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