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