1 /*
   2  * Copyright (c) 1999, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "jvm.h"
  27 #include "classfile/classLoader.hpp"
  28 #include "classfile/systemDictionary.hpp"
  29 #include "classfile/vmSymbols.hpp"
  30 #include "code/icBuffer.hpp"
  31 #include "code/vtableStubs.hpp"
  32 #include "compiler/compileBroker.hpp"
  33 #include "compiler/disassembler.hpp"
  34 #include "interpreter/interpreter.hpp"
  35 #include "logging/log.hpp"
  36 #include "memory/allocation.inline.hpp"
  37 #include "memory/filemap.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_linux.inline.hpp"
  40 #include "os_share_linux.hpp"
  41 #include "osContainer_linux.hpp"
  42 #include "prims/jniFastGetField.hpp"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.inline.hpp"
  49 #include "runtime/init.hpp"
  50 #include "runtime/java.hpp"
  51 #include "runtime/javaCalls.hpp"
  52 #include "runtime/mutexLocker.hpp"
  53 #include "runtime/objectMonitor.hpp"
  54 #include "runtime/orderAccess.hpp"
  55 #include "runtime/osThread.hpp"
  56 #include "runtime/perfMemory.hpp"
  57 #include "runtime/sharedRuntime.hpp"
  58 #include "runtime/statSampler.hpp"
  59 #include "runtime/stubRoutines.hpp"
  60 #include "runtime/thread.inline.hpp"
  61 #include "runtime/threadCritical.hpp"
  62 #include "runtime/threadSMR.hpp"
  63 #include "runtime/timer.hpp"
  64 #include "semaphore_posix.hpp"
  65 #include "services/attachListener.hpp"
  66 #include "services/memTracker.hpp"
  67 #include "services/runtimeService.hpp"
  68 #include "utilities/align.hpp"
  69 #include "utilities/decoder.hpp"
  70 #include "utilities/defaultStream.hpp"
  71 #include "utilities/events.hpp"
  72 #include "utilities/elfFile.hpp"
  73 #include "utilities/growableArray.hpp"
  74 #include "utilities/macros.hpp"
  75 #include "utilities/vmError.hpp"
  76 
  77 // put OS-includes here
  78 # include <sys/types.h>
  79 # include <sys/mman.h>
  80 # include <sys/stat.h>
  81 # include <sys/select.h>
  82 # include <pthread.h>
  83 # include <signal.h>
  84 # include <errno.h>
  85 # include <dlfcn.h>
  86 # include <stdio.h>
  87 # include <unistd.h>
  88 # include <sys/resource.h>
  89 # include <pthread.h>
  90 # include <sys/stat.h>
  91 # include <sys/time.h>
  92 # include <sys/times.h>
  93 # include <sys/utsname.h>
  94 # include <sys/socket.h>
  95 # include <sys/wait.h>
  96 # include <pwd.h>
  97 # include <poll.h>
  98 # include <fcntl.h>
  99 # include <string.h>
 100 # include <syscall.h>
 101 # include <sys/sysinfo.h>
 102 # include <gnu/libc-version.h>
 103 # include <sys/ipc.h>
 104 # include <sys/shm.h>
 105 # include <link.h>
 106 # include <stdint.h>
 107 # include <inttypes.h>
 108 # include <sys/ioctl.h>
 109 
 110 #ifndef _GNU_SOURCE
 111   #define _GNU_SOURCE
 112   #include <sched.h>
 113   #undef _GNU_SOURCE
 114 #else
 115   #include <sched.h>
 116 #endif
 117 
 118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 119 // getrusage() is prepared to handle the associated failure.
 120 #ifndef RUSAGE_THREAD
 121   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 122 #endif
 123 
 124 #define MAX_PATH    (2 * K)
 125 
 126 #define MAX_SECS 100000000
 127 
 128 // for timer info max values which include all bits
 129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 130 
 131 #define LARGEPAGES_BIT (1 << 6)
 132 #define DAX_SHARED_BIT (1 << 8)
 133 ////////////////////////////////////////////////////////////////////////////////
 134 // global variables
 135 julong os::Linux::_physical_memory = 0;
 136 
 137 address   os::Linux::_initial_thread_stack_bottom = NULL;
 138 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 139 
 140 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 141 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 142 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 143 Mutex* os::Linux::_createThread_lock = NULL;
 144 pthread_t os::Linux::_main_thread;
 145 int os::Linux::_page_size = -1;
 146 bool os::Linux::_supports_fast_thread_cpu_time = false;
 147 uint32_t os::Linux::_os_version = 0;
 148 const char * os::Linux::_glibc_version = NULL;
 149 const char * os::Linux::_libpthread_version = NULL;
 150 
 151 static jlong initial_time_count=0;
 152 
 153 static int clock_tics_per_sec = 100;
 154 
 155 // If the VM might have been created on the primordial thread, we need to resolve the
 156 // primordial thread stack bounds and check if the current thread might be the
 157 // primordial thread in places. If we know that the primordial thread is never used,
 158 // such as when the VM was created by one of the standard java launchers, we can
 159 // avoid this
 160 static bool suppress_primordial_thread_resolution = false;
 161 
 162 // For diagnostics to print a message once. see run_periodic_checks
 163 static sigset_t check_signal_done;
 164 static bool check_signals = true;
 165 
 166 // Signal number used to suspend/resume a thread
 167 
 168 // do not use any signal number less than SIGSEGV, see 4355769
 169 static int SR_signum = SIGUSR2;
 170 sigset_t SR_sigset;
 171 
 172 // utility functions
 173 
 174 static int SR_initialize();
 175 
 176 julong os::available_memory() {
 177   return Linux::available_memory();
 178 }
 179 
 180 julong os::Linux::available_memory() {
 181   // values in struct sysinfo are "unsigned long"
 182   struct sysinfo si;
 183   julong avail_mem;
 184 
 185   if (OSContainer::is_containerized()) {
 186     jlong mem_limit, mem_usage;
 187     if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
 188       log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
 189                              mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
 190     }
 191     if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
 192       log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
 193     }
 194     if (mem_limit > 0 && mem_usage > 0 ) {
 195       avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
 196       log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
 197       return avail_mem;
 198     }
 199   }
 200 
 201   sysinfo(&si);
 202   avail_mem = (julong)si.freeram * si.mem_unit;
 203   log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
 204   return avail_mem;
 205 }
 206 
 207 julong os::physical_memory() {
 208   jlong phys_mem = 0;
 209   if (OSContainer::is_containerized()) {
 210     jlong mem_limit;
 211     if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
 212       log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
 213       return mem_limit;
 214     }
 215     log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
 216                             mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
 217   }
 218 
 219   phys_mem = Linux::physical_memory();
 220   log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
 221   return phys_mem;
 222 }
 223 
 224 // Return true if user is running as root.
 225 
 226 bool os::have_special_privileges() {
 227   static bool init = false;
 228   static bool privileges = false;
 229   if (!init) {
 230     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 231     init = true;
 232   }
 233   return privileges;
 234 }
 235 
 236 
 237 #ifndef SYS_gettid
 238 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 239   #ifdef __ia64__
 240     #define SYS_gettid 1105
 241   #else
 242     #ifdef __i386__
 243       #define SYS_gettid 224
 244     #else
 245       #ifdef __amd64__
 246         #define SYS_gettid 186
 247       #else
 248         #ifdef __sparc__
 249           #define SYS_gettid 143
 250         #else
 251           #error define gettid for the arch
 252         #endif
 253       #endif
 254     #endif
 255   #endif
 256 #endif
 257 
 258 
 259 // pid_t gettid()
 260 //
 261 // Returns the kernel thread id of the currently running thread. Kernel
 262 // thread id is used to access /proc.
 263 pid_t os::Linux::gettid() {
 264   int rslt = syscall(SYS_gettid);
 265   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 266   return (pid_t)rslt;
 267 }
 268 
 269 // Most versions of linux have a bug where the number of processors are
 270 // determined by looking at the /proc file system.  In a chroot environment,
 271 // the system call returns 1.  This causes the VM to act as if it is
 272 // a single processor and elide locking (see is_MP() call).
 273 static bool unsafe_chroot_detected = false;
 274 static const char *unstable_chroot_error = "/proc file system not found.\n"
 275                      "Java may be unstable running multithreaded in a chroot "
 276                      "environment on Linux when /proc filesystem is not mounted.";
 277 
 278 void os::Linux::initialize_system_info() {
 279   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 280   if (processor_count() == 1) {
 281     pid_t pid = os::Linux::gettid();
 282     char fname[32];
 283     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 284     FILE *fp = fopen(fname, "r");
 285     if (fp == NULL) {
 286       unsafe_chroot_detected = true;
 287     } else {
 288       fclose(fp);
 289     }
 290   }
 291   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 292   assert(processor_count() > 0, "linux error");
 293 }
 294 
 295 void os::init_system_properties_values() {
 296   // The next steps are taken in the product version:
 297   //
 298   // Obtain the JAVA_HOME value from the location of libjvm.so.
 299   // This library should be located at:
 300   // <JAVA_HOME>/lib/{client|server}/libjvm.so.
 301   //
 302   // If "/jre/lib/" appears at the right place in the path, then we
 303   // assume libjvm.so is installed in a JDK and we use this path.
 304   //
 305   // Otherwise exit with message: "Could not create the Java virtual machine."
 306   //
 307   // The following extra steps are taken in the debugging version:
 308   //
 309   // If "/jre/lib/" does NOT appear at the right place in the path
 310   // instead of exit check for $JAVA_HOME environment variable.
 311   //
 312   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 313   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 314   // it looks like libjvm.so is installed there
 315   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 316   //
 317   // Otherwise exit.
 318   //
 319   // Important note: if the location of libjvm.so changes this
 320   // code needs to be changed accordingly.
 321 
 322   // See ld(1):
 323   //      The linker uses the following search paths to locate required
 324   //      shared libraries:
 325   //        1: ...
 326   //        ...
 327   //        7: The default directories, normally /lib and /usr/lib.
 328 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
 329   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 330 #else
 331   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 332 #endif
 333 
 334 // Base path of extensions installed on the system.
 335 #define SYS_EXT_DIR     "/usr/java/packages"
 336 #define EXTENSIONS_DIR  "/lib/ext"
 337 
 338   // Buffer that fits several sprintfs.
 339   // Note that the space for the colon and the trailing null are provided
 340   // by the nulls included by the sizeof operator.
 341   const size_t bufsize =
 342     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 343          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 344   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 345 
 346   // sysclasspath, java_home, dll_dir
 347   {
 348     char *pslash;
 349     os::jvm_path(buf, bufsize);
 350 
 351     // Found the full path to libjvm.so.
 352     // Now cut the path to <java_home>/jre if we can.
 353     pslash = strrchr(buf, '/');
 354     if (pslash != NULL) {
 355       *pslash = '\0';            // Get rid of /libjvm.so.
 356     }
 357     pslash = strrchr(buf, '/');
 358     if (pslash != NULL) {
 359       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 360     }
 361     Arguments::set_dll_dir(buf);
 362 
 363     if (pslash != NULL) {
 364       pslash = strrchr(buf, '/');
 365       if (pslash != NULL) {
 366         *pslash = '\0';        // Get rid of /lib.
 367       }
 368     }
 369     Arguments::set_java_home(buf);
 370     if (!set_boot_path('/', ':')) {
 371       vm_exit_during_initialization("Failed setting boot class path.", NULL);
 372     }
 373   }
 374 
 375   // Where to look for native libraries.
 376   //
 377   // Note: Due to a legacy implementation, most of the library path
 378   // is set in the launcher. This was to accomodate linking restrictions
 379   // on legacy Linux implementations (which are no longer supported).
 380   // Eventually, all the library path setting will be done here.
 381   //
 382   // However, to prevent the proliferation of improperly built native
 383   // libraries, the new path component /usr/java/packages is added here.
 384   // Eventually, all the library path setting will be done here.
 385   {
 386     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 387     // should always exist (until the legacy problem cited above is
 388     // addressed).
 389     const char *v = ::getenv("LD_LIBRARY_PATH");
 390     const char *v_colon = ":";
 391     if (v == NULL) { v = ""; v_colon = ""; }
 392     // That's +1 for the colon and +1 for the trailing '\0'.
 393     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 394                                                      strlen(v) + 1 +
 395                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
 396                                                      mtInternal);
 397     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
 398     Arguments::set_library_path(ld_library_path);
 399     FREE_C_HEAP_ARRAY(char, ld_library_path);
 400   }
 401 
 402   // Extensions directories.
 403   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 404   Arguments::set_ext_dirs(buf);
 405 
 406   FREE_C_HEAP_ARRAY(char, buf);
 407 
 408 #undef DEFAULT_LIBPATH
 409 #undef SYS_EXT_DIR
 410 #undef EXTENSIONS_DIR
 411 }
 412 
 413 ////////////////////////////////////////////////////////////////////////////////
 414 // breakpoint support
 415 
 416 void os::breakpoint() {
 417   BREAKPOINT;
 418 }
 419 
 420 extern "C" void breakpoint() {
 421   // use debugger to set breakpoint here
 422 }
 423 
 424 ////////////////////////////////////////////////////////////////////////////////
 425 // signal support
 426 
 427 debug_only(static bool signal_sets_initialized = false);
 428 static sigset_t unblocked_sigs, vm_sigs;
 429 
 430 void os::Linux::signal_sets_init() {
 431   // Should also have an assertion stating we are still single-threaded.
 432   assert(!signal_sets_initialized, "Already initialized");
 433   // Fill in signals that are necessarily unblocked for all threads in
 434   // the VM. Currently, we unblock the following signals:
 435   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 436   //                         by -Xrs (=ReduceSignalUsage));
 437   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 438   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 439   // the dispositions or masks wrt these signals.
 440   // Programs embedding the VM that want to use the above signals for their
 441   // own purposes must, at this time, use the "-Xrs" option to prevent
 442   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 443   // (See bug 4345157, and other related bugs).
 444   // In reality, though, unblocking these signals is really a nop, since
 445   // these signals are not blocked by default.
 446   sigemptyset(&unblocked_sigs);
 447   sigaddset(&unblocked_sigs, SIGILL);
 448   sigaddset(&unblocked_sigs, SIGSEGV);
 449   sigaddset(&unblocked_sigs, SIGBUS);
 450   sigaddset(&unblocked_sigs, SIGFPE);
 451 #if defined(PPC64)
 452   sigaddset(&unblocked_sigs, SIGTRAP);
 453 #endif
 454   sigaddset(&unblocked_sigs, SR_signum);
 455 
 456   if (!ReduceSignalUsage) {
 457     if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 458       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 459     }
 460     if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 461       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 462     }
 463     if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 464       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 465     }
 466   }
 467   // Fill in signals that are blocked by all but the VM thread.
 468   sigemptyset(&vm_sigs);
 469   if (!ReduceSignalUsage) {
 470     sigaddset(&vm_sigs, BREAK_SIGNAL);
 471   }
 472   debug_only(signal_sets_initialized = true);
 473 
 474 }
 475 
 476 // These are signals that are unblocked while a thread is running Java.
 477 // (For some reason, they get blocked by default.)
 478 sigset_t* os::Linux::unblocked_signals() {
 479   assert(signal_sets_initialized, "Not initialized");
 480   return &unblocked_sigs;
 481 }
 482 
 483 // These are the signals that are blocked while a (non-VM) thread is
 484 // running Java. Only the VM thread handles these signals.
 485 sigset_t* os::Linux::vm_signals() {
 486   assert(signal_sets_initialized, "Not initialized");
 487   return &vm_sigs;
 488 }
 489 
 490 void os::Linux::hotspot_sigmask(Thread* thread) {
 491 
 492   //Save caller's signal mask before setting VM signal mask
 493   sigset_t caller_sigmask;
 494   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 495 
 496   OSThread* osthread = thread->osthread();
 497   osthread->set_caller_sigmask(caller_sigmask);
 498 
 499   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 500 
 501   if (!ReduceSignalUsage) {
 502     if (thread->is_VM_thread()) {
 503       // Only the VM thread handles BREAK_SIGNAL ...
 504       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 505     } else {
 506       // ... all other threads block BREAK_SIGNAL
 507       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 508     }
 509   }
 510 }
 511 
 512 //////////////////////////////////////////////////////////////////////////////
 513 // detecting pthread library
 514 
 515 void os::Linux::libpthread_init() {
 516   // Save glibc and pthread version strings.
 517 #if !defined(_CS_GNU_LIBC_VERSION) || \
 518     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 519   #error "glibc too old (< 2.3.2)"
 520 #endif
 521 
 522   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 523   assert(n > 0, "cannot retrieve glibc version");
 524   char *str = (char *)malloc(n, mtInternal);
 525   confstr(_CS_GNU_LIBC_VERSION, str, n);
 526   os::Linux::set_glibc_version(str);
 527 
 528   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 529   assert(n > 0, "cannot retrieve pthread version");
 530   str = (char *)malloc(n, mtInternal);
 531   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 532   os::Linux::set_libpthread_version(str);
 533 }
 534 
 535 /////////////////////////////////////////////////////////////////////////////
 536 // thread stack expansion
 537 
 538 // os::Linux::manually_expand_stack() takes care of expanding the thread
 539 // stack. Note that this is normally not needed: pthread stacks allocate
 540 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 541 // committed. Therefore it is not necessary to expand the stack manually.
 542 //
 543 // Manually expanding the stack was historically needed on LinuxThreads
 544 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 545 // it is kept to deal with very rare corner cases:
 546 //
 547 // For one, user may run the VM on an own implementation of threads
 548 // whose stacks are - like the old LinuxThreads - implemented using
 549 // mmap(MAP_GROWSDOWN).
 550 //
 551 // Also, this coding may be needed if the VM is running on the primordial
 552 // thread. Normally we avoid running on the primordial thread; however,
 553 // user may still invoke the VM on the primordial thread.
 554 //
 555 // The following historical comment describes the details about running
 556 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 557 
 558 
 559 // Force Linux kernel to expand current thread stack. If "bottom" is close
 560 // to the stack guard, caller should block all signals.
 561 //
 562 // MAP_GROWSDOWN:
 563 //   A special mmap() flag that is used to implement thread stacks. It tells
 564 //   kernel that the memory region should extend downwards when needed. This
 565 //   allows early versions of LinuxThreads to only mmap the first few pages
 566 //   when creating a new thread. Linux kernel will automatically expand thread
 567 //   stack as needed (on page faults).
 568 //
 569 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 570 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 571 //   region, it's hard to tell if the fault is due to a legitimate stack
 572 //   access or because of reading/writing non-exist memory (e.g. buffer
 573 //   overrun). As a rule, if the fault happens below current stack pointer,
 574 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 575 //   application (see Linux kernel fault.c).
 576 //
 577 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 578 //   stack overflow detection.
 579 //
 580 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 581 //   not use MAP_GROWSDOWN.
 582 //
 583 // To get around the problem and allow stack banging on Linux, we need to
 584 // manually expand thread stack after receiving the SIGSEGV.
 585 //
 586 // There are two ways to expand thread stack to address "bottom", we used
 587 // both of them in JVM before 1.5:
 588 //   1. adjust stack pointer first so that it is below "bottom", and then
 589 //      touch "bottom"
 590 //   2. mmap() the page in question
 591 //
 592 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 593 // if current sp is already near the lower end of page 101, and we need to
 594 // call mmap() to map page 100, it is possible that part of the mmap() frame
 595 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 596 // That will destroy the mmap() frame and cause VM to crash.
 597 //
 598 // The following code works by adjusting sp first, then accessing the "bottom"
 599 // page to force a page fault. Linux kernel will then automatically expand the
 600 // stack mapping.
 601 //
 602 // _expand_stack_to() assumes its frame size is less than page size, which
 603 // should always be true if the function is not inlined.
 604 
 605 static void NOINLINE _expand_stack_to(address bottom) {
 606   address sp;
 607   size_t size;
 608   volatile char *p;
 609 
 610   // Adjust bottom to point to the largest address within the same page, it
 611   // gives us a one-page buffer if alloca() allocates slightly more memory.
 612   bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
 613   bottom += os::Linux::page_size() - 1;
 614 
 615   // sp might be slightly above current stack pointer; if that's the case, we
 616   // will alloca() a little more space than necessary, which is OK. Don't use
 617   // os::current_stack_pointer(), as its result can be slightly below current
 618   // stack pointer, causing us to not alloca enough to reach "bottom".
 619   sp = (address)&sp;
 620 
 621   if (sp > bottom) {
 622     size = sp - bottom;
 623     p = (volatile char *)alloca(size);
 624     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 625     p[0] = '\0';
 626   }
 627 }
 628 
 629 void os::Linux::expand_stack_to(address bottom) {
 630   _expand_stack_to(bottom);
 631 }
 632 
 633 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 634   assert(t!=NULL, "just checking");
 635   assert(t->osthread()->expanding_stack(), "expand should be set");
 636   assert(t->stack_base() != NULL, "stack_base was not initialized");
 637 
 638   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 639     sigset_t mask_all, old_sigset;
 640     sigfillset(&mask_all);
 641     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 642     _expand_stack_to(addr);
 643     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 644     return true;
 645   }
 646   return false;
 647 }
 648 
 649 //////////////////////////////////////////////////////////////////////////////
 650 // create new thread
 651 
 652 // Thread start routine for all newly created threads
 653 static void *thread_native_entry(Thread *thread) {
 654   // Try to randomize the cache line index of hot stack frames.
 655   // This helps when threads of the same stack traces evict each other's
 656   // cache lines. The threads can be either from the same JVM instance, or
 657   // from different JVM instances. The benefit is especially true for
 658   // processors with hyperthreading technology.
 659   static int counter = 0;
 660   int pid = os::current_process_id();
 661   alloca(((pid ^ counter++) & 7) * 128);
 662 
 663   thread->initialize_thread_current();
 664 
 665   OSThread* osthread = thread->osthread();
 666   Monitor* sync = osthread->startThread_lock();
 667 
 668   osthread->set_thread_id(os::current_thread_id());
 669 
 670   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 671     os::current_thread_id(), (uintx) pthread_self());
 672 
 673   if (UseNUMA) {
 674     int lgrp_id = os::numa_get_group_id();
 675     if (lgrp_id != -1) {
 676       thread->set_lgrp_id(lgrp_id);
 677     }
 678   }
 679   // initialize signal mask for this thread
 680   os::Linux::hotspot_sigmask(thread);
 681 
 682   // initialize floating point control register
 683   os::Linux::init_thread_fpu_state();
 684 
 685   // handshaking with parent thread
 686   {
 687     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 688 
 689     // notify parent thread
 690     osthread->set_state(INITIALIZED);
 691     sync->notify_all();
 692 
 693     // wait until os::start_thread()
 694     while (osthread->get_state() == INITIALIZED) {
 695       sync->wait(Mutex::_no_safepoint_check_flag);
 696     }
 697   }
 698 
 699   // call one more level start routine
 700   thread->run();
 701 
 702   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 703     os::current_thread_id(), (uintx) pthread_self());
 704 
 705   // If a thread has not deleted itself ("delete this") as part of its
 706   // termination sequence, we have to ensure thread-local-storage is
 707   // cleared before we actually terminate. No threads should ever be
 708   // deleted asynchronously with respect to their termination.
 709   if (Thread::current_or_null_safe() != NULL) {
 710     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 711     thread->clear_thread_current();
 712   }
 713 
 714   return 0;
 715 }
 716 
 717 bool os::create_thread(Thread* thread, ThreadType thr_type,
 718                        size_t req_stack_size) {
 719   assert(thread->osthread() == NULL, "caller responsible");
 720 
 721   // Allocate the OSThread object
 722   OSThread* osthread = new OSThread(NULL, NULL);
 723   if (osthread == NULL) {
 724     return false;
 725   }
 726 
 727   // set the correct thread state
 728   osthread->set_thread_type(thr_type);
 729 
 730   // Initial state is ALLOCATED but not INITIALIZED
 731   osthread->set_state(ALLOCATED);
 732 
 733   thread->set_osthread(osthread);
 734 
 735   // init thread attributes
 736   pthread_attr_t attr;
 737   pthread_attr_init(&attr);
 738   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 739 
 740   // Calculate stack size if it's not specified by caller.
 741   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 742   // In the Linux NPTL pthread implementation the guard size mechanism
 743   // is not implemented properly. The posix standard requires adding
 744   // the size of the guard pages to the stack size, instead Linux
 745   // takes the space out of 'stacksize'. Thus we adapt the requested
 746   // stack_size by the size of the guard pages to mimick proper
 747   // behaviour. However, be careful not to end up with a size
 748   // of zero due to overflow. Don't add the guard page in that case.
 749   size_t guard_size = os::Linux::default_guard_size(thr_type);
 750   if (stack_size <= SIZE_MAX - guard_size) {
 751     stack_size += guard_size;
 752   }
 753   assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
 754 
 755   int status = pthread_attr_setstacksize(&attr, stack_size);
 756   assert_status(status == 0, status, "pthread_attr_setstacksize");
 757 
 758   // Configure glibc guard page.
 759   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 760 
 761   ThreadState state;
 762 
 763   {
 764     pthread_t tid;
 765     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 766 
 767     char buf[64];
 768     if (ret == 0) {
 769       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 770         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 771     } else {
 772       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 773         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 774     }
 775 
 776     pthread_attr_destroy(&attr);
 777 
 778     if (ret != 0) {
 779       // Need to clean up stuff we've allocated so far
 780       thread->set_osthread(NULL);
 781       delete osthread;
 782       return false;
 783     }
 784 
 785     // Store pthread info into the OSThread
 786     osthread->set_pthread_id(tid);
 787 
 788     // Wait until child thread is either initialized or aborted
 789     {
 790       Monitor* sync_with_child = osthread->startThread_lock();
 791       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 792       while ((state = osthread->get_state()) == ALLOCATED) {
 793         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 794       }
 795     }
 796   }
 797 
 798   // Aborted due to thread limit being reached
 799   if (state == ZOMBIE) {
 800     thread->set_osthread(NULL);
 801     delete osthread;
 802     return false;
 803   }
 804 
 805   // The thread is returned suspended (in state INITIALIZED),
 806   // and is started higher up in the call chain
 807   assert(state == INITIALIZED, "race condition");
 808   return true;
 809 }
 810 
 811 /////////////////////////////////////////////////////////////////////////////
 812 // attach existing thread
 813 
 814 // bootstrap the main thread
 815 bool os::create_main_thread(JavaThread* thread) {
 816   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 817   return create_attached_thread(thread);
 818 }
 819 
 820 bool os::create_attached_thread(JavaThread* thread) {
 821 #ifdef ASSERT
 822   thread->verify_not_published();
 823 #endif
 824 
 825   // Allocate the OSThread object
 826   OSThread* osthread = new OSThread(NULL, NULL);
 827 
 828   if (osthread == NULL) {
 829     return false;
 830   }
 831 
 832   // Store pthread info into the OSThread
 833   osthread->set_thread_id(os::Linux::gettid());
 834   osthread->set_pthread_id(::pthread_self());
 835 
 836   // initialize floating point control register
 837   os::Linux::init_thread_fpu_state();
 838 
 839   // Initial thread state is RUNNABLE
 840   osthread->set_state(RUNNABLE);
 841 
 842   thread->set_osthread(osthread);
 843 
 844   if (UseNUMA) {
 845     int lgrp_id = os::numa_get_group_id();
 846     if (lgrp_id != -1) {
 847       thread->set_lgrp_id(lgrp_id);
 848     }
 849   }
 850 
 851   if (os::is_primordial_thread()) {
 852     // If current thread is primordial thread, its stack is mapped on demand,
 853     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 854     // the entire stack region to avoid SEGV in stack banging.
 855     // It is also useful to get around the heap-stack-gap problem on SuSE
 856     // kernel (see 4821821 for details). We first expand stack to the top
 857     // of yellow zone, then enable stack yellow zone (order is significant,
 858     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 859     // is no gap between the last two virtual memory regions.
 860 
 861     JavaThread *jt = (JavaThread *)thread;
 862     address addr = jt->stack_reserved_zone_base();
 863     assert(addr != NULL, "initialization problem?");
 864     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 865 
 866     osthread->set_expanding_stack();
 867     os::Linux::manually_expand_stack(jt, addr);
 868     osthread->clear_expanding_stack();
 869   }
 870 
 871   // initialize signal mask for this thread
 872   // and save the caller's signal mask
 873   os::Linux::hotspot_sigmask(thread);
 874 
 875   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 876     os::current_thread_id(), (uintx) pthread_self());
 877 
 878   return true;
 879 }
 880 
 881 void os::pd_start_thread(Thread* thread) {
 882   OSThread * osthread = thread->osthread();
 883   assert(osthread->get_state() != INITIALIZED, "just checking");
 884   Monitor* sync_with_child = osthread->startThread_lock();
 885   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 886   sync_with_child->notify();
 887 }
 888 
 889 // Free Linux resources related to the OSThread
 890 void os::free_thread(OSThread* osthread) {
 891   assert(osthread != NULL, "osthread not set");
 892 
 893   // We are told to free resources of the argument thread,
 894   // but we can only really operate on the current thread.
 895   assert(Thread::current()->osthread() == osthread,
 896          "os::free_thread but not current thread");
 897 
 898 #ifdef ASSERT
 899   sigset_t current;
 900   sigemptyset(&current);
 901   pthread_sigmask(SIG_SETMASK, NULL, &current);
 902   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 903 #endif
 904 
 905   // Restore caller's signal mask
 906   sigset_t sigmask = osthread->caller_sigmask();
 907   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 908 
 909   delete osthread;
 910 }
 911 
 912 //////////////////////////////////////////////////////////////////////////////
 913 // primordial thread
 914 
 915 // Check if current thread is the primordial thread, similar to Solaris thr_main.
 916 bool os::is_primordial_thread(void) {
 917   if (suppress_primordial_thread_resolution) {
 918     return false;
 919   }
 920   char dummy;
 921   // If called before init complete, thread stack bottom will be null.
 922   // Can be called if fatal error occurs before initialization.
 923   if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
 924   assert(os::Linux::initial_thread_stack_bottom() != NULL &&
 925          os::Linux::initial_thread_stack_size()   != 0,
 926          "os::init did not locate primordial thread's stack region");
 927   if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
 928       (address)&dummy < os::Linux::initial_thread_stack_bottom() +
 929                         os::Linux::initial_thread_stack_size()) {
 930     return true;
 931   } else {
 932     return false;
 933   }
 934 }
 935 
 936 // Find the virtual memory area that contains addr
 937 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 938   FILE *fp = fopen("/proc/self/maps", "r");
 939   if (fp) {
 940     address low, high;
 941     while (!feof(fp)) {
 942       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 943         if (low <= addr && addr < high) {
 944           if (vma_low)  *vma_low  = low;
 945           if (vma_high) *vma_high = high;
 946           fclose(fp);
 947           return true;
 948         }
 949       }
 950       for (;;) {
 951         int ch = fgetc(fp);
 952         if (ch == EOF || ch == (int)'\n') break;
 953       }
 954     }
 955     fclose(fp);
 956   }
 957   return false;
 958 }
 959 
 960 // Locate primordial thread stack. This special handling of primordial thread stack
 961 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 962 // bogus value for the primordial process thread. While the launcher has created
 963 // the VM in a new thread since JDK 6, we still have to allow for the use of the
 964 // JNI invocation API from a primordial thread.
 965 void os::Linux::capture_initial_stack(size_t max_size) {
 966 
 967   // max_size is either 0 (which means accept OS default for thread stacks) or
 968   // a user-specified value known to be at least the minimum needed. If we
 969   // are actually on the primordial thread we can make it appear that we have a
 970   // smaller max_size stack by inserting the guard pages at that location. But we
 971   // cannot do anything to emulate a larger stack than what has been provided by
 972   // the OS or threading library. In fact if we try to use a stack greater than
 973   // what is set by rlimit then we will crash the hosting process.
 974 
 975   // Maximum stack size is the easy part, get it from RLIMIT_STACK.
 976   // If this is "unlimited" then it will be a huge value.
 977   struct rlimit rlim;
 978   getrlimit(RLIMIT_STACK, &rlim);
 979   size_t stack_size = rlim.rlim_cur;
 980 
 981   // 6308388: a bug in ld.so will relocate its own .data section to the
 982   //   lower end of primordial stack; reduce ulimit -s value a little bit
 983   //   so we won't install guard page on ld.so's data section.
 984   //   But ensure we don't underflow the stack size - allow 1 page spare
 985   if (stack_size >= (size_t)(3 * page_size())) {
 986     stack_size -= 2 * page_size();
 987   }
 988 
 989   // Try to figure out where the stack base (top) is. This is harder.
 990   //
 991   // When an application is started, glibc saves the initial stack pointer in
 992   // a global variable "__libc_stack_end", which is then used by system
 993   // libraries. __libc_stack_end should be pretty close to stack top. The
 994   // variable is available since the very early days. However, because it is
 995   // a private interface, it could disappear in the future.
 996   //
 997   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 998   // to __libc_stack_end, it is very close to stack top, but isn't the real
 999   // stack top. Note that /proc may not exist if VM is running as a chroot
1000   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1001   // /proc/<pid>/stat could change in the future (though unlikely).
1002   //
1003   // We try __libc_stack_end first. If that doesn't work, look for
1004   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1005   // as a hint, which should work well in most cases.
1006 
1007   uintptr_t stack_start;
1008 
1009   // try __libc_stack_end first
1010   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1011   if (p && *p) {
1012     stack_start = *p;
1013   } else {
1014     // see if we can get the start_stack field from /proc/self/stat
1015     FILE *fp;
1016     int pid;
1017     char state;
1018     int ppid;
1019     int pgrp;
1020     int session;
1021     int nr;
1022     int tpgrp;
1023     unsigned long flags;
1024     unsigned long minflt;
1025     unsigned long cminflt;
1026     unsigned long majflt;
1027     unsigned long cmajflt;
1028     unsigned long utime;
1029     unsigned long stime;
1030     long cutime;
1031     long cstime;
1032     long prio;
1033     long nice;
1034     long junk;
1035     long it_real;
1036     uintptr_t start;
1037     uintptr_t vsize;
1038     intptr_t rss;
1039     uintptr_t rsslim;
1040     uintptr_t scodes;
1041     uintptr_t ecode;
1042     int i;
1043 
1044     // Figure what the primordial thread stack base is. Code is inspired
1045     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1046     // followed by command name surrounded by parentheses, state, etc.
1047     char stat[2048];
1048     int statlen;
1049 
1050     fp = fopen("/proc/self/stat", "r");
1051     if (fp) {
1052       statlen = fread(stat, 1, 2047, fp);
1053       stat[statlen] = '\0';
1054       fclose(fp);
1055 
1056       // Skip pid and the command string. Note that we could be dealing with
1057       // weird command names, e.g. user could decide to rename java launcher
1058       // to "java 1.4.2 :)", then the stat file would look like
1059       //                1234 (java 1.4.2 :)) R ... ...
1060       // We don't really need to know the command string, just find the last
1061       // occurrence of ")" and then start parsing from there. See bug 4726580.
1062       char * s = strrchr(stat, ')');
1063 
1064       i = 0;
1065       if (s) {
1066         // Skip blank chars
1067         do { s++; } while (s && isspace(*s));
1068 
1069 #define _UFM UINTX_FORMAT
1070 #define _DFM INTX_FORMAT
1071 
1072         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1073         //              3  4  5  6  7  8   9   0   1   2   3   4   5   6   7   8   9   0   1    2    3    4    5    6    7    8
1074         i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1075                    &state,          // 3  %c
1076                    &ppid,           // 4  %d
1077                    &pgrp,           // 5  %d
1078                    &session,        // 6  %d
1079                    &nr,             // 7  %d
1080                    &tpgrp,          // 8  %d
1081                    &flags,          // 9  %lu
1082                    &minflt,         // 10 %lu
1083                    &cminflt,        // 11 %lu
1084                    &majflt,         // 12 %lu
1085                    &cmajflt,        // 13 %lu
1086                    &utime,          // 14 %lu
1087                    &stime,          // 15 %lu
1088                    &cutime,         // 16 %ld
1089                    &cstime,         // 17 %ld
1090                    &prio,           // 18 %ld
1091                    &nice,           // 19 %ld
1092                    &junk,           // 20 %ld
1093                    &it_real,        // 21 %ld
1094                    &start,          // 22 UINTX_FORMAT
1095                    &vsize,          // 23 UINTX_FORMAT
1096                    &rss,            // 24 INTX_FORMAT
1097                    &rsslim,         // 25 UINTX_FORMAT
1098                    &scodes,         // 26 UINTX_FORMAT
1099                    &ecode,          // 27 UINTX_FORMAT
1100                    &stack_start);   // 28 UINTX_FORMAT
1101       }
1102 
1103 #undef _UFM
1104 #undef _DFM
1105 
1106       if (i != 28 - 2) {
1107         assert(false, "Bad conversion from /proc/self/stat");
1108         // product mode - assume we are the primordial thread, good luck in the
1109         // embedded case.
1110         warning("Can't detect primordial thread stack location - bad conversion");
1111         stack_start = (uintptr_t) &rlim;
1112       }
1113     } else {
1114       // For some reason we can't open /proc/self/stat (for example, running on
1115       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1116       // most cases, so don't abort:
1117       warning("Can't detect primordial thread stack location - no /proc/self/stat");
1118       stack_start = (uintptr_t) &rlim;
1119     }
1120   }
1121 
1122   // Now we have a pointer (stack_start) very close to the stack top, the
1123   // next thing to do is to figure out the exact location of stack top. We
1124   // can find out the virtual memory area that contains stack_start by
1125   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1126   // and its upper limit is the real stack top. (again, this would fail if
1127   // running inside chroot, because /proc may not exist.)
1128 
1129   uintptr_t stack_top;
1130   address low, high;
1131   if (find_vma((address)stack_start, &low, &high)) {
1132     // success, "high" is the true stack top. (ignore "low", because initial
1133     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1134     stack_top = (uintptr_t)high;
1135   } else {
1136     // failed, likely because /proc/self/maps does not exist
1137     warning("Can't detect primordial thread stack location - find_vma failed");
1138     // best effort: stack_start is normally within a few pages below the real
1139     // stack top, use it as stack top, and reduce stack size so we won't put
1140     // guard page outside stack.
1141     stack_top = stack_start;
1142     stack_size -= 16 * page_size();
1143   }
1144 
1145   // stack_top could be partially down the page so align it
1146   stack_top = align_up(stack_top, page_size());
1147 
1148   // Allowed stack value is minimum of max_size and what we derived from rlimit
1149   if (max_size > 0) {
1150     _initial_thread_stack_size = MIN2(max_size, stack_size);
1151   } else {
1152     // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1153     // clamp it at 8MB as we do on Solaris
1154     _initial_thread_stack_size = MIN2(stack_size, 8*M);
1155   }
1156   _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1157   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1158 
1159   assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1160 
1161   if (log_is_enabled(Info, os, thread)) {
1162     // See if we seem to be on primordial process thread
1163     bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1164                       uintptr_t(&rlim) < stack_top;
1165 
1166     log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1167                          SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1168                          primordial ? "primordial" : "user", max_size / K,  _initial_thread_stack_size / K,
1169                          stack_top, intptr_t(_initial_thread_stack_bottom));
1170   }
1171 }
1172 
1173 ////////////////////////////////////////////////////////////////////////////////
1174 // time support
1175 
1176 // Time since start-up in seconds to a fine granularity.
1177 // Used by VMSelfDestructTimer and the MemProfiler.
1178 double os::elapsedTime() {
1179 
1180   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1181 }
1182 
1183 jlong os::elapsed_counter() {
1184   return javaTimeNanos() - initial_time_count;
1185 }
1186 
1187 jlong os::elapsed_frequency() {
1188   return NANOSECS_PER_SEC; // nanosecond resolution
1189 }
1190 
1191 bool os::supports_vtime() { return true; }
1192 bool os::enable_vtime()   { return false; }
1193 bool os::vtime_enabled()  { return false; }
1194 
1195 double os::elapsedVTime() {
1196   struct rusage usage;
1197   int retval = getrusage(RUSAGE_THREAD, &usage);
1198   if (retval == 0) {
1199     return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1200   } else {
1201     // better than nothing, but not much
1202     return elapsedTime();
1203   }
1204 }
1205 
1206 jlong os::javaTimeMillis() {
1207   timeval time;
1208   int status = gettimeofday(&time, NULL);
1209   assert(status != -1, "linux error");
1210   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1211 }
1212 
1213 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1214   timeval time;
1215   int status = gettimeofday(&time, NULL);
1216   assert(status != -1, "linux error");
1217   seconds = jlong(time.tv_sec);
1218   nanos = jlong(time.tv_usec) * 1000;
1219 }
1220 
1221 
1222 #ifndef CLOCK_MONOTONIC
1223   #define CLOCK_MONOTONIC (1)
1224 #endif
1225 
1226 void os::Linux::clock_init() {
1227   // we do dlopen's in this particular order due to bug in linux
1228   // dynamical loader (see 6348968) leading to crash on exit
1229   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1230   if (handle == NULL) {
1231     handle = dlopen("librt.so", RTLD_LAZY);
1232   }
1233 
1234   if (handle) {
1235     int (*clock_getres_func)(clockid_t, struct timespec*) =
1236            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1237     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1238            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1239     if (clock_getres_func && clock_gettime_func) {
1240       // See if monotonic clock is supported by the kernel. Note that some
1241       // early implementations simply return kernel jiffies (updated every
1242       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1243       // for nano time (though the monotonic property is still nice to have).
1244       // It's fixed in newer kernels, however clock_getres() still returns
1245       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1246       // resolution for now. Hopefully as people move to new kernels, this
1247       // won't be a problem.
1248       struct timespec res;
1249       struct timespec tp;
1250       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1251           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1252         // yes, monotonic clock is supported
1253         _clock_gettime = clock_gettime_func;
1254         return;
1255       } else {
1256         // close librt if there is no monotonic clock
1257         dlclose(handle);
1258       }
1259     }
1260   }
1261   warning("No monotonic clock was available - timed services may " \
1262           "be adversely affected if the time-of-day clock changes");
1263 }
1264 
1265 #ifndef SYS_clock_getres
1266   #if defined(X86) || defined(PPC64) || defined(S390)
1267     #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1268     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1269   #else
1270     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1271     #define sys_clock_getres(x,y)  -1
1272   #endif
1273 #else
1274   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1275 #endif
1276 
1277 void os::Linux::fast_thread_clock_init() {
1278   if (!UseLinuxPosixThreadCPUClocks) {
1279     return;
1280   }
1281   clockid_t clockid;
1282   struct timespec tp;
1283   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1284       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1285 
1286   // Switch to using fast clocks for thread cpu time if
1287   // the sys_clock_getres() returns 0 error code.
1288   // Note, that some kernels may support the current thread
1289   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1290   // returned by the pthread_getcpuclockid().
1291   // If the fast Posix clocks are supported then the sys_clock_getres()
1292   // must return at least tp.tv_sec == 0 which means a resolution
1293   // better than 1 sec. This is extra check for reliability.
1294 
1295   if (pthread_getcpuclockid_func &&
1296       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1297       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1298     _supports_fast_thread_cpu_time = true;
1299     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1300   }
1301 }
1302 
1303 jlong os::javaTimeNanos() {
1304   if (os::supports_monotonic_clock()) {
1305     struct timespec tp;
1306     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1307     assert(status == 0, "gettime error");
1308     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1309     return result;
1310   } else {
1311     timeval time;
1312     int status = gettimeofday(&time, NULL);
1313     assert(status != -1, "linux error");
1314     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1315     return 1000 * usecs;
1316   }
1317 }
1318 
1319 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1320   if (os::supports_monotonic_clock()) {
1321     info_ptr->max_value = ALL_64_BITS;
1322 
1323     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1324     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1325     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1326   } else {
1327     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1328     info_ptr->max_value = ALL_64_BITS;
1329 
1330     // gettimeofday is a real time clock so it skips
1331     info_ptr->may_skip_backward = true;
1332     info_ptr->may_skip_forward = true;
1333   }
1334 
1335   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1336 }
1337 
1338 // Return the real, user, and system times in seconds from an
1339 // arbitrary fixed point in the past.
1340 bool os::getTimesSecs(double* process_real_time,
1341                       double* process_user_time,
1342                       double* process_system_time) {
1343   struct tms ticks;
1344   clock_t real_ticks = times(&ticks);
1345 
1346   if (real_ticks == (clock_t) (-1)) {
1347     return false;
1348   } else {
1349     double ticks_per_second = (double) clock_tics_per_sec;
1350     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1351     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1352     *process_real_time = ((double) real_ticks) / ticks_per_second;
1353 
1354     return true;
1355   }
1356 }
1357 
1358 
1359 char * os::local_time_string(char *buf, size_t buflen) {
1360   struct tm t;
1361   time_t long_time;
1362   time(&long_time);
1363   localtime_r(&long_time, &t);
1364   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1365                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1366                t.tm_hour, t.tm_min, t.tm_sec);
1367   return buf;
1368 }
1369 
1370 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1371   return localtime_r(clock, res);
1372 }
1373 
1374 ////////////////////////////////////////////////////////////////////////////////
1375 // runtime exit support
1376 
1377 // Note: os::shutdown() might be called very early during initialization, or
1378 // called from signal handler. Before adding something to os::shutdown(), make
1379 // sure it is async-safe and can handle partially initialized VM.
1380 void os::shutdown() {
1381 
1382   // allow PerfMemory to attempt cleanup of any persistent resources
1383   perfMemory_exit();
1384 
1385   // needs to remove object in file system
1386   AttachListener::abort();
1387 
1388   // flush buffered output, finish log files
1389   ostream_abort();
1390 
1391   // Check for abort hook
1392   abort_hook_t abort_hook = Arguments::abort_hook();
1393   if (abort_hook != NULL) {
1394     abort_hook();
1395   }
1396 
1397 }
1398 
1399 // Note: os::abort() might be called very early during initialization, or
1400 // called from signal handler. Before adding something to os::abort(), make
1401 // sure it is async-safe and can handle partially initialized VM.
1402 void os::abort(bool dump_core, void* siginfo, const void* context) {
1403   os::shutdown();
1404   if (dump_core) {
1405 #ifndef PRODUCT
1406     fdStream out(defaultStream::output_fd());
1407     out.print_raw("Current thread is ");
1408     char buf[16];
1409     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1410     out.print_raw_cr(buf);
1411     out.print_raw_cr("Dumping core ...");
1412 #endif
1413     ::abort(); // dump core
1414   }
1415 
1416   ::exit(1);
1417 }
1418 
1419 // Die immediately, no exit hook, no abort hook, no cleanup.
1420 void os::die() {
1421   ::abort();
1422 }
1423 
1424 // thread_id is kernel thread id (similar to Solaris LWP id)
1425 intx os::current_thread_id() { return os::Linux::gettid(); }
1426 int os::current_process_id() {
1427   return ::getpid();
1428 }
1429 
1430 // DLL functions
1431 
1432 const char* os::dll_file_extension() { return ".so"; }
1433 
1434 // This must be hard coded because it's the system's temporary
1435 // directory not the java application's temp directory, ala java.io.tmpdir.
1436 const char* os::get_temp_directory() { return "/tmp"; }
1437 
1438 static bool file_exists(const char* filename) {
1439   struct stat statbuf;
1440   if (filename == NULL || strlen(filename) == 0) {
1441     return false;
1442   }
1443   return os::stat(filename, &statbuf) == 0;
1444 }
1445 
1446 // check if addr is inside libjvm.so
1447 bool os::address_is_in_vm(address addr) {
1448   static address libjvm_base_addr;
1449   Dl_info dlinfo;
1450 
1451   if (libjvm_base_addr == NULL) {
1452     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1453       libjvm_base_addr = (address)dlinfo.dli_fbase;
1454     }
1455     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1456   }
1457 
1458   if (dladdr((void *)addr, &dlinfo) != 0) {
1459     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1460   }
1461 
1462   return false;
1463 }
1464 
1465 bool os::dll_address_to_function_name(address addr, char *buf,
1466                                       int buflen, int *offset,
1467                                       bool demangle) {
1468   // buf is not optional, but offset is optional
1469   assert(buf != NULL, "sanity check");
1470 
1471   Dl_info dlinfo;
1472 
1473   if (dladdr((void*)addr, &dlinfo) != 0) {
1474     // see if we have a matching symbol
1475     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1476       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1477         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1478       }
1479       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1480       return true;
1481     }
1482     // no matching symbol so try for just file info
1483     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1484       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1485                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1486         return true;
1487       }
1488     }
1489   }
1490 
1491   buf[0] = '\0';
1492   if (offset != NULL) *offset = -1;
1493   return false;
1494 }
1495 
1496 struct _address_to_library_name {
1497   address addr;          // input : memory address
1498   size_t  buflen;        //         size of fname
1499   char*   fname;         // output: library name
1500   address base;          //         library base addr
1501 };
1502 
1503 static int address_to_library_name_callback(struct dl_phdr_info *info,
1504                                             size_t size, void *data) {
1505   int i;
1506   bool found = false;
1507   address libbase = NULL;
1508   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1509 
1510   // iterate through all loadable segments
1511   for (i = 0; i < info->dlpi_phnum; i++) {
1512     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1513     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1514       // base address of a library is the lowest address of its loaded
1515       // segments.
1516       if (libbase == NULL || libbase > segbase) {
1517         libbase = segbase;
1518       }
1519       // see if 'addr' is within current segment
1520       if (segbase <= d->addr &&
1521           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1522         found = true;
1523       }
1524     }
1525   }
1526 
1527   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1528   // so dll_address_to_library_name() can fall through to use dladdr() which
1529   // can figure out executable name from argv[0].
1530   if (found && info->dlpi_name && info->dlpi_name[0]) {
1531     d->base = libbase;
1532     if (d->fname) {
1533       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1534     }
1535     return 1;
1536   }
1537   return 0;
1538 }
1539 
1540 bool os::dll_address_to_library_name(address addr, char* buf,
1541                                      int buflen, int* offset) {
1542   // buf is not optional, but offset is optional
1543   assert(buf != NULL, "sanity check");
1544 
1545   Dl_info dlinfo;
1546   struct _address_to_library_name data;
1547 
1548   // There is a bug in old glibc dladdr() implementation that it could resolve
1549   // to wrong library name if the .so file has a base address != NULL. Here
1550   // we iterate through the program headers of all loaded libraries to find
1551   // out which library 'addr' really belongs to. This workaround can be
1552   // removed once the minimum requirement for glibc is moved to 2.3.x.
1553   data.addr = addr;
1554   data.fname = buf;
1555   data.buflen = buflen;
1556   data.base = NULL;
1557   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1558 
1559   if (rslt) {
1560     // buf already contains library name
1561     if (offset) *offset = addr - data.base;
1562     return true;
1563   }
1564   if (dladdr((void*)addr, &dlinfo) != 0) {
1565     if (dlinfo.dli_fname != NULL) {
1566       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1567     }
1568     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1569       *offset = addr - (address)dlinfo.dli_fbase;
1570     }
1571     return true;
1572   }
1573 
1574   buf[0] = '\0';
1575   if (offset) *offset = -1;
1576   return false;
1577 }
1578 
1579 // Loads .dll/.so and
1580 // in case of error it checks if .dll/.so was built for the
1581 // same architecture as Hotspot is running on
1582 
1583 
1584 // Remember the stack's state. The Linux dynamic linker will change
1585 // the stack to 'executable' at most once, so we must safepoint only once.
1586 bool os::Linux::_stack_is_executable = false;
1587 
1588 // VM operation that loads a library.  This is necessary if stack protection
1589 // of the Java stacks can be lost during loading the library.  If we
1590 // do not stop the Java threads, they can stack overflow before the stacks
1591 // are protected again.
1592 class VM_LinuxDllLoad: public VM_Operation {
1593  private:
1594   const char *_filename;
1595   char *_ebuf;
1596   int _ebuflen;
1597   void *_lib;
1598  public:
1599   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1600     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1601   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1602   void doit() {
1603     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1604     os::Linux::_stack_is_executable = true;
1605   }
1606   void* loaded_library() { return _lib; }
1607 };
1608 
1609 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1610   void * result = NULL;
1611   bool load_attempted = false;
1612 
1613   // Check whether the library to load might change execution rights
1614   // of the stack. If they are changed, the protection of the stack
1615   // guard pages will be lost. We need a safepoint to fix this.
1616   //
1617   // See Linux man page execstack(8) for more info.
1618   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1619     if (!ElfFile::specifies_noexecstack(filename)) {
1620       if (!is_init_completed()) {
1621         os::Linux::_stack_is_executable = true;
1622         // This is OK - No Java threads have been created yet, and hence no
1623         // stack guard pages to fix.
1624         //
1625         // This should happen only when you are building JDK7 using a very
1626         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1627         //
1628         // Dynamic loader will make all stacks executable after
1629         // this function returns, and will not do that again.
1630         assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1631       } else {
1632         warning("You have loaded library %s which might have disabled stack guard. "
1633                 "The VM will try to fix the stack guard now.\n"
1634                 "It's highly recommended that you fix the library with "
1635                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1636                 filename);
1637 
1638         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1639         JavaThread *jt = JavaThread::current();
1640         if (jt->thread_state() != _thread_in_native) {
1641           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1642           // that requires ExecStack. Cannot enter safe point. Let's give up.
1643           warning("Unable to fix stack guard. Giving up.");
1644         } else {
1645           if (!LoadExecStackDllInVMThread) {
1646             // This is for the case where the DLL has an static
1647             // constructor function that executes JNI code. We cannot
1648             // load such DLLs in the VMThread.
1649             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1650           }
1651 
1652           ThreadInVMfromNative tiv(jt);
1653           debug_only(VMNativeEntryWrapper vew;)
1654 
1655           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1656           VMThread::execute(&op);
1657           if (LoadExecStackDllInVMThread) {
1658             result = op.loaded_library();
1659           }
1660           load_attempted = true;
1661         }
1662       }
1663     }
1664   }
1665 
1666   if (!load_attempted) {
1667     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1668   }
1669 
1670   if (result != NULL) {
1671     // Successful loading
1672     return result;
1673   }
1674 
1675   Elf32_Ehdr elf_head;
1676   int diag_msg_max_length=ebuflen-strlen(ebuf);
1677   char* diag_msg_buf=ebuf+strlen(ebuf);
1678 
1679   if (diag_msg_max_length==0) {
1680     // No more space in ebuf for additional diagnostics message
1681     return NULL;
1682   }
1683 
1684 
1685   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1686 
1687   if (file_descriptor < 0) {
1688     // Can't open library, report dlerror() message
1689     return NULL;
1690   }
1691 
1692   bool failed_to_read_elf_head=
1693     (sizeof(elf_head)!=
1694      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1695 
1696   ::close(file_descriptor);
1697   if (failed_to_read_elf_head) {
1698     // file i/o error - report dlerror() msg
1699     return NULL;
1700   }
1701 
1702   typedef struct {
1703     Elf32_Half    code;         // Actual value as defined in elf.h
1704     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1705     unsigned char elf_class;    // 32 or 64 bit
1706     unsigned char endianess;    // MSB or LSB
1707     char*         name;         // String representation
1708   } arch_t;
1709 
1710 #ifndef EM_486
1711   #define EM_486          6               /* Intel 80486 */
1712 #endif
1713 #ifndef EM_AARCH64
1714   #define EM_AARCH64    183               /* ARM AARCH64 */
1715 #endif
1716 
1717   static const arch_t arch_array[]={
1718     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1719     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1720     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1721     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1722     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1723     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1724     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1725     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1726 #if defined(VM_LITTLE_ENDIAN)
1727     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1728     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1729 #else
1730     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1731     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1732 #endif
1733     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1734     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1735     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1736     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1737     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1738     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1739     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1740     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1741   };
1742 
1743 #if  (defined IA32)
1744   static  Elf32_Half running_arch_code=EM_386;
1745 #elif   (defined AMD64) || (defined X32)
1746   static  Elf32_Half running_arch_code=EM_X86_64;
1747 #elif  (defined IA64)
1748   static  Elf32_Half running_arch_code=EM_IA_64;
1749 #elif  (defined __sparc) && (defined _LP64)
1750   static  Elf32_Half running_arch_code=EM_SPARCV9;
1751 #elif  (defined __sparc) && (!defined _LP64)
1752   static  Elf32_Half running_arch_code=EM_SPARC;
1753 #elif  (defined __powerpc64__)
1754   static  Elf32_Half running_arch_code=EM_PPC64;
1755 #elif  (defined __powerpc__)
1756   static  Elf32_Half running_arch_code=EM_PPC;
1757 #elif  (defined AARCH64)
1758   static  Elf32_Half running_arch_code=EM_AARCH64;
1759 #elif  (defined ARM)
1760   static  Elf32_Half running_arch_code=EM_ARM;
1761 #elif  (defined S390)
1762   static  Elf32_Half running_arch_code=EM_S390;
1763 #elif  (defined ALPHA)
1764   static  Elf32_Half running_arch_code=EM_ALPHA;
1765 #elif  (defined MIPSEL)
1766   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1767 #elif  (defined PARISC)
1768   static  Elf32_Half running_arch_code=EM_PARISC;
1769 #elif  (defined MIPS)
1770   static  Elf32_Half running_arch_code=EM_MIPS;
1771 #elif  (defined M68K)
1772   static  Elf32_Half running_arch_code=EM_68K;
1773 #elif  (defined SH)
1774   static  Elf32_Half running_arch_code=EM_SH;
1775 #else
1776     #error Method os::dll_load requires that one of following is defined:\
1777         AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1778 #endif
1779 
1780   // Identify compatability class for VM's architecture and library's architecture
1781   // Obtain string descriptions for architectures
1782 
1783   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1784   int running_arch_index=-1;
1785 
1786   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1787     if (running_arch_code == arch_array[i].code) {
1788       running_arch_index    = i;
1789     }
1790     if (lib_arch.code == arch_array[i].code) {
1791       lib_arch.compat_class = arch_array[i].compat_class;
1792       lib_arch.name         = arch_array[i].name;
1793     }
1794   }
1795 
1796   assert(running_arch_index != -1,
1797          "Didn't find running architecture code (running_arch_code) in arch_array");
1798   if (running_arch_index == -1) {
1799     // Even though running architecture detection failed
1800     // we may still continue with reporting dlerror() message
1801     return NULL;
1802   }
1803 
1804   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1805     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1806     return NULL;
1807   }
1808 
1809 #ifndef S390
1810   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1811     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1812     return NULL;
1813   }
1814 #endif // !S390
1815 
1816   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1817     if (lib_arch.name!=NULL) {
1818       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1819                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1820                  lib_arch.name, arch_array[running_arch_index].name);
1821     } else {
1822       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1823                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1824                  lib_arch.code,
1825                  arch_array[running_arch_index].name);
1826     }
1827   }
1828 
1829   return NULL;
1830 }
1831 
1832 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1833                                 int ebuflen) {
1834   void * result = ::dlopen(filename, RTLD_LAZY);
1835   if (result == NULL) {
1836     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1837     ebuf[ebuflen-1] = '\0';
1838   }
1839   return result;
1840 }
1841 
1842 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1843                                        int ebuflen) {
1844   void * result = NULL;
1845   if (LoadExecStackDllInVMThread) {
1846     result = dlopen_helper(filename, ebuf, ebuflen);
1847   }
1848 
1849   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1850   // library that requires an executable stack, or which does not have this
1851   // stack attribute set, dlopen changes the stack attribute to executable. The
1852   // read protection of the guard pages gets lost.
1853   //
1854   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1855   // may have been queued at the same time.
1856 
1857   if (!_stack_is_executable) {
1858     for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1859       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1860           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1861         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1862           warning("Attempt to reguard stack yellow zone failed.");
1863         }
1864       }
1865     }
1866   }
1867 
1868   return result;
1869 }
1870 
1871 void* os::dll_lookup(void* handle, const char* name) {
1872   void* res = dlsym(handle, name);
1873   return res;
1874 }
1875 
1876 void* os::get_default_process_handle() {
1877   return (void*)::dlopen(NULL, RTLD_LAZY);
1878 }
1879 
1880 static bool _print_ascii_file(const char* filename, outputStream* st) {
1881   int fd = ::open(filename, O_RDONLY);
1882   if (fd == -1) {
1883     return false;
1884   }
1885 
1886   char buf[33];
1887   int bytes;
1888   buf[32] = '\0';
1889   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1890     st->print_raw(buf, bytes);
1891   }
1892 
1893   ::close(fd);
1894 
1895   return true;
1896 }
1897 
1898 void os::print_dll_info(outputStream *st) {
1899   st->print_cr("Dynamic libraries:");
1900 
1901   char fname[32];
1902   pid_t pid = os::Linux::gettid();
1903 
1904   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1905 
1906   if (!_print_ascii_file(fname, st)) {
1907     st->print("Can not get library information for pid = %d\n", pid);
1908   }
1909 }
1910 
1911 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1912   FILE *procmapsFile = NULL;
1913 
1914   // Open the procfs maps file for the current process
1915   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1916     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1917     char line[PATH_MAX + 100];
1918 
1919     // Read line by line from 'file'
1920     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1921       u8 base, top, offset, inode;
1922       char permissions[5];
1923       char device[6];
1924       char name[PATH_MAX + 1];
1925 
1926       // Parse fields from line
1927       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1928              &base, &top, permissions, &offset, device, &inode, name);
1929 
1930       // Filter by device id '00:00' so that we only get file system mapped files.
1931       if (strcmp(device, "00:00") != 0) {
1932 
1933         // Call callback with the fields of interest
1934         if(callback(name, (address)base, (address)top, param)) {
1935           // Oops abort, callback aborted
1936           fclose(procmapsFile);
1937           return 1;
1938         }
1939       }
1940     }
1941     fclose(procmapsFile);
1942   }
1943   return 0;
1944 }
1945 
1946 void os::print_os_info_brief(outputStream* st) {
1947   os::Linux::print_distro_info(st);
1948 
1949   os::Posix::print_uname_info(st);
1950 
1951   os::Linux::print_libversion_info(st);
1952 
1953 }
1954 
1955 void os::print_os_info(outputStream* st) {
1956   st->print("OS:");
1957 
1958   os::Linux::print_distro_info(st);
1959 
1960   os::Posix::print_uname_info(st);
1961 
1962   // Print warning if unsafe chroot environment detected
1963   if (unsafe_chroot_detected) {
1964     st->print("WARNING!! ");
1965     st->print_cr("%s", unstable_chroot_error);
1966   }
1967 
1968   os::Linux::print_libversion_info(st);
1969 
1970   os::Posix::print_rlimit_info(st);
1971 
1972   os::Posix::print_load_average(st);
1973 
1974   os::Linux::print_full_memory_info(st);
1975 
1976   os::Linux::print_proc_sys_info(st);
1977 
1978   os::Linux::print_container_info(st);
1979 }
1980 
1981 // Try to identify popular distros.
1982 // Most Linux distributions have a /etc/XXX-release file, which contains
1983 // the OS version string. Newer Linux distributions have a /etc/lsb-release
1984 // file that also contains the OS version string. Some have more than one
1985 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
1986 // /etc/redhat-release.), so the order is important.
1987 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
1988 // their own specific XXX-release file as well as a redhat-release file.
1989 // Because of this the XXX-release file needs to be searched for before the
1990 // redhat-release file.
1991 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
1992 // search for redhat-release / SuSE-release needs to be before lsb-release.
1993 // Since the lsb-release file is the new standard it needs to be searched
1994 // before the older style release files.
1995 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
1996 // next to last resort.  The os-release file is a new standard that contains
1997 // distribution information and the system-release file seems to be an old
1998 // standard that has been replaced by the lsb-release and os-release files.
1999 // Searching for the debian_version file is the last resort.  It contains
2000 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2001 // "Debian " is printed before the contents of the debian_version file.
2002 
2003 const char* distro_files[] = {
2004   "/etc/oracle-release",
2005   "/etc/mandriva-release",
2006   "/etc/mandrake-release",
2007   "/etc/sun-release",
2008   "/etc/redhat-release",
2009   "/etc/SuSE-release",
2010   "/etc/lsb-release",
2011   "/etc/turbolinux-release",
2012   "/etc/gentoo-release",
2013   "/etc/ltib-release",
2014   "/etc/angstrom-version",
2015   "/etc/system-release",
2016   "/etc/os-release",
2017   NULL };
2018 
2019 void os::Linux::print_distro_info(outputStream* st) {
2020   for (int i = 0;; i++) {
2021     const char* file = distro_files[i];
2022     if (file == NULL) {
2023       break;  // done
2024     }
2025     // If file prints, we found it.
2026     if (_print_ascii_file(file, st)) {
2027       return;
2028     }
2029   }
2030 
2031   if (file_exists("/etc/debian_version")) {
2032     st->print("Debian ");
2033     _print_ascii_file("/etc/debian_version", st);
2034   } else {
2035     st->print("Linux");
2036   }
2037   st->cr();
2038 }
2039 
2040 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2041   char buf[256];
2042   while (fgets(buf, sizeof(buf), fp)) {
2043     // Edit out extra stuff in expected format
2044     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2045       char* ptr = strstr(buf, "\"");  // the name is in quotes
2046       if (ptr != NULL) {
2047         ptr++; // go beyond first quote
2048         char* nl = strchr(ptr, '\"');
2049         if (nl != NULL) *nl = '\0';
2050         strncpy(distro, ptr, length);
2051       } else {
2052         ptr = strstr(buf, "=");
2053         ptr++; // go beyond equals then
2054         char* nl = strchr(ptr, '\n');
2055         if (nl != NULL) *nl = '\0';
2056         strncpy(distro, ptr, length);
2057       }
2058       return;
2059     } else if (get_first_line) {
2060       char* nl = strchr(buf, '\n');
2061       if (nl != NULL) *nl = '\0';
2062       strncpy(distro, buf, length);
2063       return;
2064     }
2065   }
2066   // print last line and close
2067   char* nl = strchr(buf, '\n');
2068   if (nl != NULL) *nl = '\0';
2069   strncpy(distro, buf, length);
2070 }
2071 
2072 static void parse_os_info(char* distro, size_t length, const char* file) {
2073   FILE* fp = fopen(file, "r");
2074   if (fp != NULL) {
2075     // if suse format, print out first line
2076     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2077     parse_os_info_helper(fp, distro, length, get_first_line);
2078     fclose(fp);
2079   }
2080 }
2081 
2082 void os::get_summary_os_info(char* buf, size_t buflen) {
2083   for (int i = 0;; i++) {
2084     const char* file = distro_files[i];
2085     if (file == NULL) {
2086       break; // ran out of distro_files
2087     }
2088     if (file_exists(file)) {
2089       parse_os_info(buf, buflen, file);
2090       return;
2091     }
2092   }
2093   // special case for debian
2094   if (file_exists("/etc/debian_version")) {
2095     strncpy(buf, "Debian ", buflen);
2096     if (buflen > 7) {
2097       parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2098     }
2099   } else {
2100     strncpy(buf, "Linux", buflen);
2101   }
2102 }
2103 
2104 void os::Linux::print_libversion_info(outputStream* st) {
2105   // libc, pthread
2106   st->print("libc:");
2107   st->print("%s ", os::Linux::glibc_version());
2108   st->print("%s ", os::Linux::libpthread_version());
2109   st->cr();
2110 }
2111 
2112 void os::Linux::print_proc_sys_info(outputStream* st) {
2113   st->cr();
2114   st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2115   _print_ascii_file("/proc/sys/kernel/threads-max", st);
2116   st->cr();
2117   st->cr();
2118 
2119   st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2120   _print_ascii_file("/proc/sys/vm/max_map_count", st);
2121   st->cr();
2122   st->cr();
2123 
2124   st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2125   _print_ascii_file("/proc/sys/kernel/pid_max", st);
2126   st->cr();
2127   st->cr();
2128 }
2129 
2130 void os::Linux::print_full_memory_info(outputStream* st) {
2131   st->print("\n/proc/meminfo:\n");
2132   _print_ascii_file("/proc/meminfo", st);
2133   st->cr();
2134 }
2135 
2136 void os::Linux::print_container_info(outputStream* st) {
2137   if (!OSContainer::is_containerized()) {
2138     return;
2139   }
2140 
2141   st->print("container (cgroup) information:\n");
2142 
2143   const char *p_ct = OSContainer::container_type();
2144   st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2145 
2146   char *p = OSContainer::cpu_cpuset_cpus();
2147   st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2148   free(p);
2149 
2150   p = OSContainer::cpu_cpuset_memory_nodes();
2151   st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2152   free(p);
2153 
2154   int i = OSContainer::active_processor_count();
2155   if (i > 0) {
2156     st->print("active_processor_count: %d\n", i);
2157   } else {
2158     st->print("active_processor_count: failed\n");
2159   }
2160 
2161   i = OSContainer::cpu_quota();
2162   st->print("cpu_quota: %d\n", i);
2163 
2164   i = OSContainer::cpu_period();
2165   st->print("cpu_period: %d\n", i);
2166 
2167   i = OSContainer::cpu_shares();
2168   st->print("cpu_shares: %d\n", i);
2169 
2170   jlong j = OSContainer::memory_limit_in_bytes();
2171   st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2172 
2173   j = OSContainer::memory_and_swap_limit_in_bytes();
2174   st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2175 
2176   j = OSContainer::memory_soft_limit_in_bytes();
2177   st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2178 
2179   j = OSContainer::OSContainer::memory_usage_in_bytes();
2180   st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2181 
2182   j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2183   st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2184   st->cr();
2185 }
2186 
2187 void os::print_memory_info(outputStream* st) {
2188 
2189   st->print("Memory:");
2190   st->print(" %dk page", os::vm_page_size()>>10);
2191 
2192   // values in struct sysinfo are "unsigned long"
2193   struct sysinfo si;
2194   sysinfo(&si);
2195 
2196   st->print(", physical " UINT64_FORMAT "k",
2197             os::physical_memory() >> 10);
2198   st->print("(" UINT64_FORMAT "k free)",
2199             os::available_memory() >> 10);
2200   st->print(", swap " UINT64_FORMAT "k",
2201             ((jlong)si.totalswap * si.mem_unit) >> 10);
2202   st->print("(" UINT64_FORMAT "k free)",
2203             ((jlong)si.freeswap * si.mem_unit) >> 10);
2204   st->cr();
2205 }
2206 
2207 // Print the first "model name" line and the first "flags" line
2208 // that we find and nothing more. We assume "model name" comes
2209 // before "flags" so if we find a second "model name", then the
2210 // "flags" field is considered missing.
2211 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2212 #if defined(IA32) || defined(AMD64)
2213   // Other platforms have less repetitive cpuinfo files
2214   FILE *fp = fopen("/proc/cpuinfo", "r");
2215   if (fp) {
2216     while (!feof(fp)) {
2217       if (fgets(buf, buflen, fp)) {
2218         // Assume model name comes before flags
2219         bool model_name_printed = false;
2220         if (strstr(buf, "model name") != NULL) {
2221           if (!model_name_printed) {
2222             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2223             st->print_raw(buf);
2224             model_name_printed = true;
2225           } else {
2226             // model name printed but not flags?  Odd, just return
2227             fclose(fp);
2228             return true;
2229           }
2230         }
2231         // print the flags line too
2232         if (strstr(buf, "flags") != NULL) {
2233           st->print_raw(buf);
2234           fclose(fp);
2235           return true;
2236         }
2237       }
2238     }
2239     fclose(fp);
2240   }
2241 #endif // x86 platforms
2242   return false;
2243 }
2244 
2245 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2246   // Only print the model name if the platform provides this as a summary
2247   if (!print_model_name_and_flags(st, buf, buflen)) {
2248     st->print("\n/proc/cpuinfo:\n");
2249     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2250       st->print_cr("  <Not Available>");
2251     }
2252   }
2253 }
2254 
2255 #if defined(AMD64) || defined(IA32) || defined(X32)
2256 const char* search_string = "model name";
2257 #elif defined(M68K)
2258 const char* search_string = "CPU";
2259 #elif defined(PPC64)
2260 const char* search_string = "cpu";
2261 #elif defined(S390)
2262 const char* search_string = "processor";
2263 #elif defined(SPARC)
2264 const char* search_string = "cpu";
2265 #else
2266 const char* search_string = "Processor";
2267 #endif
2268 
2269 // Parses the cpuinfo file for string representing the model name.
2270 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2271   FILE* fp = fopen("/proc/cpuinfo", "r");
2272   if (fp != NULL) {
2273     while (!feof(fp)) {
2274       char buf[256];
2275       if (fgets(buf, sizeof(buf), fp)) {
2276         char* start = strstr(buf, search_string);
2277         if (start != NULL) {
2278           char *ptr = start + strlen(search_string);
2279           char *end = buf + strlen(buf);
2280           while (ptr != end) {
2281              // skip whitespace and colon for the rest of the name.
2282              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2283                break;
2284              }
2285              ptr++;
2286           }
2287           if (ptr != end) {
2288             // reasonable string, get rid of newline and keep the rest
2289             char* nl = strchr(buf, '\n');
2290             if (nl != NULL) *nl = '\0';
2291             strncpy(cpuinfo, ptr, length);
2292             fclose(fp);
2293             return;
2294           }
2295         }
2296       }
2297     }
2298     fclose(fp);
2299   }
2300   // cpuinfo not found or parsing failed, just print generic string.  The entire
2301   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2302 #if   defined(AARCH64)
2303   strncpy(cpuinfo, "AArch64", length);
2304 #elif defined(AMD64)
2305   strncpy(cpuinfo, "x86_64", length);
2306 #elif defined(ARM)  // Order wrt. AARCH64 is relevant!
2307   strncpy(cpuinfo, "ARM", length);
2308 #elif defined(IA32)
2309   strncpy(cpuinfo, "x86_32", length);
2310 #elif defined(IA64)
2311   strncpy(cpuinfo, "IA64", length);
2312 #elif defined(PPC)
2313   strncpy(cpuinfo, "PPC64", length);
2314 #elif defined(S390)
2315   strncpy(cpuinfo, "S390", length);
2316 #elif defined(SPARC)
2317   strncpy(cpuinfo, "sparcv9", length);
2318 #elif defined(ZERO_LIBARCH)
2319   strncpy(cpuinfo, ZERO_LIBARCH, length);
2320 #else
2321   strncpy(cpuinfo, "unknown", length);
2322 #endif
2323 }
2324 
2325 static void print_signal_handler(outputStream* st, int sig,
2326                                  char* buf, size_t buflen);
2327 
2328 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2329   st->print_cr("Signal Handlers:");
2330   print_signal_handler(st, SIGSEGV, buf, buflen);
2331   print_signal_handler(st, SIGBUS , buf, buflen);
2332   print_signal_handler(st, SIGFPE , buf, buflen);
2333   print_signal_handler(st, SIGPIPE, buf, buflen);
2334   print_signal_handler(st, SIGXFSZ, buf, buflen);
2335   print_signal_handler(st, SIGILL , buf, buflen);
2336   print_signal_handler(st, SR_signum, buf, buflen);
2337   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2338   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2339   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2340   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2341 #if defined(PPC64)
2342   print_signal_handler(st, SIGTRAP, buf, buflen);
2343 #endif
2344 }
2345 
2346 static char saved_jvm_path[MAXPATHLEN] = {0};
2347 
2348 // Find the full path to the current module, libjvm.so
2349 void os::jvm_path(char *buf, jint buflen) {
2350   // Error checking.
2351   if (buflen < MAXPATHLEN) {
2352     assert(false, "must use a large-enough buffer");
2353     buf[0] = '\0';
2354     return;
2355   }
2356   // Lazy resolve the path to current module.
2357   if (saved_jvm_path[0] != 0) {
2358     strcpy(buf, saved_jvm_path);
2359     return;
2360   }
2361 
2362   char dli_fname[MAXPATHLEN];
2363   bool ret = dll_address_to_library_name(
2364                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2365                                          dli_fname, sizeof(dli_fname), NULL);
2366   assert(ret, "cannot locate libjvm");
2367   char *rp = NULL;
2368   if (ret && dli_fname[0] != '\0') {
2369     rp = os::Posix::realpath(dli_fname, buf, buflen);
2370   }
2371   if (rp == NULL) {
2372     return;
2373   }
2374 
2375   if (Arguments::sun_java_launcher_is_altjvm()) {
2376     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2377     // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2378     // If "/jre/lib/" appears at the right place in the string, then
2379     // assume we are installed in a JDK and we're done. Otherwise, check
2380     // for a JAVA_HOME environment variable and fix up the path so it
2381     // looks like libjvm.so is installed there (append a fake suffix
2382     // hotspot/libjvm.so).
2383     const char *p = buf + strlen(buf) - 1;
2384     for (int count = 0; p > buf && count < 5; ++count) {
2385       for (--p; p > buf && *p != '/'; --p)
2386         /* empty */ ;
2387     }
2388 
2389     if (strncmp(p, "/jre/lib/", 9) != 0) {
2390       // Look for JAVA_HOME in the environment.
2391       char* java_home_var = ::getenv("JAVA_HOME");
2392       if (java_home_var != NULL && java_home_var[0] != 0) {
2393         char* jrelib_p;
2394         int len;
2395 
2396         // Check the current module name "libjvm.so".
2397         p = strrchr(buf, '/');
2398         if (p == NULL) {
2399           return;
2400         }
2401         assert(strstr(p, "/libjvm") == p, "invalid library name");
2402 
2403         rp = os::Posix::realpath(java_home_var, buf, buflen);
2404         if (rp == NULL) {
2405           return;
2406         }
2407 
2408         // determine if this is a legacy image or modules image
2409         // modules image doesn't have "jre" subdirectory
2410         len = strlen(buf);
2411         assert(len < buflen, "Ran out of buffer room");
2412         jrelib_p = buf + len;
2413         snprintf(jrelib_p, buflen-len, "/jre/lib");
2414         if (0 != access(buf, F_OK)) {
2415           snprintf(jrelib_p, buflen-len, "/lib");
2416         }
2417 
2418         if (0 == access(buf, F_OK)) {
2419           // Use current module name "libjvm.so"
2420           len = strlen(buf);
2421           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2422         } else {
2423           // Go back to path of .so
2424           rp = os::Posix::realpath(dli_fname, buf, buflen);
2425           if (rp == NULL) {
2426             return;
2427           }
2428         }
2429       }
2430     }
2431   }
2432 
2433   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2434   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2435 }
2436 
2437 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2438   // no prefix required, not even "_"
2439 }
2440 
2441 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2442   // no suffix required
2443 }
2444 
2445 ////////////////////////////////////////////////////////////////////////////////
2446 // sun.misc.Signal support
2447 
2448 static volatile jint sigint_count = 0;
2449 
2450 static void UserHandler(int sig, void *siginfo, void *context) {
2451   // 4511530 - sem_post is serialized and handled by the manager thread. When
2452   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2453   // don't want to flood the manager thread with sem_post requests.
2454   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2455     return;
2456   }
2457 
2458   // Ctrl-C is pressed during error reporting, likely because the error
2459   // handler fails to abort. Let VM die immediately.
2460   if (sig == SIGINT && VMError::is_error_reported()) {
2461     os::die();
2462   }
2463 
2464   os::signal_notify(sig);
2465 }
2466 
2467 void* os::user_handler() {
2468   return CAST_FROM_FN_PTR(void*, UserHandler);
2469 }
2470 
2471 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2472   struct timespec ts;
2473   // Semaphore's are always associated with CLOCK_REALTIME
2474   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2475   // see os_posix.cpp for discussion on overflow checking
2476   if (sec >= MAX_SECS) {
2477     ts.tv_sec += MAX_SECS;
2478     ts.tv_nsec = 0;
2479   } else {
2480     ts.tv_sec += sec;
2481     ts.tv_nsec += nsec;
2482     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2483       ts.tv_nsec -= NANOSECS_PER_SEC;
2484       ++ts.tv_sec; // note: this must be <= max_secs
2485     }
2486   }
2487 
2488   return ts;
2489 }
2490 
2491 extern "C" {
2492   typedef void (*sa_handler_t)(int);
2493   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2494 }
2495 
2496 void* os::signal(int signal_number, void* handler) {
2497   struct sigaction sigAct, oldSigAct;
2498 
2499   sigfillset(&(sigAct.sa_mask));
2500   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2501   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2502 
2503   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2504     // -1 means registration failed
2505     return (void *)-1;
2506   }
2507 
2508   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2509 }
2510 
2511 void os::signal_raise(int signal_number) {
2512   ::raise(signal_number);
2513 }
2514 
2515 // The following code is moved from os.cpp for making this
2516 // code platform specific, which it is by its very nature.
2517 
2518 // Will be modified when max signal is changed to be dynamic
2519 int os::sigexitnum_pd() {
2520   return NSIG;
2521 }
2522 
2523 // a counter for each possible signal value
2524 static volatile jint pending_signals[NSIG+1] = { 0 };
2525 
2526 // Linux(POSIX) specific hand shaking semaphore.
2527 static Semaphore* sig_sem = NULL;
2528 static PosixSemaphore sr_semaphore;
2529 
2530 static void jdk_misc_signal_init() {
2531   // Initialize signal structures
2532   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2533 
2534   // Initialize signal semaphore
2535   sig_sem = new Semaphore();
2536 }
2537 
2538 void os::signal_notify(int sig) {
2539   if (sig_sem != NULL) {
2540     Atomic::inc(&pending_signals[sig]);
2541     sig_sem->signal();
2542   } else {
2543     // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2544     // initialization isn't called.
2545     assert(ReduceSignalUsage, "signal semaphore should be created");
2546   }
2547 }
2548 
2549 static int check_pending_signals() {
2550   Atomic::store(0, &sigint_count);
2551   for (;;) {
2552     for (int i = 0; i < NSIG + 1; i++) {
2553       jint n = pending_signals[i];
2554       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2555         return i;
2556       }
2557     }
2558     JavaThread *thread = JavaThread::current();
2559     ThreadBlockInVM tbivm(thread);
2560 
2561     bool threadIsSuspended;
2562     do {
2563       thread->set_suspend_equivalent();
2564       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2565       sig_sem->wait();
2566 
2567       // were we externally suspended while we were waiting?
2568       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2569       if (threadIsSuspended) {
2570         // The semaphore has been incremented, but while we were waiting
2571         // another thread suspended us. We don't want to continue running
2572         // while suspended because that would surprise the thread that
2573         // suspended us.
2574         sig_sem->signal();
2575 
2576         thread->java_suspend_self();
2577       }
2578     } while (threadIsSuspended);
2579   }
2580 }
2581 
2582 int os::signal_wait() {
2583   return check_pending_signals();
2584 }
2585 
2586 ////////////////////////////////////////////////////////////////////////////////
2587 // Virtual Memory
2588 
2589 int os::vm_page_size() {
2590   // Seems redundant as all get out
2591   assert(os::Linux::page_size() != -1, "must call os::init");
2592   return os::Linux::page_size();
2593 }
2594 
2595 // Solaris allocates memory by pages.
2596 int os::vm_allocation_granularity() {
2597   assert(os::Linux::page_size() != -1, "must call os::init");
2598   return os::Linux::page_size();
2599 }
2600 
2601 // Rationale behind this function:
2602 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2603 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2604 //  samples for JITted code. Here we create private executable mapping over the code cache
2605 //  and then we can use standard (well, almost, as mapping can change) way to provide
2606 //  info for the reporting script by storing timestamp and location of symbol
2607 void linux_wrap_code(char* base, size_t size) {
2608   static volatile jint cnt = 0;
2609 
2610   if (!UseOprofile) {
2611     return;
2612   }
2613 
2614   char buf[PATH_MAX+1];
2615   int num = Atomic::add(1, &cnt);
2616 
2617   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2618            os::get_temp_directory(), os::current_process_id(), num);
2619   unlink(buf);
2620 
2621   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2622 
2623   if (fd != -1) {
2624     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2625     if (rv != (off_t)-1) {
2626       if (::write(fd, "", 1) == 1) {
2627         mmap(base, size,
2628              PROT_READ|PROT_WRITE|PROT_EXEC,
2629              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2630       }
2631     }
2632     ::close(fd);
2633     unlink(buf);
2634   }
2635 }
2636 
2637 static bool recoverable_mmap_error(int err) {
2638   // See if the error is one we can let the caller handle. This
2639   // list of errno values comes from JBS-6843484. I can't find a
2640   // Linux man page that documents this specific set of errno
2641   // values so while this list currently matches Solaris, it may
2642   // change as we gain experience with this failure mode.
2643   switch (err) {
2644   case EBADF:
2645   case EINVAL:
2646   case ENOTSUP:
2647     // let the caller deal with these errors
2648     return true;
2649 
2650   default:
2651     // Any remaining errors on this OS can cause our reserved mapping
2652     // to be lost. That can cause confusion where different data
2653     // structures think they have the same memory mapped. The worst
2654     // scenario is if both the VM and a library think they have the
2655     // same memory mapped.
2656     return false;
2657   }
2658 }
2659 
2660 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2661                                     int err) {
2662   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2663           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2664           os::strerror(err), err);
2665 }
2666 
2667 static void warn_fail_commit_memory(char* addr, size_t size,
2668                                     size_t alignment_hint, bool exec,
2669                                     int err) {
2670   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2671           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2672           alignment_hint, exec, os::strerror(err), err);
2673 }
2674 
2675 // NOTE: Linux kernel does not really reserve the pages for us.
2676 //       All it does is to check if there are enough free pages
2677 //       left at the time of mmap(). This could be a potential
2678 //       problem.
2679 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2680   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2681   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2682                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2683   if (res != (uintptr_t) MAP_FAILED) {
2684     if (UseNUMAInterleaving) {
2685       numa_make_global(addr, size);
2686     }
2687     return 0;
2688   }
2689 
2690   int err = errno;  // save errno from mmap() call above
2691 
2692   if (!recoverable_mmap_error(err)) {
2693     warn_fail_commit_memory(addr, size, exec, err);
2694     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2695   }
2696 
2697   return err;
2698 }
2699 
2700 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2701   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2702 }
2703 
2704 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2705                                   const char* mesg) {
2706   assert(mesg != NULL, "mesg must be specified");
2707   int err = os::Linux::commit_memory_impl(addr, size, exec);
2708   if (err != 0) {
2709     // the caller wants all commit errors to exit with the specified mesg:
2710     warn_fail_commit_memory(addr, size, exec, err);
2711     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2712   }
2713 }
2714 
2715 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2716 #ifndef MAP_HUGETLB
2717   #define MAP_HUGETLB 0x40000
2718 #endif
2719 
2720 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2721 #ifndef MADV_HUGEPAGE
2722   #define MADV_HUGEPAGE 14
2723 #endif
2724 
2725 int os::Linux::commit_memory_impl(char* addr, size_t size,
2726                                   size_t alignment_hint, bool exec) {
2727   int err = os::Linux::commit_memory_impl(addr, size, exec);
2728   if (err == 0) {
2729     realign_memory(addr, size, alignment_hint);
2730   }
2731   return err;
2732 }
2733 
2734 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2735                           bool exec) {
2736   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2737 }
2738 
2739 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2740                                   size_t alignment_hint, bool exec,
2741                                   const char* mesg) {
2742   assert(mesg != NULL, "mesg must be specified");
2743   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2744   if (err != 0) {
2745     // the caller wants all commit errors to exit with the specified mesg:
2746     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2747     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2748   }
2749 }
2750 
2751 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2752   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2753     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2754     // be supported or the memory may already be backed by huge pages.
2755     ::madvise(addr, bytes, MADV_HUGEPAGE);
2756   }
2757 }
2758 
2759 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2760   // This method works by doing an mmap over an existing mmaping and effectively discarding
2761   // the existing pages. However it won't work for SHM-based large pages that cannot be
2762   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2763   // small pages on top of the SHM segment. This method always works for small pages, so we
2764   // allow that in any case.
2765   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2766     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2767   }
2768 }
2769 
2770 void os::numa_make_global(char *addr, size_t bytes) {
2771   Linux::numa_interleave_memory(addr, bytes);
2772 }
2773 
2774 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2775 // bind policy to MPOL_PREFERRED for the current thread.
2776 #define USE_MPOL_PREFERRED 0
2777 
2778 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2779   // To make NUMA and large pages more robust when both enabled, we need to ease
2780   // the requirements on where the memory should be allocated. MPOL_BIND is the
2781   // default policy and it will force memory to be allocated on the specified
2782   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2783   // the specified node, but will not force it. Using this policy will prevent
2784   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2785   // free large pages.
2786   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2787   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2788 }
2789 
2790 bool os::numa_topology_changed() { return false; }
2791 
2792 size_t os::numa_get_groups_num() {
2793   // Return just the number of nodes in which it's possible to allocate memory
2794   // (in numa terminology, configured nodes).
2795   return Linux::numa_num_configured_nodes();
2796 }
2797 
2798 int os::numa_get_group_id() {
2799   int cpu_id = Linux::sched_getcpu();
2800   if (cpu_id != -1) {
2801     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2802     if (lgrp_id != -1) {
2803       return lgrp_id;
2804     }
2805   }
2806   return 0;
2807 }
2808 
2809 int os::Linux::get_existing_num_nodes() {
2810   int node;
2811   int highest_node_number = Linux::numa_max_node();
2812   int num_nodes = 0;
2813 
2814   // Get the total number of nodes in the system including nodes without memory.
2815   for (node = 0; node <= highest_node_number; node++) {
2816     if (isnode_in_existing_nodes(node)) {
2817       num_nodes++;
2818     }
2819   }
2820   return num_nodes;
2821 }
2822 
2823 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2824   int highest_node_number = Linux::numa_max_node();
2825   size_t i = 0;
2826 
2827   // Map all node ids in which it is possible to allocate memory. Also nodes are
2828   // not always consecutively available, i.e. available from 0 to the highest
2829   // node number. If the nodes have been bound explicitly using numactl membind,
2830   // then allocate memory from those nodes only.
2831   for (int node = 0; node <= highest_node_number; node++) {
2832     if (Linux::isnode_in_bound_nodes((unsigned int)node)) {
2833       ids[i++] = node;
2834     }
2835   }
2836   return i;
2837 }
2838 
2839 bool os::get_page_info(char *start, page_info* info) {
2840   return false;
2841 }
2842 
2843 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2844                      page_info* page_found) {
2845   return end;
2846 }
2847 
2848 
2849 int os::Linux::sched_getcpu_syscall(void) {
2850   unsigned int cpu = 0;
2851   int retval = -1;
2852 
2853 #if defined(IA32)
2854   #ifndef SYS_getcpu
2855     #define SYS_getcpu 318
2856   #endif
2857   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2858 #elif defined(AMD64)
2859 // Unfortunately we have to bring all these macros here from vsyscall.h
2860 // to be able to compile on old linuxes.
2861   #define __NR_vgetcpu 2
2862   #define VSYSCALL_START (-10UL << 20)
2863   #define VSYSCALL_SIZE 1024
2864   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2865   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2866   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2867   retval = vgetcpu(&cpu, NULL, NULL);
2868 #endif
2869 
2870   return (retval == -1) ? retval : cpu;
2871 }
2872 
2873 void os::Linux::sched_getcpu_init() {
2874   // sched_getcpu() should be in libc.
2875   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2876                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2877 
2878   // If it's not, try a direct syscall.
2879   if (sched_getcpu() == -1) {
2880     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2881                                     (void*)&sched_getcpu_syscall));
2882   }
2883 
2884   if (sched_getcpu() == -1) {
2885     vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
2886   }
2887 }
2888 
2889 // Something to do with the numa-aware allocator needs these symbols
2890 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2891 extern "C" JNIEXPORT void numa_error(char *where) { }
2892 
2893 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2894 // load symbol from base version instead.
2895 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2896   void *f = dlvsym(handle, name, "libnuma_1.1");
2897   if (f == NULL) {
2898     f = dlsym(handle, name);
2899   }
2900   return f;
2901 }
2902 
2903 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2904 // Return NULL if the symbol is not defined in this particular version.
2905 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2906   return dlvsym(handle, name, "libnuma_1.2");
2907 }
2908 
2909 bool os::Linux::libnuma_init() {
2910   if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2911     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2912     if (handle != NULL) {
2913       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2914                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2915       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2916                                        libnuma_dlsym(handle, "numa_max_node")));
2917       set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2918                                                    libnuma_dlsym(handle, "numa_num_configured_nodes")));
2919       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2920                                         libnuma_dlsym(handle, "numa_available")));
2921       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2922                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2923       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2924                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2925       set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2926                                                 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2927       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2928                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2929       set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2930                                                libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2931       set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2932                                        libnuma_dlsym(handle, "numa_distance")));
2933       set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
2934                                           libnuma_v2_dlsym(handle, "numa_get_membind")));
2935 
2936       if (numa_available() != -1) {
2937         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2938         set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2939         set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2940         // Create an index -> node mapping, since nodes are not always consecutive
2941         _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2942         rebuild_nindex_to_node_map();
2943         // Create a cpu -> node mapping
2944         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2945         rebuild_cpu_to_node_map();
2946         return true;
2947       }
2948     }
2949   }
2950   return false;
2951 }
2952 
2953 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2954   // Creating guard page is very expensive. Java thread has HotSpot
2955   // guard pages, only enable glibc guard page for non-Java threads.
2956   // (Remember: compiler thread is a Java thread, too!)
2957   return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2958 }
2959 
2960 void os::Linux::rebuild_nindex_to_node_map() {
2961   int highest_node_number = Linux::numa_max_node();
2962 
2963   nindex_to_node()->clear();
2964   for (int node = 0; node <= highest_node_number; node++) {
2965     if (Linux::isnode_in_existing_nodes(node)) {
2966       nindex_to_node()->append(node);
2967     }
2968   }
2969 }
2970 
2971 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2972 // The table is later used in get_node_by_cpu().
2973 void os::Linux::rebuild_cpu_to_node_map() {
2974   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2975                               // in libnuma (possible values are starting from 16,
2976                               // and continuing up with every other power of 2, but less
2977                               // than the maximum number of CPUs supported by kernel), and
2978                               // is a subject to change (in libnuma version 2 the requirements
2979                               // are more reasonable) we'll just hardcode the number they use
2980                               // in the library.
2981   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2982 
2983   size_t cpu_num = processor_count();
2984   size_t cpu_map_size = NCPUS / BitsPerCLong;
2985   size_t cpu_map_valid_size =
2986     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2987 
2988   cpu_to_node()->clear();
2989   cpu_to_node()->at_grow(cpu_num - 1);
2990 
2991   size_t node_num = get_existing_num_nodes();
2992 
2993   int distance = 0;
2994   int closest_distance = INT_MAX;
2995   int closest_node = 0;
2996   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2997   for (size_t i = 0; i < node_num; i++) {
2998     // Check if node is configured (not a memory-less node). If it is not, find
2999     // the closest configured node. Check also if node is bound, i.e. it's allowed
3000     // to allocate memory from the node. If it's not allowed, map cpus in that node
3001     // to the closest node from which memory allocation is allowed.
3002     if (!isnode_in_configured_nodes(nindex_to_node()->at(i)) ||
3003         !isnode_in_bound_nodes(nindex_to_node()->at(i))) {
3004       closest_distance = INT_MAX;
3005       // Check distance from all remaining nodes in the system. Ignore distance
3006       // from itself, from another non-configured node, and from another non-bound
3007       // node.
3008       for (size_t m = 0; m < node_num; m++) {
3009         if (m != i &&
3010             isnode_in_configured_nodes(nindex_to_node()->at(m)) &&
3011             isnode_in_bound_nodes(nindex_to_node()->at(m))) {
3012           distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3013           // If a closest node is found, update. There is always at least one
3014           // configured and bound node in the system so there is always at least
3015           // one node close.
3016           if (distance != 0 && distance < closest_distance) {
3017             closest_distance = distance;
3018             closest_node = nindex_to_node()->at(m);
3019           }
3020         }
3021       }
3022      } else {
3023        // Current node is already a configured node.
3024        closest_node = nindex_to_node()->at(i);
3025      }
3026 
3027     // Get cpus from the original node and map them to the closest node. If node
3028     // is a configured node (not a memory-less node), then original node and
3029     // closest node are the same.
3030     if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3031       for (size_t j = 0; j < cpu_map_valid_size; j++) {
3032         if (cpu_map[j] != 0) {
3033           for (size_t k = 0; k < BitsPerCLong; k++) {
3034             if (cpu_map[j] & (1UL << k)) {
3035               cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3036             }
3037           }
3038         }
3039       }
3040     }
3041   }
3042   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3043 }
3044 
3045 int os::Linux::get_node_by_cpu(int cpu_id) {
3046   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3047     return cpu_to_node()->at(cpu_id);
3048   }
3049   return -1;
3050 }
3051 
3052 GrowableArray<int>* os::Linux::_cpu_to_node;
3053 GrowableArray<int>* os::Linux::_nindex_to_node;
3054 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3055 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3056 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3057 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3058 os::Linux::numa_available_func_t os::Linux::_numa_available;
3059 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3060 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3061 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3062 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3063 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3064 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3065 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3066 unsigned long* os::Linux::_numa_all_nodes;
3067 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3068 struct bitmask* os::Linux::_numa_nodes_ptr;
3069 
3070 bool os::pd_uncommit_memory(char* addr, size_t size) {
3071   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3072                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3073   return res  != (uintptr_t) MAP_FAILED;
3074 }
3075 
3076 static address get_stack_commited_bottom(address bottom, size_t size) {
3077   address nbot = bottom;
3078   address ntop = bottom + size;
3079 
3080   size_t page_sz = os::vm_page_size();
3081   unsigned pages = size / page_sz;
3082 
3083   unsigned char vec[1];
3084   unsigned imin = 1, imax = pages + 1, imid;
3085   int mincore_return_value = 0;
3086 
3087   assert(imin <= imax, "Unexpected page size");
3088 
3089   while (imin < imax) {
3090     imid = (imax + imin) / 2;
3091     nbot = ntop - (imid * page_sz);
3092 
3093     // Use a trick with mincore to check whether the page is mapped or not.
3094     // mincore sets vec to 1 if page resides in memory and to 0 if page
3095     // is swapped output but if page we are asking for is unmapped
3096     // it returns -1,ENOMEM
3097     mincore_return_value = mincore(nbot, page_sz, vec);
3098 
3099     if (mincore_return_value == -1) {
3100       // Page is not mapped go up
3101       // to find first mapped page
3102       if (errno != EAGAIN) {
3103         assert(errno == ENOMEM, "Unexpected mincore errno");
3104         imax = imid;
3105       }
3106     } else {
3107       // Page is mapped go down
3108       // to find first not mapped page
3109       imin = imid + 1;
3110     }
3111   }
3112 
3113   nbot = nbot + page_sz;
3114 
3115   // Adjust stack bottom one page up if last checked page is not mapped
3116   if (mincore_return_value == -1) {
3117     nbot = nbot + page_sz;
3118   }
3119 
3120   return nbot;
3121 }
3122 
3123 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3124   int mincore_return_value;
3125   const size_t stripe = 1024;  // query this many pages each time
3126   unsigned char vec[stripe + 1];
3127   // set a guard
3128   vec[stripe] = 'X';
3129 
3130   const size_t page_sz = os::vm_page_size();
3131   size_t pages = size / page_sz;
3132 
3133   assert(is_aligned(start, page_sz), "Start address must be page aligned");
3134   assert(is_aligned(size, page_sz), "Size must be page aligned");
3135 
3136   committed_start = NULL;
3137 
3138   int loops = (pages + stripe - 1) / stripe;
3139   int committed_pages = 0;
3140   address loop_base = start;
3141   bool found_range = false;
3142 
3143   for (int index = 0; index < loops && !found_range; index ++) {
3144     assert(pages > 0, "Nothing to do");
3145     int pages_to_query = (pages >= stripe) ? stripe : pages;
3146     pages -= pages_to_query;
3147 
3148     // Get stable read
3149     while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3150 
3151     // During shutdown, some memory goes away without properly notifying NMT,
3152     // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3153     // Bailout and return as not committed for now.
3154     if (mincore_return_value == -1 && errno == ENOMEM) {
3155       return false;
3156     }
3157 
3158     assert(vec[stripe] == 'X', "overflow guard");
3159     assert(mincore_return_value == 0, "Range must be valid");
3160     // Process this stripe
3161     for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3162       if ((vec[vecIdx] & 0x01) == 0) { // not committed
3163         // End of current contiguous region
3164         if (committed_start != NULL) {
3165           found_range = true;
3166           break;
3167         }
3168       } else { // committed
3169         // Start of region
3170         if (committed_start == NULL) {
3171           committed_start = loop_base + page_sz * vecIdx;
3172         }
3173         committed_pages ++;
3174       }
3175     }
3176 
3177     loop_base += pages_to_query * page_sz;
3178   }
3179 
3180   if (committed_start != NULL) {
3181     assert(committed_pages > 0, "Must have committed region");
3182     assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3183     assert(committed_start >= start && committed_start < start + size, "Out of range");
3184     committed_size = page_sz * committed_pages;
3185     return true;
3186   } else {
3187     assert(committed_pages == 0, "Should not have committed region");
3188     return false;
3189   }
3190 }
3191 
3192 
3193 // Linux uses a growable mapping for the stack, and if the mapping for
3194 // the stack guard pages is not removed when we detach a thread the
3195 // stack cannot grow beyond the pages where the stack guard was
3196 // mapped.  If at some point later in the process the stack expands to
3197 // that point, the Linux kernel cannot expand the stack any further
3198 // because the guard pages are in the way, and a segfault occurs.
3199 //
3200 // However, it's essential not to split the stack region by unmapping
3201 // a region (leaving a hole) that's already part of the stack mapping,
3202 // so if the stack mapping has already grown beyond the guard pages at
3203 // the time we create them, we have to truncate the stack mapping.
3204 // So, we need to know the extent of the stack mapping when
3205 // create_stack_guard_pages() is called.
3206 
3207 // We only need this for stacks that are growable: at the time of
3208 // writing thread stacks don't use growable mappings (i.e. those
3209 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3210 // only applies to the main thread.
3211 
3212 // If the (growable) stack mapping already extends beyond the point
3213 // where we're going to put our guard pages, truncate the mapping at
3214 // that point by munmap()ping it.  This ensures that when we later
3215 // munmap() the guard pages we don't leave a hole in the stack
3216 // mapping. This only affects the main/primordial thread
3217 
3218 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3219   if (os::is_primordial_thread()) {
3220     // As we manually grow stack up to bottom inside create_attached_thread(),
3221     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3222     // we don't need to do anything special.
3223     // Check it first, before calling heavy function.
3224     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3225     unsigned char vec[1];
3226 
3227     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3228       // Fallback to slow path on all errors, including EAGAIN
3229       stack_extent = (uintptr_t) get_stack_commited_bottom(
3230                                                            os::Linux::initial_thread_stack_bottom(),
3231                                                            (size_t)addr - stack_extent);
3232     }
3233 
3234     if (stack_extent < (uintptr_t)addr) {
3235       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3236     }
3237   }
3238 
3239   return os::commit_memory(addr, size, !ExecMem);
3240 }
3241 
3242 // If this is a growable mapping, remove the guard pages entirely by
3243 // munmap()ping them.  If not, just call uncommit_memory(). This only
3244 // affects the main/primordial thread, but guard against future OS changes.
3245 // It's safe to always unmap guard pages for primordial thread because we
3246 // always place it right after end of the mapped region.
3247 
3248 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3249   uintptr_t stack_extent, stack_base;
3250 
3251   if (os::is_primordial_thread()) {
3252     return ::munmap(addr, size) == 0;
3253   }
3254 
3255   return os::uncommit_memory(addr, size);
3256 }
3257 
3258 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3259 // at 'requested_addr'. If there are existing memory mappings at the same
3260 // location, however, they will be overwritten. If 'fixed' is false,
3261 // 'requested_addr' is only treated as a hint, the return value may or
3262 // may not start from the requested address. Unlike Linux mmap(), this
3263 // function returns NULL to indicate failure.
3264 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3265   char * addr;
3266   int flags;
3267 
3268   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3269   if (fixed) {
3270     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3271     flags |= MAP_FIXED;
3272   }
3273 
3274   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3275   // touch an uncommitted page. Otherwise, the read/write might
3276   // succeed if we have enough swap space to back the physical page.
3277   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3278                        flags, -1, 0);
3279 
3280   return addr == MAP_FAILED ? NULL : addr;
3281 }
3282 
3283 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3284 //   (req_addr != NULL) or with a given alignment.
3285 //  - bytes shall be a multiple of alignment.
3286 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3287 //  - alignment sets the alignment at which memory shall be allocated.
3288 //     It must be a multiple of allocation granularity.
3289 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3290 //  req_addr or NULL.
3291 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3292 
3293   size_t extra_size = bytes;
3294   if (req_addr == NULL && alignment > 0) {
3295     extra_size += alignment;
3296   }
3297 
3298   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3299     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3300     -1, 0);
3301   if (start == MAP_FAILED) {
3302     start = NULL;
3303   } else {
3304     if (req_addr != NULL) {
3305       if (start != req_addr) {
3306         ::munmap(start, extra_size);
3307         start = NULL;
3308       }
3309     } else {
3310       char* const start_aligned = align_up(start, alignment);
3311       char* const end_aligned = start_aligned + bytes;
3312       char* const end = start + extra_size;
3313       if (start_aligned > start) {
3314         ::munmap(start, start_aligned - start);
3315       }
3316       if (end_aligned < end) {
3317         ::munmap(end_aligned, end - end_aligned);
3318       }
3319       start = start_aligned;
3320     }
3321   }
3322   return start;
3323 }
3324 
3325 static int anon_munmap(char * addr, size_t size) {
3326   return ::munmap(addr, size) == 0;
3327 }
3328 
3329 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3330                             size_t alignment_hint) {
3331   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3332 }
3333 
3334 bool os::pd_release_memory(char* addr, size_t size) {
3335   return anon_munmap(addr, size);
3336 }
3337 
3338 static bool linux_mprotect(char* addr, size_t size, int prot) {
3339   // Linux wants the mprotect address argument to be page aligned.
3340   char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3341 
3342   // According to SUSv3, mprotect() should only be used with mappings
3343   // established by mmap(), and mmap() always maps whole pages. Unaligned
3344   // 'addr' likely indicates problem in the VM (e.g. trying to change
3345   // protection of malloc'ed or statically allocated memory). Check the
3346   // caller if you hit this assert.
3347   assert(addr == bottom, "sanity check");
3348 
3349   size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3350   return ::mprotect(bottom, size, prot) == 0;
3351 }
3352 
3353 // Set protections specified
3354 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3355                         bool is_committed) {
3356   unsigned int p = 0;
3357   switch (prot) {
3358   case MEM_PROT_NONE: p = PROT_NONE; break;
3359   case MEM_PROT_READ: p = PROT_READ; break;
3360   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3361   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3362   default:
3363     ShouldNotReachHere();
3364   }
3365   // is_committed is unused.
3366   return linux_mprotect(addr, bytes, p);
3367 }
3368 
3369 bool os::guard_memory(char* addr, size_t size) {
3370   return linux_mprotect(addr, size, PROT_NONE);
3371 }
3372 
3373 bool os::unguard_memory(char* addr, size_t size) {
3374   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3375 }
3376 
3377 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3378                                                     size_t page_size) {
3379   bool result = false;
3380   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3381                  MAP_ANONYMOUS|MAP_PRIVATE,
3382                  -1, 0);
3383   if (p != MAP_FAILED) {
3384     void *aligned_p = align_up(p, page_size);
3385 
3386     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3387 
3388     munmap(p, page_size * 2);
3389   }
3390 
3391   if (warn && !result) {
3392     warning("TransparentHugePages is not supported by the operating system.");
3393   }
3394 
3395   return result;
3396 }
3397 
3398 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3399   bool result = false;
3400   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3401                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3402                  -1, 0);
3403 
3404   if (p != MAP_FAILED) {
3405     // We don't know if this really is a huge page or not.
3406     FILE *fp = fopen("/proc/self/maps", "r");
3407     if (fp) {
3408       while (!feof(fp)) {
3409         char chars[257];
3410         long x = 0;
3411         if (fgets(chars, sizeof(chars), fp)) {
3412           if (sscanf(chars, "%lx-%*x", &x) == 1
3413               && x == (long)p) {
3414             if (strstr (chars, "hugepage")) {
3415               result = true;
3416               break;
3417             }
3418           }
3419         }
3420       }
3421       fclose(fp);
3422     }
3423     munmap(p, page_size);
3424   }
3425 
3426   if (warn && !result) {
3427     warning("HugeTLBFS is not supported by the operating system.");
3428   }
3429 
3430   return result;
3431 }
3432 
3433 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3434 //
3435 // From the coredump_filter documentation:
3436 //
3437 // - (bit 0) anonymous private memory
3438 // - (bit 1) anonymous shared memory
3439 // - (bit 2) file-backed private memory
3440 // - (bit 3) file-backed shared memory
3441 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3442 //           effective only if the bit 2 is cleared)
3443 // - (bit 5) hugetlb private memory
3444 // - (bit 6) hugetlb shared memory
3445 // - (bit 7) dax private memory
3446 // - (bit 8) dax shared memory
3447 //
3448 static void set_coredump_filter(bool largepages, bool dax_shared) {
3449   FILE *f;
3450   long cdm;
3451   bool filter_changed = false;
3452 
3453   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3454     return;
3455   }
3456 
3457   if (fscanf(f, "%lx", &cdm) != 1) {
3458     fclose(f);
3459     return;
3460   }
3461 
3462   rewind(f);
3463 
3464   if (largepages && (cdm & LARGEPAGES_BIT) == 0) {
3465     cdm |= LARGEPAGES_BIT;
3466     filter_changed = true;
3467   }
3468   if (dax_shared && (cdm & DAX_SHARED_BIT) == 0) {
3469     cdm |= DAX_SHARED_BIT;
3470     filter_changed = true;
3471   }
3472   if (filter_changed) {
3473     fprintf(f, "%#lx", cdm);
3474   }
3475 
3476   fclose(f);
3477 }
3478 
3479 // Large page support
3480 
3481 static size_t _large_page_size = 0;
3482 
3483 size_t os::Linux::find_large_page_size() {
3484   size_t large_page_size = 0;
3485 
3486   // large_page_size on Linux is used to round up heap size. x86 uses either
3487   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3488   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3489   // page as large as 256M.
3490   //
3491   // Here we try to figure out page size by parsing /proc/meminfo and looking
3492   // for a line with the following format:
3493   //    Hugepagesize:     2048 kB
3494   //
3495   // If we can't determine the value (e.g. /proc is not mounted, or the text
3496   // format has been changed), we'll use the largest page size supported by
3497   // the processor.
3498 
3499 #ifndef ZERO
3500   large_page_size =
3501     AARCH64_ONLY(2 * M)
3502     AMD64_ONLY(2 * M)
3503     ARM32_ONLY(2 * M)
3504     IA32_ONLY(4 * M)
3505     IA64_ONLY(256 * M)
3506     PPC_ONLY(4 * M)
3507     S390_ONLY(1 * M)
3508     SPARC_ONLY(4 * M);
3509 #endif // ZERO
3510 
3511   FILE *fp = fopen("/proc/meminfo", "r");
3512   if (fp) {
3513     while (!feof(fp)) {
3514       int x = 0;
3515       char buf[16];
3516       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3517         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3518           large_page_size = x * K;
3519           break;
3520         }
3521       } else {
3522         // skip to next line
3523         for (;;) {
3524           int ch = fgetc(fp);
3525           if (ch == EOF || ch == (int)'\n') break;
3526         }
3527       }
3528     }
3529     fclose(fp);
3530   }
3531 
3532   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3533     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3534             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3535             proper_unit_for_byte_size(large_page_size));
3536   }
3537 
3538   return large_page_size;
3539 }
3540 
3541 size_t os::Linux::setup_large_page_size() {
3542   _large_page_size = Linux::find_large_page_size();
3543   const size_t default_page_size = (size_t)Linux::page_size();
3544   if (_large_page_size > default_page_size) {
3545     _page_sizes[0] = _large_page_size;
3546     _page_sizes[1] = default_page_size;
3547     _page_sizes[2] = 0;
3548   }
3549 
3550   return _large_page_size;
3551 }
3552 
3553 bool os::Linux::setup_large_page_type(size_t page_size) {
3554   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3555       FLAG_IS_DEFAULT(UseSHM) &&
3556       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3557 
3558     // The type of large pages has not been specified by the user.
3559 
3560     // Try UseHugeTLBFS and then UseSHM.
3561     UseHugeTLBFS = UseSHM = true;
3562 
3563     // Don't try UseTransparentHugePages since there are known
3564     // performance issues with it turned on. This might change in the future.
3565     UseTransparentHugePages = false;
3566   }
3567 
3568   if (UseTransparentHugePages) {
3569     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3570     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3571       UseHugeTLBFS = false;
3572       UseSHM = false;
3573       return true;
3574     }
3575     UseTransparentHugePages = false;
3576   }
3577 
3578   if (UseHugeTLBFS) {
3579     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3580     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3581       UseSHM = false;
3582       return true;
3583     }
3584     UseHugeTLBFS = false;
3585   }
3586 
3587   return UseSHM;
3588 }
3589 
3590 void os::large_page_init() {
3591   if (!UseLargePages &&
3592       !UseTransparentHugePages &&
3593       !UseHugeTLBFS &&
3594       !UseSHM) {
3595     // Not using large pages.
3596     return;
3597   }
3598 
3599   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3600     // The user explicitly turned off large pages.
3601     // Ignore the rest of the large pages flags.
3602     UseTransparentHugePages = false;
3603     UseHugeTLBFS = false;
3604     UseSHM = false;
3605     return;
3606   }
3607 
3608   size_t large_page_size = Linux::setup_large_page_size();
3609   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3610 
3611   set_coredump_filter(true /*largepages*/, false /*dax_shared*/);
3612 }
3613 
3614 #ifndef SHM_HUGETLB
3615   #define SHM_HUGETLB 04000
3616 #endif
3617 
3618 #define shm_warning_format(format, ...)              \
3619   do {                                               \
3620     if (UseLargePages &&                             \
3621         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3622          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3623          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3624       warning(format, __VA_ARGS__);                  \
3625     }                                                \
3626   } while (0)
3627 
3628 #define shm_warning(str) shm_warning_format("%s", str)
3629 
3630 #define shm_warning_with_errno(str)                \
3631   do {                                             \
3632     int err = errno;                               \
3633     shm_warning_format(str " (error = %d)", err);  \
3634   } while (0)
3635 
3636 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3637   assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3638 
3639   if (!is_aligned(alignment, SHMLBA)) {
3640     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3641     return NULL;
3642   }
3643 
3644   // To ensure that we get 'alignment' aligned memory from shmat,
3645   // we pre-reserve aligned virtual memory and then attach to that.
3646 
3647   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3648   if (pre_reserved_addr == NULL) {
3649     // Couldn't pre-reserve aligned memory.
3650     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3651     return NULL;
3652   }
3653 
3654   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3655   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3656 
3657   if ((intptr_t)addr == -1) {
3658     int err = errno;
3659     shm_warning_with_errno("Failed to attach shared memory.");
3660 
3661     assert(err != EACCES, "Unexpected error");
3662     assert(err != EIDRM,  "Unexpected error");
3663     assert(err != EINVAL, "Unexpected error");
3664 
3665     // Since we don't know if the kernel unmapped the pre-reserved memory area
3666     // we can't unmap it, since that would potentially unmap memory that was
3667     // mapped from other threads.
3668     return NULL;
3669   }
3670 
3671   return addr;
3672 }
3673 
3674 static char* shmat_at_address(int shmid, char* req_addr) {
3675   if (!is_aligned(req_addr, SHMLBA)) {
3676     assert(false, "Requested address needs to be SHMLBA aligned");
3677     return NULL;
3678   }
3679 
3680   char* addr = (char*)shmat(shmid, req_addr, 0);
3681 
3682   if ((intptr_t)addr == -1) {
3683     shm_warning_with_errno("Failed to attach shared memory.");
3684     return NULL;
3685   }
3686 
3687   return addr;
3688 }
3689 
3690 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3691   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3692   if (req_addr != NULL) {
3693     assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3694     assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3695     return shmat_at_address(shmid, req_addr);
3696   }
3697 
3698   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3699   // return large page size aligned memory addresses when req_addr == NULL.
3700   // However, if the alignment is larger than the large page size, we have
3701   // to manually ensure that the memory returned is 'alignment' aligned.
3702   if (alignment > os::large_page_size()) {
3703     assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3704     return shmat_with_alignment(shmid, bytes, alignment);
3705   } else {
3706     return shmat_at_address(shmid, NULL);
3707   }
3708 }
3709 
3710 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3711                                             char* req_addr, bool exec) {
3712   // "exec" is passed in but not used.  Creating the shared image for
3713   // the code cache doesn't have an SHM_X executable permission to check.
3714   assert(UseLargePages && UseSHM, "only for SHM large pages");
3715   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3716   assert(is_aligned(req_addr, alignment), "Unaligned address");
3717 
3718   if (!is_aligned(bytes, os::large_page_size())) {
3719     return NULL; // Fallback to small pages.
3720   }
3721 
3722   // Create a large shared memory region to attach to based on size.
3723   // Currently, size is the total size of the heap.
3724   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3725   if (shmid == -1) {
3726     // Possible reasons for shmget failure:
3727     // 1. shmmax is too small for Java heap.
3728     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3729     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3730     // 2. not enough large page memory.
3731     //    > check available large pages: cat /proc/meminfo
3732     //    > increase amount of large pages:
3733     //          echo new_value > /proc/sys/vm/nr_hugepages
3734     //      Note 1: different Linux may use different name for this property,
3735     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3736     //      Note 2: it's possible there's enough physical memory available but
3737     //            they are so fragmented after a long run that they can't
3738     //            coalesce into large pages. Try to reserve large pages when
3739     //            the system is still "fresh".
3740     shm_warning_with_errno("Failed to reserve shared memory.");
3741     return NULL;
3742   }
3743 
3744   // Attach to the region.
3745   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3746 
3747   // Remove shmid. If shmat() is successful, the actual shared memory segment
3748   // will be deleted when it's detached by shmdt() or when the process
3749   // terminates. If shmat() is not successful this will remove the shared
3750   // segment immediately.
3751   shmctl(shmid, IPC_RMID, NULL);
3752 
3753   return addr;
3754 }
3755 
3756 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3757                                         int error) {
3758   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3759 
3760   bool warn_on_failure = UseLargePages &&
3761       (!FLAG_IS_DEFAULT(UseLargePages) ||
3762        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3763        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3764 
3765   if (warn_on_failure) {
3766     char msg[128];
3767     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3768                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3769     warning("%s", msg);
3770   }
3771 }
3772 
3773 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3774                                                         char* req_addr,
3775                                                         bool exec) {
3776   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3777   assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3778   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3779 
3780   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3781   char* addr = (char*)::mmap(req_addr, bytes, prot,
3782                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3783                              -1, 0);
3784 
3785   if (addr == MAP_FAILED) {
3786     warn_on_large_pages_failure(req_addr, bytes, errno);
3787     return NULL;
3788   }
3789 
3790   assert(is_aligned(addr, os::large_page_size()), "Must be");
3791 
3792   return addr;
3793 }
3794 
3795 // Reserve memory using mmap(MAP_HUGETLB).
3796 //  - bytes shall be a multiple of alignment.
3797 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3798 //  - alignment sets the alignment at which memory shall be allocated.
3799 //     It must be a multiple of allocation granularity.
3800 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3801 //  req_addr or NULL.
3802 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3803                                                          size_t alignment,
3804                                                          char* req_addr,
3805                                                          bool exec) {
3806   size_t large_page_size = os::large_page_size();
3807   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3808 
3809   assert(is_aligned(req_addr, alignment), "Must be");
3810   assert(is_aligned(bytes, alignment), "Must be");
3811 
3812   // First reserve - but not commit - the address range in small pages.
3813   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3814 
3815   if (start == NULL) {
3816     return NULL;
3817   }
3818 
3819   assert(is_aligned(start, alignment), "Must be");
3820 
3821   char* end = start + bytes;
3822 
3823   // Find the regions of the allocated chunk that can be promoted to large pages.
3824   char* lp_start = align_up(start, large_page_size);
3825   char* lp_end   = align_down(end, large_page_size);
3826 
3827   size_t lp_bytes = lp_end - lp_start;
3828 
3829   assert(is_aligned(lp_bytes, large_page_size), "Must be");
3830 
3831   if (lp_bytes == 0) {
3832     // The mapped region doesn't even span the start and the end of a large page.
3833     // Fall back to allocate a non-special area.
3834     ::munmap(start, end - start);
3835     return NULL;
3836   }
3837 
3838   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3839 
3840   void* result;
3841 
3842   // Commit small-paged leading area.
3843   if (start != lp_start) {
3844     result = ::mmap(start, lp_start - start, prot,
3845                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3846                     -1, 0);
3847     if (result == MAP_FAILED) {
3848       ::munmap(lp_start, end - lp_start);
3849       return NULL;
3850     }
3851   }
3852 
3853   // Commit large-paged area.
3854   result = ::mmap(lp_start, lp_bytes, prot,
3855                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3856                   -1, 0);
3857   if (result == MAP_FAILED) {
3858     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3859     // If the mmap above fails, the large pages region will be unmapped and we
3860     // have regions before and after with small pages. Release these regions.
3861     //
3862     // |  mapped  |  unmapped  |  mapped  |
3863     // ^          ^            ^          ^
3864     // start      lp_start     lp_end     end
3865     //
3866     ::munmap(start, lp_start - start);
3867     ::munmap(lp_end, end - lp_end);
3868     return NULL;
3869   }
3870 
3871   // Commit small-paged trailing area.
3872   if (lp_end != end) {
3873     result = ::mmap(lp_end, end - lp_end, prot,
3874                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3875                     -1, 0);
3876     if (result == MAP_FAILED) {
3877       ::munmap(start, lp_end - start);
3878       return NULL;
3879     }
3880   }
3881 
3882   return start;
3883 }
3884 
3885 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3886                                                    size_t alignment,
3887                                                    char* req_addr,
3888                                                    bool exec) {
3889   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3890   assert(is_aligned(req_addr, alignment), "Must be");
3891   assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3892   assert(is_power_of_2(os::large_page_size()), "Must be");
3893   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3894 
3895   if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3896     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3897   } else {
3898     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3899   }
3900 }
3901 
3902 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3903                                  char* req_addr, bool exec) {
3904   assert(UseLargePages, "only for large pages");
3905 
3906   char* addr;
3907   if (UseSHM) {
3908     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3909   } else {
3910     assert(UseHugeTLBFS, "must be");
3911     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3912   }
3913 
3914   if (addr != NULL) {
3915     if (UseNUMAInterleaving) {
3916       numa_make_global(addr, bytes);
3917     }
3918 
3919     // The memory is committed
3920     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3921   }
3922 
3923   return addr;
3924 }
3925 
3926 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3927   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3928   return shmdt(base) == 0;
3929 }
3930 
3931 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3932   return pd_release_memory(base, bytes);
3933 }
3934 
3935 bool os::release_memory_special(char* base, size_t bytes) {
3936   bool res;
3937   if (MemTracker::tracking_level() > NMT_minimal) {
3938     Tracker tkr(Tracker::release);
3939     res = os::Linux::release_memory_special_impl(base, bytes);
3940     if (res) {
3941       tkr.record((address)base, bytes);
3942     }
3943 
3944   } else {
3945     res = os::Linux::release_memory_special_impl(base, bytes);
3946   }
3947   return res;
3948 }
3949 
3950 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3951   assert(UseLargePages, "only for large pages");
3952   bool res;
3953 
3954   if (UseSHM) {
3955     res = os::Linux::release_memory_special_shm(base, bytes);
3956   } else {
3957     assert(UseHugeTLBFS, "must be");
3958     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3959   }
3960   return res;
3961 }
3962 
3963 size_t os::large_page_size() {
3964   return _large_page_size;
3965 }
3966 
3967 // With SysV SHM the entire memory region must be allocated as shared
3968 // memory.
3969 // HugeTLBFS allows application to commit large page memory on demand.
3970 // However, when committing memory with HugeTLBFS fails, the region
3971 // that was supposed to be committed will lose the old reservation
3972 // and allow other threads to steal that memory region. Because of this
3973 // behavior we can't commit HugeTLBFS memory.
3974 bool os::can_commit_large_page_memory() {
3975   return UseTransparentHugePages;
3976 }
3977 
3978 bool os::can_execute_large_page_memory() {
3979   return UseTransparentHugePages || UseHugeTLBFS;
3980 }
3981 
3982 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
3983   assert(file_desc >= 0, "file_desc is not valid");
3984   char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
3985   if (result != NULL) {
3986     if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
3987       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
3988     }
3989   }
3990   return result;
3991 }
3992 
3993 // Reserve memory at an arbitrary address, only if that area is
3994 // available (and not reserved for something else).
3995 
3996 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3997   const int max_tries = 10;
3998   char* base[max_tries];
3999   size_t size[max_tries];
4000   const size_t gap = 0x000000;
4001 
4002   // Assert only that the size is a multiple of the page size, since
4003   // that's all that mmap requires, and since that's all we really know
4004   // about at this low abstraction level.  If we need higher alignment,
4005   // we can either pass an alignment to this method or verify alignment
4006   // in one of the methods further up the call chain.  See bug 5044738.
4007   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4008 
4009   // Repeatedly allocate blocks until the block is allocated at the
4010   // right spot.
4011 
4012   // Linux mmap allows caller to pass an address as hint; give it a try first,
4013   // if kernel honors the hint then we can return immediately.
4014   char * addr = anon_mmap(requested_addr, bytes, false);
4015   if (addr == requested_addr) {
4016     return requested_addr;
4017   }
4018 
4019   if (addr != NULL) {
4020     // mmap() is successful but it fails to reserve at the requested address
4021     anon_munmap(addr, bytes);
4022   }
4023 
4024   int i;
4025   for (i = 0; i < max_tries; ++i) {
4026     base[i] = reserve_memory(bytes);
4027 
4028     if (base[i] != NULL) {
4029       // Is this the block we wanted?
4030       if (base[i] == requested_addr) {
4031         size[i] = bytes;
4032         break;
4033       }
4034 
4035       // Does this overlap the block we wanted? Give back the overlapped
4036       // parts and try again.
4037 
4038       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
4039       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
4040         unmap_memory(base[i], top_overlap);
4041         base[i] += top_overlap;
4042         size[i] = bytes - top_overlap;
4043       } else {
4044         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
4045         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
4046           unmap_memory(requested_addr, bottom_overlap);
4047           size[i] = bytes - bottom_overlap;
4048         } else {
4049           size[i] = bytes;
4050         }
4051       }
4052     }
4053   }
4054 
4055   // Give back the unused reserved pieces.
4056 
4057   for (int j = 0; j < i; ++j) {
4058     if (base[j] != NULL) {
4059       unmap_memory(base[j], size[j]);
4060     }
4061   }
4062 
4063   if (i < max_tries) {
4064     return requested_addr;
4065   } else {
4066     return NULL;
4067   }
4068 }
4069 
4070 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4071   return ::read(fd, buf, nBytes);
4072 }
4073 
4074 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4075   return ::pread(fd, buf, nBytes, offset);
4076 }
4077 
4078 // Short sleep, direct OS call.
4079 //
4080 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4081 // sched_yield(2) will actually give up the CPU:
4082 //
4083 //   * Alone on this pariticular CPU, keeps running.
4084 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
4085 //     (pre 2.6.39).
4086 //
4087 // So calling this with 0 is an alternative.
4088 //
4089 void os::naked_short_sleep(jlong ms) {
4090   struct timespec req;
4091 
4092   assert(ms < 1000, "Un-interruptable sleep, short time use only");
4093   req.tv_sec = 0;
4094   if (ms > 0) {
4095     req.tv_nsec = (ms % 1000) * 1000000;
4096   } else {
4097     req.tv_nsec = 1;
4098   }
4099 
4100   nanosleep(&req, NULL);
4101 
4102   return;
4103 }
4104 
4105 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4106 void os::infinite_sleep() {
4107   while (true) {    // sleep forever ...
4108     ::sleep(100);   // ... 100 seconds at a time
4109   }
4110 }
4111 
4112 // Used to convert frequent JVM_Yield() to nops
4113 bool os::dont_yield() {
4114   return DontYieldALot;
4115 }
4116 
4117 void os::naked_yield() {
4118   sched_yield();
4119 }
4120 
4121 ////////////////////////////////////////////////////////////////////////////////
4122 // thread priority support
4123 
4124 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4125 // only supports dynamic priority, static priority must be zero. For real-time
4126 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4127 // However, for large multi-threaded applications, SCHED_RR is not only slower
4128 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4129 // of 5 runs - Sep 2005).
4130 //
4131 // The following code actually changes the niceness of kernel-thread/LWP. It
4132 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4133 // not the entire user process, and user level threads are 1:1 mapped to kernel
4134 // threads. It has always been the case, but could change in the future. For
4135 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4136 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4137 
4138 int os::java_to_os_priority[CriticalPriority + 1] = {
4139   19,              // 0 Entry should never be used
4140 
4141    4,              // 1 MinPriority
4142    3,              // 2
4143    2,              // 3
4144 
4145    1,              // 4
4146    0,              // 5 NormPriority
4147   -1,              // 6
4148 
4149   -2,              // 7
4150   -3,              // 8
4151   -4,              // 9 NearMaxPriority
4152 
4153   -5,              // 10 MaxPriority
4154 
4155   -5               // 11 CriticalPriority
4156 };
4157 
4158 static int prio_init() {
4159   if (ThreadPriorityPolicy == 1) {
4160     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4161     // if effective uid is not root. Perhaps, a more elegant way of doing
4162     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4163     if (geteuid() != 0) {
4164       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4165         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4166       }
4167       ThreadPriorityPolicy = 0;
4168     }
4169   }
4170   if (UseCriticalJavaThreadPriority) {
4171     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4172   }
4173   return 0;
4174 }
4175 
4176 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4177   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4178 
4179   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4180   return (ret == 0) ? OS_OK : OS_ERR;
4181 }
4182 
4183 OSReturn os::get_native_priority(const Thread* const thread,
4184                                  int *priority_ptr) {
4185   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4186     *priority_ptr = java_to_os_priority[NormPriority];
4187     return OS_OK;
4188   }
4189 
4190   errno = 0;
4191   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4192   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4193 }
4194 
4195 ////////////////////////////////////////////////////////////////////////////////
4196 // suspend/resume support
4197 
4198 //  The low-level signal-based suspend/resume support is a remnant from the
4199 //  old VM-suspension that used to be for java-suspension, safepoints etc,
4200 //  within hotspot. Currently used by JFR's OSThreadSampler
4201 //
4202 //  The remaining code is greatly simplified from the more general suspension
4203 //  code that used to be used.
4204 //
4205 //  The protocol is quite simple:
4206 //  - suspend:
4207 //      - sends a signal to the target thread
4208 //      - polls the suspend state of the osthread using a yield loop
4209 //      - target thread signal handler (SR_handler) sets suspend state
4210 //        and blocks in sigsuspend until continued
4211 //  - resume:
4212 //      - sets target osthread state to continue
4213 //      - sends signal to end the sigsuspend loop in the SR_handler
4214 //
4215 //  Note that the SR_lock plays no role in this suspend/resume protocol,
4216 //  but is checked for NULL in SR_handler as a thread termination indicator.
4217 //  The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4218 //
4219 //  Note that resume_clear_context() and suspend_save_context() are needed
4220 //  by SR_handler(), so that fetch_frame_from_ucontext() works,
4221 //  which in part is used by:
4222 //    - Forte Analyzer: AsyncGetCallTrace()
4223 //    - StackBanging: get_frame_at_stack_banging_point()
4224 
4225 static void resume_clear_context(OSThread *osthread) {
4226   osthread->set_ucontext(NULL);
4227   osthread->set_siginfo(NULL);
4228 }
4229 
4230 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4231                                  ucontext_t* context) {
4232   osthread->set_ucontext(context);
4233   osthread->set_siginfo(siginfo);
4234 }
4235 
4236 // Handler function invoked when a thread's execution is suspended or
4237 // resumed. We have to be careful that only async-safe functions are
4238 // called here (Note: most pthread functions are not async safe and
4239 // should be avoided.)
4240 //
4241 // Note: sigwait() is a more natural fit than sigsuspend() from an
4242 // interface point of view, but sigwait() prevents the signal hander
4243 // from being run. libpthread would get very confused by not having
4244 // its signal handlers run and prevents sigwait()'s use with the
4245 // mutex granting granting signal.
4246 //
4247 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4248 //
4249 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4250   // Save and restore errno to avoid confusing native code with EINTR
4251   // after sigsuspend.
4252   int old_errno = errno;
4253 
4254   Thread* thread = Thread::current_or_null_safe();
4255   assert(thread != NULL, "Missing current thread in SR_handler");
4256 
4257   // On some systems we have seen signal delivery get "stuck" until the signal
4258   // mask is changed as part of thread termination. Check that the current thread
4259   // has not already terminated (via SR_lock()) - else the following assertion
4260   // will fail because the thread is no longer a JavaThread as the ~JavaThread
4261   // destructor has completed.
4262 
4263   if (thread->SR_lock() == NULL) {
4264     return;
4265   }
4266 
4267   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4268 
4269   OSThread* osthread = thread->osthread();
4270 
4271   os::SuspendResume::State current = osthread->sr.state();
4272   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4273     suspend_save_context(osthread, siginfo, context);
4274 
4275     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4276     os::SuspendResume::State state = osthread->sr.suspended();
4277     if (state == os::SuspendResume::SR_SUSPENDED) {
4278       sigset_t suspend_set;  // signals for sigsuspend()
4279       sigemptyset(&suspend_set);
4280       // get current set of blocked signals and unblock resume signal
4281       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4282       sigdelset(&suspend_set, SR_signum);
4283 
4284       sr_semaphore.signal();
4285       // wait here until we are resumed
4286       while (1) {
4287         sigsuspend(&suspend_set);
4288 
4289         os::SuspendResume::State result = osthread->sr.running();
4290         if (result == os::SuspendResume::SR_RUNNING) {
4291           sr_semaphore.signal();
4292           break;
4293         }
4294       }
4295 
4296     } else if (state == os::SuspendResume::SR_RUNNING) {
4297       // request was cancelled, continue
4298     } else {
4299       ShouldNotReachHere();
4300     }
4301 
4302     resume_clear_context(osthread);
4303   } else if (current == os::SuspendResume::SR_RUNNING) {
4304     // request was cancelled, continue
4305   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4306     // ignore
4307   } else {
4308     // ignore
4309   }
4310 
4311   errno = old_errno;
4312 }
4313 
4314 static int SR_initialize() {
4315   struct sigaction act;
4316   char *s;
4317 
4318   // Get signal number to use for suspend/resume
4319   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4320     int sig = ::strtol(s, 0, 10);
4321     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4322         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4323       SR_signum = sig;
4324     } else {
4325       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4326               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4327     }
4328   }
4329 
4330   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4331          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4332 
4333   sigemptyset(&SR_sigset);
4334   sigaddset(&SR_sigset, SR_signum);
4335 
4336   // Set up signal handler for suspend/resume
4337   act.sa_flags = SA_RESTART|SA_SIGINFO;
4338   act.sa_handler = (void (*)(int)) SR_handler;
4339 
4340   // SR_signum is blocked by default.
4341   // 4528190 - We also need to block pthread restart signal (32 on all
4342   // supported Linux platforms). Note that LinuxThreads need to block
4343   // this signal for all threads to work properly. So we don't have
4344   // to use hard-coded signal number when setting up the mask.
4345   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4346 
4347   if (sigaction(SR_signum, &act, 0) == -1) {
4348     return -1;
4349   }
4350 
4351   // Save signal flag
4352   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4353   return 0;
4354 }
4355 
4356 static int sr_notify(OSThread* osthread) {
4357   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4358   assert_status(status == 0, status, "pthread_kill");
4359   return status;
4360 }
4361 
4362 // "Randomly" selected value for how long we want to spin
4363 // before bailing out on suspending a thread, also how often
4364 // we send a signal to a thread we want to resume
4365 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4366 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4367 
4368 // returns true on success and false on error - really an error is fatal
4369 // but this seems the normal response to library errors
4370 static bool do_suspend(OSThread* osthread) {
4371   assert(osthread->sr.is_running(), "thread should be running");
4372   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4373 
4374   // mark as suspended and send signal
4375   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4376     // failed to switch, state wasn't running?
4377     ShouldNotReachHere();
4378     return false;
4379   }
4380 
4381   if (sr_notify(osthread) != 0) {
4382     ShouldNotReachHere();
4383   }
4384 
4385   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4386   while (true) {
4387     if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4388       break;
4389     } else {
4390       // timeout
4391       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4392       if (cancelled == os::SuspendResume::SR_RUNNING) {
4393         return false;
4394       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4395         // make sure that we consume the signal on the semaphore as well
4396         sr_semaphore.wait();
4397         break;
4398       } else {
4399         ShouldNotReachHere();
4400         return false;
4401       }
4402     }
4403   }
4404 
4405   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4406   return true;
4407 }
4408 
4409 static void do_resume(OSThread* osthread) {
4410   assert(osthread->sr.is_suspended(), "thread should be suspended");
4411   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4412 
4413   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4414     // failed to switch to WAKEUP_REQUEST
4415     ShouldNotReachHere();
4416     return;
4417   }
4418 
4419   while (true) {
4420     if (sr_notify(osthread) == 0) {
4421       if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
4422         if (osthread->sr.is_running()) {
4423           return;
4424         }
4425       }
4426     } else {
4427       ShouldNotReachHere();
4428     }
4429   }
4430 
4431   guarantee(osthread->sr.is_running(), "Must be running!");
4432 }
4433 
4434 ///////////////////////////////////////////////////////////////////////////////////
4435 // signal handling (except suspend/resume)
4436 
4437 // This routine may be used by user applications as a "hook" to catch signals.
4438 // The user-defined signal handler must pass unrecognized signals to this
4439 // routine, and if it returns true (non-zero), then the signal handler must
4440 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4441 // routine will never retun false (zero), but instead will execute a VM panic
4442 // routine kill the process.
4443 //
4444 // If this routine returns false, it is OK to call it again.  This allows
4445 // the user-defined signal handler to perform checks either before or after
4446 // the VM performs its own checks.  Naturally, the user code would be making
4447 // a serious error if it tried to handle an exception (such as a null check
4448 // or breakpoint) that the VM was generating for its own correct operation.
4449 //
4450 // This routine may recognize any of the following kinds of signals:
4451 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4452 // It should be consulted by handlers for any of those signals.
4453 //
4454 // The caller of this routine must pass in the three arguments supplied
4455 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4456 // field of the structure passed to sigaction().  This routine assumes that
4457 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4458 //
4459 // Note that the VM will print warnings if it detects conflicting signal
4460 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4461 //
4462 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4463                                                  siginfo_t* siginfo,
4464                                                  void* ucontext,
4465                                                  int abort_if_unrecognized);
4466 
4467 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4468   assert(info != NULL && uc != NULL, "it must be old kernel");
4469   int orig_errno = errno;  // Preserve errno value over signal handler.
4470   JVM_handle_linux_signal(sig, info, uc, true);
4471   errno = orig_errno;
4472 }
4473 
4474 
4475 // This boolean allows users to forward their own non-matching signals
4476 // to JVM_handle_linux_signal, harmlessly.
4477 bool os::Linux::signal_handlers_are_installed = false;
4478 
4479 // For signal-chaining
4480 struct sigaction sigact[NSIG];
4481 uint64_t sigs = 0;
4482 #if (64 < NSIG-1)
4483 #error "Not all signals can be encoded in sigs. Adapt its type!"
4484 #endif
4485 bool os::Linux::libjsig_is_loaded = false;
4486 typedef struct sigaction *(*get_signal_t)(int);
4487 get_signal_t os::Linux::get_signal_action = NULL;
4488 
4489 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4490   struct sigaction *actp = NULL;
4491 
4492   if (libjsig_is_loaded) {
4493     // Retrieve the old signal handler from libjsig
4494     actp = (*get_signal_action)(sig);
4495   }
4496   if (actp == NULL) {
4497     // Retrieve the preinstalled signal handler from jvm
4498     actp = get_preinstalled_handler(sig);
4499   }
4500 
4501   return actp;
4502 }
4503 
4504 static bool call_chained_handler(struct sigaction *actp, int sig,
4505                                  siginfo_t *siginfo, void *context) {
4506   // Call the old signal handler
4507   if (actp->sa_handler == SIG_DFL) {
4508     // It's more reasonable to let jvm treat it as an unexpected exception
4509     // instead of taking the default action.
4510     return false;
4511   } else if (actp->sa_handler != SIG_IGN) {
4512     if ((actp->sa_flags & SA_NODEFER) == 0) {
4513       // automaticlly block the signal
4514       sigaddset(&(actp->sa_mask), sig);
4515     }
4516 
4517     sa_handler_t hand = NULL;
4518     sa_sigaction_t sa = NULL;
4519     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4520     // retrieve the chained handler
4521     if (siginfo_flag_set) {
4522       sa = actp->sa_sigaction;
4523     } else {
4524       hand = actp->sa_handler;
4525     }
4526 
4527     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4528       actp->sa_handler = SIG_DFL;
4529     }
4530 
4531     // try to honor the signal mask
4532     sigset_t oset;
4533     sigemptyset(&oset);
4534     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4535 
4536     // call into the chained handler
4537     if (siginfo_flag_set) {
4538       (*sa)(sig, siginfo, context);
4539     } else {
4540       (*hand)(sig);
4541     }
4542 
4543     // restore the signal mask
4544     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4545   }
4546   // Tell jvm's signal handler the signal is taken care of.
4547   return true;
4548 }
4549 
4550 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4551   bool chained = false;
4552   // signal-chaining
4553   if (UseSignalChaining) {
4554     struct sigaction *actp = get_chained_signal_action(sig);
4555     if (actp != NULL) {
4556       chained = call_chained_handler(actp, sig, siginfo, context);
4557     }
4558   }
4559   return chained;
4560 }
4561 
4562 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4563   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4564     return &sigact[sig];
4565   }
4566   return NULL;
4567 }
4568 
4569 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4570   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4571   sigact[sig] = oldAct;
4572   sigs |= (uint64_t)1 << (sig-1);
4573 }
4574 
4575 // for diagnostic
4576 int sigflags[NSIG];
4577 
4578 int os::Linux::get_our_sigflags(int sig) {
4579   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4580   return sigflags[sig];
4581 }
4582 
4583 void os::Linux::set_our_sigflags(int sig, int flags) {
4584   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4585   if (sig > 0 && sig < NSIG) {
4586     sigflags[sig] = flags;
4587   }
4588 }
4589 
4590 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4591   // Check for overwrite.
4592   struct sigaction oldAct;
4593   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4594 
4595   void* oldhand = oldAct.sa_sigaction
4596                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4597                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4598   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4599       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4600       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4601     if (AllowUserSignalHandlers || !set_installed) {
4602       // Do not overwrite; user takes responsibility to forward to us.
4603       return;
4604     } else if (UseSignalChaining) {
4605       // save the old handler in jvm
4606       save_preinstalled_handler(sig, oldAct);
4607       // libjsig also interposes the sigaction() call below and saves the
4608       // old sigaction on it own.
4609     } else {
4610       fatal("Encountered unexpected pre-existing sigaction handler "
4611             "%#lx for signal %d.", (long)oldhand, sig);
4612     }
4613   }
4614 
4615   struct sigaction sigAct;
4616   sigfillset(&(sigAct.sa_mask));
4617   sigAct.sa_handler = SIG_DFL;
4618   if (!set_installed) {
4619     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4620   } else {
4621     sigAct.sa_sigaction = signalHandler;
4622     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4623   }
4624   // Save flags, which are set by ours
4625   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4626   sigflags[sig] = sigAct.sa_flags;
4627 
4628   int ret = sigaction(sig, &sigAct, &oldAct);
4629   assert(ret == 0, "check");
4630 
4631   void* oldhand2  = oldAct.sa_sigaction
4632                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4633                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4634   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4635 }
4636 
4637 // install signal handlers for signals that HotSpot needs to
4638 // handle in order to support Java-level exception handling.
4639 
4640 void os::Linux::install_signal_handlers() {
4641   if (!signal_handlers_are_installed) {
4642     signal_handlers_are_installed = true;
4643 
4644     // signal-chaining
4645     typedef void (*signal_setting_t)();
4646     signal_setting_t begin_signal_setting = NULL;
4647     signal_setting_t end_signal_setting = NULL;
4648     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4649                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4650     if (begin_signal_setting != NULL) {
4651       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4652                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4653       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4654                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4655       libjsig_is_loaded = true;
4656       assert(UseSignalChaining, "should enable signal-chaining");
4657     }
4658     if (libjsig_is_loaded) {
4659       // Tell libjsig jvm is setting signal handlers
4660       (*begin_signal_setting)();
4661     }
4662 
4663     set_signal_handler(SIGSEGV, true);
4664     set_signal_handler(SIGPIPE, true);
4665     set_signal_handler(SIGBUS, true);
4666     set_signal_handler(SIGILL, true);
4667     set_signal_handler(SIGFPE, true);
4668 #if defined(PPC64)
4669     set_signal_handler(SIGTRAP, true);
4670 #endif
4671     set_signal_handler(SIGXFSZ, true);
4672 
4673     if (libjsig_is_loaded) {
4674       // Tell libjsig jvm finishes setting signal handlers
4675       (*end_signal_setting)();
4676     }
4677 
4678     // We don't activate signal checker if libjsig is in place, we trust ourselves
4679     // and if UserSignalHandler is installed all bets are off.
4680     // Log that signal checking is off only if -verbose:jni is specified.
4681     if (CheckJNICalls) {
4682       if (libjsig_is_loaded) {
4683         if (PrintJNIResolving) {
4684           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4685         }
4686         check_signals = false;
4687       }
4688       if (AllowUserSignalHandlers) {
4689         if (PrintJNIResolving) {
4690           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4691         }
4692         check_signals = false;
4693       }
4694     }
4695   }
4696 }
4697 
4698 // This is the fastest way to get thread cpu time on Linux.
4699 // Returns cpu time (user+sys) for any thread, not only for current.
4700 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4701 // It might work on 2.6.10+ with a special kernel/glibc patch.
4702 // For reference, please, see IEEE Std 1003.1-2004:
4703 //   http://www.unix.org/single_unix_specification
4704 
4705 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4706   struct timespec tp;
4707   int rc = os::Linux::clock_gettime(clockid, &tp);
4708   assert(rc == 0, "clock_gettime is expected to return 0 code");
4709 
4710   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4711 }
4712 
4713 void os::Linux::initialize_os_info() {
4714   assert(_os_version == 0, "OS info already initialized");
4715 
4716   struct utsname _uname;
4717 
4718   uint32_t major;
4719   uint32_t minor;
4720   uint32_t fix;
4721 
4722   int rc;
4723 
4724   // Kernel version is unknown if
4725   // verification below fails.
4726   _os_version = 0x01000000;
4727 
4728   rc = uname(&_uname);
4729   if (rc != -1) {
4730 
4731     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4732     if (rc == 3) {
4733 
4734       if (major < 256 && minor < 256 && fix < 256) {
4735         // Kernel version format is as expected,
4736         // set it overriding unknown state.
4737         _os_version = (major << 16) |
4738                       (minor << 8 ) |
4739                       (fix   << 0 ) ;
4740       }
4741     }
4742   }
4743 }
4744 
4745 uint32_t os::Linux::os_version() {
4746   assert(_os_version != 0, "not initialized");
4747   return _os_version & 0x00FFFFFF;
4748 }
4749 
4750 bool os::Linux::os_version_is_known() {
4751   assert(_os_version != 0, "not initialized");
4752   return _os_version & 0x01000000 ? false : true;
4753 }
4754 
4755 /////
4756 // glibc on Linux platform uses non-documented flag
4757 // to indicate, that some special sort of signal
4758 // trampoline is used.
4759 // We will never set this flag, and we should
4760 // ignore this flag in our diagnostic
4761 #ifdef SIGNIFICANT_SIGNAL_MASK
4762   #undef SIGNIFICANT_SIGNAL_MASK
4763 #endif
4764 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4765 
4766 static const char* get_signal_handler_name(address handler,
4767                                            char* buf, int buflen) {
4768   int offset = 0;
4769   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4770   if (found) {
4771     // skip directory names
4772     const char *p1, *p2;
4773     p1 = buf;
4774     size_t len = strlen(os::file_separator());
4775     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4776     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4777   } else {
4778     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4779   }
4780   return buf;
4781 }
4782 
4783 static void print_signal_handler(outputStream* st, int sig,
4784                                  char* buf, size_t buflen) {
4785   struct sigaction sa;
4786 
4787   sigaction(sig, NULL, &sa);
4788 
4789   // See comment for SIGNIFICANT_SIGNAL_MASK define
4790   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4791 
4792   st->print("%s: ", os::exception_name(sig, buf, buflen));
4793 
4794   address handler = (sa.sa_flags & SA_SIGINFO)
4795     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4796     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4797 
4798   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4799     st->print("SIG_DFL");
4800   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4801     st->print("SIG_IGN");
4802   } else {
4803     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4804   }
4805 
4806   st->print(", sa_mask[0]=");
4807   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4808 
4809   address rh = VMError::get_resetted_sighandler(sig);
4810   // May be, handler was resetted by VMError?
4811   if (rh != NULL) {
4812     handler = rh;
4813     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4814   }
4815 
4816   st->print(", sa_flags=");
4817   os::Posix::print_sa_flags(st, sa.sa_flags);
4818 
4819   // Check: is it our handler?
4820   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4821       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4822     // It is our signal handler
4823     // check for flags, reset system-used one!
4824     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4825       st->print(
4826                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4827                 os::Linux::get_our_sigflags(sig));
4828     }
4829   }
4830   st->cr();
4831 }
4832 
4833 
4834 #define DO_SIGNAL_CHECK(sig)                      \
4835   do {                                            \
4836     if (!sigismember(&check_signal_done, sig)) {  \
4837       os::Linux::check_signal_handler(sig);       \
4838     }                                             \
4839   } while (0)
4840 
4841 // This method is a periodic task to check for misbehaving JNI applications
4842 // under CheckJNI, we can add any periodic checks here
4843 
4844 void os::run_periodic_checks() {
4845   if (check_signals == false) return;
4846 
4847   // SEGV and BUS if overridden could potentially prevent
4848   // generation of hs*.log in the event of a crash, debugging
4849   // such a case can be very challenging, so we absolutely
4850   // check the following for a good measure:
4851   DO_SIGNAL_CHECK(SIGSEGV);
4852   DO_SIGNAL_CHECK(SIGILL);
4853   DO_SIGNAL_CHECK(SIGFPE);
4854   DO_SIGNAL_CHECK(SIGBUS);
4855   DO_SIGNAL_CHECK(SIGPIPE);
4856   DO_SIGNAL_CHECK(SIGXFSZ);
4857 #if defined(PPC64)
4858   DO_SIGNAL_CHECK(SIGTRAP);
4859 #endif
4860 
4861   // ReduceSignalUsage allows the user to override these handlers
4862   // see comments at the very top and jvm_md.h
4863   if (!ReduceSignalUsage) {
4864     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4865     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4866     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4867     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4868   }
4869 
4870   DO_SIGNAL_CHECK(SR_signum);
4871 }
4872 
4873 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4874 
4875 static os_sigaction_t os_sigaction = NULL;
4876 
4877 void os::Linux::check_signal_handler(int sig) {
4878   char buf[O_BUFLEN];
4879   address jvmHandler = NULL;
4880 
4881 
4882   struct sigaction act;
4883   if (os_sigaction == NULL) {
4884     // only trust the default sigaction, in case it has been interposed
4885     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4886     if (os_sigaction == NULL) return;
4887   }
4888 
4889   os_sigaction(sig, (struct sigaction*)NULL, &act);
4890 
4891 
4892   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4893 
4894   address thisHandler = (act.sa_flags & SA_SIGINFO)
4895     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4896     : CAST_FROM_FN_PTR(address, act.sa_handler);
4897 
4898 
4899   switch (sig) {
4900   case SIGSEGV:
4901   case SIGBUS:
4902   case SIGFPE:
4903   case SIGPIPE:
4904   case SIGILL:
4905   case SIGXFSZ:
4906     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4907     break;
4908 
4909   case SHUTDOWN1_SIGNAL:
4910   case SHUTDOWN2_SIGNAL:
4911   case SHUTDOWN3_SIGNAL:
4912   case BREAK_SIGNAL:
4913     jvmHandler = (address)user_handler();
4914     break;
4915 
4916   default:
4917     if (sig == SR_signum) {
4918       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4919     } else {
4920       return;
4921     }
4922     break;
4923   }
4924 
4925   if (thisHandler != jvmHandler) {
4926     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4927     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4928     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4929     // No need to check this sig any longer
4930     sigaddset(&check_signal_done, sig);
4931     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4932     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4933       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4934                     exception_name(sig, buf, O_BUFLEN));
4935     }
4936   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4937     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4938     tty->print("expected:");
4939     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4940     tty->cr();
4941     tty->print("  found:");
4942     os::Posix::print_sa_flags(tty, act.sa_flags);
4943     tty->cr();
4944     // No need to check this sig any longer
4945     sigaddset(&check_signal_done, sig);
4946   }
4947 
4948   // Dump all the signal
4949   if (sigismember(&check_signal_done, sig)) {
4950     print_signal_handlers(tty, buf, O_BUFLEN);
4951   }
4952 }
4953 
4954 extern void report_error(char* file_name, int line_no, char* title,
4955                          char* format, ...);
4956 
4957 // this is called _before_ most of the global arguments have been parsed
4958 void os::init(void) {
4959   char dummy;   // used to get a guess on initial stack address
4960 
4961   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4962 
4963   init_random(1234567);
4964 
4965   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4966   if (Linux::page_size() == -1) {
4967     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4968           os::strerror(errno));
4969   }
4970   init_page_sizes((size_t) Linux::page_size());
4971 
4972   Linux::initialize_system_info();
4973 
4974   Linux::initialize_os_info();
4975 
4976   // _main_thread points to the thread that created/loaded the JVM.
4977   Linux::_main_thread = pthread_self();
4978 
4979   Linux::clock_init();
4980   initial_time_count = javaTimeNanos();
4981 
4982   // retrieve entry point for pthread_setname_np
4983   Linux::_pthread_setname_np =
4984     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4985 
4986   os::Posix::init();
4987 }
4988 
4989 // To install functions for atexit system call
4990 extern "C" {
4991   static void perfMemory_exit_helper() {
4992     perfMemory_exit();
4993   }
4994 }
4995 
4996 void os::pd_init_container_support() {
4997   OSContainer::init();
4998 }
4999 
5000 // this is called _after_ the global arguments have been parsed
5001 jint os::init_2(void) {
5002 
5003   os::Posix::init_2();
5004 
5005   Linux::fast_thread_clock_init();
5006 
5007   // initialize suspend/resume support - must do this before signal_sets_init()
5008   if (SR_initialize() != 0) {
5009     perror("SR_initialize failed");
5010     return JNI_ERR;
5011   }
5012 
5013   Linux::signal_sets_init();
5014   Linux::install_signal_handlers();
5015   // Initialize data for jdk.internal.misc.Signal
5016   if (!ReduceSignalUsage) {
5017     jdk_misc_signal_init();
5018   }
5019 
5020   // Check and sets minimum stack sizes against command line options
5021   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5022     return JNI_ERR;
5023   }
5024 
5025   suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5026   if (!suppress_primordial_thread_resolution) {
5027     Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5028   }
5029 
5030 #if defined(IA32)
5031   workaround_expand_exec_shield_cs_limit();
5032 #endif
5033 
5034   Linux::libpthread_init();
5035   Linux::sched_getcpu_init();
5036   log_info(os)("HotSpot is running with %s, %s",
5037                Linux::glibc_version(), Linux::libpthread_version());
5038 
5039   if (UseNUMA) {
5040     if (!Linux::libnuma_init()) {
5041       UseNUMA = false;
5042     } else {
5043       if ((Linux::numa_max_node() < 1) || Linux::isbound_to_single_node()) {
5044         // If there's only one node (they start from 0) or if the process
5045         // is bound explicitly to a single node using membind, disable NUMA.
5046         UseNUMA = false;
5047       }
5048     }
5049 
5050     if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5051       // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5052       // we can make the adaptive lgrp chunk resizing work. If the user specified both
5053       // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5054       // and disable adaptive resizing.
5055       if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5056         warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5057                 "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5058         UseAdaptiveSizePolicy = false;
5059         UseAdaptiveNUMAChunkSizing = false;
5060       }
5061     }
5062 
5063     if (!UseNUMA && ForceNUMA) {
5064       UseNUMA = true;
5065     }
5066   }
5067 
5068   if (MaxFDLimit) {
5069     // set the number of file descriptors to max. print out error
5070     // if getrlimit/setrlimit fails but continue regardless.
5071     struct rlimit nbr_files;
5072     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5073     if (status != 0) {
5074       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5075     } else {
5076       nbr_files.rlim_cur = nbr_files.rlim_max;
5077       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5078       if (status != 0) {
5079         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5080       }
5081     }
5082   }
5083 
5084   // Initialize lock used to serialize thread creation (see os::create_thread)
5085   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5086 
5087   // at-exit methods are called in the reverse order of their registration.
5088   // atexit functions are called on return from main or as a result of a
5089   // call to exit(3C). There can be only 32 of these functions registered
5090   // and atexit() does not set errno.
5091 
5092   if (PerfAllowAtExitRegistration) {
5093     // only register atexit functions if PerfAllowAtExitRegistration is set.
5094     // atexit functions can be delayed until process exit time, which
5095     // can be problematic for embedded VM situations. Embedded VMs should
5096     // call DestroyJavaVM() to assure that VM resources are released.
5097 
5098     // note: perfMemory_exit_helper atexit function may be removed in
5099     // the future if the appropriate cleanup code can be added to the
5100     // VM_Exit VMOperation's doit method.
5101     if (atexit(perfMemory_exit_helper) != 0) {
5102       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5103     }
5104   }
5105 
5106   // initialize thread priority policy
5107   prio_init();
5108 
5109   if (!FLAG_IS_DEFAULT(AllocateHeapAt)) {
5110     set_coredump_filter(false /*largepages*/, true /*dax_shared*/);
5111   }
5112   return JNI_OK;
5113 }
5114 
5115 // Mark the polling page as unreadable
5116 void os::make_polling_page_unreadable(void) {
5117   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5118     fatal("Could not disable polling page");
5119   }
5120 }
5121 
5122 // Mark the polling page as readable
5123 void os::make_polling_page_readable(void) {
5124   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5125     fatal("Could not enable polling page");
5126   }
5127 }
5128 
5129 // older glibc versions don't have this macro (which expands to
5130 // an optimized bit-counting function) so we have to roll our own
5131 #ifndef CPU_COUNT
5132 
5133 static int _cpu_count(const cpu_set_t* cpus) {
5134   int count = 0;
5135   // only look up to the number of configured processors
5136   for (int i = 0; i < os::processor_count(); i++) {
5137     if (CPU_ISSET(i, cpus)) {
5138       count++;
5139     }
5140   }
5141   return count;
5142 }
5143 
5144 #define CPU_COUNT(cpus) _cpu_count(cpus)
5145 
5146 #endif // CPU_COUNT
5147 
5148 // Get the current number of available processors for this process.
5149 // This value can change at any time during a process's lifetime.
5150 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5151 // If it appears there may be more than 1024 processors then we do a
5152 // dynamic check - see 6515172 for details.
5153 // If anything goes wrong we fallback to returning the number of online
5154 // processors - which can be greater than the number available to the process.
5155 int os::Linux::active_processor_count() {
5156   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
5157   cpu_set_t* cpus_p = &cpus;
5158   int cpus_size = sizeof(cpu_set_t);
5159 
5160   int configured_cpus = os::processor_count();  // upper bound on available cpus
5161   int cpu_count = 0;
5162 
5163 // old build platforms may not support dynamic cpu sets
5164 #ifdef CPU_ALLOC
5165 
5166   // To enable easy testing of the dynamic path on different platforms we
5167   // introduce a diagnostic flag: UseCpuAllocPath
5168   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5169     // kernel may use a mask bigger than cpu_set_t
5170     log_trace(os)("active_processor_count: using dynamic path %s"
5171                   "- configured processors: %d",
5172                   UseCpuAllocPath ? "(forced) " : "",
5173                   configured_cpus);
5174     cpus_p = CPU_ALLOC(configured_cpus);
5175     if (cpus_p != NULL) {
5176       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5177       // zero it just to be safe
5178       CPU_ZERO_S(cpus_size, cpus_p);
5179     }
5180     else {
5181        // failed to allocate so fallback to online cpus
5182        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5183        log_trace(os)("active_processor_count: "
5184                      "CPU_ALLOC failed (%s) - using "
5185                      "online processor count: %d",
5186                      os::strerror(errno), online_cpus);
5187        return online_cpus;
5188     }
5189   }
5190   else {
5191     log_trace(os)("active_processor_count: using static path - configured processors: %d",
5192                   configured_cpus);
5193   }
5194 #else // CPU_ALLOC
5195 // these stubs won't be executed
5196 #define CPU_COUNT_S(size, cpus) -1
5197 #define CPU_FREE(cpus)
5198 
5199   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5200                 configured_cpus);
5201 #endif // CPU_ALLOC
5202 
5203   // pid 0 means the current thread - which we have to assume represents the process
5204   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5205     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5206       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5207     }
5208     else {
5209       cpu_count = CPU_COUNT(cpus_p);
5210     }
5211     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5212   }
5213   else {
5214     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5215     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5216             "which may exceed available processors", os::strerror(errno), cpu_count);
5217   }
5218 
5219   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5220     CPU_FREE(cpus_p);
5221   }
5222 
5223   assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5224   return cpu_count;
5225 }
5226 
5227 // Determine the active processor count from one of
5228 // three different sources:
5229 //
5230 // 1. User option -XX:ActiveProcessorCount
5231 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5232 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5233 //
5234 // Option 1, if specified, will always override.
5235 // If the cgroup subsystem is active and configured, we
5236 // will return the min of the cgroup and option 2 results.
5237 // This is required since tools, such as numactl, that
5238 // alter cpu affinity do not update cgroup subsystem
5239 // cpuset configuration files.
5240 int os::active_processor_count() {
5241   // User has overridden the number of active processors
5242   if (ActiveProcessorCount > 0) {
5243     log_trace(os)("active_processor_count: "
5244                   "active processor count set by user : %d",
5245                   ActiveProcessorCount);
5246     return ActiveProcessorCount;
5247   }
5248 
5249   int active_cpus;
5250   if (OSContainer::is_containerized()) {
5251     active_cpus = OSContainer::active_processor_count();
5252     log_trace(os)("active_processor_count: determined by OSContainer: %d",
5253                    active_cpus);
5254   } else {
5255     active_cpus = os::Linux::active_processor_count();
5256   }
5257 
5258   return active_cpus;
5259 }
5260 
5261 uint os::processor_id() {
5262   const int id = Linux::sched_getcpu();
5263   assert(id >= 0 && id < _processor_count, "Invalid processor id");
5264   return (uint)id;
5265 }
5266 
5267 void os::set_native_thread_name(const char *name) {
5268   if (Linux::_pthread_setname_np) {
5269     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5270     snprintf(buf, sizeof(buf), "%s", name);
5271     buf[sizeof(buf) - 1] = '\0';
5272     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5273     // ERANGE should not happen; all other errors should just be ignored.
5274     assert(rc != ERANGE, "pthread_setname_np failed");
5275   }
5276 }
5277 
5278 bool os::distribute_processes(uint length, uint* distribution) {
5279   // Not yet implemented.
5280   return false;
5281 }
5282 
5283 bool os::bind_to_processor(uint processor_id) {
5284   // Not yet implemented.
5285   return false;
5286 }
5287 
5288 ///
5289 
5290 void os::SuspendedThreadTask::internal_do_task() {
5291   if (do_suspend(_thread->osthread())) {
5292     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5293     do_task(context);
5294     do_resume(_thread->osthread());
5295   }
5296 }
5297 
5298 ////////////////////////////////////////////////////////////////////////////////
5299 // debug support
5300 
5301 bool os::find(address addr, outputStream* st) {
5302   Dl_info dlinfo;
5303   memset(&dlinfo, 0, sizeof(dlinfo));
5304   if (dladdr(addr, &dlinfo) != 0) {
5305     st->print(PTR_FORMAT ": ", p2i(addr));
5306     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5307       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5308                 p2i(addr) - p2i(dlinfo.dli_saddr));
5309     } else if (dlinfo.dli_fbase != NULL) {
5310       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5311     } else {
5312       st->print("<absolute address>");
5313     }
5314     if (dlinfo.dli_fname != NULL) {
5315       st->print(" in %s", dlinfo.dli_fname);
5316     }
5317     if (dlinfo.dli_fbase != NULL) {
5318       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5319     }
5320     st->cr();
5321 
5322     if (Verbose) {
5323       // decode some bytes around the PC
5324       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5325       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5326       address       lowest = (address) dlinfo.dli_sname;
5327       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5328       if (begin < lowest)  begin = lowest;
5329       Dl_info dlinfo2;
5330       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5331           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5332         end = (address) dlinfo2.dli_saddr;
5333       }
5334       Disassembler::decode(begin, end, st);
5335     }
5336     return true;
5337   }
5338   return false;
5339 }
5340 
5341 ////////////////////////////////////////////////////////////////////////////////
5342 // misc
5343 
5344 // This does not do anything on Linux. This is basically a hook for being
5345 // able to use structured exception handling (thread-local exception filters)
5346 // on, e.g., Win32.
5347 void
5348 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5349                          JavaCallArguments* args, Thread* thread) {
5350   f(value, method, args, thread);
5351 }
5352 
5353 void os::print_statistics() {
5354 }
5355 
5356 bool os::message_box(const char* title, const char* message) {
5357   int i;
5358   fdStream err(defaultStream::error_fd());
5359   for (i = 0; i < 78; i++) err.print_raw("=");
5360   err.cr();
5361   err.print_raw_cr(title);
5362   for (i = 0; i < 78; i++) err.print_raw("-");
5363   err.cr();
5364   err.print_raw_cr(message);
5365   for (i = 0; i < 78; i++) err.print_raw("=");
5366   err.cr();
5367 
5368   char buf[16];
5369   // Prevent process from exiting upon "read error" without consuming all CPU
5370   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5371 
5372   return buf[0] == 'y' || buf[0] == 'Y';
5373 }
5374 
5375 // Is a (classpath) directory empty?
5376 bool os::dir_is_empty(const char* path) {
5377   DIR *dir = NULL;
5378   struct dirent *ptr;
5379 
5380   dir = opendir(path);
5381   if (dir == NULL) return true;
5382 
5383   // Scan the directory
5384   bool result = true;
5385   while (result && (ptr = readdir(dir)) != NULL) {
5386     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5387       result = false;
5388     }
5389   }
5390   closedir(dir);
5391   return result;
5392 }
5393 
5394 // This code originates from JDK's sysOpen and open64_w
5395 // from src/solaris/hpi/src/system_md.c
5396 
5397 int os::open(const char *path, int oflag, int mode) {
5398   if (strlen(path) > MAX_PATH - 1) {
5399     errno = ENAMETOOLONG;
5400     return -1;
5401   }
5402 
5403   // All file descriptors that are opened in the Java process and not
5404   // specifically destined for a subprocess should have the close-on-exec
5405   // flag set.  If we don't set it, then careless 3rd party native code
5406   // might fork and exec without closing all appropriate file descriptors
5407   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5408   // turn might:
5409   //
5410   // - cause end-of-file to fail to be detected on some file
5411   //   descriptors, resulting in mysterious hangs, or
5412   //
5413   // - might cause an fopen in the subprocess to fail on a system
5414   //   suffering from bug 1085341.
5415   //
5416   // (Yes, the default setting of the close-on-exec flag is a Unix
5417   // design flaw)
5418   //
5419   // See:
5420   // 1085341: 32-bit stdio routines should support file descriptors >255
5421   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5422   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5423   //
5424   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5425   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5426   // because it saves a system call and removes a small window where the flag
5427   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5428   // and we fall back to using FD_CLOEXEC (see below).
5429 #ifdef O_CLOEXEC
5430   oflag |= O_CLOEXEC;
5431 #endif
5432 
5433   int fd = ::open64(path, oflag, mode);
5434   if (fd == -1) return -1;
5435 
5436   //If the open succeeded, the file might still be a directory
5437   {
5438     struct stat64 buf64;
5439     int ret = ::fstat64(fd, &buf64);
5440     int st_mode = buf64.st_mode;
5441 
5442     if (ret != -1) {
5443       if ((st_mode & S_IFMT) == S_IFDIR) {
5444         errno = EISDIR;
5445         ::close(fd);
5446         return -1;
5447       }
5448     } else {
5449       ::close(fd);
5450       return -1;
5451     }
5452   }
5453 
5454 #ifdef FD_CLOEXEC
5455   // Validate that the use of the O_CLOEXEC flag on open above worked.
5456   // With recent kernels, we will perform this check exactly once.
5457   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5458   if (!O_CLOEXEC_is_known_to_work) {
5459     int flags = ::fcntl(fd, F_GETFD);
5460     if (flags != -1) {
5461       if ((flags & FD_CLOEXEC) != 0)
5462         O_CLOEXEC_is_known_to_work = 1;
5463       else
5464         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5465     }
5466   }
5467 #endif
5468 
5469   return fd;
5470 }
5471 
5472 
5473 // create binary file, rewriting existing file if required
5474 int os::create_binary_file(const char* path, bool rewrite_existing) {
5475   int oflags = O_WRONLY | O_CREAT;
5476   if (!rewrite_existing) {
5477     oflags |= O_EXCL;
5478   }
5479   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5480 }
5481 
5482 // return current position of file pointer
5483 jlong os::current_file_offset(int fd) {
5484   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5485 }
5486 
5487 // move file pointer to the specified offset
5488 jlong os::seek_to_file_offset(int fd, jlong offset) {
5489   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5490 }
5491 
5492 // This code originates from JDK's sysAvailable
5493 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5494 
5495 int os::available(int fd, jlong *bytes) {
5496   jlong cur, end;
5497   int mode;
5498   struct stat64 buf64;
5499 
5500   if (::fstat64(fd, &buf64) >= 0) {
5501     mode = buf64.st_mode;
5502     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5503       int n;
5504       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5505         *bytes = n;
5506         return 1;
5507       }
5508     }
5509   }
5510   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5511     return 0;
5512   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5513     return 0;
5514   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5515     return 0;
5516   }
5517   *bytes = end - cur;
5518   return 1;
5519 }
5520 
5521 // Map a block of memory.
5522 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5523                         char *addr, size_t bytes, bool read_only,
5524                         bool allow_exec) {
5525   int prot;
5526   int flags = MAP_PRIVATE;
5527 
5528   if (read_only) {
5529     prot = PROT_READ;
5530   } else {
5531     prot = PROT_READ | PROT_WRITE;
5532   }
5533 
5534   if (allow_exec) {
5535     prot |= PROT_EXEC;
5536   }
5537 
5538   if (addr != NULL) {
5539     flags |= MAP_FIXED;
5540   }
5541 
5542   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5543                                      fd, file_offset);
5544   if (mapped_address == MAP_FAILED) {
5545     return NULL;
5546   }
5547   return mapped_address;
5548 }
5549 
5550 
5551 // Remap a block of memory.
5552 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5553                           char *addr, size_t bytes, bool read_only,
5554                           bool allow_exec) {
5555   // same as map_memory() on this OS
5556   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5557                         allow_exec);
5558 }
5559 
5560 
5561 // Unmap a block of memory.
5562 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5563   return munmap(addr, bytes) == 0;
5564 }
5565 
5566 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5567 
5568 static jlong fast_cpu_time(Thread *thread) {
5569     clockid_t clockid;
5570     int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5571                                               &clockid);
5572     if (rc == 0) {
5573       return os::Linux::fast_thread_cpu_time(clockid);
5574     } else {
5575       // It's possible to encounter a terminated native thread that failed
5576       // to detach itself from the VM - which should result in ESRCH.
5577       assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5578       return -1;
5579     }
5580 }
5581 
5582 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5583 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5584 // of a thread.
5585 //
5586 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5587 // the fast estimate available on the platform.
5588 
5589 jlong os::current_thread_cpu_time() {
5590   if (os::Linux::supports_fast_thread_cpu_time()) {
5591     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5592   } else {
5593     // return user + sys since the cost is the same
5594     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5595   }
5596 }
5597 
5598 jlong os::thread_cpu_time(Thread* thread) {
5599   // consistent with what current_thread_cpu_time() returns
5600   if (os::Linux::supports_fast_thread_cpu_time()) {
5601     return fast_cpu_time(thread);
5602   } else {
5603     return slow_thread_cpu_time(thread, true /* user + sys */);
5604   }
5605 }
5606 
5607 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5608   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5609     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5610   } else {
5611     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5612   }
5613 }
5614 
5615 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5616   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5617     return fast_cpu_time(thread);
5618   } else {
5619     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5620   }
5621 }
5622 
5623 //  -1 on error.
5624 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5625   pid_t  tid = thread->osthread()->thread_id();
5626   char *s;
5627   char stat[2048];
5628   int statlen;
5629   char proc_name[64];
5630   int count;
5631   long sys_time, user_time;
5632   char cdummy;
5633   int idummy;
5634   long ldummy;
5635   FILE *fp;
5636 
5637   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5638   fp = fopen(proc_name, "r");
5639   if (fp == NULL) return -1;
5640   statlen = fread(stat, 1, 2047, fp);
5641   stat[statlen] = '\0';
5642   fclose(fp);
5643 
5644   // Skip pid and the command string. Note that we could be dealing with
5645   // weird command names, e.g. user could decide to rename java launcher
5646   // to "java 1.4.2 :)", then the stat file would look like
5647   //                1234 (java 1.4.2 :)) R ... ...
5648   // We don't really need to know the command string, just find the last
5649   // occurrence of ")" and then start parsing from there. See bug 4726580.
5650   s = strrchr(stat, ')');
5651   if (s == NULL) return -1;
5652 
5653   // Skip blank chars
5654   do { s++; } while (s && isspace(*s));
5655 
5656   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5657                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5658                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5659                  &user_time, &sys_time);
5660   if (count != 13) return -1;
5661   if (user_sys_cpu_time) {
5662     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5663   } else {
5664     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5665   }
5666 }
5667 
5668 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5669   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5670   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5671   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5672   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5673 }
5674 
5675 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5676   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5677   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5678   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5679   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5680 }
5681 
5682 bool os::is_thread_cpu_time_supported() {
5683   return true;
5684 }
5685 
5686 // System loadavg support.  Returns -1 if load average cannot be obtained.
5687 // Linux doesn't yet have a (official) notion of processor sets,
5688 // so just return the system wide load average.
5689 int os::loadavg(double loadavg[], int nelem) {
5690   return ::getloadavg(loadavg, nelem);
5691 }
5692 
5693 void os::pause() {
5694   char filename[MAX_PATH];
5695   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5696     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5697   } else {
5698     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5699   }
5700 
5701   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5702   if (fd != -1) {
5703     struct stat buf;
5704     ::close(fd);
5705     while (::stat(filename, &buf) == 0) {
5706       (void)::poll(NULL, 0, 100);
5707     }
5708   } else {
5709     jio_fprintf(stderr,
5710                 "Could not open pause file '%s', continuing immediately.\n", filename);
5711   }
5712 }
5713 
5714 extern char** environ;
5715 
5716 // Run the specified command in a separate process. Return its exit value,
5717 // or -1 on failure (e.g. can't fork a new process).
5718 // Unlike system(), this function can be called from signal handler. It
5719 // doesn't block SIGINT et al.
5720 int os::fork_and_exec(char* cmd) {
5721   const char * argv[4] = {"sh", "-c", cmd, NULL};
5722 
5723   pid_t pid = fork();
5724 
5725   if (pid < 0) {
5726     // fork failed
5727     return -1;
5728 
5729   } else if (pid == 0) {
5730     // child process
5731 
5732     execve("/bin/sh", (char* const*)argv, environ);
5733 
5734     // execve failed
5735     _exit(-1);
5736 
5737   } else  {
5738     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5739     // care about the actual exit code, for now.
5740 
5741     int status;
5742 
5743     // Wait for the child process to exit.  This returns immediately if
5744     // the child has already exited. */
5745     while (waitpid(pid, &status, 0) < 0) {
5746       switch (errno) {
5747       case ECHILD: return 0;
5748       case EINTR: break;
5749       default: return -1;
5750       }
5751     }
5752 
5753     if (WIFEXITED(status)) {
5754       // The child exited normally; get its exit code.
5755       return WEXITSTATUS(status);
5756     } else if (WIFSIGNALED(status)) {
5757       // The child exited because of a signal
5758       // The best value to return is 0x80 + signal number,
5759       // because that is what all Unix shells do, and because
5760       // it allows callers to distinguish between process exit and
5761       // process death by signal.
5762       return 0x80 + WTERMSIG(status);
5763     } else {
5764       // Unknown exit code; pass it through
5765       return status;
5766     }
5767   }
5768 }
5769 
5770 // Get the default path to the core file
5771 // Returns the length of the string
5772 int os::get_core_path(char* buffer, size_t bufferSize) {
5773   /*
5774    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5775    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5776    */
5777   const int core_pattern_len = 129;
5778   char core_pattern[core_pattern_len] = {0};
5779 
5780   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5781   if (core_pattern_file == -1) {
5782     return -1;
5783   }
5784 
5785   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5786   ::close(core_pattern_file);
5787   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5788     return -1;
5789   }
5790   if (core_pattern[ret-1] == '\n') {
5791     core_pattern[ret-1] = '\0';
5792   } else {
5793     core_pattern[ret] = '\0';
5794   }
5795 
5796   // Replace the %p in the core pattern with the process id. NOTE: we do this
5797   // only if the pattern doesn't start with "|", and we support only one %p in
5798   // the pattern.
5799   char *pid_pos = strstr(core_pattern, "%p");
5800   const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : "";  // skip over the "%p"
5801   int written;
5802 
5803   if (core_pattern[0] == '/') {
5804     if (pid_pos != NULL) {
5805       *pid_pos = '\0';
5806       written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5807                              current_process_id(), tail);
5808     } else {
5809       written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5810     }
5811   } else {
5812     char cwd[PATH_MAX];
5813 
5814     const char* p = get_current_directory(cwd, PATH_MAX);
5815     if (p == NULL) {
5816       return -1;
5817     }
5818 
5819     if (core_pattern[0] == '|') {
5820       written = jio_snprintf(buffer, bufferSize,
5821                              "\"%s\" (or dumping to %s/core.%d)",
5822                              &core_pattern[1], p, current_process_id());
5823     } else if (pid_pos != NULL) {
5824       *pid_pos = '\0';
5825       written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5826                              current_process_id(), tail);
5827     } else {
5828       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5829     }
5830   }
5831 
5832   if (written < 0) {
5833     return -1;
5834   }
5835 
5836   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5837     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5838 
5839     if (core_uses_pid_file != -1) {
5840       char core_uses_pid = 0;
5841       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5842       ::close(core_uses_pid_file);
5843 
5844       if (core_uses_pid == '1') {
5845         jio_snprintf(buffer + written, bufferSize - written,
5846                                           ".%d", current_process_id());
5847       }
5848     }
5849   }
5850 
5851   return strlen(buffer);
5852 }
5853 
5854 bool os::start_debugging(char *buf, int buflen) {
5855   int len = (int)strlen(buf);
5856   char *p = &buf[len];
5857 
5858   jio_snprintf(p, buflen-len,
5859                "\n\n"
5860                "Do you want to debug the problem?\n\n"
5861                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5862                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5863                "Otherwise, press RETURN to abort...",
5864                os::current_process_id(), os::current_process_id(),
5865                os::current_thread_id(), os::current_thread_id());
5866 
5867   bool yes = os::message_box("Unexpected Error", buf);
5868 
5869   if (yes) {
5870     // yes, user asked VM to launch debugger
5871     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5872                  os::current_process_id(), os::current_process_id());
5873 
5874     os::fork_and_exec(buf);
5875     yes = false;
5876   }
5877   return yes;
5878 }
5879 
5880 
5881 // Java/Compiler thread:
5882 //
5883 //   Low memory addresses
5884 // P0 +------------------------+
5885 //    |                        |\  Java thread created by VM does not have glibc
5886 //    |    glibc guard page    | - guard page, attached Java thread usually has
5887 //    |                        |/  1 glibc guard page.
5888 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5889 //    |                        |\
5890 //    |  HotSpot Guard Pages   | - red, yellow and reserved pages
5891 //    |                        |/
5892 //    +------------------------+ JavaThread::stack_reserved_zone_base()
5893 //    |                        |\
5894 //    |      Normal Stack      | -
5895 //    |                        |/
5896 // P2 +------------------------+ Thread::stack_base()
5897 //
5898 // Non-Java thread:
5899 //
5900 //   Low memory addresses
5901 // P0 +------------------------+
5902 //    |                        |\
5903 //    |  glibc guard page      | - usually 1 page
5904 //    |                        |/
5905 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5906 //    |                        |\
5907 //    |      Normal Stack      | -
5908 //    |                        |/
5909 // P2 +------------------------+ Thread::stack_base()
5910 //
5911 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5912 //    returned from pthread_attr_getstack().
5913 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5914 //    of the stack size given in pthread_attr. We work around this for
5915 //    threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5916 //
5917 #ifndef ZERO
5918 static void current_stack_region(address * bottom, size_t * size) {
5919   if (os::is_primordial_thread()) {
5920     // primordial thread needs special handling because pthread_getattr_np()
5921     // may return bogus value.
5922     *bottom = os::Linux::initial_thread_stack_bottom();
5923     *size   = os::Linux::initial_thread_stack_size();
5924   } else {
5925     pthread_attr_t attr;
5926 
5927     int rslt = pthread_getattr_np(pthread_self(), &attr);
5928 
5929     // JVM needs to know exact stack location, abort if it fails
5930     if (rslt != 0) {
5931       if (rslt == ENOMEM) {
5932         vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5933       } else {
5934         fatal("pthread_getattr_np failed with error = %d", rslt);
5935       }
5936     }
5937 
5938     if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5939       fatal("Cannot locate current stack attributes!");
5940     }
5941 
5942     // Work around NPTL stack guard error.
5943     size_t guard_size = 0;
5944     rslt = pthread_attr_getguardsize(&attr, &guard_size);
5945     if (rslt != 0) {
5946       fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5947     }
5948     *bottom += guard_size;
5949     *size   -= guard_size;
5950 
5951     pthread_attr_destroy(&attr);
5952 
5953   }
5954   assert(os::current_stack_pointer() >= *bottom &&
5955          os::current_stack_pointer() < *bottom + *size, "just checking");
5956 }
5957 
5958 address os::current_stack_base() {
5959   address bottom;
5960   size_t size;
5961   current_stack_region(&bottom, &size);
5962   return (bottom + size);
5963 }
5964 
5965 size_t os::current_stack_size() {
5966   // This stack size includes the usable stack and HotSpot guard pages
5967   // (for the threads that have Hotspot guard pages).
5968   address bottom;
5969   size_t size;
5970   current_stack_region(&bottom, &size);
5971   return size;
5972 }
5973 #endif
5974 
5975 static inline struct timespec get_mtime(const char* filename) {
5976   struct stat st;
5977   int ret = os::stat(filename, &st);
5978   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5979   return st.st_mtim;
5980 }
5981 
5982 int os::compare_file_modified_times(const char* file1, const char* file2) {
5983   struct timespec filetime1 = get_mtime(file1);
5984   struct timespec filetime2 = get_mtime(file2);
5985   int diff = filetime1.tv_sec - filetime2.tv_sec;
5986   if (diff == 0) {
5987     return filetime1.tv_nsec - filetime2.tv_nsec;
5988   }
5989   return diff;
5990 }
5991 
5992 /////////////// Unit tests ///////////////
5993 
5994 #ifndef PRODUCT
5995 
5996 #define test_log(...)              \
5997   do {                             \
5998     if (VerboseInternalVMTests) {  \
5999       tty->print_cr(__VA_ARGS__);  \
6000       tty->flush();                \
6001     }                              \
6002   } while (false)
6003 
6004 class TestReserveMemorySpecial : AllStatic {
6005  public:
6006   static void small_page_write(void* addr, size_t size) {
6007     size_t page_size = os::vm_page_size();
6008 
6009     char* end = (char*)addr + size;
6010     for (char* p = (char*)addr; p < end; p += page_size) {
6011       *p = 1;
6012     }
6013   }
6014 
6015   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6016     if (!UseHugeTLBFS) {
6017       return;
6018     }
6019 
6020     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6021 
6022     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6023 
6024     if (addr != NULL) {
6025       small_page_write(addr, size);
6026 
6027       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6028     }
6029   }
6030 
6031   static void test_reserve_memory_special_huge_tlbfs_only() {
6032     if (!UseHugeTLBFS) {
6033       return;
6034     }
6035 
6036     size_t lp = os::large_page_size();
6037 
6038     for (size_t size = lp; size <= lp * 10; size += lp) {
6039       test_reserve_memory_special_huge_tlbfs_only(size);
6040     }
6041   }
6042 
6043   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6044     size_t lp = os::large_page_size();
6045     size_t ag = os::vm_allocation_granularity();
6046 
6047     // sizes to test
6048     const size_t sizes[] = {
6049       lp, lp + ag, lp + lp / 2, lp * 2,
6050       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6051       lp * 10, lp * 10 + lp / 2
6052     };
6053     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6054 
6055     // For each size/alignment combination, we test three scenarios:
6056     // 1) with req_addr == NULL
6057     // 2) with a non-null req_addr at which we expect to successfully allocate
6058     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6059     //    expect the allocation to either fail or to ignore req_addr
6060 
6061     // Pre-allocate two areas; they shall be as large as the largest allocation
6062     //  and aligned to the largest alignment we will be testing.
6063     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6064     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6065       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6066       -1, 0);
6067     assert(mapping1 != MAP_FAILED, "should work");
6068 
6069     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6070       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6071       -1, 0);
6072     assert(mapping2 != MAP_FAILED, "should work");
6073 
6074     // Unmap the first mapping, but leave the second mapping intact: the first
6075     // mapping will serve as a value for a "good" req_addr (case 2). The second
6076     // mapping, still intact, as "bad" req_addr (case 3).
6077     ::munmap(mapping1, mapping_size);
6078 
6079     // Case 1
6080     test_log("%s, req_addr NULL:", __FUNCTION__);
6081     test_log("size            align           result");
6082 
6083     for (int i = 0; i < num_sizes; i++) {
6084       const size_t size = sizes[i];
6085       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6086         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6087         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6088                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6089         if (p != NULL) {
6090           assert(is_aligned(p, alignment), "must be");
6091           small_page_write(p, size);
6092           os::Linux::release_memory_special_huge_tlbfs(p, size);
6093         }
6094       }
6095     }
6096 
6097     // Case 2
6098     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6099     test_log("size            align           req_addr         result");
6100 
6101     for (int i = 0; i < num_sizes; i++) {
6102       const size_t size = sizes[i];
6103       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6104         char* const req_addr = align_up(mapping1, alignment);
6105         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6106         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6107                  size, alignment, p2i(req_addr), p2i(p),
6108                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6109         if (p != NULL) {
6110           assert(p == req_addr, "must be");
6111           small_page_write(p, size);
6112           os::Linux::release_memory_special_huge_tlbfs(p, size);
6113         }
6114       }
6115     }
6116 
6117     // Case 3
6118     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6119     test_log("size            align           req_addr         result");
6120 
6121     for (int i = 0; i < num_sizes; i++) {
6122       const size_t size = sizes[i];
6123       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6124         char* const req_addr = align_up(mapping2, alignment);
6125         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6126         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6127                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6128         // as the area around req_addr contains already existing mappings, the API should always
6129         // return NULL (as per contract, it cannot return another address)
6130         assert(p == NULL, "must be");
6131       }
6132     }
6133 
6134     ::munmap(mapping2, mapping_size);
6135 
6136   }
6137 
6138   static void test_reserve_memory_special_huge_tlbfs() {
6139     if (!UseHugeTLBFS) {
6140       return;
6141     }
6142 
6143     test_reserve_memory_special_huge_tlbfs_only();
6144     test_reserve_memory_special_huge_tlbfs_mixed();
6145   }
6146 
6147   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6148     if (!UseSHM) {
6149       return;
6150     }
6151 
6152     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6153 
6154     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6155 
6156     if (addr != NULL) {
6157       assert(is_aligned(addr, alignment), "Check");
6158       assert(is_aligned(addr, os::large_page_size()), "Check");
6159 
6160       small_page_write(addr, size);
6161 
6162       os::Linux::release_memory_special_shm(addr, size);
6163     }
6164   }
6165 
6166   static void test_reserve_memory_special_shm() {
6167     size_t lp = os::large_page_size();
6168     size_t ag = os::vm_allocation_granularity();
6169 
6170     for (size_t size = ag; size < lp * 3; size += ag) {
6171       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6172         test_reserve_memory_special_shm(size, alignment);
6173       }
6174     }
6175   }
6176 
6177   static void test() {
6178     test_reserve_memory_special_huge_tlbfs();
6179     test_reserve_memory_special_shm();
6180   }
6181 };
6182 
6183 void TestReserveMemorySpecial_test() {
6184   TestReserveMemorySpecial::test();
6185 }
6186 
6187 #endif