
2405       // Look for JAVA_HOME in the environment.
2406       char* java_home_var = ::getenv("JAVA_HOME");
2407       if (java_home_var != NULL && java_home_var[0] != 0) {
2408         char* jrelib_p;
2409         int len;
2410 
2411         // Check the current module name "libjvm.so".
2412         p = strrchr(buf, '/');
2413         if (p == NULL) {
2414           return;
2415         }
2416         assert(strstr(p, "/libjvm") == p, "invalid library name");
2417 
2418         rp = os::Posix::realpath(java_home_var, buf, buflen);
2419         if (rp == NULL) {
2420           return;
2421         }
2422 
2423         // determine if this is a legacy image or modules image
2424         // modules image doesn't have "jre" subdirectory
2425         len = strlen(buf);
2426         assert(len < buflen, "Ran out of buffer room");
2427         jrelib_p = buf + len;
2428         snprintf(jrelib_p, buflen-len, "/jre/lib");
2429         if (0 != access(buf, F_OK)) {
2430           snprintf(jrelib_p, buflen-len, "/lib");
2431         }
2432 
2433         if (0 == access(buf, F_OK)) {
2434           // Use current module name "libjvm.so"
2435           len = strlen(buf);
2436           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2437         } else {
2438           // Go back to path of .so
2439           rp = os::Posix::realpath(dli_fname, buf, buflen);
2440           if (rp == NULL) {
2441             return;
2442           }
2443         }
2444       }
2445     }
2446   }
2447 
2448   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2449   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2450 }
2451 
2452 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2453   // no prefix required, not even "_"
2454 }
2455 
2456 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2457   // no suffix required
2458 }
2459 
2460 ////////////////////////////////////////////////////////////////////////////////
2461 // sun.misc.Signal support
2462 
2463 static volatile jint sigint_count = 0;
2464 
2465 static void UserHandler(int sig, void *siginfo, void *context) {
2466   // 4511530 - sem_post is serialized and handled by the manager thread. When
2467   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2468   // don't want to flood the manager thread with sem_post requests.
2469   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2470     return;
2471   }
2472 
2473   // Ctrl-C is pressed during error reporting, likely because the error
2474   // handler fails to abort. Let VM die immediately.
2475   if (sig == SIGINT && VMError::is_error_reported()) {
2476     os::die();
2477   }
2478 
2479   os::signal_notify(sig);
2480 }
2481 
2482 void* os::user_handler() {
2483   return CAST_FROM_FN_PTR(void*, UserHandler);
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 static void jdk_misc_signal_init() {
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 jdk_misc_signal_init
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   int node;
2806   int 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 (is_node_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   int highest_node_number = Linux::numa_max_node();
2820   size_t i = 0;
2821 
2822   // Map all node ids in which it is possible to allocate memory. Also nodes are
2823   // not always consecutively available, i.e. available from 0 to the highest
2824   // node number. If the nodes have been bound explicitly using numactl membind,
2825   // then allocate memory from those nodes only.
2826   for (int node = 0; node <= highest_node_number; node++) {
2827     if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
2828       ids[i++] = node;
2829     }
2830   }
2831   return i;
2832 }
2833 
2834 bool os::get_page_info(char *start, page_info* info) {
2835   return false;
2836 }
2837 
2838 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2839                      page_info* page_found) {
2840   return end;
2841 }
2842 
2843 
2844 int os::Linux::sched_getcpu_syscall(void) {
2845   unsigned int cpu = 0;
2846   int retval = -1;
2847 
2848 #if defined(IA32)
2849   #ifndef SYS_getcpu
2850     #define SYS_getcpu 318
2851   #endif
2852   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2853 #elif defined(AMD64)
2854 // Unfortunately we have to bring all these macros here from vsyscall.h
2855 // to be able to compile on old linuxes.
2856   #define __NR_vgetcpu 2
2857   #define VSYSCALL_START (-10UL << 20)
2858   #define VSYSCALL_SIZE 1024
2859   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2860   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2861   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2862   retval = vgetcpu(&cpu, NULL, NULL);
2863 #endif
2864 
2865   return (retval == -1) ? retval : cpu;
2866 }
2867 
2868 void os::Linux::sched_getcpu_init() {
2869   // sched_getcpu() should be in libc.
2870   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2871                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2872 
2873   // If it's not, try a direct syscall.
2874   if (sched_getcpu() == -1) {
2875     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2876                                     (void*)&sched_getcpu_syscall));
2877   }
2878 
2879   if (sched_getcpu() == -1) {
2880     vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
2881   }
2882 }
2883 
2884 // Something to do with the numa-aware allocator needs these symbols
2885 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2886 extern "C" JNIEXPORT void numa_error(char *where) { }
2887 
2888 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2889 // load symbol from base version instead.
2890 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2891   void *f = dlvsym(handle, name, "libnuma_1.1");
2892   if (f == NULL) {
2893     f = dlsym(handle, name);
2894   }
2895   return f;
2896 }
2897 
2898 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2899 // Return NULL if the symbol is not defined in this particular version.
2900 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2901   return dlvsym(handle, name, "libnuma_1.2");
2902 }
2903 
2904 bool os::Linux::libnuma_init() {
2905   if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2906     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2907     if (handle != NULL) {
2908       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2909                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2910       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2911                                        libnuma_dlsym(handle, "numa_max_node")));
2912       set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2913                                                    libnuma_dlsym(handle, "numa_num_configured_nodes")));
2914       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2915                                         libnuma_dlsym(handle, "numa_available")));
2916       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2917                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2918       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2919                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2920       set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2921                                                 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2922       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2923                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2924       set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2925                                                libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2926       set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2927                                        libnuma_dlsym(handle, "numa_distance")));
2928       set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
2929                                           libnuma_v2_dlsym(handle, "numa_get_membind")));
2930       set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
2931                                                   libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
2932 
2933       if (numa_available() != -1) {
2934         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2935         set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2936         set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2937         set_numa_interleave_bitmask(_numa_get_interleave_mask());
2938         set_numa_membind_bitmask(_numa_get_membind());
2939         // Create an index -> node mapping, since nodes are not always consecutive
2940         _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2941         rebuild_nindex_to_node_map();
2942         // Create a cpu -> node mapping
2943         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2944         rebuild_cpu_to_node_map();
2945         return true;
2946       }
2947     }
2948   }
2949   return false;
2950 }
2951 
2952 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2953   // Creating guard page is very expensive. Java thread has HotSpot
2954   // guard pages, only enable glibc guard page for non-Java threads.
2955   // (Remember: compiler thread is a Java thread, too!)
2956   return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2957 }
2958 
2959 void os::Linux::rebuild_nindex_to_node_map() {
2960   int highest_node_number = Linux::numa_max_node();
2961 
2962   nindex_to_node()->clear();
2963   for (int node = 0; node <= highest_node_number; node++) {
2964     if (Linux::is_node_in_existing_nodes(node)) {
2965       nindex_to_node()->append(node);
2966     }
2967   }
2968 }
2969 
2970 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2971 // The table is later used in get_node_by_cpu().
2972 void os::Linux::rebuild_cpu_to_node_map() {
2973   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2974                               // in libnuma (possible values are starting from 16,
2975                               // and continuing up with every other power of 2, but less
2976                               // than the maximum number of CPUs supported by kernel), and
2977                               // is a subject to change (in libnuma version 2 the requirements
2978                               // are more reasonable) we'll just hardcode the number they use
2979                               // in the library.
2980   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2981 
2982   size_t cpu_num = processor_count();
2983   size_t cpu_map_size = NCPUS / BitsPerCLong;
2984   size_t cpu_map_valid_size =
2985     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2986 
2987   cpu_to_node()->clear();
2988   cpu_to_node()->at_grow(cpu_num - 1);
2989 
2990   size_t node_num = get_existing_num_nodes();
2991 
2992   int distance = 0;
2993   int closest_distance = INT_MAX;
2994   int closest_node = 0;
2995   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2996   for (size_t i = 0; i < node_num; i++) {
2997     // Check if node is configured (not a memory-less node). If it is not, find
2998     // the closest configured node. Check also if node is bound, i.e. it's allowed
2999     // to allocate memory from the node. If it's not allowed, map cpus in that node
3000     // to the closest node from which memory allocation is allowed.
3001     if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
3002         !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
3003       closest_distance = INT_MAX;
3004       // Check distance from all remaining nodes in the system. Ignore distance
3005       // from itself, from another non-configured node, and from another non-bound
3006       // node.
3007       for (size_t m = 0; m < node_num; m++) {
3008         if (m != i &&
3009             is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
3010             is_node_in_bound_nodes(nindex_to_node()->at(m))) {
3011           distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3012           // If a closest node is found, update. There is always at least one
3013           // configured and bound node in the system so there is always at least
3014           // one node close.
3015           if (distance != 0 && distance < closest_distance) {
3016             closest_distance = distance;
3017             closest_node = nindex_to_node()->at(m);
3018           }
3019         }
3020       }
3021      } else {
3022        // Current node is already a configured node.
3023        closest_node = nindex_to_node()->at(i);
3024      }
3025 
3026     // Get cpus from the original node and map them to the closest node. If node
3027     // is a configured node (not a memory-less node), then original node and
3028     // closest node are the same.
3029     if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3030       for (size_t j = 0; j < cpu_map_valid_size; j++) {
3031         if (cpu_map[j] != 0) {
3032           for (size_t k = 0; k < BitsPerCLong; k++) {
3033             if (cpu_map[j] & (1UL << k)) {
3034               cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3035             }
3036           }
3037         }
3038       }
3039     }
3040   }
3041   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3042 }
3043 
3044 int os::Linux::get_node_by_cpu(int cpu_id) {
3045   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3046     return cpu_to_node()->at(cpu_id);
3047   }
3048   return -1;
3049 }
3050 
3051 GrowableArray<int>* os::Linux::_cpu_to_node;
3052 GrowableArray<int>* os::Linux::_nindex_to_node;
3053 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3054 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3055 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3056 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3057 os::Linux::numa_available_func_t os::Linux::_numa_available;
3058 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3059 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3060 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3061 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3062 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3063 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3064 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3065 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3066 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy;
3067 unsigned long* os::Linux::_numa_all_nodes;
3068 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3069 struct bitmask* os::Linux::_numa_nodes_ptr;
3070 struct bitmask* os::Linux::_numa_interleave_bitmask;
3071 struct bitmask* os::Linux::_numa_membind_bitmask;
3072 
3073 bool os::pd_uncommit_memory(char* addr, size_t size) {
3074   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3075                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3076   return res  != (uintptr_t) MAP_FAILED;
3077 }
3078 
3079 static address get_stack_commited_bottom(address bottom, size_t size) {
3080   address nbot = bottom;
3081   address ntop = bottom + size;
3082 
3083   size_t page_sz = os::vm_page_size();
3084   unsigned pages = size / page_sz;
3085 
3086   unsigned char vec[1];
3087   unsigned imin = 1, imax = pages + 1, imid;
3088   int mincore_return_value = 0;
3089 
3090   assert(imin <= imax, "Unexpected page size");
3091 
3092   while (imin < imax) {
3093     imid = (imax + imin) / 2;
3094     nbot = ntop - (imid * page_sz);
3095 
3096     // Use a trick with mincore to check whether the page is mapped or not.
3097     // mincore sets vec to 1 if page resides in memory and to 0 if page
3098     // is swapped output but if page we are asking for is unmapped
3099     // it returns -1,ENOMEM
3100     mincore_return_value = mincore(nbot, page_sz, vec);
3101 
3102     if (mincore_return_value == -1) {
3103       // Page is not mapped go up
3104       // to find first mapped page
3105       if (errno != EAGAIN) {
3106         assert(errno == ENOMEM, "Unexpected mincore errno");
3107         imax = imid;
3108       }
3109     } else {
3110       // Page is mapped go down
3111       // to find first not mapped page
3112       imin = imid + 1;
3113     }
3114   }
3115 
3116   nbot = nbot + page_sz;
3117 
3118   // Adjust stack bottom one page up if last checked page is not mapped
3119   if (mincore_return_value == -1) {
3120     nbot = nbot + page_sz;
3121   }
3122 
3123   return nbot;
3124 }
3125 
3126 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3127   int mincore_return_value;
3128   const size_t stripe = 1024;  // query this many pages each time
3129   unsigned char vec[stripe + 1];
3130   // set a guard
3131   vec[stripe] = 'X';
3132 
3133   const size_t page_sz = os::vm_page_size();
3134   size_t pages = size / page_sz;
3135 
3136   assert(is_aligned(start, page_sz), "Start address must be page aligned");
3137   assert(is_aligned(size, page_sz), "Size must be page aligned");
3138 
3139   committed_start = NULL;
3140 
3141   int loops = (pages + stripe - 1) / stripe;
3142   int committed_pages = 0;
3143   address loop_base = start;
3144   bool found_range = false;
3145 
3146   for (int index = 0; index < loops && !found_range; index ++) {
3147     assert(pages > 0, "Nothing to do");
3148     int pages_to_query = (pages >= stripe) ? stripe : pages;
3149     pages -= pages_to_query;
3150 
3151     // Get stable read
3152     while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3153 
3154     // During shutdown, some memory goes away without properly notifying NMT,
3155     // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3156     // Bailout and return as not committed for now.
3157     if (mincore_return_value == -1 && errno == ENOMEM) {
3158       return false;
3159     }
3160 
3161     assert(vec[stripe] == 'X', "overflow guard");
3162     assert(mincore_return_value == 0, "Range must be valid");
3163     // Process this stripe
3164     for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3165       if ((vec[vecIdx] & 0x01) == 0) { // not committed
3166         // End of current contiguous region
3167         if (committed_start != NULL) {
3168           found_range = true;
3169           break;
3170         }
3171       } else { // committed
3172         // Start of region
3173         if (committed_start == NULL) {
3174           committed_start = loop_base + page_sz * vecIdx;
3175         }
3176         committed_pages ++;
3177       }
3178     }
3179 
3180     loop_base += pages_to_query * page_sz;
3181   }
3182 
3183   if (committed_start != NULL) {
3184     assert(committed_pages > 0, "Must have committed region");
3185     assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3186     assert(committed_start >= start && committed_start < start + size, "Out of range");
3187     committed_size = page_sz * committed_pages;
3188     return true;
3189   } else {
3190     assert(committed_pages == 0, "Should not have committed region");
3191     return false;
3192   }
3193 }
3194 
3195 
3196 // Linux uses a growable mapping for the stack, and if the mapping for
3197 // the stack guard pages is not removed when we detach a thread the
3198 // stack cannot grow beyond the pages where the stack guard was
3199 // mapped.  If at some point later in the process the stack expands to
3200 // that point, the Linux kernel cannot expand the stack any further
3201 // because the guard pages are in the way, and a segfault occurs.
3202 //
3203 // However, it's essential not to split the stack region by unmapping
3204 // a region (leaving a hole) that's already part of the stack mapping,
3205 // so if the stack mapping has already grown beyond the guard pages at
3206 // the time we create them, we have to truncate the stack mapping.
3207 // So, we need to know the extent of the stack mapping when
3208 // create_stack_guard_pages() is called.
3209 
3210 // We only need this for stacks that are growable: at the time of
3211 // writing thread stacks don't use growable mappings (i.e. those
3212 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3213 // only applies to the main thread.
3214 
3215 // If the (growable) stack mapping already extends beyond the point
3216 // where we're going to put our guard pages, truncate the mapping at
3217 // that point by munmap()ping it.  This ensures that when we later
3218 // munmap() the guard pages we don't leave a hole in the stack
3219 // mapping. This only affects the main/primordial thread
3220 
3221 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3222   if (os::is_primordial_thread()) {
3223     // As we manually grow stack up to bottom inside create_attached_thread(),
3224     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3225     // we don't need to do anything special.
3226     // Check it first, before calling heavy function.
3227     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3228     unsigned char vec[1];
3229 
3230     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3231       // Fallback to slow path on all errors, including EAGAIN
3232       stack_extent = (uintptr_t) get_stack_commited_bottom(
3233                                                            os::Linux::initial_thread_stack_bottom(),
3234                                                            (size_t)addr - stack_extent);
3235     }
3236 
3237     if (stack_extent < (uintptr_t)addr) {
3238       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3239     }
3240   }
3241 
3242   return os::commit_memory(addr, size, !ExecMem);
3243 }
3244 
3245 // If this is a growable mapping, remove the guard pages entirely by
3246 // munmap()ping them.  If not, just call uncommit_memory(). This only
3247 // affects the main/primordial thread, but guard against future OS changes.
3248 // It's safe to always unmap guard pages for primordial thread because we
3249 // always place it right after end of the mapped region.
3250 
3251 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3252   uintptr_t stack_extent, stack_base;
3253 
3254   if (os::is_primordial_thread()) {
3255     return ::munmap(addr, size) == 0;
3256   }
3257 
3258   return os::uncommit_memory(addr, size);
3259 }
3260 
3261 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3262 // at 'requested_addr'. If there are existing memory mappings at the same
3263 // location, however, they will be overwritten. If 'fixed' is false,
3264 // 'requested_addr' is only treated as a hint, the return value may or
3265 // may not start from the requested address. Unlike Linux mmap(), this
3266 // function returns NULL to indicate failure.
3267 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3268   char * addr;
3269   int flags;
3270 
3271   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3272   if (fixed) {
3273     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3274     flags |= MAP_FIXED;
3275   }
3276 
3277   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3278   // touch an uncommitted page. Otherwise, the read/write might
3279   // succeed if we have enough swap space to back the physical page.
3280   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3281                        flags, -1, 0);
3282 
3283   return addr == MAP_FAILED ? NULL : addr;
3284 }
3285 
3286 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3287 //   (req_addr != NULL) or with a given alignment.
3288 //  - bytes shall be a multiple of alignment.
3289 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3290 //  - alignment sets the alignment at which memory shall be allocated.
3291 //     It must be a multiple of allocation granularity.
3292 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3293 //  req_addr or NULL.
3294 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3295 
3296   size_t extra_size = bytes;
3297   if (req_addr == NULL && alignment > 0) {
3298     extra_size += alignment;
3299   }
3300 
3301   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3302     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3303     -1, 0);
3304   if (start == MAP_FAILED) {
3305     start = NULL;
3306   } else {
3307     if (req_addr != NULL) {
3308       if (start != req_addr) {
3309         ::munmap(start, extra_size);
3310         start = NULL;
3311       }
3312     } else {
3313       char* const start_aligned = align_up(start, alignment);
3314       char* const end_aligned = start_aligned + bytes;
3315       char* const end = start + extra_size;
3316       if (start_aligned > start) {
3317         ::munmap(start, start_aligned - start);
3318       }
3319       if (end_aligned < end) {
3320         ::munmap(end_aligned, end - end_aligned);
3321       }
3322       start = start_aligned;
3323     }
3324   }
3325   return start;
3326 }
3327 
3328 static int anon_munmap(char * addr, size_t size) {
3329   return ::munmap(addr, size) == 0;
3330 }
3331 
3332 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3333                             size_t alignment_hint) {
3334   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3335 }
3336 
3337 bool os::pd_release_memory(char* addr, size_t size) {
3338   return anon_munmap(addr, size);
3339 }
3340 
3341 static bool linux_mprotect(char* addr, size_t size, int prot) {
3342   // Linux wants the mprotect address argument to be page aligned.
3343   char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3344 
3345   // According to SUSv3, mprotect() should only be used with mappings
3346   // established by mmap(), and mmap() always maps whole pages. Unaligned
3347   // 'addr' likely indicates problem in the VM (e.g. trying to change
3348   // protection of malloc'ed or statically allocated memory). Check the
3349   // caller if you hit this assert.
3350   assert(addr == bottom, "sanity check");
3351 
3352   size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3353   return ::mprotect(bottom, size, prot) == 0;
3354 }
3355 
3356 // Set protections specified
3357 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3358                         bool is_committed) {
3359   unsigned int p = 0;
3360   switch (prot) {
3361   case MEM_PROT_NONE: p = PROT_NONE; break;
3362   case MEM_PROT_READ: p = PROT_READ; break;
3363   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3364   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3365   default:
3366     ShouldNotReachHere();
3367   }
3368   // is_committed is unused.
3369   return linux_mprotect(addr, bytes, p);
3370 }
3371 
3372 bool os::guard_memory(char* addr, size_t size) {
3373   return linux_mprotect(addr, size, PROT_NONE);
3374 }
3375 
3376 bool os::unguard_memory(char* addr, size_t size) {
3377   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3378 }
3379 
3380 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3381                                                     size_t page_size) {
3382   bool result = false;
3383   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3384                  MAP_ANONYMOUS|MAP_PRIVATE,
3385                  -1, 0);
3386   if (p != MAP_FAILED) {
3387     void *aligned_p = align_up(p, page_size);
3388 
3389     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3390 
3391     munmap(p, page_size * 2);
3392   }
3393 
3394   if (warn && !result) {
3395     warning("TransparentHugePages is not supported by the operating system.");
3396   }
3397 
3398   return result;
3399 }
3400 
3401 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3402   bool result = false;
3403   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3404                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3405                  -1, 0);
3406 
3407   if (p != MAP_FAILED) {
3408     // We don't know if this really is a huge page or not.
3409     FILE *fp = fopen("/proc/self/maps", "r");
3410     if (fp) {
3411       while (!feof(fp)) {
3412         char chars[257];
3413         long x = 0;
3414         if (fgets(chars, sizeof(chars), fp)) {
3415           if (sscanf(chars, "%lx-%*x", &x) == 1
3416               && x == (long)p) {
3417             if (strstr (chars, "hugepage")) {
3418               result = true;
3419               break;
3420             }
3421           }
3422         }
3423       }
3424       fclose(fp);
3425     }
3426     munmap(p, page_size);
3427   }
3428 
3429   if (warn && !result) {
3430     warning("HugeTLBFS is not supported by the operating system.");
3431   }
3432 
3433   return result;
3434 }
3435 
3436 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3437 //
3438 // From the coredump_filter documentation:
3439 //
3440 // - (bit 0) anonymous private memory
3441 // - (bit 1) anonymous shared memory
3442 // - (bit 2) file-backed private memory
3443 // - (bit 3) file-backed shared memory
3444 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3445 //           effective only if the bit 2 is cleared)
3446 // - (bit 5) hugetlb private memory
3447 // - (bit 6) hugetlb shared memory
3448 // - (bit 7) dax private memory
3449 // - (bit 8) dax shared memory
3450 //
3451 static void set_coredump_filter(CoredumpFilterBit bit) {
3452   FILE *f;
3453   long cdm;
3454 
3455   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3456     return;
3457   }
3458 
3459   if (fscanf(f, "%lx", &cdm) != 1) {
3460     fclose(f);
3461     return;
3462   }
3463 
3464   long saved_cdm = cdm;
3465   rewind(f);
3466   cdm |= bit;
3467 
3468   if (cdm != saved_cdm) {
3469     fprintf(f, "%#lx", cdm);
3470   }
3471 
3472   fclose(f);
3473 }
3474 
3475 // Large page support
3476 
3477 static size_t _large_page_size = 0;
3478 
3479 size_t os::Linux::find_large_page_size() {
3480   size_t large_page_size = 0;
3481 
3482   // large_page_size on Linux is used to round up heap size. x86 uses either
3483   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3484   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3485   // page as large as 256M.
3486   //
3487   // Here we try to figure out page size by parsing /proc/meminfo and looking
3488   // for a line with the following format:
3489   //    Hugepagesize:     2048 kB
3490   //
3491   // If we can't determine the value (e.g. /proc is not mounted, or the text
3492   // format has been changed), we'll use the largest page size supported by
3493   // the processor.
3494 
3495 #ifndef ZERO
3496   large_page_size =
3497     AARCH64_ONLY(2 * M)
3498     AMD64_ONLY(2 * M)
3499     ARM32_ONLY(2 * M)
3500     IA32_ONLY(4 * M)
3501     IA64_ONLY(256 * M)
3502     PPC_ONLY(4 * M)
3503     S390_ONLY(1 * M)
3504     SPARC_ONLY(4 * M);
3505 #endif // ZERO
3506 
3507   FILE *fp = fopen("/proc/meminfo", "r");
3508   if (fp) {
3509     while (!feof(fp)) {
3510       int x = 0;
3511       char buf[16];
3512       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3513         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3514           large_page_size = x * K;
3515           break;
3516         }
3517       } else {
3518         // skip to next line
3519         for (;;) {
3520           int ch = fgetc(fp);
3521           if (ch == EOF || ch == (int)'\n') break;
3522         }
3523       }
3524     }
3525     fclose(fp);
3526   }
3527 
3528   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3529     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3530             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3531             proper_unit_for_byte_size(large_page_size));
3532   }
3533 
3534   return large_page_size;
3535 }
3536 
3537 size_t os::Linux::setup_large_page_size() {
3538   _large_page_size = Linux::find_large_page_size();
3539   const size_t default_page_size = (size_t)Linux::page_size();
3540   if (_large_page_size > default_page_size) {
3541     _page_sizes[0] = _large_page_size;
3542     _page_sizes[1] = default_page_size;
3543     _page_sizes[2] = 0;
3544   }
3545 
3546   return _large_page_size;
3547 }
3548 
3549 bool os::Linux::setup_large_page_type(size_t page_size) {
3550   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3551       FLAG_IS_DEFAULT(UseSHM) &&
3552       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3553 
3554     // The type of large pages has not been specified by the user.
3555 
3556     // Try UseHugeTLBFS and then UseSHM.
3557     UseHugeTLBFS = UseSHM = true;
3558 
3559     // Don't try UseTransparentHugePages since there are known
3560     // performance issues with it turned on. This might change in the future.
3561     UseTransparentHugePages = false;
3562   }
3563 
3564   if (UseTransparentHugePages) {
3565     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3566     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3567       UseHugeTLBFS = false;
3568       UseSHM = false;
3569       return true;
3570     }
3571     UseTransparentHugePages = false;
3572   }
3573 
3574   if (UseHugeTLBFS) {
3575     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3576     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3577       UseSHM = false;
3578       return true;
3579     }
3580     UseHugeTLBFS = false;
3581   }
3582 
3583   return UseSHM;
3584 }
3585 
3586 void os::large_page_init() {
3587   if (!UseLargePages &&
3588       !UseTransparentHugePages &&
3589       !UseHugeTLBFS &&
3590       !UseSHM) {
3591     // Not using large pages.
3592     return;
3593   }
3594 
3595   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3596     // The user explicitly turned off large pages.
3597     // Ignore the rest of the large pages flags.
3598     UseTransparentHugePages = false;
3599     UseHugeTLBFS = false;
3600     UseSHM = false;
3601     return;
3602   }
3603 
3604   size_t large_page_size = Linux::setup_large_page_size();
3605   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3606 
3607   set_coredump_filter(LARGEPAGES_BIT);
3608 }
3609 
3610 #ifndef SHM_HUGETLB
3611   #define SHM_HUGETLB 04000
3612 #endif
3613 
3614 #define shm_warning_format(format, ...)              \
3615   do {                                               \
3616     if (UseLargePages &&                             \
3617         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3618          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3619          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3620       warning(format, __VA_ARGS__);                  \
3621     }                                                \
3622   } while (0)
3623 
3624 #define shm_warning(str) shm_warning_format("%s", str)
3625 
3626 #define shm_warning_with_errno(str)                \
3627   do {                                             \
3628     int err = errno;                               \
3629     shm_warning_format(str " (error = %d)", err);  \
3630   } while (0)
3631 
3632 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3633   assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3634 
3635   if (!is_aligned(alignment, SHMLBA)) {
3636     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3637     return NULL;
3638   }
3639 
3640   // To ensure that we get 'alignment' aligned memory from shmat,
3641   // we pre-reserve aligned virtual memory and then attach to that.
3642 
3643   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3644   if (pre_reserved_addr == NULL) {
3645     // Couldn't pre-reserve aligned memory.
3646     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3647     return NULL;
3648   }
3649 
3650   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3651   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3652 
3653   if ((intptr_t)addr == -1) {
3654     int err = errno;
3655     shm_warning_with_errno("Failed to attach shared memory.");
3656 
3657     assert(err != EACCES, "Unexpected error");
3658     assert(err != EIDRM,  "Unexpected error");
3659     assert(err != EINVAL, "Unexpected error");
3660 
3661     // Since we don't know if the kernel unmapped the pre-reserved memory area
3662     // we can't unmap it, since that would potentially unmap memory that was
3663     // mapped from other threads.
3664     return NULL;
3665   }
3666 
3667   return addr;
3668 }
3669 
3670 static char* shmat_at_address(int shmid, char* req_addr) {
3671   if (!is_aligned(req_addr, SHMLBA)) {
3672     assert(false, "Requested address needs to be SHMLBA aligned");
3673     return NULL;
3674   }
3675 
3676   char* addr = (char*)shmat(shmid, req_addr, 0);
3677 
3678   if ((intptr_t)addr == -1) {
3679     shm_warning_with_errno("Failed to attach shared memory.");
3680     return NULL;
3681   }
3682 
3683   return addr;
3684 }
3685 
3686 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3687   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3688   if (req_addr != NULL) {
3689     assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3690     assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3691     return shmat_at_address(shmid, req_addr);
3692   }
3693 
3694   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3695   // return large page size aligned memory addresses when req_addr == NULL.
3696   // However, if the alignment is larger than the large page size, we have
3697   // to manually ensure that the memory returned is 'alignment' aligned.
3698   if (alignment > os::large_page_size()) {
3699     assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3700     return shmat_with_alignment(shmid, bytes, alignment);
3701   } else {
3702     return shmat_at_address(shmid, NULL);
3703   }
3704 }
3705 
3706 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3707                                             char* req_addr, bool exec) {
3708   // "exec" is passed in but not used.  Creating the shared image for
3709   // the code cache doesn't have an SHM_X executable permission to check.
3710   assert(UseLargePages && UseSHM, "only for SHM large pages");
3711   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3712   assert(is_aligned(req_addr, alignment), "Unaligned address");
3713 
3714   if (!is_aligned(bytes, os::large_page_size())) {
3715     return NULL; // Fallback to small pages.
3716   }
3717 
3718   // Create a large shared memory region to attach to based on size.
3719   // Currently, size is the total size of the heap.
3720   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3721   if (shmid == -1) {
3722     // Possible reasons for shmget failure:
3723     // 1. shmmax is too small for Java heap.
3724     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3725     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3726     // 2. not enough large page memory.
3727     //    > check available large pages: cat /proc/meminfo
3728     //    > increase amount of large pages:
3729     //          echo new_value > /proc/sys/vm/nr_hugepages
3730     //      Note 1: different Linux may use different name for this property,
3731     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3732     //      Note 2: it's possible there's enough physical memory available but
3733     //            they are so fragmented after a long run that they can't
3734     //            coalesce into large pages. Try to reserve large pages when
3735     //            the system is still "fresh".
3736     shm_warning_with_errno("Failed to reserve shared memory.");
3737     return NULL;
3738   }
3739 
3740   // Attach to the region.
3741   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3742 
3743   // Remove shmid. If shmat() is successful, the actual shared memory segment
3744   // will be deleted when it's detached by shmdt() or when the process
3745   // terminates. If shmat() is not successful this will remove the shared
3746   // segment immediately.
3747   shmctl(shmid, IPC_RMID, NULL);
3748 
3749   return addr;
3750 }
3751 
3752 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3753                                         int error) {
3754   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3755 
3756   bool warn_on_failure = UseLargePages &&
3757       (!FLAG_IS_DEFAULT(UseLargePages) ||
3758        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3759        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3760 
3761   if (warn_on_failure) {
3762     char msg[128];
3763     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3764                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3765     warning("%s", msg);
3766   }
3767 }
3768 
3769 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3770                                                         char* req_addr,
3771                                                         bool exec) {
3772   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3773   assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3774   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3775 
3776   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3777   char* addr = (char*)::mmap(req_addr, bytes, prot,
3778                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3779                              -1, 0);
3780 
3781   if (addr == MAP_FAILED) {
3782     warn_on_large_pages_failure(req_addr, bytes, errno);
3783     return NULL;
3784   }
3785 
3786   assert(is_aligned(addr, os::large_page_size()), "Must be");
3787 
3788   return addr;
3789 }
3790 
3791 // Reserve memory using mmap(MAP_HUGETLB).
3792 //  - bytes shall be a multiple of alignment.
3793 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3794 //  - alignment sets the alignment at which memory shall be allocated.
3795 //     It must be a multiple of allocation granularity.
3796 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3797 //  req_addr or NULL.
3798 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3799                                                          size_t alignment,
3800                                                          char* req_addr,
3801                                                          bool exec) {
3802   size_t large_page_size = os::large_page_size();
3803   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3804 
3805   assert(is_aligned(req_addr, alignment), "Must be");
3806   assert(is_aligned(bytes, alignment), "Must be");
3807 
3808   // First reserve - but not commit - the address range in small pages.
3809   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3810 
3811   if (start == NULL) {
3812     return NULL;
3813   }
3814 
3815   assert(is_aligned(start, alignment), "Must be");
3816 
3817   char* end = start + bytes;
3818 
3819   // Find the regions of the allocated chunk that can be promoted to large pages.
3820   char* lp_start = align_up(start, large_page_size);
3821   char* lp_end   = align_down(end, large_page_size);
3822 
3823   size_t lp_bytes = lp_end - lp_start;
3824 
3825   assert(is_aligned(lp_bytes, large_page_size), "Must be");
3826 
3827   if (lp_bytes == 0) {
3828     // The mapped region doesn't even span the start and the end of a large page.
3829     // Fall back to allocate a non-special area.
3830     ::munmap(start, end - start);
3831     return NULL;
3832   }
3833 
3834   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3835 
3836   void* result;
3837 
3838   // Commit small-paged leading area.
3839   if (start != lp_start) {
3840     result = ::mmap(start, lp_start - start, prot,
3841                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3842                     -1, 0);
3843     if (result == MAP_FAILED) {
3844       ::munmap(lp_start, end - lp_start);
3845       return NULL;
3846     }
3847   }
3848 
3849   // Commit large-paged area.
3850   result = ::mmap(lp_start, lp_bytes, prot,
3851                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3852                   -1, 0);
3853   if (result == MAP_FAILED) {
3854     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3855     // If the mmap above fails, the large pages region will be unmapped and we
3856     // have regions before and after with small pages. Release these regions.
3857     //
3858     // |  mapped  |  unmapped  |  mapped  |
3859     // ^          ^            ^          ^
3860     // start      lp_start     lp_end     end
3861     //
3862     ::munmap(start, lp_start - start);
3863     ::munmap(lp_end, end - lp_end);
3864     return NULL;
3865   }
3866 
3867   // Commit small-paged trailing area.
3868   if (lp_end != end) {
3869     result = ::mmap(lp_end, end - lp_end, prot,
3870                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3871                     -1, 0);
3872     if (result == MAP_FAILED) {
3873       ::munmap(start, lp_end - start);
3874       return NULL;
3875     }
3876   }
3877 
3878   return start;
3879 }
3880 
3881 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3882                                                    size_t alignment,
3883                                                    char* req_addr,
3884                                                    bool exec) {
3885   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3886   assert(is_aligned(req_addr, alignment), "Must be");
3887   assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3888   assert(is_power_of_2(os::large_page_size()), "Must be");
3889   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3890 
3891   if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3892     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3893   } else {
3894     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3895   }
3896 }
3897 
3898 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3899                                  char* req_addr, bool exec) {
3900   assert(UseLargePages, "only for large pages");
3901 
3902   char* addr;
3903   if (UseSHM) {
3904     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3905   } else {
3906     assert(UseHugeTLBFS, "must be");
3907     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3908   }
3909 
3910   if (addr != NULL) {
3911     if (UseNUMAInterleaving) {
3912       numa_make_global(addr, bytes);
3913     }
3914 
3915     // The memory is committed
3916     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3917   }
3918 
3919   return addr;
3920 }
3921 
3922 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3923   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3924   return shmdt(base) == 0;
3925 }
3926 
3927 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3928   return pd_release_memory(base, bytes);
3929 }
3930 
3931 bool os::release_memory_special(char* base, size_t bytes) {
3932   bool res;
3933   if (MemTracker::tracking_level() > NMT_minimal) {
3934     Tracker tkr(Tracker::release);
3935     res = os::Linux::release_memory_special_impl(base, bytes);
3936     if (res) {
3937       tkr.record((address)base, bytes);
3938     }
3939 
3940   } else {
3941     res = os::Linux::release_memory_special_impl(base, bytes);
3942   }
3943   return res;
3944 }
3945 
3946 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3947   assert(UseLargePages, "only for large pages");
3948   bool res;
3949 
3950   if (UseSHM) {
3951     res = os::Linux::release_memory_special_shm(base, bytes);
3952   } else {
3953     assert(UseHugeTLBFS, "must be");
3954     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3955   }
3956   return res;
3957 }
3958 
3959 size_t os::large_page_size() {
3960   return _large_page_size;
3961 }
3962 
3963 // With SysV SHM the entire memory region must be allocated as shared
3964 // memory.
3965 // HugeTLBFS allows application to commit large page memory on demand.
3966 // However, when committing memory with HugeTLBFS fails, the region
3967 // that was supposed to be committed will lose the old reservation
3968 // and allow other threads to steal that memory region. Because of this
3969 // behavior we can't commit HugeTLBFS memory.
3970 bool os::can_commit_large_page_memory() {
3971   return UseTransparentHugePages;
3972 }
3973 
3974 bool os::can_execute_large_page_memory() {
3975   return UseTransparentHugePages || UseHugeTLBFS;
3976 }
3977 
3978 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3979   assert(file_desc >= 0, "file_desc is not valid");
3980   char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3981   if (result != NULL) {
3982     if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3983       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3984     }
3985   }
3986   return result;
3987 }
3988 
3989 // Reserve memory at an arbitrary address, only if that area is
3990 // available (and not reserved for something else).
3991 
3992 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3993   const int max_tries = 10;
3994   char* base[max_tries];
3995   size_t size[max_tries];
3996   const size_t gap = 0x000000;
3997 
3998   // Assert only that the size is a multiple of the page size, since
3999   // that's all that mmap requires, and since that's all we really know
4000   // about at this low abstraction level.  If we need higher alignment,
4001   // we can either pass an alignment to this method or verify alignment
4002   // in one of the methods further up the call chain.  See bug 5044738.
4003   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4004 
4005   // Repeatedly allocate blocks until the block is allocated at the
4006   // right spot.
4007 
4008   // Linux mmap allows caller to pass an address as hint; give it a try first,
4009   // if kernel honors the hint then we can return immediately.
4010   char * addr = anon_mmap(requested_addr, bytes, false);
4011   if (addr == requested_addr) {
4012     return requested_addr;
4013   }
4014 
4015   if (addr != NULL) {
4016     // mmap() is successful but it fails to reserve at the requested address
4017     anon_munmap(addr, bytes);
4018   }
4019 
4020   int i;
4021   for (i = 0; i < max_tries; ++i) {
4022     base[i] = reserve_memory(bytes);
4023 
4024     if (base[i] != NULL) {
4025       // Is this the block we wanted?
4026       if (base[i] == requested_addr) {
4027         size[i] = bytes;
4028         break;
4029       }
4030 
4031       // Does this overlap the block we wanted? Give back the overlapped
4032       // parts and try again.
4033 
4034       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
4035       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
4036         unmap_memory(base[i], top_overlap);
4037         base[i] += top_overlap;
4038         size[i] = bytes - top_overlap;
4039       } else {
4040         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
4041         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
4042           unmap_memory(requested_addr, bottom_overlap);
4043           size[i] = bytes - bottom_overlap;
4044         } else {
4045           size[i] = bytes;
4046         }
4047       }
4048     }
4049   }
4050 
4051   // Give back the unused reserved pieces.
4052 
4053   for (int j = 0; j < i; ++j) {
4054     if (base[j] != NULL) {
4055       unmap_memory(base[j], size[j]);
4056     }
4057   }
4058 
4059   if (i < max_tries) {
4060     return requested_addr;
4061   } else {
4062     return NULL;
4063   }
4064 }
4065 
4066 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4067 void os::infinite_sleep() {
4068   while (true) {    // sleep forever ...
4069     ::sleep(100);   // ... 100 seconds at a time
4070   }
4071 }
4072 
4073 // Used to convert frequent JVM_Yield() to nops
4074 bool os::dont_yield() {
4075   return DontYieldALot;
4076 }
4077 
4078 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4079 // actually give up the CPU. Since skip buddy (v2.6.28):
4080 //
4081 // * Sets the yielding task as skip buddy for current CPU's run queue.
4082 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4083 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4084 //
4085 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4086 // getting re-scheduled immediately.
4087 //
4088 void os::naked_yield() {
4089   sched_yield();
4090 }
4091 
4092 ////////////////////////////////////////////////////////////////////////////////
4093 // thread priority support
4094 
4095 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4096 // only supports dynamic priority, static priority must be zero. For real-time
4097 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4098 // However, for large multi-threaded applications, SCHED_RR is not only slower
4099 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4100 // of 5 runs - Sep 2005).
4101 //
4102 // The following code actually changes the niceness of kernel-thread/LWP. It
4103 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4104 // not the entire user process, and user level threads are 1:1 mapped to kernel
4105 // threads. It has always been the case, but could change in the future. For
4106 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4107 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4108 // (e.g., root privilege or CAP_SYS_NICE capability).
4109 
4110 int os::java_to_os_priority[CriticalPriority + 1] = {
4111   19,              // 0 Entry should never be used
4112 
4113    4,              // 1 MinPriority
4114    3,              // 2
4115    2,              // 3
4116 
4117    1,              // 4
4118    0,              // 5 NormPriority
4119   -1,              // 6
4120 
4121   -2,              // 7
4122   -3,              // 8
4123   -4,              // 9 NearMaxPriority
4124 
4125   -5,              // 10 MaxPriority
4126 
4127   -5               // 11 CriticalPriority
4128 };
4129 
4130 static int prio_init() {
4131   if (ThreadPriorityPolicy == 1) {
4132     if (geteuid() != 0) {
4133       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4134         warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4135                 "e.g., being the root user. If the necessary permission is not " \
4136                 "possessed, changes to priority will be silently ignored.");
4137       }
4138     }
4139   }
4140   if (UseCriticalJavaThreadPriority) {
4141     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4142   }
4143   return 0;
4144 }
4145 
4146 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4147   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4148 
4149   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4150   return (ret == 0) ? OS_OK : OS_ERR;
4151 }
4152 
4153 OSReturn os::get_native_priority(const Thread* const thread,
4154                                  int *priority_ptr) {
4155   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4156     *priority_ptr = java_to_os_priority[NormPriority];
4157     return OS_OK;
4158   }
4159 
4160   errno = 0;
4161   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4162   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4163 }
4164 
4165 ////////////////////////////////////////////////////////////////////////////////
4166 // suspend/resume support
4167 
4168 //  The low-level signal-based suspend/resume support is a remnant from the
4169 //  old VM-suspension that used to be for java-suspension, safepoints etc,
4170 //  within hotspot. Currently used by JFR's OSThreadSampler
4171 //
4172 //  The remaining code is greatly simplified from the more general suspension
4173 //  code that used to be used.
4174 //
4175 //  The protocol is quite simple:
4176 //  - suspend:
4177 //      - sends a signal to the target thread
4178 //      - polls the suspend state of the osthread using a yield loop
4179 //      - target thread signal handler (SR_handler) sets suspend state
4180 //        and blocks in sigsuspend until continued
4181 //  - resume:
4182 //      - sets target osthread state to continue
4183 //      - sends signal to end the sigsuspend loop in the SR_handler
4184 //
4185 //  Note that the SR_lock plays no role in this suspend/resume protocol,
4186 //  but is checked for NULL in SR_handler as a thread termination indicator.
4187 //  The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4188 //
4189 //  Note that resume_clear_context() and suspend_save_context() are needed
4190 //  by SR_handler(), so that fetch_frame_from_ucontext() works,
4191 //  which in part is used by:
4192 //    - Forte Analyzer: AsyncGetCallTrace()
4193 //    - StackBanging: get_frame_at_stack_banging_point()
4194 
4195 static void resume_clear_context(OSThread *osthread) {
4196   osthread->set_ucontext(NULL);
4197   osthread->set_siginfo(NULL);
4198 }
4199 
4200 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4201                                  ucontext_t* context) {
4202   osthread->set_ucontext(context);
4203   osthread->set_siginfo(siginfo);
4204 }
4205 
4206 // Handler function invoked when a thread's execution is suspended or
4207 // resumed. We have to be careful that only async-safe functions are
4208 // called here (Note: most pthread functions are not async safe and
4209 // should be avoided.)
4210 //
4211 // Note: sigwait() is a more natural fit than sigsuspend() from an
4212 // interface point of view, but sigwait() prevents the signal hander
4213 // from being run. libpthread would get very confused by not having
4214 // its signal handlers run and prevents sigwait()'s use with the
4215 // mutex granting granting signal.
4216 //
4217 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4218 //
4219 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4220   // Save and restore errno to avoid confusing native code with EINTR
4221   // after sigsuspend.
4222   int old_errno = errno;
4223 
4224   Thread* thread = Thread::current_or_null_safe();
4225   assert(thread != NULL, "Missing current thread in SR_handler");
4226 
4227   // On some systems we have seen signal delivery get "stuck" until the signal
4228   // mask is changed as part of thread termination. Check that the current thread
4229   // has not already terminated (via SR_lock()) - else the following assertion
4230   // will fail because the thread is no longer a JavaThread as the ~JavaThread
4231   // destructor has completed.
4232 
4233   if (thread->SR_lock() == NULL) {
4234     return;
4235   }
4236 
4237   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4238 
4239   OSThread* osthread = thread->osthread();
4240 
4241   os::SuspendResume::State current = osthread->sr.state();
4242   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4243     suspend_save_context(osthread, siginfo, context);
4244 
4245     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4246     os::SuspendResume::State state = osthread->sr.suspended();
4247     if (state == os::SuspendResume::SR_SUSPENDED) {
4248       sigset_t suspend_set;  // signals for sigsuspend()
4249       sigemptyset(&suspend_set);
4250       // get current set of blocked signals and unblock resume signal
4251       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4252       sigdelset(&suspend_set, SR_signum);
4253 
4254       sr_semaphore.signal();
4255       // wait here until we are resumed
4256       while (1) {
4257         sigsuspend(&suspend_set);
4258 
4259         os::SuspendResume::State result = osthread->sr.running();
4260         if (result == os::SuspendResume::SR_RUNNING) {
4261           sr_semaphore.signal();
4262           break;
4263         }
4264       }
4265 
4266     } else if (state == os::SuspendResume::SR_RUNNING) {
4267       // request was cancelled, continue
4268     } else {
4269       ShouldNotReachHere();
4270     }
4271 
4272     resume_clear_context(osthread);
4273   } else if (current == os::SuspendResume::SR_RUNNING) {
4274     // request was cancelled, continue
4275   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4276     // ignore
4277   } else {
4278     // ignore
4279   }
4280 
4281   errno = old_errno;
4282 }
4283 
4284 static int SR_initialize() {
4285   struct sigaction act;
4286   char *s;
4287 
4288   // Get signal number to use for suspend/resume
4289   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4290     int sig = ::strtol(s, 0, 10);
4291     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4292         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4293       SR_signum = sig;
4294     } else {
4295       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4296               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4297     }
4298   }
4299 
4300   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4301          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4302 
4303   sigemptyset(&SR_sigset);
4304   sigaddset(&SR_sigset, SR_signum);
4305 
4306   // Set up signal handler for suspend/resume
4307   act.sa_flags = SA_RESTART|SA_SIGINFO;
4308   act.sa_handler = (void (*)(int)) SR_handler;
4309 
4310   // SR_signum is blocked by default.
4311   // 4528190 - We also need to block pthread restart signal (32 on all
4312   // supported Linux platforms). Note that LinuxThreads need to block
4313   // this signal for all threads to work properly. So we don't have
4314   // to use hard-coded signal number when setting up the mask.
4315   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4316 
4317   if (sigaction(SR_signum, &act, 0) == -1) {
4318     return -1;
4319   }
4320 
4321   // Save signal flag
4322   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4323   return 0;
4324 }
4325 
4326 static int sr_notify(OSThread* osthread) {
4327   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4328   assert_status(status == 0, status, "pthread_kill");
4329   return status;
4330 }
4331 
4332 // "Randomly" selected value for how long we want to spin
4333 // before bailing out on suspending a thread, also how often
4334 // we send a signal to a thread we want to resume
4335 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4336 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4337 
4338 // returns true on success and false on error - really an error is fatal
4339 // but this seems the normal response to library errors
4340 static bool do_suspend(OSThread* osthread) {
4341   assert(osthread->sr.is_running(), "thread should be running");
4342   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4343 
4344   // mark as suspended and send signal
4345   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4346     // failed to switch, state wasn't running?
4347     ShouldNotReachHere();
4348     return false;
4349   }
4350 
4351   if (sr_notify(osthread) != 0) {
4352     ShouldNotReachHere();
4353   }
4354 
4355   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4356   while (true) {
4357     if (sr_semaphore.timedwait(2)) {
4358       break;
4359     } else {
4360       // timeout
4361       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4362       if (cancelled == os::SuspendResume::SR_RUNNING) {
4363         return false;
4364       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4365         // make sure that we consume the signal on the semaphore as well
4366         sr_semaphore.wait();
4367         break;
4368       } else {
4369         ShouldNotReachHere();
4370         return false;
4371       }
4372     }
4373   }
4374 
4375   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4376   return true;
4377 }
4378 
4379 static void do_resume(OSThread* osthread) {
4380   assert(osthread->sr.is_suspended(), "thread should be suspended");
4381   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4382 
4383   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4384     // failed to switch to WAKEUP_REQUEST
4385     ShouldNotReachHere();
4386     return;
4387   }
4388 
4389   while (true) {
4390     if (sr_notify(osthread) == 0) {
4391       if (sr_semaphore.timedwait(2)) {
4392         if (osthread->sr.is_running()) {
4393           return;
4394         }
4395       }
4396     } else {
4397       ShouldNotReachHere();
4398     }
4399   }
4400 
4401   guarantee(osthread->sr.is_running(), "Must be running!");
4402 }
4403 
4404 ///////////////////////////////////////////////////////////////////////////////////
4405 // signal handling (except suspend/resume)
4406 
4407 // This routine may be used by user applications as a "hook" to catch signals.
4408 // The user-defined signal handler must pass unrecognized signals to this
4409 // routine, and if it returns true (non-zero), then the signal handler must
4410 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4411 // routine will never retun false (zero), but instead will execute a VM panic
4412 // routine kill the process.
4413 //
4414 // If this routine returns false, it is OK to call it again.  This allows
4415 // the user-defined signal handler to perform checks either before or after
4416 // the VM performs its own checks.  Naturally, the user code would be making
4417 // a serious error if it tried to handle an exception (such as a null check
4418 // or breakpoint) that the VM was generating for its own correct operation.
4419 //
4420 // This routine may recognize any of the following kinds of signals:
4421 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4422 // It should be consulted by handlers for any of those signals.
4423 //
4424 // The caller of this routine must pass in the three arguments supplied
4425 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4426 // field of the structure passed to sigaction().  This routine assumes that
4427 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4428 //
4429 // Note that the VM will print warnings if it detects conflicting signal
4430 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4431 //
4432 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4433                                                  siginfo_t* siginfo,
4434                                                  void* ucontext,
4435                                                  int abort_if_unrecognized);
4436 
4437 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4438   assert(info != NULL && uc != NULL, "it must be old kernel");
4439   int orig_errno = errno;  // Preserve errno value over signal handler.
4440   JVM_handle_linux_signal(sig, info, uc, true);
4441   errno = orig_errno;
4442 }
4443 
4444 
4445 // This boolean allows users to forward their own non-matching signals
4446 // to JVM_handle_linux_signal, harmlessly.
4447 bool os::Linux::signal_handlers_are_installed = false;
4448 
4449 // For signal-chaining
4450 struct sigaction sigact[NSIG];
4451 uint64_t sigs = 0;
4452 #if (64 < NSIG-1)
4453 #error "Not all signals can be encoded in sigs. Adapt its type!"
4454 #endif
4455 bool os::Linux::libjsig_is_loaded = false;
4456 typedef struct sigaction *(*get_signal_t)(int);
4457 get_signal_t os::Linux::get_signal_action = NULL;
4458 
4459 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4460   struct sigaction *actp = NULL;
4461 
4462   if (libjsig_is_loaded) {
4463     // Retrieve the old signal handler from libjsig
4464     actp = (*get_signal_action)(sig);
4465   }
4466   if (actp == NULL) {
4467     // Retrieve the preinstalled signal handler from jvm
4468     actp = get_preinstalled_handler(sig);
4469   }
4470 
4471   return actp;
4472 }
4473 
4474 static bool call_chained_handler(struct sigaction *actp, int sig,
4475                                  siginfo_t *siginfo, void *context) {
4476   // Call the old signal handler
4477   if (actp->sa_handler == SIG_DFL) {
4478     // It's more reasonable to let jvm treat it as an unexpected exception
4479     // instead of taking the default action.
4480     return false;
4481   } else if (actp->sa_handler != SIG_IGN) {
4482     if ((actp->sa_flags & SA_NODEFER) == 0) {
4483       // automaticlly block the signal
4484       sigaddset(&(actp->sa_mask), sig);
4485     }
4486 
4487     sa_handler_t hand = NULL;
4488     sa_sigaction_t sa = NULL;
4489     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4490     // retrieve the chained handler
4491     if (siginfo_flag_set) {
4492       sa = actp->sa_sigaction;
4493     } else {
4494       hand = actp->sa_handler;
4495     }
4496 
4497     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4498       actp->sa_handler = SIG_DFL;
4499     }
4500 
4501     // try to honor the signal mask
4502     sigset_t oset;
4503     sigemptyset(&oset);
4504     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4505 
4506     // call into the chained handler
4507     if (siginfo_flag_set) {
4508       (*sa)(sig, siginfo, context);
4509     } else {
4510       (*hand)(sig);
4511     }
4512 
4513     // restore the signal mask
4514     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4515   }
4516   // Tell jvm's signal handler the signal is taken care of.
4517   return true;
4518 }
4519 
4520 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4521   bool chained = false;
4522   // signal-chaining
4523   if (UseSignalChaining) {
4524     struct sigaction *actp = get_chained_signal_action(sig);
4525     if (actp != NULL) {
4526       chained = call_chained_handler(actp, sig, siginfo, context);
4527     }
4528   }
4529   return chained;
4530 }
4531 
4532 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4533   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4534     return &sigact[sig];
4535   }
4536   return NULL;
4537 }
4538 
4539 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4540   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4541   sigact[sig] = oldAct;
4542   sigs |= (uint64_t)1 << (sig-1);
4543 }
4544 
4545 // for diagnostic
4546 int sigflags[NSIG];
4547 
4548 int os::Linux::get_our_sigflags(int sig) {
4549   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4550   return sigflags[sig];
4551 }
4552 
4553 void os::Linux::set_our_sigflags(int sig, int flags) {
4554   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4555   if (sig > 0 && sig < NSIG) {
4556     sigflags[sig] = flags;
4557   }
4558 }
4559 
4560 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4561   // Check for overwrite.
4562   struct sigaction oldAct;
4563   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4564 
4565   void* oldhand = oldAct.sa_sigaction
4566                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4567                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4568   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4569       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4570       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4571     if (AllowUserSignalHandlers || !set_installed) {
4572       // Do not overwrite; user takes responsibility to forward to us.
4573       return;
4574     } else if (UseSignalChaining) {
4575       // save the old handler in jvm
4576       save_preinstalled_handler(sig, oldAct);
4577       // libjsig also interposes the sigaction() call below and saves the
4578       // old sigaction on it own.
4579     } else {
4580       fatal("Encountered unexpected pre-existing sigaction handler "
4581             "%#lx for signal %d.", (long)oldhand, sig);
4582     }
4583   }
4584 
4585   struct sigaction sigAct;
4586   sigfillset(&(sigAct.sa_mask));
4587   sigAct.sa_handler = SIG_DFL;
4588   if (!set_installed) {
4589     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4590   } else {
4591     sigAct.sa_sigaction = signalHandler;
4592     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4593   }
4594   // Save flags, which are set by ours
4595   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4596   sigflags[sig] = sigAct.sa_flags;
4597 
4598   int ret = sigaction(sig, &sigAct, &oldAct);
4599   assert(ret == 0, "check");
4600 
4601   void* oldhand2  = oldAct.sa_sigaction
4602                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4603                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4604   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4605 }
4606 
4607 // install signal handlers for signals that HotSpot needs to
4608 // handle in order to support Java-level exception handling.
4609 
4610 void os::Linux::install_signal_handlers() {
4611   if (!signal_handlers_are_installed) {
4612     signal_handlers_are_installed = true;
4613 
4614     // signal-chaining
4615     typedef void (*signal_setting_t)();
4616     signal_setting_t begin_signal_setting = NULL;
4617     signal_setting_t end_signal_setting = NULL;
4618     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4619                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4620     if (begin_signal_setting != NULL) {
4621       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4622                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4623       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4624                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4625       libjsig_is_loaded = true;
4626       assert(UseSignalChaining, "should enable signal-chaining");
4627     }
4628     if (libjsig_is_loaded) {
4629       // Tell libjsig jvm is setting signal handlers
4630       (*begin_signal_setting)();
4631     }
4632 
4633     set_signal_handler(SIGSEGV, true);
4634     set_signal_handler(SIGPIPE, true);
4635     set_signal_handler(SIGBUS, true);
4636     set_signal_handler(SIGILL, true);
4637     set_signal_handler(SIGFPE, true);
4638 #if defined(PPC64)
4639     set_signal_handler(SIGTRAP, true);
4640 #endif
4641     set_signal_handler(SIGXFSZ, true);
4642 
4643     if (libjsig_is_loaded) {
4644       // Tell libjsig jvm finishes setting signal handlers
4645       (*end_signal_setting)();
4646     }
4647 
4648     // We don't activate signal checker if libjsig is in place, we trust ourselves
4649     // and if UserSignalHandler is installed all bets are off.
4650     // Log that signal checking is off only if -verbose:jni is specified.
4651     if (CheckJNICalls) {
4652       if (libjsig_is_loaded) {
4653         if (PrintJNIResolving) {
4654           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4655         }
4656         check_signals = false;
4657       }
4658       if (AllowUserSignalHandlers) {
4659         if (PrintJNIResolving) {
4660           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4661         }
4662         check_signals = false;
4663       }
4664     }
4665   }
4666 }
4667 
4668 // This is the fastest way to get thread cpu time on Linux.
4669 // Returns cpu time (user+sys) for any thread, not only for current.
4670 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4671 // It might work on 2.6.10+ with a special kernel/glibc patch.
4672 // For reference, please, see IEEE Std 1003.1-2004:
4673 //   http://www.unix.org/single_unix_specification
4674 
4675 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4676   struct timespec tp;
4677   int rc = os::Posix::clock_gettime(clockid, &tp);
4678   assert(rc == 0, "clock_gettime is expected to return 0 code");
4679 
4680   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4681 }
4682 
4683 void os::Linux::initialize_os_info() {
4684   assert(_os_version == 0, "OS info already initialized");
4685 
4686   struct utsname _uname;
4687 
4688   uint32_t major;
4689   uint32_t minor;
4690   uint32_t fix;
4691 
4692   int rc;
4693 
4694   // Kernel version is unknown if
4695   // verification below fails.
4696   _os_version = 0x01000000;
4697 
4698   rc = uname(&_uname);
4699   if (rc != -1) {
4700 
4701     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4702     if (rc == 3) {
4703 
4704       if (major < 256 && minor < 256 && fix < 256) {
4705         // Kernel version format is as expected,
4706         // set it overriding unknown state.
4707         _os_version = (major << 16) |
4708                       (minor << 8 ) |
4709                       (fix   << 0 ) ;
4710       }
4711     }
4712   }
4713 }
4714 
4715 uint32_t os::Linux::os_version() {
4716   assert(_os_version != 0, "not initialized");
4717   return _os_version & 0x00FFFFFF;
4718 }
4719 
4720 bool os::Linux::os_version_is_known() {
4721   assert(_os_version != 0, "not initialized");
4722   return _os_version & 0x01000000 ? false : true;
4723 }
4724 
4725 /////
4726 // glibc on Linux platform uses non-documented flag
4727 // to indicate, that some special sort of signal
4728 // trampoline is used.
4729 // We will never set this flag, and we should
4730 // ignore this flag in our diagnostic
4731 #ifdef SIGNIFICANT_SIGNAL_MASK
4732   #undef SIGNIFICANT_SIGNAL_MASK
4733 #endif
4734 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4735 
4736 static const char* get_signal_handler_name(address handler,
4737                                            char* buf, int buflen) {
4738   int offset = 0;
4739   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4740   if (found) {
4741     // skip directory names
4742     const char *p1, *p2;
4743     p1 = buf;
4744     size_t len = strlen(os::file_separator());
4745     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4746     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4747   } else {
4748     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4749   }
4750   return buf;
4751 }
4752 
4753 static void print_signal_handler(outputStream* st, int sig,
4754                                  char* buf, size_t buflen) {
4755   struct sigaction sa;
4756 
4757   sigaction(sig, NULL, &sa);
4758 
4759   // See comment for SIGNIFICANT_SIGNAL_MASK define
4760   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4761 
4762   st->print("%s: ", os::exception_name(sig, buf, buflen));
4763 
4764   address handler = (sa.sa_flags & SA_SIGINFO)
4765     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4766     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4767 
4768   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4769     st->print("SIG_DFL");
4770   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4771     st->print("SIG_IGN");
4772   } else {
4773     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4774   }
4775 
4776   st->print(", sa_mask[0]=");
4777   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4778 
4779   address rh = VMError::get_resetted_sighandler(sig);
4780   // May be, handler was resetted by VMError?
4781   if (rh != NULL) {
4782     handler = rh;
4783     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4784   }
4785 
4786   st->print(", sa_flags=");
4787   os::Posix::print_sa_flags(st, sa.sa_flags);
4788 
4789   // Check: is it our handler?
4790   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4791       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4792     // It is our signal handler
4793     // check for flags, reset system-used one!
4794     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4795       st->print(
4796                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4797                 os::Linux::get_our_sigflags(sig));
4798     }
4799   }
4800   st->cr();
4801 }
4802 
4803 
4804 #define DO_SIGNAL_CHECK(sig)                      \
4805   do {                                            \
4806     if (!sigismember(&check_signal_done, sig)) {  \
4807       os::Linux::check_signal_handler(sig);       \
4808     }                                             \
4809   } while (0)
4810 
4811 // This method is a periodic task to check for misbehaving JNI applications
4812 // under CheckJNI, we can add any periodic checks here
4813 
4814 void os::run_periodic_checks() {
4815   if (check_signals == false) return;
4816 
4817   // SEGV and BUS if overridden could potentially prevent
4818   // generation of hs*.log in the event of a crash, debugging
4819   // such a case can be very challenging, so we absolutely
4820   // check the following for a good measure:
4821   DO_SIGNAL_CHECK(SIGSEGV);
4822   DO_SIGNAL_CHECK(SIGILL);
4823   DO_SIGNAL_CHECK(SIGFPE);
4824   DO_SIGNAL_CHECK(SIGBUS);
4825   DO_SIGNAL_CHECK(SIGPIPE);
4826   DO_SIGNAL_CHECK(SIGXFSZ);
4827 #if defined(PPC64)
4828   DO_SIGNAL_CHECK(SIGTRAP);
4829 #endif
4830 
4831   // ReduceSignalUsage allows the user to override these handlers
4832   // see comments at the very top and jvm_md.h
4833   if (!ReduceSignalUsage) {
4834     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4835     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4836     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4837     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4838   }
4839 
4840   DO_SIGNAL_CHECK(SR_signum);
4841 }
4842 
4843 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4844 
4845 static os_sigaction_t os_sigaction = NULL;
4846 
4847 void os::Linux::check_signal_handler(int sig) {
4848   char buf[O_BUFLEN];
4849   address jvmHandler = NULL;
4850 
4851 
4852   struct sigaction act;
4853   if (os_sigaction == NULL) {
4854     // only trust the default sigaction, in case it has been interposed
4855     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4856     if (os_sigaction == NULL) return;
4857   }
4858 
4859   os_sigaction(sig, (struct sigaction*)NULL, &act);
4860 
4861 
4862   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4863 
4864   address thisHandler = (act.sa_flags & SA_SIGINFO)
4865     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4866     : CAST_FROM_FN_PTR(address, act.sa_handler);
4867 
4868 
4869   switch (sig) {
4870   case SIGSEGV:
4871   case SIGBUS:
4872   case SIGFPE:
4873   case SIGPIPE:
4874   case SIGILL:
4875   case SIGXFSZ:
4876     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4877     break;
4878 
4879   case SHUTDOWN1_SIGNAL:
4880   case SHUTDOWN2_SIGNAL:
4881   case SHUTDOWN3_SIGNAL:
4882   case BREAK_SIGNAL:
4883     jvmHandler = (address)user_handler();
4884     break;
4885 
4886   default:
4887     if (sig == SR_signum) {
4888       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4889     } else {
4890       return;
4891     }
4892     break;
4893   }
4894 
4895   if (thisHandler != jvmHandler) {
4896     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4897     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4898     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4899     // No need to check this sig any longer
4900     sigaddset(&check_signal_done, sig);
4901     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4902     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4903       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4904                     exception_name(sig, buf, O_BUFLEN));
4905     }
4906   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4907     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4908     tty->print("expected:");
4909     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4910     tty->cr();
4911     tty->print("  found:");
4912     os::Posix::print_sa_flags(tty, act.sa_flags);
4913     tty->cr();
4914     // No need to check this sig any longer
4915     sigaddset(&check_signal_done, sig);
4916   }
4917 
4918   // Dump all the signal
4919   if (sigismember(&check_signal_done, sig)) {
4920     print_signal_handlers(tty, buf, O_BUFLEN);
4921   }
4922 }
4923 
4924 extern void report_error(char* file_name, int line_no, char* title,
4925                          char* format, ...);
4926 
4927 // this is called _before_ most of the global arguments have been parsed
4928 void os::init(void) {
4929   char dummy;   // used to get a guess on initial stack address
4930 
4931   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4932 
4933   init_random(1234567);
4934 
4935   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4936   if (Linux::page_size() == -1) {
4937     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4938           os::strerror(errno));
4939   }
4940   init_page_sizes((size_t) Linux::page_size());
4941 
4942   Linux::initialize_system_info();
4943 
4944   Linux::initialize_os_info();
4945 
4946   // _main_thread points to the thread that created/loaded the JVM.
4947   Linux::_main_thread = pthread_self();
4948 
4949   // retrieve entry point for pthread_setname_np
4950   Linux::_pthread_setname_np =
4951     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4952 
4953   os::Posix::init();
4954 
4955   initial_time_count = javaTimeNanos();
4956 
4957   // Always warn if no monotonic clock available
4958   if (!os::Posix::supports_monotonic_clock()) {
4959     warning("No monotonic clock was available - timed services may "    \
4960             "be adversely affected if the time-of-day clock changes");
4961   }
4962 }
4963 
4964 // To install functions for atexit system call
4965 extern "C" {
4966   static void perfMemory_exit_helper() {
4967     perfMemory_exit();
4968   }
4969 }
4970 
4971 void os::pd_init_container_support() {
4972   OSContainer::init();
4973 }
4974 
4975 void os::Linux::numa_init() {
4976 
4977   // Java can be invoked as
4978   // 1. Without numactl and heap will be allocated/configured on all nodes as
4979   //    per the system policy.
4980   // 2. With numactl --interleave:
4981   //      Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
4982   //      API for membind case bitmask is reset.
4983   //      Interleave is only hint and Kernel can fallback to other nodes if
4984   //      no memory is available on the target nodes.
4985   // 3. With numactl --membind:
4986   //      Use numa_get_membind(v2) API to get nodes bitmask. The same API for
4987   //      interleave case returns bitmask of all nodes.
4988   // numa_all_nodes_ptr holds bitmask of all nodes.
4989   // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
4990   // bitmask when externally configured to run on all or fewer nodes.
4991 
4992   if (!Linux::libnuma_init()) {
4993     UseNUMA = false;
4994   } else {
4995     if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
4996       // If there's only one node (they start from 0) or if the process
4997       // is bound explicitly to a single node using membind, disable NUMA.
4998       UseNUMA = false;
4999     } else {
5000 
5001       LogTarget(Info,os) log;
5002       LogStream ls(log);
5003 
5004       Linux::set_configured_numa_policy(Linux::identify_numa_policy());
5005 
5006       struct bitmask* bmp = Linux::_numa_membind_bitmask;
5007       const char* numa_mode = "membind";
5008 
5009       if (Linux::is_running_in_interleave_mode()) {
5010         bmp = Linux::_numa_interleave_bitmask;
5011         numa_mode = "interleave";
5012       }
5013 
5014       ls.print("UseNUMA is enabled and invoked in '%s' mode."
5015                " Heap will be configured using NUMA memory nodes:", numa_mode);
5016 
5017       for (int node = 0; node <= Linux::numa_max_node(); node++) {
5018         if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5019           ls.print(" %d", node);
5020         }
5021       }
5022     }
5023   }
5024 
5025   if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5026     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5027     // we can make the adaptive lgrp chunk resizing work. If the user specified both
5028     // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5029     // and disable adaptive resizing.
5030     if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5031       warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5032               "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5033       UseAdaptiveSizePolicy = false;
5034       UseAdaptiveNUMAChunkSizing = false;
5035     }
5036   }
5037 
5038   if (!UseNUMA && ForceNUMA) {
5039     UseNUMA = true;
5040   }
5041 }
5042 
5043 // this is called _after_ the global arguments have been parsed
5044 jint os::init_2(void) {
5045 
5046   // This could be set after os::Posix::init() but all platforms
5047   // have to set it the same so we have to mirror Solaris.
5048   DEBUG_ONLY(os::set_mutex_init_done();)
5049 
5050   os::Posix::init_2();
5051 
5052   Linux::fast_thread_clock_init();
5053 
5054   // initialize suspend/resume support - must do this before signal_sets_init()
5055   if (SR_initialize() != 0) {
5056     perror("SR_initialize failed");
5057     return JNI_ERR;
5058   }
5059 
5060   Linux::signal_sets_init();
5061   Linux::install_signal_handlers();
5062   // Initialize data for jdk.internal.misc.Signal
5063   if (!ReduceSignalUsage) {
5064     jdk_misc_signal_init();
5065   }
5066 
5067   // Check and sets minimum stack sizes against command line options
5068   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5069     return JNI_ERR;
5070   }
5071 
5072   suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5073   if (!suppress_primordial_thread_resolution) {
5074     Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5075   }
5076 
5077 #if defined(IA32)
5078   workaround_expand_exec_shield_cs_limit();
5079 #endif
5080 
5081   Linux::libpthread_init();
5082   Linux::sched_getcpu_init();
5083   log_info(os)("HotSpot is running with %s, %s",
5084                Linux::glibc_version(), Linux::libpthread_version());
5085 
5086   if (UseNUMA) {
5087     Linux::numa_init();
5088   }
5089 
5090   if (MaxFDLimit) {
5091     // set the number of file descriptors to max. print out error
5092     // if getrlimit/setrlimit fails but continue regardless.
5093     struct rlimit nbr_files;
5094     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5095     if (status != 0) {
5096       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5097     } else {
5098       nbr_files.rlim_cur = nbr_files.rlim_max;
5099       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5100       if (status != 0) {
5101         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5102       }
5103     }
5104   }
5105 
5106   // Initialize lock used to serialize thread creation (see os::create_thread)
5107   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5108 
5109   // at-exit methods are called in the reverse order of their registration.
5110   // atexit functions are called on return from main or as a result of a
5111   // call to exit(3C). There can be only 32 of these functions registered
5112   // and atexit() does not set errno.
5113 
5114   if (PerfAllowAtExitRegistration) {
5115     // only register atexit functions if PerfAllowAtExitRegistration is set.
5116     // atexit functions can be delayed until process exit time, which
5117     // can be problematic for embedded VM situations. Embedded VMs should
5118     // call DestroyJavaVM() to assure that VM resources are released.
5119 
5120     // note: perfMemory_exit_helper atexit function may be removed in
5121     // the future if the appropriate cleanup code can be added to the
5122     // VM_Exit VMOperation's doit method.
5123     if (atexit(perfMemory_exit_helper) != 0) {
5124       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5125     }
5126   }
5127 
5128   // initialize thread priority policy
5129   prio_init();
5130 
5131   if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) {
5132     set_coredump_filter(DAX_SHARED_BIT);
5133   }
5134 
5135   if (DumpPrivateMappingsInCore) {
5136     set_coredump_filter(FILE_BACKED_PVT_BIT);
5137   }
5138 
5139   if (DumpSharedMappingsInCore) {
5140     set_coredump_filter(FILE_BACKED_SHARED_BIT);
5141   }
5142 
5143   return JNI_OK;
5144 }
5145 
5146 // Mark the polling page as unreadable
5147 void os::make_polling_page_unreadable(void) {
5148   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5149     fatal("Could not disable polling page");
5150   }
5151 }
5152 
5153 // Mark the polling page as readable
5154 void os::make_polling_page_readable(void) {
5155   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5156     fatal("Could not enable polling page");
5157   }
5158 }
5159 
5160 // older glibc versions don't have this macro (which expands to
5161 // an optimized bit-counting function) so we have to roll our own
5162 #ifndef CPU_COUNT
5163 
5164 static int _cpu_count(const cpu_set_t* cpus) {
5165   int count = 0;
5166   // only look up to the number of configured processors
5167   for (int i = 0; i < os::processor_count(); i++) {
5168     if (CPU_ISSET(i, cpus)) {
5169       count++;
5170     }
5171   }
5172   return count;
5173 }
5174 
5175 #define CPU_COUNT(cpus) _cpu_count(cpus)
5176 
5177 #endif // CPU_COUNT
5178 
5179 // Get the current number of available processors for this process.
5180 // This value can change at any time during a process's lifetime.
5181 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5182 // If it appears there may be more than 1024 processors then we do a
5183 // dynamic check - see 6515172 for details.
5184 // If anything goes wrong we fallback to returning the number of online
5185 // processors - which can be greater than the number available to the process.
5186 int os::Linux::active_processor_count() {
5187   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
5188   cpu_set_t* cpus_p = &cpus;
5189   int cpus_size = sizeof(cpu_set_t);
5190 
5191   int configured_cpus = os::processor_count();  // upper bound on available cpus
5192   int cpu_count = 0;
5193 
5194 // old build platforms may not support dynamic cpu sets
5195 #ifdef CPU_ALLOC
5196 
5197   // To enable easy testing of the dynamic path on different platforms we
5198   // introduce a diagnostic flag: UseCpuAllocPath
5199   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5200     // kernel may use a mask bigger than cpu_set_t
5201     log_trace(os)("active_processor_count: using dynamic path %s"
5202                   "- configured processors: %d",
5203                   UseCpuAllocPath ? "(forced) " : "",
5204                   configured_cpus);
5205     cpus_p = CPU_ALLOC(configured_cpus);
5206     if (cpus_p != NULL) {
5207       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5208       // zero it just to be safe
5209       CPU_ZERO_S(cpus_size, cpus_p);
5210     }
5211     else {
5212        // failed to allocate so fallback to online cpus
5213        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5214        log_trace(os)("active_processor_count: "
5215                      "CPU_ALLOC failed (%s) - using "
5216                      "online processor count: %d",
5217                      os::strerror(errno), online_cpus);
5218        return online_cpus;
5219     }
5220   }
5221   else {
5222     log_trace(os)("active_processor_count: using static path - configured processors: %d",
5223                   configured_cpus);
5224   }
5225 #else // CPU_ALLOC
5226 // these stubs won't be executed
5227 #define CPU_COUNT_S(size, cpus) -1
5228 #define CPU_FREE(cpus)
5229 
5230   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5231                 configured_cpus);
5232 #endif // CPU_ALLOC
5233 
5234   // pid 0 means the current thread - which we have to assume represents the process
5235   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5236     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5237       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5238     }
5239     else {
5240       cpu_count = CPU_COUNT(cpus_p);
5241     }
5242     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5243   }
5244   else {
5245     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5246     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5247             "which may exceed available processors", os::strerror(errno), cpu_count);
5248   }
5249 
5250   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5251     CPU_FREE(cpus_p);
5252   }
5253 
5254   assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5255   return cpu_count;
5256 }
5257 
5258 // Determine the active processor count from one of
5259 // three different sources:
5260 //
5261 // 1. User option -XX:ActiveProcessorCount
5262 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5263 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5264 //
5265 // Option 1, if specified, will always override.
5266 // If the cgroup subsystem is active and configured, we
5267 // will return the min of the cgroup and option 2 results.
5268 // This is required since tools, such as numactl, that
5269 // alter cpu affinity do not update cgroup subsystem
5270 // cpuset configuration files.
5271 int os::active_processor_count() {
5272   // User has overridden the number of active processors
5273   if (ActiveProcessorCount > 0) {
5274     log_trace(os)("active_processor_count: "
5275                   "active processor count set by user : %d",
5276                   ActiveProcessorCount);
5277     return ActiveProcessorCount;
5278   }
5279 
5280   int active_cpus;
5281   if (OSContainer::is_containerized()) {
5282     active_cpus = OSContainer::active_processor_count();
5283     log_trace(os)("active_processor_count: determined by OSContainer: %d",
5284                    active_cpus);
5285   } else {
5286     active_cpus = os::Linux::active_processor_count();
5287   }
5288 
5289   return active_cpus;
5290 }
5291 
5292 uint os::processor_id() {
5293   const int id = Linux::sched_getcpu();
5294   assert(id >= 0 && id < _processor_count, "Invalid processor id");
5295   return (uint)id;
5296 }
5297 
5298 void os::set_native_thread_name(const char *name) {
5299   if (Linux::_pthread_setname_np) {
5300     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5301     snprintf(buf, sizeof(buf), "%s", name);
5302     buf[sizeof(buf) - 1] = '\0';
5303     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5304     // ERANGE should not happen; all other errors should just be ignored.
5305     assert(rc != ERANGE, "pthread_setname_np failed");
5306   }
5307 }
5308 
5309 bool os::distribute_processes(uint length, uint* distribution) {
5310   // Not yet implemented.
5311   return false;
5312 }
5313 
5314 bool os::bind_to_processor(uint processor_id) {
5315   // Not yet implemented.
5316   return false;
5317 }
5318 
5319 ///
5320 
5321 void os::SuspendedThreadTask::internal_do_task() {
5322   if (do_suspend(_thread->osthread())) {
5323     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5324     do_task(context);
5325     do_resume(_thread->osthread());
5326   }
5327 }
5328 
5329 ////////////////////////////////////////////////////////////////////////////////
5330 // debug support
5331 
5332 bool os::find(address addr, outputStream* st) {
5333   Dl_info dlinfo;
5334   memset(&dlinfo, 0, sizeof(dlinfo));
5335   if (dladdr(addr, &dlinfo) != 0) {
5336     st->print(PTR_FORMAT ": ", p2i(addr));
5337     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5338       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5339                 p2i(addr) - p2i(dlinfo.dli_saddr));
5340     } else if (dlinfo.dli_fbase != NULL) {
5341       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5342     } else {
5343       st->print("<absolute address>");
5344     }
5345     if (dlinfo.dli_fname != NULL) {
5346       st->print(" in %s", dlinfo.dli_fname);
5347     }
5348     if (dlinfo.dli_fbase != NULL) {
5349       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5350     }
5351     st->cr();
5352 
5353     if (Verbose) {
5354       // decode some bytes around the PC
5355       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5356       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5357       address       lowest = (address) dlinfo.dli_sname;
5358       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5359       if (begin < lowest)  begin = lowest;
5360       Dl_info dlinfo2;
5361       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5362           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5363         end = (address) dlinfo2.dli_saddr;
5364       }
5365       Disassembler::decode(begin, end, st);
5366     }
5367     return true;
5368   }
5369   return false;
5370 }
5371 
5372 ////////////////////////////////////////////////////////////////////////////////
5373 // misc
5374 
5375 // This does not do anything on Linux. This is basically a hook for being
5376 // able to use structured exception handling (thread-local exception filters)
5377 // on, e.g., Win32.
5378 void
5379 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5380                          JavaCallArguments* args, Thread* thread) {
5381   f(value, method, args, thread);
5382 }
5383 
5384 void os::print_statistics() {
5385 }
5386 
5387 bool os::message_box(const char* title, const char* message) {
5388   int i;
5389   fdStream err(defaultStream::error_fd());
5390   for (i = 0; i < 78; i++) err.print_raw("=");
5391   err.cr();
5392   err.print_raw_cr(title);
5393   for (i = 0; i < 78; i++) err.print_raw("-");
5394   err.cr();
5395   err.print_raw_cr(message);
5396   for (i = 0; i < 78; i++) err.print_raw("=");
5397   err.cr();
5398 
5399   char buf[16];
5400   // Prevent process from exiting upon "read error" without consuming all CPU
5401   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5402 
5403   return buf[0] == 'y' || buf[0] == 'Y';
5404 }
5405 
5406 // Is a (classpath) directory empty?
5407 bool os::dir_is_empty(const char* path) {
5408   DIR *dir = NULL;
5409   struct dirent *ptr;
5410 
5411   dir = opendir(path);
5412   if (dir == NULL) return true;
5413 
5414   // Scan the directory
5415   bool result = true;
5416   while (result && (ptr = readdir(dir)) != NULL) {
5417     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5418       result = false;
5419     }
5420   }
5421   closedir(dir);
5422   return result;
5423 }
5424 
5425 // This code originates from JDK's sysOpen and open64_w
5426 // from src/solaris/hpi/src/system_md.c
5427 
5428 int os::open(const char *path, int oflag, int mode) {
5429   if (strlen(path) > MAX_PATH - 1) {
5430     errno = ENAMETOOLONG;
5431     return -1;
5432   }
5433 
5434   // All file descriptors that are opened in the Java process and not
5435   // specifically destined for a subprocess should have the close-on-exec
5436   // flag set.  If we don't set it, then careless 3rd party native code
5437   // might fork and exec without closing all appropriate file descriptors
5438   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5439   // turn might:
5440   //
5441   // - cause end-of-file to fail to be detected on some file
5442   //   descriptors, resulting in mysterious hangs, or
5443   //
5444   // - might cause an fopen in the subprocess to fail on a system
5445   //   suffering from bug 1085341.
5446   //
5447   // (Yes, the default setting of the close-on-exec flag is a Unix
5448   // design flaw)
5449   //
5450   // See:
5451   // 1085341: 32-bit stdio routines should support file descriptors >255
5452   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5453   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5454   //
5455   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5456   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5457   // because it saves a system call and removes a small window where the flag
5458   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5459   // and we fall back to using FD_CLOEXEC (see below).
5460 #ifdef O_CLOEXEC
5461   oflag |= O_CLOEXEC;
5462 #endif
5463 
5464   int fd = ::open64(path, oflag, mode);
5465   if (fd == -1) return -1;
5466 
5467   //If the open succeeded, the file might still be a directory
5468   {
5469     struct stat64 buf64;
5470     int ret = ::fstat64(fd, &buf64);
5471     int st_mode = buf64.st_mode;
5472 
5473     if (ret != -1) {
5474       if ((st_mode & S_IFMT) == S_IFDIR) {
5475         errno = EISDIR;
5476         ::close(fd);
5477         return -1;
5478       }
5479     } else {
5480       ::close(fd);
5481       return -1;
5482     }
5483   }
5484 
5485 #ifdef FD_CLOEXEC
5486   // Validate that the use of the O_CLOEXEC flag on open above worked.
5487   // With recent kernels, we will perform this check exactly once.
5488   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5489   if (!O_CLOEXEC_is_known_to_work) {
5490     int flags = ::fcntl(fd, F_GETFD);
5491     if (flags != -1) {
5492       if ((flags & FD_CLOEXEC) != 0)
5493         O_CLOEXEC_is_known_to_work = 1;
5494       else
5495         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5496     }
5497   }
5498 #endif
5499 
5500   return fd;
5501 }
5502 
5503 
5504 // create binary file, rewriting existing file if required
5505 int os::create_binary_file(const char* path, bool rewrite_existing) {
5506   int oflags = O_WRONLY | O_CREAT;
5507   if (!rewrite_existing) {
5508     oflags |= O_EXCL;
5509   }
5510   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5511 }
5512 
5513 // return current position of file pointer
5514 jlong os::current_file_offset(int fd) {
5515   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5516 }
5517 
5518 // move file pointer to the specified offset
5519 jlong os::seek_to_file_offset(int fd, jlong offset) {
5520   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5521 }
5522 
5523 // This code originates from JDK's sysAvailable
5524 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5525 
5526 int os::available(int fd, jlong *bytes) {
5527   jlong cur, end;
5528   int mode;
5529   struct stat64 buf64;
5530 
5531   if (::fstat64(fd, &buf64) >= 0) {
5532     mode = buf64.st_mode;
5533     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5534       int n;
5535       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5536         *bytes = n;
5537         return 1;
5538       }
5539     }
5540   }
5541   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5542     return 0;
5543   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5544     return 0;
5545   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5546     return 0;
5547   }
5548   *bytes = end - cur;
5549   return 1;
5550 }
5551 
5552 // Map a block of memory.
5553 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5554                         char *addr, size_t bytes, bool read_only,
5555                         bool allow_exec) {
5556   int prot;
5557   int flags = MAP_PRIVATE;
5558 
5559   if (read_only) {
5560     prot = PROT_READ;
5561   } else {
5562     prot = PROT_READ | PROT_WRITE;
5563   }
5564 
5565   if (allow_exec) {
5566     prot |= PROT_EXEC;
5567   }
5568 
5569   if (addr != NULL) {
5570     flags |= MAP_FIXED;
5571   }
5572 
5573   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5574                                      fd, file_offset);
5575   if (mapped_address == MAP_FAILED) {
5576     return NULL;
5577   }
5578   return mapped_address;
5579 }
5580 
5581 
5582 // Remap a block of memory.
5583 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5584                           char *addr, size_t bytes, bool read_only,
5585                           bool allow_exec) {
5586   // same as map_memory() on this OS
5587   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5588                         allow_exec);
5589 }
5590 
5591 
5592 // Unmap a block of memory.
5593 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5594   return munmap(addr, bytes) == 0;
5595 }
5596 
5597 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5598 
5599 static jlong fast_cpu_time(Thread *thread) {
5600     clockid_t clockid;
5601     int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5602                                               &clockid);
5603     if (rc == 0) {
5604       return os::Linux::fast_thread_cpu_time(clockid);
5605     } else {
5606       // It's possible to encounter a terminated native thread that failed
5607       // to detach itself from the VM - which should result in ESRCH.
5608       assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5609       return -1;
5610     }
5611 }
5612 
5613 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5614 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5615 // of a thread.
5616 //
5617 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5618 // the fast estimate available on the platform.
5619 
5620 jlong os::current_thread_cpu_time() {
5621   if (os::Linux::supports_fast_thread_cpu_time()) {
5622     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5623   } else {
5624     // return user + sys since the cost is the same
5625     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5626   }
5627 }
5628 
5629 jlong os::thread_cpu_time(Thread* thread) {
5630   // consistent with what current_thread_cpu_time() returns
5631   if (os::Linux::supports_fast_thread_cpu_time()) {
5632     return fast_cpu_time(thread);
5633   } else {
5634     return slow_thread_cpu_time(thread, true /* user + sys */);
5635   }
5636 }
5637 
5638 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5639   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5640     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5641   } else {
5642     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5643   }
5644 }
5645 
5646 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5647   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5648     return fast_cpu_time(thread);
5649   } else {
5650     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5651   }
5652 }
5653 
5654 //  -1 on error.
5655 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5656   pid_t  tid = thread->osthread()->thread_id();
5657   char *s;
5658   char stat[2048];
5659   int statlen;
5660   char proc_name[64];
5661   int count;
5662   long sys_time, user_time;
5663   char cdummy;
5664   int idummy;
5665   long ldummy;
5666   FILE *fp;
5667 
5668   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5669   fp = fopen(proc_name, "r");
5670   if (fp == NULL) return -1;
5671   statlen = fread(stat, 1, 2047, fp);
5672   stat[statlen] = '\0';
5673   fclose(fp);
5674 
5675   // Skip pid and the command string. Note that we could be dealing with
5676   // weird command names, e.g. user could decide to rename java launcher
5677   // to "java 1.4.2 :)", then the stat file would look like
5678   //                1234 (java 1.4.2 :)) R ... ...
5679   // We don't really need to know the command string, just find the last
5680   // occurrence of ")" and then start parsing from there. See bug 4726580.
5681   s = strrchr(stat, ')');
5682   if (s == NULL) return -1;
5683 
5684   // Skip blank chars
5685   do { s++; } while (s && isspace(*s));
5686 
5687   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5688                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5689                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5690                  &user_time, &sys_time);
5691   if (count != 13) return -1;
5692   if (user_sys_cpu_time) {
5693     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5694   } else {
5695     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5696   }
5697 }
5698 
5699 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5700   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5701   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5702   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5703   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5704 }
5705 
5706 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5707   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5708   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5709   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5710   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5711 }
5712 
5713 bool os::is_thread_cpu_time_supported() {
5714   return true;
5715 }
5716 
5717 // System loadavg support.  Returns -1 if load average cannot be obtained.
5718 // Linux doesn't yet have a (official) notion of processor sets,
5719 // so just return the system wide load average.
5720 int os::loadavg(double loadavg[], int nelem) {
5721   return ::getloadavg(loadavg, nelem);
5722 }
5723 
5724 void os::pause() {
5725   char filename[MAX_PATH];
5726   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5727     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5728   } else {
5729     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5730   }
5731 
5732   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5733   if (fd != -1) {
5734     struct stat buf;
5735     ::close(fd);
5736     while (::stat(filename, &buf) == 0) {
5737       (void)::poll(NULL, 0, 100);
5738     }
5739   } else {
5740     jio_fprintf(stderr,
5741                 "Could not open pause file '%s', continuing immediately.\n", filename);
5742   }
5743 }
5744 
5745 extern char** environ;
5746 
5747 // Run the specified command in a separate process. Return its exit value,
5748 // or -1 on failure (e.g. can't fork a new process).
5749 // Unlike system(), this function can be called from signal handler. It
5750 // doesn't block SIGINT et al.
5751 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5752   const char * argv[4] = {"sh", "-c", cmd, NULL};
5753 
5754   pid_t pid ;
5755 
5756   if (use_vfork_if_available) {
5757     pid = vfork();
5758   } else {
5759     pid = fork();
5760   }
5761 
5762   if (pid < 0) {
5763     // fork failed
5764     return -1;
5765 
5766   } else if (pid == 0) {
5767     // child process
5768 
5769     execve("/bin/sh", (char* const*)argv, environ);
5770 
5771     // execve failed
5772     _exit(-1);
5773 
5774   } else  {
5775     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5776     // care about the actual exit code, for now.
5777 
5778     int status;
5779 
5780     // Wait for the child process to exit.  This returns immediately if
5781     // the child has already exited. */
5782     while (waitpid(pid, &status, 0) < 0) {
5783       switch (errno) {
5784       case ECHILD: return 0;
5785       case EINTR: break;
5786       default: return -1;
5787       }
5788     }
5789 
5790     if (WIFEXITED(status)) {
5791       // The child exited normally; get its exit code.
5792       return WEXITSTATUS(status);
5793     } else if (WIFSIGNALED(status)) {
5794       // The child exited because of a signal
5795       // The best value to return is 0x80 + signal number,
5796       // because that is what all Unix shells do, and because
5797       // it allows callers to distinguish between process exit and
5798       // process death by signal.
5799       return 0x80 + WTERMSIG(status);
5800     } else {
5801       // Unknown exit code; pass it through
5802       return status;
5803     }
5804   }
5805 }
5806 
5807 // Get the default path to the core file
5808 // Returns the length of the string
5809 int os::get_core_path(char* buffer, size_t bufferSize) {
5810   /*
5811    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5812    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5813    */
5814   const int core_pattern_len = 129;
5815   char core_pattern[core_pattern_len] = {0};
5816 
5817   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5818   if (core_pattern_file == -1) {
5819     return -1;
5820   }
5821 
5822   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5823   ::close(core_pattern_file);
5824   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5825     return -1;
5826   }
5827   if (core_pattern[ret-1] == '\n') {
5828     core_pattern[ret-1] = '\0';
5829   } else {
5830     core_pattern[ret] = '\0';
5831   }
5832 
5833   // Replace the %p in the core pattern with the process id. NOTE: we do this
5834   // only if the pattern doesn't start with "|", and we support only one %p in
5835   // the pattern.
5836   char *pid_pos = strstr(core_pattern, "%p");
5837   const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : "";  // skip over the "%p"
5838   int written;
5839 
5840   if (core_pattern[0] == '/') {
5841     if (pid_pos != NULL) {
5842       *pid_pos = '\0';
5843       written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5844                              current_process_id(), tail);
5845     } else {
5846       written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5847     }
5848   } else {
5849     char cwd[PATH_MAX];
5850 
5851     const char* p = get_current_directory(cwd, PATH_MAX);
5852     if (p == NULL) {
5853       return -1;
5854     }
5855 
5856     if (core_pattern[0] == '|') {
5857       written = jio_snprintf(buffer, bufferSize,
5858                              "\"%s\" (or dumping to %s/core.%d)",
5859                              &core_pattern[1], p, current_process_id());
5860     } else if (pid_pos != NULL) {
5861       *pid_pos = '\0';
5862       written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5863                              current_process_id(), tail);
5864     } else {
5865       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5866     }
5867   }
5868 
5869   if (written < 0) {
5870     return -1;
5871   }
5872 
5873   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5874     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5875 
5876     if (core_uses_pid_file != -1) {
5877       char core_uses_pid = 0;
5878       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5879       ::close(core_uses_pid_file);
5880 
5881       if (core_uses_pid == '1') {
5882         jio_snprintf(buffer + written, bufferSize - written,
5883                                           ".%d", current_process_id());
5884       }
5885     }
5886   }
5887 
5888   return strlen(buffer);
5889 }
5890 
5891 bool os::start_debugging(char *buf, int buflen) {
5892   int len = (int)strlen(buf);
5893   char *p = &buf[len];
5894 
5895   jio_snprintf(p, buflen-len,
5896                "\n\n"
5897                "Do you want to debug the problem?\n\n"
5898                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5899                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5900                "Otherwise, press RETURN to abort...",
5901                os::current_process_id(), os::current_process_id(),
5902                os::current_thread_id(), os::current_thread_id());
5903 
5904   bool yes = os::message_box("Unexpected Error", buf);
5905 
5906   if (yes) {
5907     // yes, user asked VM to launch debugger
5908     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5909                  os::current_process_id(), os::current_process_id());
5910 
5911     os::fork_and_exec(buf);
5912     yes = false;
5913   }
5914   return yes;
5915 }
5916 
5917 
5918 // Java/Compiler thread:
5919 //
5920 //   Low memory addresses
5921 // P0 +------------------------+
5922 //    |                        |\  Java thread created by VM does not have glibc
5923 //    |    glibc guard page    | - guard page, attached Java thread usually has
5924 //    |                        |/  1 glibc guard page.
5925 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5926 //    |                        |\
5927 //    |  HotSpot Guard Pages   | - red, yellow and reserved pages
5928 //    |                        |/
5929 //    +------------------------+ JavaThread::stack_reserved_zone_base()
5930 //    |                        |\
5931 //    |      Normal Stack      | -
5932 //    |                        |/
5933 // P2 +------------------------+ Thread::stack_base()
5934 //
5935 // Non-Java thread:
5936 //
5937 //   Low memory addresses
5938 // P0 +------------------------+
5939 //    |                        |\
5940 //    |  glibc guard page      | - usually 1 page
5941 //    |                        |/
5942 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5943 //    |                        |\
5944 //    |      Normal Stack      | -
5945 //    |                        |/
5946 // P2 +------------------------+ Thread::stack_base()
5947 //
5948 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5949 //    returned from pthread_attr_getstack().
5950 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5951 //    of the stack size given in pthread_attr. We work around this for
5952 //    threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5953 //
5954 #ifndef ZERO
5955 static void current_stack_region(address * bottom, size_t * size) {
5956   if (os::is_primordial_thread()) {
5957     // primordial thread needs special handling because pthread_getattr_np()
5958     // may return bogus value.
5959     *bottom = os::Linux::initial_thread_stack_bottom();
5960     *size   = os::Linux::initial_thread_stack_size();
5961   } else {
5962     pthread_attr_t attr;
5963 
5964     int rslt = pthread_getattr_np(pthread_self(), &attr);
5965 
5966     // JVM needs to know exact stack location, abort if it fails
5967     if (rslt != 0) {
5968       if (rslt == ENOMEM) {
5969         vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5970       } else {
5971         fatal("pthread_getattr_np failed with error = %d", rslt);
5972       }
5973     }
5974 
5975     if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5976       fatal("Cannot locate current stack attributes!");
5977     }
5978 
5979     // Work around NPTL stack guard error.
5980     size_t guard_size = 0;
5981     rslt = pthread_attr_getguardsize(&attr, &guard_size);
5982     if (rslt != 0) {
5983       fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5984     }
5985     *bottom += guard_size;
5986     *size   -= guard_size;
5987 
5988     pthread_attr_destroy(&attr);
5989 
5990   }
5991   assert(os::current_stack_pointer() >= *bottom &&
5992          os::current_stack_pointer() < *bottom + *size, "just checking");
5993 }
5994 
5995 address os::current_stack_base() {
5996   address bottom;
5997   size_t size;
5998   current_stack_region(&bottom, &size);
5999   return (bottom + size);
6000 }
6001 
6002 size_t os::current_stack_size() {
6003   // This stack size includes the usable stack and HotSpot guard pages
6004   // (for the threads that have Hotspot guard pages).
6005   address bottom;
6006   size_t size;
6007   current_stack_region(&bottom, &size);
6008   return size;
6009 }
6010 #endif
6011 
6012 static inline struct timespec get_mtime(const char* filename) {
6013   struct stat st;
6014   int ret = os::stat(filename, &st);
6015   assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
6016   return st.st_mtim;
6017 }
6018 
6019 int os::compare_file_modified_times(const char* file1, const char* file2) {
6020   struct timespec filetime1 = get_mtime(file1);
6021   struct timespec filetime2 = get_mtime(file2);
6022   int diff = filetime1.tv_sec - filetime2.tv_sec;
6023   if (diff == 0) {
6024     return filetime1.tv_nsec - filetime2.tv_nsec;
6025   }
6026   return diff;
6027 }
6028 
6029 /////////////// Unit tests ///////////////
6030 
6031 #ifndef PRODUCT
6032 
6033 class TestReserveMemorySpecial : AllStatic {
6034  public:
6035   static void small_page_write(void* addr, size_t size) {
6036     size_t page_size = os::vm_page_size();
6037 
6038     char* end = (char*)addr + size;
6039     for (char* p = (char*)addr; p < end; p += page_size) {
6040       *p = 1;
6041     }
6042   }
6043 
6044   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6045     if (!UseHugeTLBFS) {
6046       return;
6047     }
6048 
6049     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6050 
6051     if (addr != NULL) {
6052       small_page_write(addr, size);
6053 
6054       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6055     }
6056   }
6057 
6058   static void test_reserve_memory_special_huge_tlbfs_only() {
6059     if (!UseHugeTLBFS) {
6060       return;
6061     }
6062 
6063     size_t lp = os::large_page_size();
6064 
6065     for (size_t size = lp; size <= lp * 10; size += lp) {
6066       test_reserve_memory_special_huge_tlbfs_only(size);
6067     }
6068   }
6069 
6070   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6071     size_t lp = os::large_page_size();
6072     size_t ag = os::vm_allocation_granularity();
6073 
6074     // sizes to test
6075     const size_t sizes[] = {
6076       lp, lp + ag, lp + lp / 2, lp * 2,
6077       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6078       lp * 10, lp * 10 + lp / 2
6079     };
6080     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6081 
6082     // For each size/alignment combination, we test three scenarios:
6083     // 1) with req_addr == NULL
6084     // 2) with a non-null req_addr at which we expect to successfully allocate
6085     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6086     //    expect the allocation to either fail or to ignore req_addr
6087 
6088     // Pre-allocate two areas; they shall be as large as the largest allocation
6089     //  and aligned to the largest alignment we will be testing.
6090     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6091     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6092       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6093       -1, 0);
6094     assert(mapping1 != MAP_FAILED, "should work");
6095 
6096     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6097       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6098       -1, 0);
6099     assert(mapping2 != MAP_FAILED, "should work");
6100 
6101     // Unmap the first mapping, but leave the second mapping intact: the first
6102     // mapping will serve as a value for a "good" req_addr (case 2). The second
6103     // mapping, still intact, as "bad" req_addr (case 3).
6104     ::munmap(mapping1, mapping_size);
6105 
6106     // Case 1
6107     for (int i = 0; i < num_sizes; i++) {
6108       const size_t size = sizes[i];
6109       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6110         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6111         if (p != NULL) {
6112           assert(is_aligned(p, alignment), "must be");
6113           small_page_write(p, size);
6114           os::Linux::release_memory_special_huge_tlbfs(p, size);
6115         }
6116       }
6117     }
6118 
6119     // Case 2
6120     for (int i = 0; i < num_sizes; i++) {
6121       const size_t size = sizes[i];
6122       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6123         char* const req_addr = align_up(mapping1, alignment);
6124         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6125         if (p != NULL) {
6126           assert(p == req_addr, "must be");
6127           small_page_write(p, size);
6128           os::Linux::release_memory_special_huge_tlbfs(p, size);
6129         }
6130       }
6131     }
6132 
6133     // Case 3
6134     for (int i = 0; i < num_sizes; i++) {
6135       const size_t size = sizes[i];
6136       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6137         char* const req_addr = align_up(mapping2, alignment);
6138         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6139         // as the area around req_addr contains already existing mappings, the API should always
6140         // return NULL (as per contract, it cannot return another address)
6141         assert(p == NULL, "must be");
6142       }
6143     }
6144 
6145     ::munmap(mapping2, mapping_size);
6146 
6147   }
6148 
6149   static void test_reserve_memory_special_huge_tlbfs() {
6150     if (!UseHugeTLBFS) {
6151       return;
6152     }
6153 
6154     test_reserve_memory_special_huge_tlbfs_only();
6155     test_reserve_memory_special_huge_tlbfs_mixed();
6156   }
6157 
6158   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6159     if (!UseSHM) {
6160       return;
6161     }
6162 
6163     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6164 
6165     if (addr != NULL) {
6166       assert(is_aligned(addr, alignment), "Check");
6167       assert(is_aligned(addr, os::large_page_size()), "Check");
6168 
6169       small_page_write(addr, size);
6170 
6171       os::Linux::release_memory_special_shm(addr, size);
6172     }
6173   }
6174 
6175   static void test_reserve_memory_special_shm() {
6176     size_t lp = os::large_page_size();
6177     size_t ag = os::vm_allocation_granularity();
6178 
6179     for (size_t size = ag; size < lp * 3; size += ag) {
6180       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6181         test_reserve_memory_special_shm(size, alignment);
6182       }
6183     }
6184   }
6185 
6186   static void test() {
6187     test_reserve_memory_special_huge_tlbfs();
6188     test_reserve_memory_special_shm();
6189   }
6190 };
6191 
6192 void TestReserveMemorySpecial_test() {
6193   TestReserveMemorySpecial::test();
6194 }
6195 
6196 #endif