1 /*
   2  * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "jvm.h"
  27 #include "classfile/classLoader.hpp"
  28 #include "classfile/systemDictionary.hpp"
  29 #include "classfile/vmSymbols.hpp"
  30 #include "code/icBuffer.hpp"
  31 #include "code/vtableStubs.hpp"
  32 #include "compiler/compileBroker.hpp"
  33 #include "compiler/disassembler.hpp"
  34 #include "interpreter/interpreter.hpp"
  35 #include "logging/log.hpp"
  36 #include "logging/logStream.hpp"
  37 #include "memory/allocation.inline.hpp"
  38 #include "memory/filemap.hpp"
  39 #include "oops/oop.inline.hpp"
  40 #include "os_linux.inline.hpp"
  41 #include "os_posix.inline.hpp"
  42 #include "os_share_linux.hpp"
  43 #include "osContainer_linux.hpp"
  44 #include "prims/jniFastGetField.hpp"
  45 #include "prims/jvm_misc.hpp"
  46 #include "runtime/arguments.hpp"
  47 #include "runtime/atomic.hpp"
  48 #include "runtime/extendedPC.hpp"
  49 #include "runtime/globals.hpp"
  50 #include "runtime/interfaceSupport.inline.hpp"
  51 #include "runtime/init.hpp"
  52 #include "runtime/java.hpp"
  53 #include "runtime/javaCalls.hpp"
  54 #include "runtime/mutexLocker.hpp"
  55 #include "runtime/objectMonitor.hpp"
  56 #include "runtime/orderAccess.hpp"
  57 #include "runtime/osThread.hpp"
  58 #include "runtime/perfMemory.hpp"
  59 #include "runtime/sharedRuntime.hpp"
  60 #include "runtime/statSampler.hpp"
  61 #include "runtime/stubRoutines.hpp"
  62 #include "runtime/thread.inline.hpp"
  63 #include "runtime/threadCritical.hpp"
  64 #include "runtime/threadSMR.hpp"
  65 #include "runtime/timer.hpp"
  66 #include "runtime/vm_version.hpp"
  67 #include "semaphore_posix.hpp"
  68 #include "services/attachListener.hpp"
  69 #include "services/memTracker.hpp"
  70 #include "services/runtimeService.hpp"
  71 #include "utilities/align.hpp"
  72 #include "utilities/decoder.hpp"
  73 #include "utilities/defaultStream.hpp"
  74 #include "utilities/events.hpp"
  75 #include "utilities/elfFile.hpp"
  76 #include "utilities/growableArray.hpp"
  77 #include "utilities/macros.hpp"
  78 #include "utilities/vmError.hpp"
  79 
  80 // put OS-includes here
  81 # include <sys/types.h>
  82 # include <sys/mman.h>
  83 # include <sys/stat.h>
  84 # include <sys/select.h>
  85 # include <pthread.h>
  86 # include <signal.h>
  87 # include <errno.h>
  88 # include <dlfcn.h>
  89 # include <stdio.h>
  90 # include <unistd.h>
  91 # include <sys/resource.h>
  92 # include <pthread.h>
  93 # include <sys/stat.h>
  94 # include <sys/time.h>
  95 # include <sys/times.h>
  96 # include <sys/utsname.h>
  97 # include <sys/socket.h>
  98 # include <sys/wait.h>
  99 # include <pwd.h>
 100 # include <poll.h>
 101 # include <fcntl.h>
 102 # include <string.h>
 103 # include <syscall.h>
 104 # include <sys/sysinfo.h>
 105 # include <gnu/libc-version.h>
 106 # include <sys/ipc.h>
 107 # include <sys/shm.h>
 108 # include <link.h>
 109 # include <stdint.h>
 110 # include <inttypes.h>
 111 # include <sys/ioctl.h>
 112 
 113 #ifndef _GNU_SOURCE
 114   #define _GNU_SOURCE
 115   #include <sched.h>
 116   #undef _GNU_SOURCE
 117 #else
 118   #include <sched.h>
 119 #endif
 120 
 121 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 122 // getrusage() is prepared to handle the associated failure.
 123 #ifndef RUSAGE_THREAD
 124   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 125 #endif
 126 
 127 #define MAX_PATH    (2 * K)
 128 
 129 #define MAX_SECS 100000000
 130 
 131 // for timer info max values which include all bits
 132 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 133 
 134 enum CoredumpFilterBit {
 135   FILE_BACKED_PVT_BIT = 1 << 2,
 136   FILE_BACKED_SHARED_BIT = 1 << 3,
 137   LARGEPAGES_BIT = 1 << 6,
 138   DAX_SHARED_BIT = 1 << 8
 139 };
 140 
 141 ////////////////////////////////////////////////////////////////////////////////
 142 // global variables
 143 julong os::Linux::_physical_memory = 0;
 144 
 145 address   os::Linux::_initial_thread_stack_bottom = NULL;
 146 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 147 
 148 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 149 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 150 Mutex* os::Linux::_createThread_lock = NULL;
 151 pthread_t os::Linux::_main_thread;
 152 int os::Linux::_page_size = -1;
 153 bool os::Linux::_supports_fast_thread_cpu_time = false;
 154 uint32_t os::Linux::_os_version = 0;
 155 const char * os::Linux::_glibc_version = NULL;
 156 const char * os::Linux::_libpthread_version = NULL;
 157 
 158 static jlong initial_time_count=0;
 159 
 160 static int clock_tics_per_sec = 100;
 161 
 162 // If the VM might have been created on the primordial thread, we need to resolve the
 163 // primordial thread stack bounds and check if the current thread might be the
 164 // primordial thread in places. If we know that the primordial thread is never used,
 165 // such as when the VM was created by one of the standard java launchers, we can
 166 // avoid this
 167 static bool suppress_primordial_thread_resolution = false;
 168 
 169 // For diagnostics to print a message once. see run_periodic_checks
 170 static sigset_t check_signal_done;
 171 static bool check_signals = true;
 172 
 173 // Signal number used to suspend/resume a thread
 174 
 175 // do not use any signal number less than SIGSEGV, see 4355769
 176 static int SR_signum = SIGUSR2;
 177 sigset_t SR_sigset;
 178 
 179 // utility functions
 180 
 181 static int SR_initialize();
 182 
 183 julong os::available_memory() {
 184   return Linux::available_memory();
 185 }
 186 
 187 julong os::Linux::available_memory() {
 188   // values in struct sysinfo are "unsigned long"
 189   struct sysinfo si;
 190   julong avail_mem;
 191 
 192   if (OSContainer::is_containerized()) {
 193     jlong mem_limit, mem_usage;
 194     if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
 195       log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
 196                              mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
 197     }
 198     if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
 199       log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
 200     }
 201     if (mem_limit > 0 && mem_usage > 0 ) {
 202       avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
 203       log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
 204       return avail_mem;
 205     }
 206   }
 207 
 208   sysinfo(&si);
 209   avail_mem = (julong)si.freeram * si.mem_unit;
 210   log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
 211   return avail_mem;
 212 }
 213 
 214 julong os::physical_memory() {
 215   jlong phys_mem = 0;
 216   if (OSContainer::is_containerized()) {
 217     jlong mem_limit;
 218     if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
 219       log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
 220       return mem_limit;
 221     }
 222     log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
 223                             mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
 224   }
 225 
 226   phys_mem = Linux::physical_memory();
 227   log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
 228   return phys_mem;
 229 }
 230 
 231 static uint64_t initial_total_ticks = 0;
 232 static uint64_t initial_steal_ticks = 0;
 233 static bool     has_initial_tick_info = false;
 234 
 235 static void next_line(FILE *f) {
 236   int c;
 237   do {
 238     c = fgetc(f);
 239   } while (c != '\n' && c != EOF);
 240 }
 241 
 242 bool os::Linux::get_tick_information(CPUPerfTicks* pticks, int which_logical_cpu) {
 243   FILE*         fh;
 244   uint64_t      userTicks, niceTicks, systemTicks, idleTicks;
 245   // since at least kernel 2.6 : iowait: time waiting for I/O to complete
 246   // irq: time  servicing interrupts; softirq: time servicing softirqs
 247   uint64_t      iowTicks = 0, irqTicks = 0, sirqTicks= 0;
 248   // steal (since kernel 2.6.11): time spent in other OS when running in a virtualized environment
 249   uint64_t      stealTicks = 0;
 250   // guest (since kernel 2.6.24): time spent running a virtual CPU for guest OS under the
 251   // control of the Linux kernel
 252   uint64_t      guestNiceTicks = 0;
 253   int           logical_cpu = -1;
 254   const int     required_tickinfo_count = (which_logical_cpu == -1) ? 4 : 5;
 255   int           n;
 256 
 257   memset(pticks, 0, sizeof(CPUPerfTicks));
 258 
 259   if ((fh = fopen("/proc/stat", "r")) == NULL) {
 260     return false;
 261   }
 262 
 263   if (which_logical_cpu == -1) {
 264     n = fscanf(fh, "cpu " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
 265             UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
 266             UINT64_FORMAT " " UINT64_FORMAT " ",
 267             &userTicks, &niceTicks, &systemTicks, &idleTicks,
 268             &iowTicks, &irqTicks, &sirqTicks,
 269             &stealTicks, &guestNiceTicks);
 270   } else {
 271     // Move to next line
 272     next_line(fh);
 273 
 274     // find the line for requested cpu faster to just iterate linefeeds?
 275     for (int i = 0; i < which_logical_cpu; i++) {
 276       next_line(fh);
 277     }
 278 
 279     n = fscanf(fh, "cpu%u " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
 280                UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
 281                UINT64_FORMAT " " UINT64_FORMAT " ",
 282                &logical_cpu, &userTicks, &niceTicks,
 283                &systemTicks, &idleTicks, &iowTicks, &irqTicks, &sirqTicks,
 284                &stealTicks, &guestNiceTicks);
 285   }
 286 
 287   fclose(fh);
 288   if (n < required_tickinfo_count || logical_cpu != which_logical_cpu) {
 289     return false;
 290   }
 291   pticks->used       = userTicks + niceTicks;
 292   pticks->usedKernel = systemTicks + irqTicks + sirqTicks;
 293   pticks->total      = userTicks + niceTicks + systemTicks + idleTicks +
 294                        iowTicks + irqTicks + sirqTicks + stealTicks + guestNiceTicks;
 295 
 296   if (n > required_tickinfo_count + 3) {
 297     pticks->steal = stealTicks;
 298     pticks->has_steal_ticks = true;
 299   } else {
 300     pticks->steal = 0;
 301     pticks->has_steal_ticks = false;
 302   }
 303 
 304   return true;
 305 }
 306 
 307 // Return true if user is running as root.
 308 
 309 bool os::have_special_privileges() {
 310   static bool init = false;
 311   static bool privileges = false;
 312   if (!init) {
 313     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 314     init = true;
 315   }
 316   return privileges;
 317 }
 318 
 319 
 320 #ifndef SYS_gettid
 321 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 322   #ifdef __ia64__
 323     #define SYS_gettid 1105
 324   #else
 325     #ifdef __i386__
 326       #define SYS_gettid 224
 327     #else
 328       #ifdef __amd64__
 329         #define SYS_gettid 186
 330       #else
 331         #ifdef __sparc__
 332           #define SYS_gettid 143
 333         #else
 334           #error define gettid for the arch
 335         #endif
 336       #endif
 337     #endif
 338   #endif
 339 #endif
 340 
 341 
 342 // pid_t gettid()
 343 //
 344 // Returns the kernel thread id of the currently running thread. Kernel
 345 // thread id is used to access /proc.
 346 pid_t os::Linux::gettid() {
 347   int rslt = syscall(SYS_gettid);
 348   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 349   return (pid_t)rslt;
 350 }
 351 
 352 // Most versions of linux have a bug where the number of processors are
 353 // determined by looking at the /proc file system.  In a chroot environment,
 354 // the system call returns 1.
 355 static bool unsafe_chroot_detected = false;
 356 static const char *unstable_chroot_error = "/proc file system not found.\n"
 357                      "Java may be unstable running multithreaded in a chroot "
 358                      "environment on Linux when /proc filesystem is not mounted.";
 359 
 360 void os::Linux::initialize_system_info() {
 361   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 362   if (processor_count() == 1) {
 363     pid_t pid = os::Linux::gettid();
 364     char fname[32];
 365     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 366     FILE *fp = fopen(fname, "r");
 367     if (fp == NULL) {
 368       unsafe_chroot_detected = true;
 369     } else {
 370       fclose(fp);
 371     }
 372   }
 373   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 374   assert(processor_count() > 0, "linux error");
 375 }
 376 
 377 void os::init_system_properties_values() {
 378   // The next steps are taken in the product version:
 379   //
 380   // Obtain the JAVA_HOME value from the location of libjvm.so.
 381   // This library should be located at:
 382   // <JAVA_HOME>/lib/{client|server}/libjvm.so.
 383   //
 384   // If "/jre/lib/" appears at the right place in the path, then we
 385   // assume libjvm.so is installed in a JDK and we use this path.
 386   //
 387   // Otherwise exit with message: "Could not create the Java virtual machine."
 388   //
 389   // The following extra steps are taken in the debugging version:
 390   //
 391   // If "/jre/lib/" does NOT appear at the right place in the path
 392   // instead of exit check for $JAVA_HOME environment variable.
 393   //
 394   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 395   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 396   // it looks like libjvm.so is installed there
 397   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 398   //
 399   // Otherwise exit.
 400   //
 401   // Important note: if the location of libjvm.so changes this
 402   // code needs to be changed accordingly.
 403 
 404   // See ld(1):
 405   //      The linker uses the following search paths to locate required
 406   //      shared libraries:
 407   //        1: ...
 408   //        ...
 409   //        7: The default directories, normally /lib and /usr/lib.
 410 #ifndef OVERRIDE_LIBPATH
 411   #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
 412     #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 413   #else
 414     #define DEFAULT_LIBPATH "/lib:/usr/lib"
 415   #endif
 416 #else
 417   #define DEFAULT_LIBPATH OVERRIDE_LIBPATH
 418 #endif
 419 
 420 // Base path of extensions installed on the system.
 421 #define SYS_EXT_DIR     "/usr/java/packages"
 422 #define EXTENSIONS_DIR  "/lib/ext"
 423 
 424   // Buffer that fits several sprintfs.
 425   // Note that the space for the colon and the trailing null are provided
 426   // by the nulls included by the sizeof operator.
 427   const size_t bufsize =
 428     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 429          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 430   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 431 
 432   // sysclasspath, java_home, dll_dir
 433   {
 434     char *pslash;
 435     os::jvm_path(buf, bufsize);
 436 
 437     // Found the full path to libjvm.so.
 438     // Now cut the path to <java_home>/jre if we can.
 439     pslash = strrchr(buf, '/');
 440     if (pslash != NULL) {
 441       *pslash = '\0';            // Get rid of /libjvm.so.
 442     }
 443     pslash = strrchr(buf, '/');
 444     if (pslash != NULL) {
 445       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 446     }
 447     Arguments::set_dll_dir(buf);
 448 
 449     if (pslash != NULL) {
 450       pslash = strrchr(buf, '/');
 451       if (pslash != NULL) {
 452         *pslash = '\0';        // Get rid of /lib.
 453       }
 454     }
 455     Arguments::set_java_home(buf);
 456     if (!set_boot_path('/', ':')) {
 457       vm_exit_during_initialization("Failed setting boot class path.", NULL);
 458     }
 459   }
 460 
 461   // Where to look for native libraries.
 462   //
 463   // Note: Due to a legacy implementation, most of the library path
 464   // is set in the launcher. This was to accomodate linking restrictions
 465   // on legacy Linux implementations (which are no longer supported).
 466   // Eventually, all the library path setting will be done here.
 467   //
 468   // However, to prevent the proliferation of improperly built native
 469   // libraries, the new path component /usr/java/packages is added here.
 470   // Eventually, all the library path setting will be done here.
 471   {
 472     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 473     // should always exist (until the legacy problem cited above is
 474     // addressed).
 475     const char *v = ::getenv("LD_LIBRARY_PATH");
 476     const char *v_colon = ":";
 477     if (v == NULL) { v = ""; v_colon = ""; }
 478     // That's +1 for the colon and +1 for the trailing '\0'.
 479     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 480                                                      strlen(v) + 1 +
 481                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
 482                                                      mtInternal);
 483     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
 484     Arguments::set_library_path(ld_library_path);
 485     FREE_C_HEAP_ARRAY(char, ld_library_path);
 486   }
 487 
 488   // Extensions directories.
 489   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 490   Arguments::set_ext_dirs(buf);
 491 
 492   FREE_C_HEAP_ARRAY(char, buf);
 493 
 494 #undef DEFAULT_LIBPATH
 495 #undef SYS_EXT_DIR
 496 #undef EXTENSIONS_DIR
 497 }
 498 
 499 ////////////////////////////////////////////////////////////////////////////////
 500 // breakpoint support
 501 
 502 void os::breakpoint() {
 503   BREAKPOINT;
 504 }
 505 
 506 extern "C" void breakpoint() {
 507   // use debugger to set breakpoint here
 508 }
 509 
 510 ////////////////////////////////////////////////////////////////////////////////
 511 // signal support
 512 
 513 debug_only(static bool signal_sets_initialized = false);
 514 static sigset_t unblocked_sigs, vm_sigs;
 515 
 516 void os::Linux::signal_sets_init() {
 517   // Should also have an assertion stating we are still single-threaded.
 518   assert(!signal_sets_initialized, "Already initialized");
 519   // Fill in signals that are necessarily unblocked for all threads in
 520   // the VM. Currently, we unblock the following signals:
 521   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 522   //                         by -Xrs (=ReduceSignalUsage));
 523   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 524   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 525   // the dispositions or masks wrt these signals.
 526   // Programs embedding the VM that want to use the above signals for their
 527   // own purposes must, at this time, use the "-Xrs" option to prevent
 528   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 529   // (See bug 4345157, and other related bugs).
 530   // In reality, though, unblocking these signals is really a nop, since
 531   // these signals are not blocked by default.
 532   sigemptyset(&unblocked_sigs);
 533   sigaddset(&unblocked_sigs, SIGILL);
 534   sigaddset(&unblocked_sigs, SIGSEGV);
 535   sigaddset(&unblocked_sigs, SIGBUS);
 536   sigaddset(&unblocked_sigs, SIGFPE);
 537 #if defined(PPC64)
 538   sigaddset(&unblocked_sigs, SIGTRAP);
 539 #endif
 540   sigaddset(&unblocked_sigs, SR_signum);
 541 
 542   if (!ReduceSignalUsage) {
 543     if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 544       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 545     }
 546     if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 547       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 548     }
 549     if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 550       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 551     }
 552   }
 553   // Fill in signals that are blocked by all but the VM thread.
 554   sigemptyset(&vm_sigs);
 555   if (!ReduceSignalUsage) {
 556     sigaddset(&vm_sigs, BREAK_SIGNAL);
 557   }
 558   debug_only(signal_sets_initialized = true);
 559 
 560 }
 561 
 562 // These are signals that are unblocked while a thread is running Java.
 563 // (For some reason, they get blocked by default.)
 564 sigset_t* os::Linux::unblocked_signals() {
 565   assert(signal_sets_initialized, "Not initialized");
 566   return &unblocked_sigs;
 567 }
 568 
 569 // These are the signals that are blocked while a (non-VM) thread is
 570 // running Java. Only the VM thread handles these signals.
 571 sigset_t* os::Linux::vm_signals() {
 572   assert(signal_sets_initialized, "Not initialized");
 573   return &vm_sigs;
 574 }
 575 
 576 void os::Linux::hotspot_sigmask(Thread* thread) {
 577 
 578   //Save caller's signal mask before setting VM signal mask
 579   sigset_t caller_sigmask;
 580   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 581 
 582   OSThread* osthread = thread->osthread();
 583   osthread->set_caller_sigmask(caller_sigmask);
 584 
 585   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 586 
 587   if (!ReduceSignalUsage) {
 588     if (thread->is_VM_thread()) {
 589       // Only the VM thread handles BREAK_SIGNAL ...
 590       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 591     } else {
 592       // ... all other threads block BREAK_SIGNAL
 593       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 594     }
 595   }
 596 }
 597 
 598 //////////////////////////////////////////////////////////////////////////////
 599 // detecting pthread library
 600 
 601 void os::Linux::libpthread_init() {
 602   // Save glibc and pthread version strings.
 603 #if !defined(_CS_GNU_LIBC_VERSION) || \
 604     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 605   #error "glibc too old (< 2.3.2)"
 606 #endif
 607 
 608   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 609   assert(n > 0, "cannot retrieve glibc version");
 610   char *str = (char *)malloc(n, mtInternal);
 611   confstr(_CS_GNU_LIBC_VERSION, str, n);
 612   os::Linux::set_glibc_version(str);
 613 
 614   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 615   assert(n > 0, "cannot retrieve pthread version");
 616   str = (char *)malloc(n, mtInternal);
 617   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 618   os::Linux::set_libpthread_version(str);
 619 }
 620 
 621 /////////////////////////////////////////////////////////////////////////////
 622 // thread stack expansion
 623 
 624 // os::Linux::manually_expand_stack() takes care of expanding the thread
 625 // stack. Note that this is normally not needed: pthread stacks allocate
 626 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 627 // committed. Therefore it is not necessary to expand the stack manually.
 628 //
 629 // Manually expanding the stack was historically needed on LinuxThreads
 630 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 631 // it is kept to deal with very rare corner cases:
 632 //
 633 // For one, user may run the VM on an own implementation of threads
 634 // whose stacks are - like the old LinuxThreads - implemented using
 635 // mmap(MAP_GROWSDOWN).
 636 //
 637 // Also, this coding may be needed if the VM is running on the primordial
 638 // thread. Normally we avoid running on the primordial thread; however,
 639 // user may still invoke the VM on the primordial thread.
 640 //
 641 // The following historical comment describes the details about running
 642 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 643 
 644 
 645 // Force Linux kernel to expand current thread stack. If "bottom" is close
 646 // to the stack guard, caller should block all signals.
 647 //
 648 // MAP_GROWSDOWN:
 649 //   A special mmap() flag that is used to implement thread stacks. It tells
 650 //   kernel that the memory region should extend downwards when needed. This
 651 //   allows early versions of LinuxThreads to only mmap the first few pages
 652 //   when creating a new thread. Linux kernel will automatically expand thread
 653 //   stack as needed (on page faults).
 654 //
 655 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 656 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 657 //   region, it's hard to tell if the fault is due to a legitimate stack
 658 //   access or because of reading/writing non-exist memory (e.g. buffer
 659 //   overrun). As a rule, if the fault happens below current stack pointer,
 660 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 661 //   application (see Linux kernel fault.c).
 662 //
 663 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 664 //   stack overflow detection.
 665 //
 666 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 667 //   not use MAP_GROWSDOWN.
 668 //
 669 // To get around the problem and allow stack banging on Linux, we need to
 670 // manually expand thread stack after receiving the SIGSEGV.
 671 //
 672 // There are two ways to expand thread stack to address "bottom", we used
 673 // both of them in JVM before 1.5:
 674 //   1. adjust stack pointer first so that it is below "bottom", and then
 675 //      touch "bottom"
 676 //   2. mmap() the page in question
 677 //
 678 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 679 // if current sp is already near the lower end of page 101, and we need to
 680 // call mmap() to map page 100, it is possible that part of the mmap() frame
 681 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 682 // That will destroy the mmap() frame and cause VM to crash.
 683 //
 684 // The following code works by adjusting sp first, then accessing the "bottom"
 685 // page to force a page fault. Linux kernel will then automatically expand the
 686 // stack mapping.
 687 //
 688 // _expand_stack_to() assumes its frame size is less than page size, which
 689 // should always be true if the function is not inlined.
 690 
 691 static void NOINLINE _expand_stack_to(address bottom) {
 692   address sp;
 693   size_t size;
 694   volatile char *p;
 695 
 696   // Adjust bottom to point to the largest address within the same page, it
 697   // gives us a one-page buffer if alloca() allocates slightly more memory.
 698   bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
 699   bottom += os::Linux::page_size() - 1;
 700 
 701   // sp might be slightly above current stack pointer; if that's the case, we
 702   // will alloca() a little more space than necessary, which is OK. Don't use
 703   // os::current_stack_pointer(), as its result can be slightly below current
 704   // stack pointer, causing us to not alloca enough to reach "bottom".
 705   sp = (address)&sp;
 706 
 707   if (sp > bottom) {
 708     size = sp - bottom;
 709     p = (volatile char *)alloca(size);
 710     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 711     p[0] = '\0';
 712   }
 713 }
 714 
 715 void os::Linux::expand_stack_to(address bottom) {
 716   _expand_stack_to(bottom);
 717 }
 718 
 719 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 720   assert(t!=NULL, "just checking");
 721   assert(t->osthread()->expanding_stack(), "expand should be set");
 722   assert(t->stack_base() != NULL, "stack_base was not initialized");
 723 
 724   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 725     sigset_t mask_all, old_sigset;
 726     sigfillset(&mask_all);
 727     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 728     _expand_stack_to(addr);
 729     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 730     return true;
 731   }
 732   return false;
 733 }
 734 
 735 //////////////////////////////////////////////////////////////////////////////
 736 // create new thread
 737 
 738 // Thread start routine for all newly created threads
 739 static void *thread_native_entry(Thread *thread) {
 740 
 741   thread->record_stack_base_and_size();
 742 
 743   // Try to randomize the cache line index of hot stack frames.
 744   // This helps when threads of the same stack traces evict each other's
 745   // cache lines. The threads can be either from the same JVM instance, or
 746   // from different JVM instances. The benefit is especially true for
 747   // processors with hyperthreading technology.
 748   static int counter = 0;
 749   int pid = os::current_process_id();
 750   alloca(((pid ^ counter++) & 7) * 128);
 751 
 752   thread->initialize_thread_current();
 753 
 754   OSThread* osthread = thread->osthread();
 755   Monitor* sync = osthread->startThread_lock();
 756 
 757   osthread->set_thread_id(os::current_thread_id());
 758 
 759   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 760     os::current_thread_id(), (uintx) pthread_self());
 761 
 762   if (UseNUMA) {
 763     int lgrp_id = os::numa_get_group_id();
 764     if (lgrp_id != -1) {
 765       thread->set_lgrp_id(lgrp_id);
 766     }
 767   }
 768   // initialize signal mask for this thread
 769   os::Linux::hotspot_sigmask(thread);
 770 
 771   // initialize floating point control register
 772   os::Linux::init_thread_fpu_state();
 773 
 774   // handshaking with parent thread
 775   {
 776     MutexLocker ml(sync, Mutex::_no_safepoint_check_flag);
 777 
 778     // notify parent thread
 779     osthread->set_state(INITIALIZED);
 780     sync->notify_all();
 781 
 782     // wait until os::start_thread()
 783     while (osthread->get_state() == INITIALIZED) {
 784       sync->wait_without_safepoint_check();
 785     }
 786   }
 787 
 788   assert(osthread->pthread_id() != 0, "pthread_id was not set as expected");
 789 
 790   // call one more level start routine
 791   thread->call_run();
 792 
 793   // Note: at this point the thread object may already have deleted itself.
 794   // Prevent dereferencing it from here on out.
 795   thread = NULL;
 796 
 797   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 798     os::current_thread_id(), (uintx) pthread_self());
 799 
 800   return 0;
 801 }
 802 
 803 bool os::create_thread(Thread* thread, ThreadType thr_type,
 804                        size_t req_stack_size) {
 805   assert(thread->osthread() == NULL, "caller responsible");
 806 
 807   // Allocate the OSThread object
 808   OSThread* osthread = new OSThread(NULL, NULL);
 809   if (osthread == NULL) {
 810     return false;
 811   }
 812 
 813   // set the correct thread state
 814   osthread->set_thread_type(thr_type);
 815 
 816   // Initial state is ALLOCATED but not INITIALIZED
 817   osthread->set_state(ALLOCATED);
 818 
 819   thread->set_osthread(osthread);
 820 
 821   // init thread attributes
 822   pthread_attr_t attr;
 823   pthread_attr_init(&attr);
 824   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 825 
 826   // Calculate stack size if it's not specified by caller.
 827   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 828   // In the Linux NPTL pthread implementation the guard size mechanism
 829   // is not implemented properly. The posix standard requires adding
 830   // the size of the guard pages to the stack size, instead Linux
 831   // takes the space out of 'stacksize'. Thus we adapt the requested
 832   // stack_size by the size of the guard pages to mimick proper
 833   // behaviour. However, be careful not to end up with a size
 834   // of zero due to overflow. Don't add the guard page in that case.
 835   size_t guard_size = os::Linux::default_guard_size(thr_type);
 836   if (stack_size <= SIZE_MAX - guard_size) {
 837     stack_size += guard_size;
 838   }
 839   assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
 840 
 841   int status = pthread_attr_setstacksize(&attr, stack_size);
 842   assert_status(status == 0, status, "pthread_attr_setstacksize");
 843 
 844   // Configure glibc guard page.
 845   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 846 
 847   ThreadState state;
 848 
 849   {
 850     pthread_t tid;
 851     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 852 
 853     char buf[64];
 854     if (ret == 0) {
 855       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 856         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 857     } else {
 858       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 859         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 860       // Log some OS information which might explain why creating the thread failed.
 861       log_info(os, thread)("Number of threads approx. running in the VM: %d", Threads::number_of_threads());
 862       LogStream st(Log(os, thread)::info());
 863       os::Posix::print_rlimit_info(&st);
 864       os::print_memory_info(&st);
 865       os::Linux::print_proc_sys_info(&st);
 866       os::Linux::print_container_info(&st);
 867     }
 868 
 869     pthread_attr_destroy(&attr);
 870 
 871     if (ret != 0) {
 872       // Need to clean up stuff we've allocated so far
 873       thread->set_osthread(NULL);
 874       delete osthread;
 875       return false;
 876     }
 877 
 878     // Store pthread info into the OSThread
 879     osthread->set_pthread_id(tid);
 880 
 881     // Wait until child thread is either initialized or aborted
 882     {
 883       Monitor* sync_with_child = osthread->startThread_lock();
 884       MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 885       while ((state = osthread->get_state()) == ALLOCATED) {
 886         sync_with_child->wait_without_safepoint_check();
 887       }
 888     }
 889   }
 890 
 891   // Aborted due to thread limit being reached
 892   if (state == ZOMBIE) {
 893     thread->set_osthread(NULL);
 894     delete osthread;
 895     return false;
 896   }
 897 
 898   // The thread is returned suspended (in state INITIALIZED),
 899   // and is started higher up in the call chain
 900   assert(state == INITIALIZED, "race condition");
 901   return true;
 902 }
 903 
 904 /////////////////////////////////////////////////////////////////////////////
 905 // attach existing thread
 906 
 907 // bootstrap the main thread
 908 bool os::create_main_thread(JavaThread* thread) {
 909   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 910   return create_attached_thread(thread);
 911 }
 912 
 913 bool os::create_attached_thread(JavaThread* thread) {
 914 #ifdef ASSERT
 915   thread->verify_not_published();
 916 #endif
 917 
 918   // Allocate the OSThread object
 919   OSThread* osthread = new OSThread(NULL, NULL);
 920 
 921   if (osthread == NULL) {
 922     return false;
 923   }
 924 
 925   // Store pthread info into the OSThread
 926   osthread->set_thread_id(os::Linux::gettid());
 927   osthread->set_pthread_id(::pthread_self());
 928 
 929   // initialize floating point control register
 930   os::Linux::init_thread_fpu_state();
 931 
 932   // Initial thread state is RUNNABLE
 933   osthread->set_state(RUNNABLE);
 934 
 935   thread->set_osthread(osthread);
 936 
 937   if (UseNUMA) {
 938     int lgrp_id = os::numa_get_group_id();
 939     if (lgrp_id != -1) {
 940       thread->set_lgrp_id(lgrp_id);
 941     }
 942   }
 943 
 944   if (os::is_primordial_thread()) {
 945     // If current thread is primordial thread, its stack is mapped on demand,
 946     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 947     // the entire stack region to avoid SEGV in stack banging.
 948     // It is also useful to get around the heap-stack-gap problem on SuSE
 949     // kernel (see 4821821 for details). We first expand stack to the top
 950     // of yellow zone, then enable stack yellow zone (order is significant,
 951     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 952     // is no gap between the last two virtual memory regions.
 953 
 954     JavaThread *jt = (JavaThread *)thread;
 955     address addr = jt->stack_reserved_zone_base();
 956     assert(addr != NULL, "initialization problem?");
 957     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 958 
 959     osthread->set_expanding_stack();
 960     os::Linux::manually_expand_stack(jt, addr);
 961     osthread->clear_expanding_stack();
 962   }
 963 
 964   // initialize signal mask for this thread
 965   // and save the caller's signal mask
 966   os::Linux::hotspot_sigmask(thread);
 967 
 968   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 969     os::current_thread_id(), (uintx) pthread_self());
 970 
 971   return true;
 972 }
 973 
 974 void os::pd_start_thread(Thread* thread) {
 975   OSThread * osthread = thread->osthread();
 976   assert(osthread->get_state() != INITIALIZED, "just checking");
 977   Monitor* sync_with_child = osthread->startThread_lock();
 978   MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 979   sync_with_child->notify();
 980 }
 981 
 982 // Free Linux resources related to the OSThread
 983 void os::free_thread(OSThread* osthread) {
 984   assert(osthread != NULL, "osthread not set");
 985 
 986   // We are told to free resources of the argument thread,
 987   // but we can only really operate on the current thread.
 988   assert(Thread::current()->osthread() == osthread,
 989          "os::free_thread but not current thread");
 990 
 991 #ifdef ASSERT
 992   sigset_t current;
 993   sigemptyset(&current);
 994   pthread_sigmask(SIG_SETMASK, NULL, &current);
 995   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 996 #endif
 997 
 998   // Restore caller's signal mask
 999   sigset_t sigmask = osthread->caller_sigmask();
1000   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1001 
1002   delete osthread;
1003 }
1004 
1005 //////////////////////////////////////////////////////////////////////////////
1006 // primordial thread
1007 
1008 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1009 bool os::is_primordial_thread(void) {
1010   if (suppress_primordial_thread_resolution) {
1011     return false;
1012   }
1013   char dummy;
1014   // If called before init complete, thread stack bottom will be null.
1015   // Can be called if fatal error occurs before initialization.
1016   if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1017   assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1018          os::Linux::initial_thread_stack_size()   != 0,
1019          "os::init did not locate primordial thread's stack region");
1020   if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1021       (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1022                         os::Linux::initial_thread_stack_size()) {
1023     return true;
1024   } else {
1025     return false;
1026   }
1027 }
1028 
1029 // Find the virtual memory area that contains addr
1030 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1031   FILE *fp = fopen("/proc/self/maps", "r");
1032   if (fp) {
1033     address low, high;
1034     while (!feof(fp)) {
1035       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1036         if (low <= addr && addr < high) {
1037           if (vma_low)  *vma_low  = low;
1038           if (vma_high) *vma_high = high;
1039           fclose(fp);
1040           return true;
1041         }
1042       }
1043       for (;;) {
1044         int ch = fgetc(fp);
1045         if (ch == EOF || ch == (int)'\n') break;
1046       }
1047     }
1048     fclose(fp);
1049   }
1050   return false;
1051 }
1052 
1053 // Locate primordial thread stack. This special handling of primordial thread stack
1054 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1055 // bogus value for the primordial process thread. While the launcher has created
1056 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1057 // JNI invocation API from a primordial thread.
1058 void os::Linux::capture_initial_stack(size_t max_size) {
1059 
1060   // max_size is either 0 (which means accept OS default for thread stacks) or
1061   // a user-specified value known to be at least the minimum needed. If we
1062   // are actually on the primordial thread we can make it appear that we have a
1063   // smaller max_size stack by inserting the guard pages at that location. But we
1064   // cannot do anything to emulate a larger stack than what has been provided by
1065   // the OS or threading library. In fact if we try to use a stack greater than
1066   // what is set by rlimit then we will crash the hosting process.
1067 
1068   // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1069   // If this is "unlimited" then it will be a huge value.
1070   struct rlimit rlim;
1071   getrlimit(RLIMIT_STACK, &rlim);
1072   size_t stack_size = rlim.rlim_cur;
1073 
1074   // 6308388: a bug in ld.so will relocate its own .data section to the
1075   //   lower end of primordial stack; reduce ulimit -s value a little bit
1076   //   so we won't install guard page on ld.so's data section.
1077   //   But ensure we don't underflow the stack size - allow 1 page spare
1078   if (stack_size >= (size_t)(3 * page_size())) {
1079     stack_size -= 2 * page_size();
1080   }
1081 
1082   // Try to figure out where the stack base (top) is. This is harder.
1083   //
1084   // When an application is started, glibc saves the initial stack pointer in
1085   // a global variable "__libc_stack_end", which is then used by system
1086   // libraries. __libc_stack_end should be pretty close to stack top. The
1087   // variable is available since the very early days. However, because it is
1088   // a private interface, it could disappear in the future.
1089   //
1090   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1091   // to __libc_stack_end, it is very close to stack top, but isn't the real
1092   // stack top. Note that /proc may not exist if VM is running as a chroot
1093   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1094   // /proc/<pid>/stat could change in the future (though unlikely).
1095   //
1096   // We try __libc_stack_end first. If that doesn't work, look for
1097   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1098   // as a hint, which should work well in most cases.
1099 
1100   uintptr_t stack_start;
1101 
1102   // try __libc_stack_end first
1103   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1104   if (p && *p) {
1105     stack_start = *p;
1106   } else {
1107     // see if we can get the start_stack field from /proc/self/stat
1108     FILE *fp;
1109     int pid;
1110     char state;
1111     int ppid;
1112     int pgrp;
1113     int session;
1114     int nr;
1115     int tpgrp;
1116     unsigned long flags;
1117     unsigned long minflt;
1118     unsigned long cminflt;
1119     unsigned long majflt;
1120     unsigned long cmajflt;
1121     unsigned long utime;
1122     unsigned long stime;
1123     long cutime;
1124     long cstime;
1125     long prio;
1126     long nice;
1127     long junk;
1128     long it_real;
1129     uintptr_t start;
1130     uintptr_t vsize;
1131     intptr_t rss;
1132     uintptr_t rsslim;
1133     uintptr_t scodes;
1134     uintptr_t ecode;
1135     int i;
1136 
1137     // Figure what the primordial thread stack base is. Code is inspired
1138     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1139     // followed by command name surrounded by parentheses, state, etc.
1140     char stat[2048];
1141     int statlen;
1142 
1143     fp = fopen("/proc/self/stat", "r");
1144     if (fp) {
1145       statlen = fread(stat, 1, 2047, fp);
1146       stat[statlen] = '\0';
1147       fclose(fp);
1148 
1149       // Skip pid and the command string. Note that we could be dealing with
1150       // weird command names, e.g. user could decide to rename java launcher
1151       // to "java 1.4.2 :)", then the stat file would look like
1152       //                1234 (java 1.4.2 :)) R ... ...
1153       // We don't really need to know the command string, just find the last
1154       // occurrence of ")" and then start parsing from there. See bug 4726580.
1155       char * s = strrchr(stat, ')');
1156 
1157       i = 0;
1158       if (s) {
1159         // Skip blank chars
1160         do { s++; } while (s && isspace(*s));
1161 
1162 #define _UFM UINTX_FORMAT
1163 #define _DFM INTX_FORMAT
1164 
1165         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1166         //              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
1167         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,
1168                    &state,          // 3  %c
1169                    &ppid,           // 4  %d
1170                    &pgrp,           // 5  %d
1171                    &session,        // 6  %d
1172                    &nr,             // 7  %d
1173                    &tpgrp,          // 8  %d
1174                    &flags,          // 9  %lu
1175                    &minflt,         // 10 %lu
1176                    &cminflt,        // 11 %lu
1177                    &majflt,         // 12 %lu
1178                    &cmajflt,        // 13 %lu
1179                    &utime,          // 14 %lu
1180                    &stime,          // 15 %lu
1181                    &cutime,         // 16 %ld
1182                    &cstime,         // 17 %ld
1183                    &prio,           // 18 %ld
1184                    &nice,           // 19 %ld
1185                    &junk,           // 20 %ld
1186                    &it_real,        // 21 %ld
1187                    &start,          // 22 UINTX_FORMAT
1188                    &vsize,          // 23 UINTX_FORMAT
1189                    &rss,            // 24 INTX_FORMAT
1190                    &rsslim,         // 25 UINTX_FORMAT
1191                    &scodes,         // 26 UINTX_FORMAT
1192                    &ecode,          // 27 UINTX_FORMAT
1193                    &stack_start);   // 28 UINTX_FORMAT
1194       }
1195 
1196 #undef _UFM
1197 #undef _DFM
1198 
1199       if (i != 28 - 2) {
1200         assert(false, "Bad conversion from /proc/self/stat");
1201         // product mode - assume we are the primordial thread, good luck in the
1202         // embedded case.
1203         warning("Can't detect primordial thread stack location - bad conversion");
1204         stack_start = (uintptr_t) &rlim;
1205       }
1206     } else {
1207       // For some reason we can't open /proc/self/stat (for example, running on
1208       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1209       // most cases, so don't abort:
1210       warning("Can't detect primordial thread stack location - no /proc/self/stat");
1211       stack_start = (uintptr_t) &rlim;
1212     }
1213   }
1214 
1215   // Now we have a pointer (stack_start) very close to the stack top, the
1216   // next thing to do is to figure out the exact location of stack top. We
1217   // can find out the virtual memory area that contains stack_start by
1218   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1219   // and its upper limit is the real stack top. (again, this would fail if
1220   // running inside chroot, because /proc may not exist.)
1221 
1222   uintptr_t stack_top;
1223   address low, high;
1224   if (find_vma((address)stack_start, &low, &high)) {
1225     // success, "high" is the true stack top. (ignore "low", because initial
1226     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1227     stack_top = (uintptr_t)high;
1228   } else {
1229     // failed, likely because /proc/self/maps does not exist
1230     warning("Can't detect primordial thread stack location - find_vma failed");
1231     // best effort: stack_start is normally within a few pages below the real
1232     // stack top, use it as stack top, and reduce stack size so we won't put
1233     // guard page outside stack.
1234     stack_top = stack_start;
1235     stack_size -= 16 * page_size();
1236   }
1237 
1238   // stack_top could be partially down the page so align it
1239   stack_top = align_up(stack_top, page_size());
1240 
1241   // Allowed stack value is minimum of max_size and what we derived from rlimit
1242   if (max_size > 0) {
1243     _initial_thread_stack_size = MIN2(max_size, stack_size);
1244   } else {
1245     // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1246     // clamp it at 8MB as we do on Solaris
1247     _initial_thread_stack_size = MIN2(stack_size, 8*M);
1248   }
1249   _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1250   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1251 
1252   assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1253 
1254   if (log_is_enabled(Info, os, thread)) {
1255     // See if we seem to be on primordial process thread
1256     bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1257                       uintptr_t(&rlim) < stack_top;
1258 
1259     log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1260                          SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1261                          primordial ? "primordial" : "user", max_size / K,  _initial_thread_stack_size / K,
1262                          stack_top, intptr_t(_initial_thread_stack_bottom));
1263   }
1264 }
1265 
1266 ////////////////////////////////////////////////////////////////////////////////
1267 // time support
1268 
1269 #ifndef SUPPORTS_CLOCK_MONOTONIC
1270 #error "Build platform doesn't support clock_gettime and related functionality"
1271 #endif
1272 
1273 // Time since start-up in seconds to a fine granularity.
1274 // Used by VMSelfDestructTimer and the MemProfiler.
1275 double os::elapsedTime() {
1276 
1277   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1278 }
1279 
1280 jlong os::elapsed_counter() {
1281   return javaTimeNanos() - initial_time_count;
1282 }
1283 
1284 jlong os::elapsed_frequency() {
1285   return NANOSECS_PER_SEC; // nanosecond resolution
1286 }
1287 
1288 bool os::supports_vtime() { return true; }
1289 bool os::enable_vtime()   { return false; }
1290 bool os::vtime_enabled()  { return false; }
1291 
1292 double os::elapsedVTime() {
1293   struct rusage usage;
1294   int retval = getrusage(RUSAGE_THREAD, &usage);
1295   if (retval == 0) {
1296     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);
1297   } else {
1298     // better than nothing, but not much
1299     return elapsedTime();
1300   }
1301 }
1302 
1303 jlong os::javaTimeMillis() {
1304   timeval time;
1305   int status = gettimeofday(&time, NULL);
1306   assert(status != -1, "linux error");
1307   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1308 }
1309 
1310 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1311   timeval time;
1312   int status = gettimeofday(&time, NULL);
1313   assert(status != -1, "linux error");
1314   seconds = jlong(time.tv_sec);
1315   nanos = jlong(time.tv_usec) * 1000;
1316 }
1317 
1318 void os::Linux::fast_thread_clock_init() {
1319   if (!UseLinuxPosixThreadCPUClocks) {
1320     return;
1321   }
1322   clockid_t clockid;
1323   struct timespec tp;
1324   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1325       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1326 
1327   // Switch to using fast clocks for thread cpu time if
1328   // the clock_getres() returns 0 error code.
1329   // Note, that some kernels may support the current thread
1330   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1331   // returned by the pthread_getcpuclockid().
1332   // If the fast Posix clocks are supported then the clock_getres()
1333   // must return at least tp.tv_sec == 0 which means a resolution
1334   // better than 1 sec. This is extra check for reliability.
1335 
1336   if (pthread_getcpuclockid_func &&
1337       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1338       os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1339     _supports_fast_thread_cpu_time = true;
1340     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1341   }
1342 }
1343 
1344 jlong os::javaTimeNanos() {
1345   if (os::supports_monotonic_clock()) {
1346     struct timespec tp;
1347     int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1348     assert(status == 0, "gettime error");
1349     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1350     return result;
1351   } else {
1352     timeval time;
1353     int status = gettimeofday(&time, NULL);
1354     assert(status != -1, "linux error");
1355     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1356     return 1000 * usecs;
1357   }
1358 }
1359 
1360 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1361   if (os::supports_monotonic_clock()) {
1362     info_ptr->max_value = ALL_64_BITS;
1363 
1364     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1365     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1366     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1367   } else {
1368     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1369     info_ptr->max_value = ALL_64_BITS;
1370 
1371     // gettimeofday is a real time clock so it skips
1372     info_ptr->may_skip_backward = true;
1373     info_ptr->may_skip_forward = true;
1374   }
1375 
1376   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1377 }
1378 
1379 // Return the real, user, and system times in seconds from an
1380 // arbitrary fixed point in the past.
1381 bool os::getTimesSecs(double* process_real_time,
1382                       double* process_user_time,
1383                       double* process_system_time) {
1384   struct tms ticks;
1385   clock_t real_ticks = times(&ticks);
1386 
1387   if (real_ticks == (clock_t) (-1)) {
1388     return false;
1389   } else {
1390     double ticks_per_second = (double) clock_tics_per_sec;
1391     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1392     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1393     *process_real_time = ((double) real_ticks) / ticks_per_second;
1394 
1395     return true;
1396   }
1397 }
1398 
1399 
1400 char * os::local_time_string(char *buf, size_t buflen) {
1401   struct tm t;
1402   time_t long_time;
1403   time(&long_time);
1404   localtime_r(&long_time, &t);
1405   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1406                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1407                t.tm_hour, t.tm_min, t.tm_sec);
1408   return buf;
1409 }
1410 
1411 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1412   return localtime_r(clock, res);
1413 }
1414 
1415 ////////////////////////////////////////////////////////////////////////////////
1416 // runtime exit support
1417 
1418 // Note: os::shutdown() might be called very early during initialization, or
1419 // called from signal handler. Before adding something to os::shutdown(), make
1420 // sure it is async-safe and can handle partially initialized VM.
1421 void os::shutdown() {
1422 
1423   // allow PerfMemory to attempt cleanup of any persistent resources
1424   perfMemory_exit();
1425 
1426   // needs to remove object in file system
1427   AttachListener::abort();
1428 
1429   // flush buffered output, finish log files
1430   ostream_abort();
1431 
1432   // Check for abort hook
1433   abort_hook_t abort_hook = Arguments::abort_hook();
1434   if (abort_hook != NULL) {
1435     abort_hook();
1436   }
1437 
1438 }
1439 
1440 // Note: os::abort() might be called very early during initialization, or
1441 // called from signal handler. Before adding something to os::abort(), make
1442 // sure it is async-safe and can handle partially initialized VM.
1443 void os::abort(bool dump_core, void* siginfo, const void* context) {
1444   os::shutdown();
1445   if (dump_core) {
1446     if (DumpPrivateMappingsInCore) {
1447       ClassLoader::close_jrt_image();
1448     }
1449 #ifndef PRODUCT
1450     fdStream out(defaultStream::output_fd());
1451     out.print_raw("Current thread is ");
1452     char buf[16];
1453     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1454     out.print_raw_cr(buf);
1455     out.print_raw_cr("Dumping core ...");
1456 #endif
1457     ::abort(); // dump core
1458   }
1459 
1460   ::exit(1);
1461 }
1462 
1463 // Die immediately, no exit hook, no abort hook, no cleanup.
1464 void os::die() {
1465   ::abort();
1466 }
1467 
1468 // thread_id is kernel thread id (similar to Solaris LWP id)
1469 intx os::current_thread_id() { return os::Linux::gettid(); }
1470 int os::current_process_id() {
1471   return ::getpid();
1472 }
1473 
1474 // DLL functions
1475 
1476 const char* os::dll_file_extension() { return ".so"; }
1477 
1478 // This must be hard coded because it's the system's temporary
1479 // directory not the java application's temp directory, ala java.io.tmpdir.
1480 const char* os::get_temp_directory() { return "/tmp"; }
1481 
1482 static bool file_exists(const char* filename) {
1483   struct stat statbuf;
1484   if (filename == NULL || strlen(filename) == 0) {
1485     return false;
1486   }
1487   return os::stat(filename, &statbuf) == 0;
1488 }
1489 
1490 // check if addr is inside libjvm.so
1491 bool os::address_is_in_vm(address addr) {
1492   static address libjvm_base_addr;
1493   Dl_info dlinfo;
1494 
1495   if (libjvm_base_addr == NULL) {
1496     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1497       libjvm_base_addr = (address)dlinfo.dli_fbase;
1498     }
1499     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1500   }
1501 
1502   if (dladdr((void *)addr, &dlinfo) != 0) {
1503     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1504   }
1505 
1506   return false;
1507 }
1508 
1509 bool os::dll_address_to_function_name(address addr, char *buf,
1510                                       int buflen, int *offset,
1511                                       bool demangle) {
1512   // buf is not optional, but offset is optional
1513   assert(buf != NULL, "sanity check");
1514 
1515   Dl_info dlinfo;
1516 
1517   if (dladdr((void*)addr, &dlinfo) != 0) {
1518     // see if we have a matching symbol
1519     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1520       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1521         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1522       }
1523       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1524       return true;
1525     }
1526     // no matching symbol so try for just file info
1527     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1528       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1529                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1530         return true;
1531       }
1532     }
1533   }
1534 
1535   buf[0] = '\0';
1536   if (offset != NULL) *offset = -1;
1537   return false;
1538 }
1539 
1540 struct _address_to_library_name {
1541   address addr;          // input : memory address
1542   size_t  buflen;        //         size of fname
1543   char*   fname;         // output: library name
1544   address base;          //         library base addr
1545 };
1546 
1547 static int address_to_library_name_callback(struct dl_phdr_info *info,
1548                                             size_t size, void *data) {
1549   int i;
1550   bool found = false;
1551   address libbase = NULL;
1552   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1553 
1554   // iterate through all loadable segments
1555   for (i = 0; i < info->dlpi_phnum; i++) {
1556     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1557     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1558       // base address of a library is the lowest address of its loaded
1559       // segments.
1560       if (libbase == NULL || libbase > segbase) {
1561         libbase = segbase;
1562       }
1563       // see if 'addr' is within current segment
1564       if (segbase <= d->addr &&
1565           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1566         found = true;
1567       }
1568     }
1569   }
1570 
1571   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1572   // so dll_address_to_library_name() can fall through to use dladdr() which
1573   // can figure out executable name from argv[0].
1574   if (found && info->dlpi_name && info->dlpi_name[0]) {
1575     d->base = libbase;
1576     if (d->fname) {
1577       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1578     }
1579     return 1;
1580   }
1581   return 0;
1582 }
1583 
1584 bool os::dll_address_to_library_name(address addr, char* buf,
1585                                      int buflen, int* offset) {
1586   // buf is not optional, but offset is optional
1587   assert(buf != NULL, "sanity check");
1588 
1589   Dl_info dlinfo;
1590   struct _address_to_library_name data;
1591 
1592   // There is a bug in old glibc dladdr() implementation that it could resolve
1593   // to wrong library name if the .so file has a base address != NULL. Here
1594   // we iterate through the program headers of all loaded libraries to find
1595   // out which library 'addr' really belongs to. This workaround can be
1596   // removed once the minimum requirement for glibc is moved to 2.3.x.
1597   data.addr = addr;
1598   data.fname = buf;
1599   data.buflen = buflen;
1600   data.base = NULL;
1601   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1602 
1603   if (rslt) {
1604     // buf already contains library name
1605     if (offset) *offset = addr - data.base;
1606     return true;
1607   }
1608   if (dladdr((void*)addr, &dlinfo) != 0) {
1609     if (dlinfo.dli_fname != NULL) {
1610       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1611     }
1612     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1613       *offset = addr - (address)dlinfo.dli_fbase;
1614     }
1615     return true;
1616   }
1617 
1618   buf[0] = '\0';
1619   if (offset) *offset = -1;
1620   return false;
1621 }
1622 
1623 // Loads .dll/.so and
1624 // in case of error it checks if .dll/.so was built for the
1625 // same architecture as Hotspot is running on
1626 
1627 
1628 // Remember the stack's state. The Linux dynamic linker will change
1629 // the stack to 'executable' at most once, so we must safepoint only once.
1630 bool os::Linux::_stack_is_executable = false;
1631 
1632 // VM operation that loads a library.  This is necessary if stack protection
1633 // of the Java stacks can be lost during loading the library.  If we
1634 // do not stop the Java threads, they can stack overflow before the stacks
1635 // are protected again.
1636 class VM_LinuxDllLoad: public VM_Operation {
1637  private:
1638   const char *_filename;
1639   char *_ebuf;
1640   int _ebuflen;
1641   void *_lib;
1642  public:
1643   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1644     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1645   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1646   void doit() {
1647     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1648     os::Linux::_stack_is_executable = true;
1649   }
1650   void* loaded_library() { return _lib; }
1651 };
1652 
1653 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1654   void * result = NULL;
1655   bool load_attempted = false;
1656 
1657   // Check whether the library to load might change execution rights
1658   // of the stack. If they are changed, the protection of the stack
1659   // guard pages will be lost. We need a safepoint to fix this.
1660   //
1661   // See Linux man page execstack(8) for more info.
1662   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1663     if (!ElfFile::specifies_noexecstack(filename)) {
1664       if (!is_init_completed()) {
1665         os::Linux::_stack_is_executable = true;
1666         // This is OK - No Java threads have been created yet, and hence no
1667         // stack guard pages to fix.
1668         //
1669         // Dynamic loader will make all stacks executable after
1670         // this function returns, and will not do that again.
1671         assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1672       } else {
1673         warning("You have loaded library %s which might have disabled stack guard. "
1674                 "The VM will try to fix the stack guard now.\n"
1675                 "It's highly recommended that you fix the library with "
1676                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1677                 filename);
1678 
1679         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1680         JavaThread *jt = JavaThread::current();
1681         if (jt->thread_state() != _thread_in_native) {
1682           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1683           // that requires ExecStack. Cannot enter safe point. Let's give up.
1684           warning("Unable to fix stack guard. Giving up.");
1685         } else {
1686           if (!LoadExecStackDllInVMThread) {
1687             // This is for the case where the DLL has an static
1688             // constructor function that executes JNI code. We cannot
1689             // load such DLLs in the VMThread.
1690             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1691           }
1692 
1693           ThreadInVMfromNative tiv(jt);
1694           debug_only(VMNativeEntryWrapper vew;)
1695 
1696           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1697           VMThread::execute(&op);
1698           if (LoadExecStackDllInVMThread) {
1699             result = op.loaded_library();
1700           }
1701           load_attempted = true;
1702         }
1703       }
1704     }
1705   }
1706 
1707   if (!load_attempted) {
1708     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1709   }
1710 
1711   if (result != NULL) {
1712     // Successful loading
1713     return result;
1714   }
1715 
1716   Elf32_Ehdr elf_head;
1717   int diag_msg_max_length=ebuflen-strlen(ebuf);
1718   char* diag_msg_buf=ebuf+strlen(ebuf);
1719 
1720   if (diag_msg_max_length==0) {
1721     // No more space in ebuf for additional diagnostics message
1722     return NULL;
1723   }
1724 
1725 
1726   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1727 
1728   if (file_descriptor < 0) {
1729     // Can't open library, report dlerror() message
1730     return NULL;
1731   }
1732 
1733   bool failed_to_read_elf_head=
1734     (sizeof(elf_head)!=
1735      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1736 
1737   ::close(file_descriptor);
1738   if (failed_to_read_elf_head) {
1739     // file i/o error - report dlerror() msg
1740     return NULL;
1741   }
1742 
1743   typedef struct {
1744     Elf32_Half    code;         // Actual value as defined in elf.h
1745     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1746     unsigned char elf_class;    // 32 or 64 bit
1747     unsigned char endianess;    // MSB or LSB
1748     char*         name;         // String representation
1749   } arch_t;
1750 
1751 #ifndef EM_486
1752   #define EM_486          6               /* Intel 80486 */
1753 #endif
1754 #ifndef EM_AARCH64
1755   #define EM_AARCH64    183               /* ARM AARCH64 */
1756 #endif
1757 
1758   static const arch_t arch_array[]={
1759     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1760     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1761     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1762     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1763     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1764     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1765     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1766     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1767 #if defined(VM_LITTLE_ENDIAN)
1768     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1769     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1770 #else
1771     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1772     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1773 #endif
1774     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1775     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1776     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1777     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1778     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1779     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1780     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1781     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1782   };
1783 
1784 #if  (defined IA32)
1785   static  Elf32_Half running_arch_code=EM_386;
1786 #elif   (defined AMD64) || (defined X32)
1787   static  Elf32_Half running_arch_code=EM_X86_64;
1788 #elif  (defined IA64)
1789   static  Elf32_Half running_arch_code=EM_IA_64;
1790 #elif  (defined __sparc) && (defined _LP64)
1791   static  Elf32_Half running_arch_code=EM_SPARCV9;
1792 #elif  (defined __sparc) && (!defined _LP64)
1793   static  Elf32_Half running_arch_code=EM_SPARC;
1794 #elif  (defined __powerpc64__)
1795   static  Elf32_Half running_arch_code=EM_PPC64;
1796 #elif  (defined __powerpc__)
1797   static  Elf32_Half running_arch_code=EM_PPC;
1798 #elif  (defined AARCH64)
1799   static  Elf32_Half running_arch_code=EM_AARCH64;
1800 #elif  (defined ARM)
1801   static  Elf32_Half running_arch_code=EM_ARM;
1802 #elif  (defined S390)
1803   static  Elf32_Half running_arch_code=EM_S390;
1804 #elif  (defined ALPHA)
1805   static  Elf32_Half running_arch_code=EM_ALPHA;
1806 #elif  (defined MIPSEL)
1807   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1808 #elif  (defined PARISC)
1809   static  Elf32_Half running_arch_code=EM_PARISC;
1810 #elif  (defined MIPS)
1811   static  Elf32_Half running_arch_code=EM_MIPS;
1812 #elif  (defined M68K)
1813   static  Elf32_Half running_arch_code=EM_68K;
1814 #elif  (defined SH)
1815   static  Elf32_Half running_arch_code=EM_SH;
1816 #else
1817     #error Method os::dll_load requires that one of following is defined:\
1818         AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1819 #endif
1820 
1821   // Identify compatability class for VM's architecture and library's architecture
1822   // Obtain string descriptions for architectures
1823 
1824   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1825   int running_arch_index=-1;
1826 
1827   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1828     if (running_arch_code == arch_array[i].code) {
1829       running_arch_index    = i;
1830     }
1831     if (lib_arch.code == arch_array[i].code) {
1832       lib_arch.compat_class = arch_array[i].compat_class;
1833       lib_arch.name         = arch_array[i].name;
1834     }
1835   }
1836 
1837   assert(running_arch_index != -1,
1838          "Didn't find running architecture code (running_arch_code) in arch_array");
1839   if (running_arch_index == -1) {
1840     // Even though running architecture detection failed
1841     // we may still continue with reporting dlerror() message
1842     return NULL;
1843   }
1844 
1845   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1846     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1847     return NULL;
1848   }
1849 
1850 #ifndef S390
1851   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1852     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1853     return NULL;
1854   }
1855 #endif // !S390
1856 
1857   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1858     if (lib_arch.name!=NULL) {
1859       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1860                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1861                  lib_arch.name, arch_array[running_arch_index].name);
1862     } else {
1863       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1864                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1865                  lib_arch.code,
1866                  arch_array[running_arch_index].name);
1867     }
1868   }
1869 
1870   return NULL;
1871 }
1872 
1873 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1874                                 int ebuflen) {
1875   void * result = ::dlopen(filename, RTLD_LAZY);
1876   if (result == NULL) {
1877     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1878     ebuf[ebuflen-1] = '\0';
1879   }
1880   return result;
1881 }
1882 
1883 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1884                                        int ebuflen) {
1885   void * result = NULL;
1886   if (LoadExecStackDllInVMThread) {
1887     result = dlopen_helper(filename, ebuf, ebuflen);
1888   }
1889 
1890   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1891   // library that requires an executable stack, or which does not have this
1892   // stack attribute set, dlopen changes the stack attribute to executable. The
1893   // read protection of the guard pages gets lost.
1894   //
1895   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1896   // may have been queued at the same time.
1897 
1898   if (!_stack_is_executable) {
1899     for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1900       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1901           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1902         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1903           warning("Attempt to reguard stack yellow zone failed.");
1904         }
1905       }
1906     }
1907   }
1908 
1909   return result;
1910 }
1911 
1912 void* os::dll_lookup(void* handle, const char* name) {
1913   void* res = dlsym(handle, name);
1914   return res;
1915 }
1916 
1917 void* os::get_default_process_handle() {
1918   return (void*)::dlopen(NULL, RTLD_LAZY);
1919 }
1920 
1921 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
1922   int fd = ::open(filename, O_RDONLY);
1923   if (fd == -1) {
1924     return false;
1925   }
1926 
1927   if (hdr != NULL) {
1928     st->print_cr("%s", hdr);
1929   }
1930 
1931   char buf[33];
1932   int bytes;
1933   buf[32] = '\0';
1934   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1935     st->print_raw(buf, bytes);
1936   }
1937 
1938   ::close(fd);
1939 
1940   return true;
1941 }
1942 
1943 void os::print_dll_info(outputStream *st) {
1944   st->print_cr("Dynamic libraries:");
1945 
1946   char fname[32];
1947   pid_t pid = os::Linux::gettid();
1948 
1949   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1950 
1951   if (!_print_ascii_file(fname, st)) {
1952     st->print("Can not get library information for pid = %d\n", pid);
1953   }
1954 }
1955 
1956 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1957   FILE *procmapsFile = NULL;
1958 
1959   // Open the procfs maps file for the current process
1960   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1961     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1962     char line[PATH_MAX + 100];
1963 
1964     // Read line by line from 'file'
1965     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1966       u8 base, top, offset, inode;
1967       char permissions[5];
1968       char device[6];
1969       char name[PATH_MAX + 1];
1970 
1971       // Parse fields from line
1972       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s",
1973              &base, &top, permissions, &offset, device, &inode, name);
1974 
1975       // Filter by device id '00:00' so that we only get file system mapped files.
1976       if (strcmp(device, "00:00") != 0) {
1977 
1978         // Call callback with the fields of interest
1979         if(callback(name, (address)base, (address)top, param)) {
1980           // Oops abort, callback aborted
1981           fclose(procmapsFile);
1982           return 1;
1983         }
1984       }
1985     }
1986     fclose(procmapsFile);
1987   }
1988   return 0;
1989 }
1990 
1991 void os::print_os_info_brief(outputStream* st) {
1992   os::Linux::print_distro_info(st);
1993 
1994   os::Posix::print_uname_info(st);
1995 
1996   os::Linux::print_libversion_info(st);
1997 
1998 }
1999 
2000 void os::print_os_info(outputStream* st) {
2001   st->print("OS:");
2002 
2003   os::Linux::print_distro_info(st);
2004 
2005   os::Posix::print_uname_info(st);
2006 
2007   // Print warning if unsafe chroot environment detected
2008   if (unsafe_chroot_detected) {
2009     st->print("WARNING!! ");
2010     st->print_cr("%s", unstable_chroot_error);
2011   }
2012 
2013   os::Linux::print_libversion_info(st);
2014 
2015   os::Posix::print_rlimit_info(st);
2016 
2017   os::Posix::print_load_average(st);
2018 
2019   os::Linux::print_full_memory_info(st);
2020 
2021   os::Linux::print_proc_sys_info(st);
2022 
2023   os::Linux::print_ld_preload_file(st);
2024 
2025   os::Linux::print_container_info(st);
2026 
2027   VM_Version::print_platform_virtualization_info(st);
2028 
2029   os::Linux::print_steal_info(st);
2030 }
2031 
2032 // Try to identify popular distros.
2033 // Most Linux distributions have a /etc/XXX-release file, which contains
2034 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2035 // file that also contains the OS version string. Some have more than one
2036 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2037 // /etc/redhat-release.), so the order is important.
2038 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2039 // their own specific XXX-release file as well as a redhat-release file.
2040 // Because of this the XXX-release file needs to be searched for before the
2041 // redhat-release file.
2042 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2043 // search for redhat-release / SuSE-release needs to be before lsb-release.
2044 // Since the lsb-release file is the new standard it needs to be searched
2045 // before the older style release files.
2046 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2047 // next to last resort.  The os-release file is a new standard that contains
2048 // distribution information and the system-release file seems to be an old
2049 // standard that has been replaced by the lsb-release and os-release files.
2050 // Searching for the debian_version file is the last resort.  It contains
2051 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2052 // "Debian " is printed before the contents of the debian_version file.
2053 
2054 const char* distro_files[] = {
2055   "/etc/oracle-release",
2056   "/etc/mandriva-release",
2057   "/etc/mandrake-release",
2058   "/etc/sun-release",
2059   "/etc/redhat-release",
2060   "/etc/SuSE-release",
2061   "/etc/lsb-release",
2062   "/etc/turbolinux-release",
2063   "/etc/gentoo-release",
2064   "/etc/ltib-release",
2065   "/etc/angstrom-version",
2066   "/etc/system-release",
2067   "/etc/os-release",
2068   NULL };
2069 
2070 void os::Linux::print_distro_info(outputStream* st) {
2071   for (int i = 0;; i++) {
2072     const char* file = distro_files[i];
2073     if (file == NULL) {
2074       break;  // done
2075     }
2076     // If file prints, we found it.
2077     if (_print_ascii_file(file, st)) {
2078       return;
2079     }
2080   }
2081 
2082   if (file_exists("/etc/debian_version")) {
2083     st->print("Debian ");
2084     _print_ascii_file("/etc/debian_version", st);
2085   } else {
2086     st->print("Linux");
2087   }
2088   st->cr();
2089 }
2090 
2091 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2092   char buf[256];
2093   while (fgets(buf, sizeof(buf), fp)) {
2094     // Edit out extra stuff in expected format
2095     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2096       char* ptr = strstr(buf, "\"");  // the name is in quotes
2097       if (ptr != NULL) {
2098         ptr++; // go beyond first quote
2099         char* nl = strchr(ptr, '\"');
2100         if (nl != NULL) *nl = '\0';
2101         strncpy(distro, ptr, length);
2102       } else {
2103         ptr = strstr(buf, "=");
2104         ptr++; // go beyond equals then
2105         char* nl = strchr(ptr, '\n');
2106         if (nl != NULL) *nl = '\0';
2107         strncpy(distro, ptr, length);
2108       }
2109       return;
2110     } else if (get_first_line) {
2111       char* nl = strchr(buf, '\n');
2112       if (nl != NULL) *nl = '\0';
2113       strncpy(distro, buf, length);
2114       return;
2115     }
2116   }
2117   // print last line and close
2118   char* nl = strchr(buf, '\n');
2119   if (nl != NULL) *nl = '\0';
2120   strncpy(distro, buf, length);
2121 }
2122 
2123 static void parse_os_info(char* distro, size_t length, const char* file) {
2124   FILE* fp = fopen(file, "r");
2125   if (fp != NULL) {
2126     // if suse format, print out first line
2127     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2128     parse_os_info_helper(fp, distro, length, get_first_line);
2129     fclose(fp);
2130   }
2131 }
2132 
2133 void os::get_summary_os_info(char* buf, size_t buflen) {
2134   for (int i = 0;; i++) {
2135     const char* file = distro_files[i];
2136     if (file == NULL) {
2137       break; // ran out of distro_files
2138     }
2139     if (file_exists(file)) {
2140       parse_os_info(buf, buflen, file);
2141       return;
2142     }
2143   }
2144   // special case for debian
2145   if (file_exists("/etc/debian_version")) {
2146     strncpy(buf, "Debian ", buflen);
2147     if (buflen > 7) {
2148       parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2149     }
2150   } else {
2151     strncpy(buf, "Linux", buflen);
2152   }
2153 }
2154 
2155 void os::Linux::print_libversion_info(outputStream* st) {
2156   // libc, pthread
2157   st->print("libc:");
2158   st->print("%s ", os::Linux::glibc_version());
2159   st->print("%s ", os::Linux::libpthread_version());
2160   st->cr();
2161 }
2162 
2163 void os::Linux::print_proc_sys_info(outputStream* st) {
2164   st->cr();
2165   st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2166   _print_ascii_file("/proc/sys/kernel/threads-max", st);
2167   st->cr();
2168   st->cr();
2169 
2170   st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2171   _print_ascii_file("/proc/sys/vm/max_map_count", st);
2172   st->cr();
2173   st->cr();
2174 
2175   st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2176   _print_ascii_file("/proc/sys/kernel/pid_max", st);
2177   st->cr();
2178   st->cr();
2179 }
2180 
2181 void os::Linux::print_full_memory_info(outputStream* st) {
2182   st->print("\n/proc/meminfo:\n");
2183   _print_ascii_file("/proc/meminfo", st);
2184   st->cr();
2185 }
2186 
2187 void os::Linux::print_ld_preload_file(outputStream* st) {
2188   _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2189   st->cr();
2190 }
2191 
2192 void os::Linux::print_container_info(outputStream* st) {
2193   if (!OSContainer::is_containerized()) {
2194     return;
2195   }
2196 
2197   st->print("container (cgroup) information:\n");
2198 
2199   const char *p_ct = OSContainer::container_type();
2200   st->print("container_type: %s\n", p_ct != NULL ? p_ct : "not supported");
2201 
2202   char *p = OSContainer::cpu_cpuset_cpus();
2203   st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "not supported");
2204   free(p);
2205 
2206   p = OSContainer::cpu_cpuset_memory_nodes();
2207   st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "not supported");
2208   free(p);
2209 
2210   int i = OSContainer::active_processor_count();
2211   st->print("active_processor_count: ");
2212   if (i > 0) {
2213     st->print("%d\n", i);
2214   } else {
2215     st->print("not supported\n");
2216   }
2217 
2218   i = OSContainer::cpu_quota();
2219   st->print("cpu_quota: ");
2220   if (i > 0) {
2221     st->print("%d\n", i);
2222   } else {
2223     st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no quota");
2224   }
2225 
2226   i = OSContainer::cpu_period();
2227   st->print("cpu_period: ");
2228   if (i > 0) {
2229     st->print("%d\n", i);
2230   } else {
2231     st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no period");
2232   }
2233 
2234   i = OSContainer::cpu_shares();
2235   st->print("cpu_shares: ");
2236   if (i > 0) {
2237     st->print("%d\n", i);
2238   } else {
2239     st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no shares");
2240   }
2241 
2242   jlong j = OSContainer::memory_limit_in_bytes();
2243   st->print("memory_limit_in_bytes: ");
2244   if (j > 0) {
2245     st->print(JLONG_FORMAT "\n", j);
2246   } else {
2247     st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2248   }
2249 
2250   j = OSContainer::memory_and_swap_limit_in_bytes();
2251   st->print("memory_and_swap_limit_in_bytes: ");
2252   if (j > 0) {
2253     st->print(JLONG_FORMAT "\n", j);
2254   } else {
2255     st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2256   }
2257 
2258   j = OSContainer::memory_soft_limit_in_bytes();
2259   st->print("memory_soft_limit_in_bytes: ");
2260   if (j > 0) {
2261     st->print(JLONG_FORMAT "\n", j);
2262   } else {
2263     st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2264   }
2265 
2266   j = OSContainer::OSContainer::memory_usage_in_bytes();
2267   st->print("memory_usage_in_bytes: ");
2268   if (j > 0) {
2269     st->print(JLONG_FORMAT "\n", j);
2270   } else {
2271     st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2272   }
2273 
2274   j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2275   st->print("memory_max_usage_in_bytes: ");
2276   if (j > 0) {
2277     st->print(JLONG_FORMAT "\n", j);
2278   } else {
2279     st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2280   }
2281   st->cr();
2282 }
2283 
2284 void os::Linux::print_steal_info(outputStream* st) {
2285   if (has_initial_tick_info) {
2286     CPUPerfTicks pticks;
2287     bool res = os::Linux::get_tick_information(&pticks, -1);
2288 
2289     if (res && pticks.has_steal_ticks) {
2290       uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks;
2291       uint64_t total_ticks_difference = pticks.total - initial_total_ticks;
2292       double steal_ticks_perc = 0.0;
2293       if (total_ticks_difference != 0) {
2294         steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference;
2295       }
2296       st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference);
2297       st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc);
2298     }
2299   }
2300 }
2301 
2302 void os::print_memory_info(outputStream* st) {
2303 
2304   st->print("Memory:");
2305   st->print(" %dk page", os::vm_page_size()>>10);
2306 
2307   // values in struct sysinfo are "unsigned long"
2308   struct sysinfo si;
2309   sysinfo(&si);
2310 
2311   st->print(", physical " UINT64_FORMAT "k",
2312             os::physical_memory() >> 10);
2313   st->print("(" UINT64_FORMAT "k free)",
2314             os::available_memory() >> 10);
2315   st->print(", swap " UINT64_FORMAT "k",
2316             ((jlong)si.totalswap * si.mem_unit) >> 10);
2317   st->print("(" UINT64_FORMAT "k free)",
2318             ((jlong)si.freeswap * si.mem_unit) >> 10);
2319   st->cr();
2320 }
2321 
2322 // Print the first "model name" line and the first "flags" line
2323 // that we find and nothing more. We assume "model name" comes
2324 // before "flags" so if we find a second "model name", then the
2325 // "flags" field is considered missing.
2326 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2327 #if defined(IA32) || defined(AMD64)
2328   // Other platforms have less repetitive cpuinfo files
2329   FILE *fp = fopen("/proc/cpuinfo", "r");
2330   if (fp) {
2331     while (!feof(fp)) {
2332       if (fgets(buf, buflen, fp)) {
2333         // Assume model name comes before flags
2334         bool model_name_printed = false;
2335         if (strstr(buf, "model name") != NULL) {
2336           if (!model_name_printed) {
2337             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2338             st->print_raw(buf);
2339             model_name_printed = true;
2340           } else {
2341             // model name printed but not flags?  Odd, just return
2342             fclose(fp);
2343             return true;
2344           }
2345         }
2346         // print the flags line too
2347         if (strstr(buf, "flags") != NULL) {
2348           st->print_raw(buf);
2349           fclose(fp);
2350           return true;
2351         }
2352       }
2353     }
2354     fclose(fp);
2355   }
2356 #endif // x86 platforms
2357   return false;
2358 }
2359 
2360 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2361   // Only print the model name if the platform provides this as a summary
2362   if (!print_model_name_and_flags(st, buf, buflen)) {
2363     st->print("\n/proc/cpuinfo:\n");
2364     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2365       st->print_cr("  <Not Available>");
2366     }
2367   }
2368 }
2369 
2370 #if defined(AMD64) || defined(IA32) || defined(X32)
2371 const char* search_string = "model name";
2372 #elif defined(M68K)
2373 const char* search_string = "CPU";
2374 #elif defined(PPC64)
2375 const char* search_string = "cpu";
2376 #elif defined(S390)
2377 const char* search_string = "machine =";
2378 #elif defined(SPARC)
2379 const char* search_string = "cpu";
2380 #else
2381 const char* search_string = "Processor";
2382 #endif
2383 
2384 // Parses the cpuinfo file for string representing the model name.
2385 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2386   FILE* fp = fopen("/proc/cpuinfo", "r");
2387   if (fp != NULL) {
2388     while (!feof(fp)) {
2389       char buf[256];
2390       if (fgets(buf, sizeof(buf), fp)) {
2391         char* start = strstr(buf, search_string);
2392         if (start != NULL) {
2393           char *ptr = start + strlen(search_string);
2394           char *end = buf + strlen(buf);
2395           while (ptr != end) {
2396              // skip whitespace and colon for the rest of the name.
2397              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2398                break;
2399              }
2400              ptr++;
2401           }
2402           if (ptr != end) {
2403             // reasonable string, get rid of newline and keep the rest
2404             char* nl = strchr(buf, '\n');
2405             if (nl != NULL) *nl = '\0';
2406             strncpy(cpuinfo, ptr, length);
2407             fclose(fp);
2408             return;
2409           }
2410         }
2411       }
2412     }
2413     fclose(fp);
2414   }
2415   // cpuinfo not found or parsing failed, just print generic string.  The entire
2416   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2417 #if   defined(AARCH64)
2418   strncpy(cpuinfo, "AArch64", length);
2419 #elif defined(AMD64)
2420   strncpy(cpuinfo, "x86_64", length);
2421 #elif defined(ARM)  // Order wrt. AARCH64 is relevant!
2422   strncpy(cpuinfo, "ARM", length);
2423 #elif defined(IA32)
2424   strncpy(cpuinfo, "x86_32", length);
2425 #elif defined(IA64)
2426   strncpy(cpuinfo, "IA64", length);
2427 #elif defined(PPC)
2428   strncpy(cpuinfo, "PPC64", length);
2429 #elif defined(S390)
2430   strncpy(cpuinfo, "S390", length);
2431 #elif defined(SPARC)
2432   strncpy(cpuinfo, "sparcv9", length);
2433 #elif defined(ZERO_LIBARCH)
2434   strncpy(cpuinfo, ZERO_LIBARCH, length);
2435 #else
2436   strncpy(cpuinfo, "unknown", length);
2437 #endif
2438 }
2439 
2440 static void print_signal_handler(outputStream* st, int sig,
2441                                  char* buf, size_t buflen);
2442 
2443 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2444   st->print_cr("Signal Handlers:");
2445   print_signal_handler(st, SIGSEGV, buf, buflen);
2446   print_signal_handler(st, SIGBUS , buf, buflen);
2447   print_signal_handler(st, SIGFPE , buf, buflen);
2448   print_signal_handler(st, SIGPIPE, buf, buflen);
2449   print_signal_handler(st, SIGXFSZ, buf, buflen);
2450   print_signal_handler(st, SIGILL , buf, buflen);
2451   print_signal_handler(st, SR_signum, buf, buflen);
2452   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2453   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2454   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2455   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2456 #if defined(PPC64)
2457   print_signal_handler(st, SIGTRAP, buf, buflen);
2458 #endif
2459 }
2460 
2461 static char saved_jvm_path[MAXPATHLEN] = {0};
2462 
2463 // Find the full path to the current module, libjvm.so
2464 void os::jvm_path(char *buf, jint buflen) {
2465   // Error checking.
2466   if (buflen < MAXPATHLEN) {
2467     assert(false, "must use a large-enough buffer");
2468     buf[0] = '\0';
2469     return;
2470   }
2471   // Lazy resolve the path to current module.
2472   if (saved_jvm_path[0] != 0) {
2473     strcpy(buf, saved_jvm_path);
2474     return;
2475   }
2476 
2477   char dli_fname[MAXPATHLEN];
2478   bool ret = dll_address_to_library_name(
2479                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2480                                          dli_fname, sizeof(dli_fname), NULL);
2481   assert(ret, "cannot locate libjvm");
2482   char *rp = NULL;
2483   if (ret && dli_fname[0] != '\0') {
2484     rp = os::Posix::realpath(dli_fname, buf, buflen);
2485   }
2486   if (rp == NULL) {
2487     return;
2488   }
2489 
2490   if (Arguments::sun_java_launcher_is_altjvm()) {
2491     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2492     // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2493     // If "/jre/lib/" appears at the right place in the string, then
2494     // assume we are installed in a JDK and we're done. Otherwise, check
2495     // for a JAVA_HOME environment variable and fix up the path so it
2496     // looks like libjvm.so is installed there (append a fake suffix
2497     // hotspot/libjvm.so).
2498     const char *p = buf + strlen(buf) - 1;
2499     for (int count = 0; p > buf && count < 5; ++count) {
2500       for (--p; p > buf && *p != '/'; --p)
2501         /* empty */ ;
2502     }
2503 
2504     if (strncmp(p, "/jre/lib/", 9) != 0) {
2505       // Look for JAVA_HOME in the environment.
2506       char* java_home_var = ::getenv("JAVA_HOME");
2507       if (java_home_var != NULL && java_home_var[0] != 0) {
2508         char* jrelib_p;
2509         int len;
2510 
2511         // Check the current module name "libjvm.so".
2512         p = strrchr(buf, '/');
2513         if (p == NULL) {
2514           return;
2515         }
2516         assert(strstr(p, "/libjvm") == p, "invalid library name");
2517 
2518         rp = os::Posix::realpath(java_home_var, buf, buflen);
2519         if (rp == NULL) {
2520           return;
2521         }
2522 
2523         // determine if this is a legacy image or modules image
2524         // modules image doesn't have "jre" subdirectory
2525         len = strlen(buf);
2526         assert(len < buflen, "Ran out of buffer room");
2527         jrelib_p = buf + len;
2528         snprintf(jrelib_p, buflen-len, "/jre/lib");
2529         if (0 != access(buf, F_OK)) {
2530           snprintf(jrelib_p, buflen-len, "/lib");
2531         }
2532 
2533         if (0 == access(buf, F_OK)) {
2534           // Use current module name "libjvm.so"
2535           len = strlen(buf);
2536           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2537         } else {
2538           // Go back to path of .so
2539           rp = os::Posix::realpath(dli_fname, buf, buflen);
2540           if (rp == NULL) {
2541             return;
2542           }
2543         }
2544       }
2545     }
2546   }
2547 
2548   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2549   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2550 }
2551 
2552 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2553   // no prefix required, not even "_"
2554 }
2555 
2556 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2557   // no suffix required
2558 }
2559 
2560 ////////////////////////////////////////////////////////////////////////////////
2561 // sun.misc.Signal support
2562 
2563 static volatile jint sigint_count = 0;
2564 
2565 static void UserHandler(int sig, void *siginfo, void *context) {
2566   // 4511530 - sem_post is serialized and handled by the manager thread. When
2567   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2568   // don't want to flood the manager thread with sem_post requests.
2569   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2570     return;
2571   }
2572 
2573   // Ctrl-C is pressed during error reporting, likely because the error
2574   // handler fails to abort. Let VM die immediately.
2575   if (sig == SIGINT && VMError::is_error_reported()) {
2576     os::die();
2577   }
2578 
2579   os::signal_notify(sig);
2580 }
2581 
2582 void* os::user_handler() {
2583   return CAST_FROM_FN_PTR(void*, UserHandler);
2584 }
2585 
2586 extern "C" {
2587   typedef void (*sa_handler_t)(int);
2588   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2589 }
2590 
2591 void* os::signal(int signal_number, void* handler) {
2592   struct sigaction sigAct, oldSigAct;
2593 
2594   sigfillset(&(sigAct.sa_mask));
2595   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2596   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2597 
2598   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2599     // -1 means registration failed
2600     return (void *)-1;
2601   }
2602 
2603   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2604 }
2605 
2606 void os::signal_raise(int signal_number) {
2607   ::raise(signal_number);
2608 }
2609 
2610 // The following code is moved from os.cpp for making this
2611 // code platform specific, which it is by its very nature.
2612 
2613 // Will be modified when max signal is changed to be dynamic
2614 int os::sigexitnum_pd() {
2615   return NSIG;
2616 }
2617 
2618 // a counter for each possible signal value
2619 static volatile jint pending_signals[NSIG+1] = { 0 };
2620 
2621 // Linux(POSIX) specific hand shaking semaphore.
2622 static Semaphore* sig_sem = NULL;
2623 static PosixSemaphore sr_semaphore;
2624 
2625 static void jdk_misc_signal_init() {
2626   // Initialize signal structures
2627   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2628 
2629   // Initialize signal semaphore
2630   sig_sem = new Semaphore();
2631 }
2632 
2633 void os::signal_notify(int sig) {
2634   if (sig_sem != NULL) {
2635     Atomic::inc(&pending_signals[sig]);
2636     sig_sem->signal();
2637   } else {
2638     // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2639     // initialization isn't called.
2640     assert(ReduceSignalUsage, "signal semaphore should be created");
2641   }
2642 }
2643 
2644 static int check_pending_signals() {
2645   Atomic::store(0, &sigint_count);
2646   for (;;) {
2647     for (int i = 0; i < NSIG + 1; i++) {
2648       jint n = pending_signals[i];
2649       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2650         return i;
2651       }
2652     }
2653     JavaThread *thread = JavaThread::current();
2654     ThreadBlockInVM tbivm(thread);
2655 
2656     bool threadIsSuspended;
2657     do {
2658       thread->set_suspend_equivalent();
2659       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2660       sig_sem->wait();
2661 
2662       // were we externally suspended while we were waiting?
2663       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2664       if (threadIsSuspended) {
2665         // The semaphore has been incremented, but while we were waiting
2666         // another thread suspended us. We don't want to continue running
2667         // while suspended because that would surprise the thread that
2668         // suspended us.
2669         sig_sem->signal();
2670 
2671         thread->java_suspend_self();
2672       }
2673     } while (threadIsSuspended);
2674   }
2675 }
2676 
2677 int os::signal_wait() {
2678   return check_pending_signals();
2679 }
2680 
2681 ////////////////////////////////////////////////////////////////////////////////
2682 // Virtual Memory
2683 
2684 int os::vm_page_size() {
2685   // Seems redundant as all get out
2686   assert(os::Linux::page_size() != -1, "must call os::init");
2687   return os::Linux::page_size();
2688 }
2689 
2690 // Solaris allocates memory by pages.
2691 int os::vm_allocation_granularity() {
2692   assert(os::Linux::page_size() != -1, "must call os::init");
2693   return os::Linux::page_size();
2694 }
2695 
2696 // Rationale behind this function:
2697 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2698 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2699 //  samples for JITted code. Here we create private executable mapping over the code cache
2700 //  and then we can use standard (well, almost, as mapping can change) way to provide
2701 //  info for the reporting script by storing timestamp and location of symbol
2702 void linux_wrap_code(char* base, size_t size) {
2703   static volatile jint cnt = 0;
2704 
2705   if (!UseOprofile) {
2706     return;
2707   }
2708 
2709   char buf[PATH_MAX+1];
2710   int num = Atomic::add(1, &cnt);
2711 
2712   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2713            os::get_temp_directory(), os::current_process_id(), num);
2714   unlink(buf);
2715 
2716   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2717 
2718   if (fd != -1) {
2719     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2720     if (rv != (off_t)-1) {
2721       if (::write(fd, "", 1) == 1) {
2722         mmap(base, size,
2723              PROT_READ|PROT_WRITE|PROT_EXEC,
2724              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2725       }
2726     }
2727     ::close(fd);
2728     unlink(buf);
2729   }
2730 }
2731 
2732 static bool recoverable_mmap_error(int err) {
2733   // See if the error is one we can let the caller handle. This
2734   // list of errno values comes from JBS-6843484. I can't find a
2735   // Linux man page that documents this specific set of errno
2736   // values so while this list currently matches Solaris, it may
2737   // change as we gain experience with this failure mode.
2738   switch (err) {
2739   case EBADF:
2740   case EINVAL:
2741   case ENOTSUP:
2742     // let the caller deal with these errors
2743     return true;
2744 
2745   default:
2746     // Any remaining errors on this OS can cause our reserved mapping
2747     // to be lost. That can cause confusion where different data
2748     // structures think they have the same memory mapped. The worst
2749     // scenario is if both the VM and a library think they have the
2750     // same memory mapped.
2751     return false;
2752   }
2753 }
2754 
2755 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2756                                     int err) {
2757   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2758           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2759           os::strerror(err), err);
2760 }
2761 
2762 static void warn_fail_commit_memory(char* addr, size_t size,
2763                                     size_t alignment_hint, bool exec,
2764                                     int err) {
2765   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2766           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2767           alignment_hint, exec, os::strerror(err), err);
2768 }
2769 
2770 // NOTE: Linux kernel does not really reserve the pages for us.
2771 //       All it does is to check if there are enough free pages
2772 //       left at the time of mmap(). This could be a potential
2773 //       problem.
2774 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2775   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2776   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2777                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2778   if (res != (uintptr_t) MAP_FAILED) {
2779     if (UseNUMAInterleaving) {
2780       numa_make_global(addr, size);
2781     }
2782     return 0;
2783   }
2784 
2785   int err = errno;  // save errno from mmap() call above
2786 
2787   if (!recoverable_mmap_error(err)) {
2788     warn_fail_commit_memory(addr, size, exec, err);
2789     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2790   }
2791 
2792   return err;
2793 }
2794 
2795 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2796   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2797 }
2798 
2799 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2800                                   const char* mesg) {
2801   assert(mesg != NULL, "mesg must be specified");
2802   int err = os::Linux::commit_memory_impl(addr, size, exec);
2803   if (err != 0) {
2804     // the caller wants all commit errors to exit with the specified mesg:
2805     warn_fail_commit_memory(addr, size, exec, err);
2806     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2807   }
2808 }
2809 
2810 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2811 #ifndef MAP_HUGETLB
2812   #define MAP_HUGETLB 0x40000
2813 #endif
2814 
2815 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2816 #ifndef MADV_HUGEPAGE
2817   #define MADV_HUGEPAGE 14
2818 #endif
2819 
2820 int os::Linux::commit_memory_impl(char* addr, size_t size,
2821                                   size_t alignment_hint, bool exec) {
2822   int err = os::Linux::commit_memory_impl(addr, size, exec);
2823   if (err == 0) {
2824     realign_memory(addr, size, alignment_hint);
2825   }
2826   return err;
2827 }
2828 
2829 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2830                           bool exec) {
2831   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2832 }
2833 
2834 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2835                                   size_t alignment_hint, bool exec,
2836                                   const char* mesg) {
2837   assert(mesg != NULL, "mesg must be specified");
2838   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2839   if (err != 0) {
2840     // the caller wants all commit errors to exit with the specified mesg:
2841     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2842     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2843   }
2844 }
2845 
2846 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2847   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2848     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2849     // be supported or the memory may already be backed by huge pages.
2850     ::madvise(addr, bytes, MADV_HUGEPAGE);
2851   }
2852 }
2853 
2854 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2855   // This method works by doing an mmap over an existing mmaping and effectively discarding
2856   // the existing pages. However it won't work for SHM-based large pages that cannot be
2857   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2858   // small pages on top of the SHM segment. This method always works for small pages, so we
2859   // allow that in any case.
2860   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2861     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2862   }
2863 }
2864 
2865 void os::numa_make_global(char *addr, size_t bytes) {
2866   Linux::numa_interleave_memory(addr, bytes);
2867 }
2868 
2869 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2870 // bind policy to MPOL_PREFERRED for the current thread.
2871 #define USE_MPOL_PREFERRED 0
2872 
2873 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2874   // To make NUMA and large pages more robust when both enabled, we need to ease
2875   // the requirements on where the memory should be allocated. MPOL_BIND is the
2876   // default policy and it will force memory to be allocated on the specified
2877   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2878   // the specified node, but will not force it. Using this policy will prevent
2879   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2880   // free large pages.
2881   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2882   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2883 }
2884 
2885 bool os::numa_topology_changed() { return false; }
2886 
2887 size_t os::numa_get_groups_num() {
2888   // Return just the number of nodes in which it's possible to allocate memory
2889   // (in numa terminology, configured nodes).
2890   return Linux::numa_num_configured_nodes();
2891 }
2892 
2893 int os::numa_get_group_id() {
2894   int cpu_id = Linux::sched_getcpu();
2895   if (cpu_id != -1) {
2896     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2897     if (lgrp_id != -1) {
2898       return lgrp_id;
2899     }
2900   }
2901   return 0;
2902 }
2903 
2904 int os::Linux::get_existing_num_nodes() {
2905   int node;
2906   int highest_node_number = Linux::numa_max_node();
2907   int num_nodes = 0;
2908 
2909   // Get the total number of nodes in the system including nodes without memory.
2910   for (node = 0; node <= highest_node_number; node++) {
2911     if (is_node_in_existing_nodes(node)) {
2912       num_nodes++;
2913     }
2914   }
2915   return num_nodes;
2916 }
2917 
2918 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2919   int highest_node_number = Linux::numa_max_node();
2920   size_t i = 0;
2921 
2922   // Map all node ids in which it is possible to allocate memory. Also nodes are
2923   // not always consecutively available, i.e. available from 0 to the highest
2924   // node number. If the nodes have been bound explicitly using numactl membind,
2925   // then allocate memory from those nodes only.
2926   for (int node = 0; node <= highest_node_number; node++) {
2927     if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
2928       ids[i++] = node;
2929     }
2930   }
2931   return i;
2932 }
2933 
2934 bool os::get_page_info(char *start, page_info* info) {
2935   return false;
2936 }
2937 
2938 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2939                      page_info* page_found) {
2940   return end;
2941 }
2942 
2943 
2944 int os::Linux::sched_getcpu_syscall(void) {
2945   unsigned int cpu = 0;
2946   int retval = -1;
2947 
2948 #if defined(IA32)
2949   #ifndef SYS_getcpu
2950     #define SYS_getcpu 318
2951   #endif
2952   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2953 #elif defined(AMD64)
2954 // Unfortunately we have to bring all these macros here from vsyscall.h
2955 // to be able to compile on old linuxes.
2956   #define __NR_vgetcpu 2
2957   #define VSYSCALL_START (-10UL << 20)
2958   #define VSYSCALL_SIZE 1024
2959   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2960   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2961   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2962   retval = vgetcpu(&cpu, NULL, NULL);
2963 #endif
2964 
2965   return (retval == -1) ? retval : cpu;
2966 }
2967 
2968 void os::Linux::sched_getcpu_init() {
2969   // sched_getcpu() should be in libc.
2970   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2971                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2972 
2973   // If it's not, try a direct syscall.
2974   if (sched_getcpu() == -1) {
2975     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2976                                     (void*)&sched_getcpu_syscall));
2977   }
2978 
2979   if (sched_getcpu() == -1) {
2980     vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
2981   }
2982 }
2983 
2984 // Something to do with the numa-aware allocator needs these symbols
2985 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2986 extern "C" JNIEXPORT void numa_error(char *where) { }
2987 
2988 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2989 // load symbol from base version instead.
2990 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2991   void *f = dlvsym(handle, name, "libnuma_1.1");
2992   if (f == NULL) {
2993     f = dlsym(handle, name);
2994   }
2995   return f;
2996 }
2997 
2998 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2999 // Return NULL if the symbol is not defined in this particular version.
3000 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
3001   return dlvsym(handle, name, "libnuma_1.2");
3002 }
3003 
3004 bool os::Linux::libnuma_init() {
3005   if (sched_getcpu() != -1) { // Requires sched_getcpu() support
3006     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3007     if (handle != NULL) {
3008       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3009                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
3010       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3011                                        libnuma_dlsym(handle, "numa_max_node")));
3012       set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3013                                                    libnuma_dlsym(handle, "numa_num_configured_nodes")));
3014       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3015                                         libnuma_dlsym(handle, "numa_available")));
3016       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3017                                             libnuma_dlsym(handle, "numa_tonode_memory")));
3018       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3019                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
3020       set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3021                                                 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3022       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3023                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
3024       set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3025                                                libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3026       set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3027                                        libnuma_dlsym(handle, "numa_distance")));
3028       set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
3029                                           libnuma_v2_dlsym(handle, "numa_get_membind")));
3030       set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
3031                                                   libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
3032 
3033       if (numa_available() != -1) {
3034         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3035         set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3036         set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3037         set_numa_interleave_bitmask(_numa_get_interleave_mask());
3038         set_numa_membind_bitmask(_numa_get_membind());
3039         // Create an index -> node mapping, since nodes are not always consecutive
3040         _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3041         rebuild_nindex_to_node_map();
3042         // Create a cpu -> node mapping
3043         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3044         rebuild_cpu_to_node_map();
3045         return true;
3046       }
3047     }
3048   }
3049   return false;
3050 }
3051 
3052 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
3053   // Creating guard page is very expensive. Java thread has HotSpot
3054   // guard pages, only enable glibc guard page for non-Java threads.
3055   // (Remember: compiler thread is a Java thread, too!)
3056   return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
3057 }
3058 
3059 void os::Linux::rebuild_nindex_to_node_map() {
3060   int highest_node_number = Linux::numa_max_node();
3061 
3062   nindex_to_node()->clear();
3063   for (int node = 0; node <= highest_node_number; node++) {
3064     if (Linux::is_node_in_existing_nodes(node)) {
3065       nindex_to_node()->append(node);
3066     }
3067   }
3068 }
3069 
3070 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3071 // The table is later used in get_node_by_cpu().
3072 void os::Linux::rebuild_cpu_to_node_map() {
3073   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3074                               // in libnuma (possible values are starting from 16,
3075                               // and continuing up with every other power of 2, but less
3076                               // than the maximum number of CPUs supported by kernel), and
3077                               // is a subject to change (in libnuma version 2 the requirements
3078                               // are more reasonable) we'll just hardcode the number they use
3079                               // in the library.
3080   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3081 
3082   size_t cpu_num = processor_count();
3083   size_t cpu_map_size = NCPUS / BitsPerCLong;
3084   size_t cpu_map_valid_size =
3085     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3086 
3087   cpu_to_node()->clear();
3088   cpu_to_node()->at_grow(cpu_num - 1);
3089 
3090   size_t node_num = get_existing_num_nodes();
3091 
3092   int distance = 0;
3093   int closest_distance = INT_MAX;
3094   int closest_node = 0;
3095   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3096   for (size_t i = 0; i < node_num; i++) {
3097     // Check if node is configured (not a memory-less node). If it is not, find
3098     // the closest configured node. Check also if node is bound, i.e. it's allowed
3099     // to allocate memory from the node. If it's not allowed, map cpus in that node
3100     // to the closest node from which memory allocation is allowed.
3101     if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
3102         !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
3103       closest_distance = INT_MAX;
3104       // Check distance from all remaining nodes in the system. Ignore distance
3105       // from itself, from another non-configured node, and from another non-bound
3106       // node.
3107       for (size_t m = 0; m < node_num; m++) {
3108         if (m != i &&
3109             is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
3110             is_node_in_bound_nodes(nindex_to_node()->at(m))) {
3111           distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3112           // If a closest node is found, update. There is always at least one
3113           // configured and bound node in the system so there is always at least
3114           // one node close.
3115           if (distance != 0 && distance < closest_distance) {
3116             closest_distance = distance;
3117             closest_node = nindex_to_node()->at(m);
3118           }
3119         }
3120       }
3121      } else {
3122        // Current node is already a configured node.
3123        closest_node = nindex_to_node()->at(i);
3124      }
3125 
3126     // Get cpus from the original node and map them to the closest node. If node
3127     // is a configured node (not a memory-less node), then original node and
3128     // closest node are the same.
3129     if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3130       for (size_t j = 0; j < cpu_map_valid_size; j++) {
3131         if (cpu_map[j] != 0) {
3132           for (size_t k = 0; k < BitsPerCLong; k++) {
3133             if (cpu_map[j] & (1UL << k)) {
3134               cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3135             }
3136           }
3137         }
3138       }
3139     }
3140   }
3141   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3142 }
3143 
3144 int os::Linux::get_node_by_cpu(int cpu_id) {
3145   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3146     return cpu_to_node()->at(cpu_id);
3147   }
3148   return -1;
3149 }
3150 
3151 GrowableArray<int>* os::Linux::_cpu_to_node;
3152 GrowableArray<int>* os::Linux::_nindex_to_node;
3153 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3154 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3155 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3156 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3157 os::Linux::numa_available_func_t os::Linux::_numa_available;
3158 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3159 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3160 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3161 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3162 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3163 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3164 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3165 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3166 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy;
3167 unsigned long* os::Linux::_numa_all_nodes;
3168 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3169 struct bitmask* os::Linux::_numa_nodes_ptr;
3170 struct bitmask* os::Linux::_numa_interleave_bitmask;
3171 struct bitmask* os::Linux::_numa_membind_bitmask;
3172 
3173 bool os::pd_uncommit_memory(char* addr, size_t size) {
3174   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3175                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3176   return res  != (uintptr_t) MAP_FAILED;
3177 }
3178 
3179 static address get_stack_commited_bottom(address bottom, size_t size) {
3180   address nbot = bottom;
3181   address ntop = bottom + size;
3182 
3183   size_t page_sz = os::vm_page_size();
3184   unsigned pages = size / page_sz;
3185 
3186   unsigned char vec[1];
3187   unsigned imin = 1, imax = pages + 1, imid;
3188   int mincore_return_value = 0;
3189 
3190   assert(imin <= imax, "Unexpected page size");
3191 
3192   while (imin < imax) {
3193     imid = (imax + imin) / 2;
3194     nbot = ntop - (imid * page_sz);
3195 
3196     // Use a trick with mincore to check whether the page is mapped or not.
3197     // mincore sets vec to 1 if page resides in memory and to 0 if page
3198     // is swapped output but if page we are asking for is unmapped
3199     // it returns -1,ENOMEM
3200     mincore_return_value = mincore(nbot, page_sz, vec);
3201 
3202     if (mincore_return_value == -1) {
3203       // Page is not mapped go up
3204       // to find first mapped page
3205       if (errno != EAGAIN) {
3206         assert(errno == ENOMEM, "Unexpected mincore errno");
3207         imax = imid;
3208       }
3209     } else {
3210       // Page is mapped go down
3211       // to find first not mapped page
3212       imin = imid + 1;
3213     }
3214   }
3215 
3216   nbot = nbot + page_sz;
3217 
3218   // Adjust stack bottom one page up if last checked page is not mapped
3219   if (mincore_return_value == -1) {
3220     nbot = nbot + page_sz;
3221   }
3222 
3223   return nbot;
3224 }
3225 
3226 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3227   int mincore_return_value;
3228   const size_t stripe = 1024;  // query this many pages each time
3229   unsigned char vec[stripe + 1];
3230   // set a guard
3231   vec[stripe] = 'X';
3232 
3233   const size_t page_sz = os::vm_page_size();
3234   size_t pages = size / page_sz;
3235 
3236   assert(is_aligned(start, page_sz), "Start address must be page aligned");
3237   assert(is_aligned(size, page_sz), "Size must be page aligned");
3238 
3239   committed_start = NULL;
3240 
3241   int loops = (pages + stripe - 1) / stripe;
3242   int committed_pages = 0;
3243   address loop_base = start;
3244   bool found_range = false;
3245 
3246   for (int index = 0; index < loops && !found_range; index ++) {
3247     assert(pages > 0, "Nothing to do");
3248     int pages_to_query = (pages >= stripe) ? stripe : pages;
3249     pages -= pages_to_query;
3250 
3251     // Get stable read
3252     while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3253 
3254     // During shutdown, some memory goes away without properly notifying NMT,
3255     // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3256     // Bailout and return as not committed for now.
3257     if (mincore_return_value == -1 && errno == ENOMEM) {
3258       return false;
3259     }
3260 
3261     assert(vec[stripe] == 'X', "overflow guard");
3262     assert(mincore_return_value == 0, "Range must be valid");
3263     // Process this stripe
3264     for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3265       if ((vec[vecIdx] & 0x01) == 0) { // not committed
3266         // End of current contiguous region
3267         if (committed_start != NULL) {
3268           found_range = true;
3269           break;
3270         }
3271       } else { // committed
3272         // Start of region
3273         if (committed_start == NULL) {
3274           committed_start = loop_base + page_sz * vecIdx;
3275         }
3276         committed_pages ++;
3277       }
3278     }
3279 
3280     loop_base += pages_to_query * page_sz;
3281   }
3282 
3283   if (committed_start != NULL) {
3284     assert(committed_pages > 0, "Must have committed region");
3285     assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3286     assert(committed_start >= start && committed_start < start + size, "Out of range");
3287     committed_size = page_sz * committed_pages;
3288     return true;
3289   } else {
3290     assert(committed_pages == 0, "Should not have committed region");
3291     return false;
3292   }
3293 }
3294 
3295 
3296 // Linux uses a growable mapping for the stack, and if the mapping for
3297 // the stack guard pages is not removed when we detach a thread the
3298 // stack cannot grow beyond the pages where the stack guard was
3299 // mapped.  If at some point later in the process the stack expands to
3300 // that point, the Linux kernel cannot expand the stack any further
3301 // because the guard pages are in the way, and a segfault occurs.
3302 //
3303 // However, it's essential not to split the stack region by unmapping
3304 // a region (leaving a hole) that's already part of the stack mapping,
3305 // so if the stack mapping has already grown beyond the guard pages at
3306 // the time we create them, we have to truncate the stack mapping.
3307 // So, we need to know the extent of the stack mapping when
3308 // create_stack_guard_pages() is called.
3309 
3310 // We only need this for stacks that are growable: at the time of
3311 // writing thread stacks don't use growable mappings (i.e. those
3312 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3313 // only applies to the main thread.
3314 
3315 // If the (growable) stack mapping already extends beyond the point
3316 // where we're going to put our guard pages, truncate the mapping at
3317 // that point by munmap()ping it.  This ensures that when we later
3318 // munmap() the guard pages we don't leave a hole in the stack
3319 // mapping. This only affects the main/primordial thread
3320 
3321 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3322   if (os::is_primordial_thread()) {
3323     // As we manually grow stack up to bottom inside create_attached_thread(),
3324     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3325     // we don't need to do anything special.
3326     // Check it first, before calling heavy function.
3327     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3328     unsigned char vec[1];
3329 
3330     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3331       // Fallback to slow path on all errors, including EAGAIN
3332       stack_extent = (uintptr_t) get_stack_commited_bottom(
3333                                                            os::Linux::initial_thread_stack_bottom(),
3334                                                            (size_t)addr - stack_extent);
3335     }
3336 
3337     if (stack_extent < (uintptr_t)addr) {
3338       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3339     }
3340   }
3341 
3342   return os::commit_memory(addr, size, !ExecMem);
3343 }
3344 
3345 // If this is a growable mapping, remove the guard pages entirely by
3346 // munmap()ping them.  If not, just call uncommit_memory(). This only
3347 // affects the main/primordial thread, but guard against future OS changes.
3348 // It's safe to always unmap guard pages for primordial thread because we
3349 // always place it right after end of the mapped region.
3350 
3351 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3352   uintptr_t stack_extent, stack_base;
3353 
3354   if (os::is_primordial_thread()) {
3355     return ::munmap(addr, size) == 0;
3356   }
3357 
3358   return os::uncommit_memory(addr, size);
3359 }
3360 
3361 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3362 // at 'requested_addr'. If there are existing memory mappings at the same
3363 // location, however, they will be overwritten. If 'fixed' is false,
3364 // 'requested_addr' is only treated as a hint, the return value may or
3365 // may not start from the requested address. Unlike Linux mmap(), this
3366 // function returns NULL to indicate failure.
3367 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3368   char * addr;
3369   int flags;
3370 
3371   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3372   if (fixed) {
3373     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3374     flags |= MAP_FIXED;
3375   }
3376 
3377   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3378   // touch an uncommitted page. Otherwise, the read/write might
3379   // succeed if we have enough swap space to back the physical page.
3380   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3381                        flags, -1, 0);
3382 
3383   return addr == MAP_FAILED ? NULL : addr;
3384 }
3385 
3386 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3387 //   (req_addr != NULL) or with a given alignment.
3388 //  - bytes shall be a multiple of alignment.
3389 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3390 //  - alignment sets the alignment at which memory shall be allocated.
3391 //     It must be a multiple of allocation granularity.
3392 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3393 //  req_addr or NULL.
3394 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3395 
3396   size_t extra_size = bytes;
3397   if (req_addr == NULL && alignment > 0) {
3398     extra_size += alignment;
3399   }
3400 
3401   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3402     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3403     -1, 0);
3404   if (start == MAP_FAILED) {
3405     start = NULL;
3406   } else {
3407     if (req_addr != NULL) {
3408       if (start != req_addr) {
3409         ::munmap(start, extra_size);
3410         start = NULL;
3411       }
3412     } else {
3413       char* const start_aligned = align_up(start, alignment);
3414       char* const end_aligned = start_aligned + bytes;
3415       char* const end = start + extra_size;
3416       if (start_aligned > start) {
3417         ::munmap(start, start_aligned - start);
3418       }
3419       if (end_aligned < end) {
3420         ::munmap(end_aligned, end - end_aligned);
3421       }
3422       start = start_aligned;
3423     }
3424   }
3425   return start;
3426 }
3427 
3428 static int anon_munmap(char * addr, size_t size) {
3429   return ::munmap(addr, size) == 0;
3430 }
3431 
3432 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3433                             size_t alignment_hint) {
3434   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3435 }
3436 
3437 bool os::pd_release_memory(char* addr, size_t size) {
3438   return anon_munmap(addr, size);
3439 }
3440 
3441 static bool linux_mprotect(char* addr, size_t size, int prot) {
3442   // Linux wants the mprotect address argument to be page aligned.
3443   char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3444 
3445   // According to SUSv3, mprotect() should only be used with mappings
3446   // established by mmap(), and mmap() always maps whole pages. Unaligned
3447   // 'addr' likely indicates problem in the VM (e.g. trying to change
3448   // protection of malloc'ed or statically allocated memory). Check the
3449   // caller if you hit this assert.
3450   assert(addr == bottom, "sanity check");
3451 
3452   size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3453   return ::mprotect(bottom, size, prot) == 0;
3454 }
3455 
3456 // Set protections specified
3457 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3458                         bool is_committed) {
3459   unsigned int p = 0;
3460   switch (prot) {
3461   case MEM_PROT_NONE: p = PROT_NONE; break;
3462   case MEM_PROT_READ: p = PROT_READ; break;
3463   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3464   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3465   default:
3466     ShouldNotReachHere();
3467   }
3468   // is_committed is unused.
3469   return linux_mprotect(addr, bytes, p);
3470 }
3471 
3472 bool os::guard_memory(char* addr, size_t size) {
3473   return linux_mprotect(addr, size, PROT_NONE);
3474 }
3475 
3476 bool os::unguard_memory(char* addr, size_t size) {
3477   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3478 }
3479 
3480 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3481                                                     size_t page_size) {
3482   bool result = false;
3483   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3484                  MAP_ANONYMOUS|MAP_PRIVATE,
3485                  -1, 0);
3486   if (p != MAP_FAILED) {
3487     void *aligned_p = align_up(p, page_size);
3488 
3489     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3490 
3491     munmap(p, page_size * 2);
3492   }
3493 
3494   if (warn && !result) {
3495     warning("TransparentHugePages is not supported by the operating system.");
3496   }
3497 
3498   return result;
3499 }
3500 
3501 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3502   bool result = false;
3503   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3504                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3505                  -1, 0);
3506 
3507   if (p != MAP_FAILED) {
3508     // We don't know if this really is a huge page or not.
3509     FILE *fp = fopen("/proc/self/maps", "r");
3510     if (fp) {
3511       while (!feof(fp)) {
3512         char chars[257];
3513         long x = 0;
3514         if (fgets(chars, sizeof(chars), fp)) {
3515           if (sscanf(chars, "%lx-%*x", &x) == 1
3516               && x == (long)p) {
3517             if (strstr (chars, "hugepage")) {
3518               result = true;
3519               break;
3520             }
3521           }
3522         }
3523       }
3524       fclose(fp);
3525     }
3526     munmap(p, page_size);
3527   }
3528 
3529   if (warn && !result) {
3530     warning("HugeTLBFS is not supported by the operating system.");
3531   }
3532 
3533   return result;
3534 }
3535 
3536 // From the coredump_filter documentation:
3537 //
3538 // - (bit 0) anonymous private memory
3539 // - (bit 1) anonymous shared memory
3540 // - (bit 2) file-backed private memory
3541 // - (bit 3) file-backed shared memory
3542 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3543 //           effective only if the bit 2 is cleared)
3544 // - (bit 5) hugetlb private memory
3545 // - (bit 6) hugetlb shared memory
3546 // - (bit 7) dax private memory
3547 // - (bit 8) dax shared memory
3548 //
3549 static void set_coredump_filter(CoredumpFilterBit bit) {
3550   FILE *f;
3551   long cdm;
3552 
3553   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3554     return;
3555   }
3556 
3557   if (fscanf(f, "%lx", &cdm) != 1) {
3558     fclose(f);
3559     return;
3560   }
3561 
3562   long saved_cdm = cdm;
3563   rewind(f);
3564   cdm |= bit;
3565 
3566   if (cdm != saved_cdm) {
3567     fprintf(f, "%#lx", cdm);
3568   }
3569 
3570   fclose(f);
3571 }
3572 
3573 // Large page support
3574 
3575 static size_t _large_page_size = 0;
3576 
3577 size_t os::Linux::find_large_page_size() {
3578   size_t large_page_size = 0;
3579 
3580   // large_page_size on Linux is used to round up heap size. x86 uses either
3581   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3582   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3583   // page as large as 256M.
3584   //
3585   // Here we try to figure out page size by parsing /proc/meminfo and looking
3586   // for a line with the following format:
3587   //    Hugepagesize:     2048 kB
3588   //
3589   // If we can't determine the value (e.g. /proc is not mounted, or the text
3590   // format has been changed), we'll use the largest page size supported by
3591   // the processor.
3592 
3593 #ifndef ZERO
3594   large_page_size =
3595     AARCH64_ONLY(2 * M)
3596     AMD64_ONLY(2 * M)
3597     ARM32_ONLY(2 * M)
3598     IA32_ONLY(4 * M)
3599     IA64_ONLY(256 * M)
3600     PPC_ONLY(4 * M)
3601     S390_ONLY(1 * M)
3602     SPARC_ONLY(4 * M);
3603 #endif // ZERO
3604 
3605   FILE *fp = fopen("/proc/meminfo", "r");
3606   if (fp) {
3607     while (!feof(fp)) {
3608       int x = 0;
3609       char buf[16];
3610       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3611         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3612           large_page_size = x * K;
3613           break;
3614         }
3615       } else {
3616         // skip to next line
3617         for (;;) {
3618           int ch = fgetc(fp);
3619           if (ch == EOF || ch == (int)'\n') break;
3620         }
3621       }
3622     }
3623     fclose(fp);
3624   }
3625 
3626   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3627     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3628             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3629             proper_unit_for_byte_size(large_page_size));
3630   }
3631 
3632   return large_page_size;
3633 }
3634 
3635 size_t os::Linux::setup_large_page_size() {
3636   _large_page_size = Linux::find_large_page_size();
3637   const size_t default_page_size = (size_t)Linux::page_size();
3638   if (_large_page_size > default_page_size) {
3639     _page_sizes[0] = _large_page_size;
3640     _page_sizes[1] = default_page_size;
3641     _page_sizes[2] = 0;
3642   }
3643 
3644   return _large_page_size;
3645 }
3646 
3647 bool os::Linux::setup_large_page_type(size_t page_size) {
3648   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3649       FLAG_IS_DEFAULT(UseSHM) &&
3650       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3651 
3652     // The type of large pages has not been specified by the user.
3653 
3654     // Try UseHugeTLBFS and then UseSHM.
3655     UseHugeTLBFS = UseSHM = true;
3656 
3657     // Don't try UseTransparentHugePages since there are known
3658     // performance issues with it turned on. This might change in the future.
3659     UseTransparentHugePages = false;
3660   }
3661 
3662   if (UseTransparentHugePages) {
3663     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3664     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3665       UseHugeTLBFS = false;
3666       UseSHM = false;
3667       return true;
3668     }
3669     UseTransparentHugePages = false;
3670   }
3671 
3672   if (UseHugeTLBFS) {
3673     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3674     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3675       UseSHM = false;
3676       return true;
3677     }
3678     UseHugeTLBFS = false;
3679   }
3680 
3681   return UseSHM;
3682 }
3683 
3684 void os::large_page_init() {
3685   if (!UseLargePages &&
3686       !UseTransparentHugePages &&
3687       !UseHugeTLBFS &&
3688       !UseSHM) {
3689     // Not using large pages.
3690     return;
3691   }
3692 
3693   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3694     // The user explicitly turned off large pages.
3695     // Ignore the rest of the large pages flags.
3696     UseTransparentHugePages = false;
3697     UseHugeTLBFS = false;
3698     UseSHM = false;
3699     return;
3700   }
3701 
3702   size_t large_page_size = Linux::setup_large_page_size();
3703   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3704 
3705   set_coredump_filter(LARGEPAGES_BIT);
3706 }
3707 
3708 #ifndef SHM_HUGETLB
3709   #define SHM_HUGETLB 04000
3710 #endif
3711 
3712 #define shm_warning_format(format, ...)              \
3713   do {                                               \
3714     if (UseLargePages &&                             \
3715         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3716          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3717          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3718       warning(format, __VA_ARGS__);                  \
3719     }                                                \
3720   } while (0)
3721 
3722 #define shm_warning(str) shm_warning_format("%s", str)
3723 
3724 #define shm_warning_with_errno(str)                \
3725   do {                                             \
3726     int err = errno;                               \
3727     shm_warning_format(str " (error = %d)", err);  \
3728   } while (0)
3729 
3730 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3731   assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3732 
3733   if (!is_aligned(alignment, SHMLBA)) {
3734     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3735     return NULL;
3736   }
3737 
3738   // To ensure that we get 'alignment' aligned memory from shmat,
3739   // we pre-reserve aligned virtual memory and then attach to that.
3740 
3741   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3742   if (pre_reserved_addr == NULL) {
3743     // Couldn't pre-reserve aligned memory.
3744     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3745     return NULL;
3746   }
3747 
3748   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3749   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3750 
3751   if ((intptr_t)addr == -1) {
3752     int err = errno;
3753     shm_warning_with_errno("Failed to attach shared memory.");
3754 
3755     assert(err != EACCES, "Unexpected error");
3756     assert(err != EIDRM,  "Unexpected error");
3757     assert(err != EINVAL, "Unexpected error");
3758 
3759     // Since we don't know if the kernel unmapped the pre-reserved memory area
3760     // we can't unmap it, since that would potentially unmap memory that was
3761     // mapped from other threads.
3762     return NULL;
3763   }
3764 
3765   return addr;
3766 }
3767 
3768 static char* shmat_at_address(int shmid, char* req_addr) {
3769   if (!is_aligned(req_addr, SHMLBA)) {
3770     assert(false, "Requested address needs to be SHMLBA aligned");
3771     return NULL;
3772   }
3773 
3774   char* addr = (char*)shmat(shmid, req_addr, 0);
3775 
3776   if ((intptr_t)addr == -1) {
3777     shm_warning_with_errno("Failed to attach shared memory.");
3778     return NULL;
3779   }
3780 
3781   return addr;
3782 }
3783 
3784 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3785   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3786   if (req_addr != NULL) {
3787     assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3788     assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3789     return shmat_at_address(shmid, req_addr);
3790   }
3791 
3792   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3793   // return large page size aligned memory addresses when req_addr == NULL.
3794   // However, if the alignment is larger than the large page size, we have
3795   // to manually ensure that the memory returned is 'alignment' aligned.
3796   if (alignment > os::large_page_size()) {
3797     assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3798     return shmat_with_alignment(shmid, bytes, alignment);
3799   } else {
3800     return shmat_at_address(shmid, NULL);
3801   }
3802 }
3803 
3804 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3805                                             char* req_addr, bool exec) {
3806   // "exec" is passed in but not used.  Creating the shared image for
3807   // the code cache doesn't have an SHM_X executable permission to check.
3808   assert(UseLargePages && UseSHM, "only for SHM large pages");
3809   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3810   assert(is_aligned(req_addr, alignment), "Unaligned address");
3811 
3812   if (!is_aligned(bytes, os::large_page_size())) {
3813     return NULL; // Fallback to small pages.
3814   }
3815 
3816   // Create a large shared memory region to attach to based on size.
3817   // Currently, size is the total size of the heap.
3818   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3819   if (shmid == -1) {
3820     // Possible reasons for shmget failure:
3821     // 1. shmmax is too small for Java heap.
3822     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3823     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3824     // 2. not enough large page memory.
3825     //    > check available large pages: cat /proc/meminfo
3826     //    > increase amount of large pages:
3827     //          echo new_value > /proc/sys/vm/nr_hugepages
3828     //      Note 1: different Linux may use different name for this property,
3829     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3830     //      Note 2: it's possible there's enough physical memory available but
3831     //            they are so fragmented after a long run that they can't
3832     //            coalesce into large pages. Try to reserve large pages when
3833     //            the system is still "fresh".
3834     shm_warning_with_errno("Failed to reserve shared memory.");
3835     return NULL;
3836   }
3837 
3838   // Attach to the region.
3839   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3840 
3841   // Remove shmid. If shmat() is successful, the actual shared memory segment
3842   // will be deleted when it's detached by shmdt() or when the process
3843   // terminates. If shmat() is not successful this will remove the shared
3844   // segment immediately.
3845   shmctl(shmid, IPC_RMID, NULL);
3846 
3847   return addr;
3848 }
3849 
3850 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3851                                         int error) {
3852   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3853 
3854   bool warn_on_failure = UseLargePages &&
3855       (!FLAG_IS_DEFAULT(UseLargePages) ||
3856        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3857        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3858 
3859   if (warn_on_failure) {
3860     char msg[128];
3861     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3862                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3863     warning("%s", msg);
3864   }
3865 }
3866 
3867 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3868                                                         char* req_addr,
3869                                                         bool exec) {
3870   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3871   assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3872   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3873 
3874   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3875   char* addr = (char*)::mmap(req_addr, bytes, prot,
3876                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3877                              -1, 0);
3878 
3879   if (addr == MAP_FAILED) {
3880     warn_on_large_pages_failure(req_addr, bytes, errno);
3881     return NULL;
3882   }
3883 
3884   assert(is_aligned(addr, os::large_page_size()), "Must be");
3885 
3886   return addr;
3887 }
3888 
3889 // Reserve memory using mmap(MAP_HUGETLB).
3890 //  - bytes shall be a multiple of alignment.
3891 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3892 //  - alignment sets the alignment at which memory shall be allocated.
3893 //     It must be a multiple of allocation granularity.
3894 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3895 //  req_addr or NULL.
3896 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3897                                                          size_t alignment,
3898                                                          char* req_addr,
3899                                                          bool exec) {
3900   size_t large_page_size = os::large_page_size();
3901   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3902 
3903   assert(is_aligned(req_addr, alignment), "Must be");
3904   assert(is_aligned(bytes, alignment), "Must be");
3905 
3906   // First reserve - but not commit - the address range in small pages.
3907   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3908 
3909   if (start == NULL) {
3910     return NULL;
3911   }
3912 
3913   assert(is_aligned(start, alignment), "Must be");
3914 
3915   char* end = start + bytes;
3916 
3917   // Find the regions of the allocated chunk that can be promoted to large pages.
3918   char* lp_start = align_up(start, large_page_size);
3919   char* lp_end   = align_down(end, large_page_size);
3920 
3921   size_t lp_bytes = lp_end - lp_start;
3922 
3923   assert(is_aligned(lp_bytes, large_page_size), "Must be");
3924 
3925   if (lp_bytes == 0) {
3926     // The mapped region doesn't even span the start and the end of a large page.
3927     // Fall back to allocate a non-special area.
3928     ::munmap(start, end - start);
3929     return NULL;
3930   }
3931 
3932   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3933 
3934   void* result;
3935 
3936   // Commit small-paged leading area.
3937   if (start != lp_start) {
3938     result = ::mmap(start, lp_start - start, prot,
3939                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3940                     -1, 0);
3941     if (result == MAP_FAILED) {
3942       ::munmap(lp_start, end - lp_start);
3943       return NULL;
3944     }
3945   }
3946 
3947   // Commit large-paged area.
3948   result = ::mmap(lp_start, lp_bytes, prot,
3949                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3950                   -1, 0);
3951   if (result == MAP_FAILED) {
3952     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3953     // If the mmap above fails, the large pages region will be unmapped and we
3954     // have regions before and after with small pages. Release these regions.
3955     //
3956     // |  mapped  |  unmapped  |  mapped  |
3957     // ^          ^            ^          ^
3958     // start      lp_start     lp_end     end
3959     //
3960     ::munmap(start, lp_start - start);
3961     ::munmap(lp_end, end - lp_end);
3962     return NULL;
3963   }
3964 
3965   // Commit small-paged trailing area.
3966   if (lp_end != end) {
3967     result = ::mmap(lp_end, end - lp_end, prot,
3968                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3969                     -1, 0);
3970     if (result == MAP_FAILED) {
3971       ::munmap(start, lp_end - start);
3972       return NULL;
3973     }
3974   }
3975 
3976   return start;
3977 }
3978 
3979 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3980                                                    size_t alignment,
3981                                                    char* req_addr,
3982                                                    bool exec) {
3983   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3984   assert(is_aligned(req_addr, alignment), "Must be");
3985   assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3986   assert(is_power_of_2(os::large_page_size()), "Must be");
3987   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3988 
3989   if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3990     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3991   } else {
3992     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3993   }
3994 }
3995 
3996 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3997                                  char* req_addr, bool exec) {
3998   assert(UseLargePages, "only for large pages");
3999 
4000   char* addr;
4001   if (UseSHM) {
4002     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
4003   } else {
4004     assert(UseHugeTLBFS, "must be");
4005     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
4006   }
4007 
4008   if (addr != NULL) {
4009     if (UseNUMAInterleaving) {
4010       numa_make_global(addr, bytes);
4011     }
4012 
4013     // The memory is committed
4014     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
4015   }
4016 
4017   return addr;
4018 }
4019 
4020 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
4021   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
4022   return shmdt(base) == 0;
4023 }
4024 
4025 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
4026   return pd_release_memory(base, bytes);
4027 }
4028 
4029 bool os::release_memory_special(char* base, size_t bytes) {
4030   bool res;
4031   if (MemTracker::tracking_level() > NMT_minimal) {
4032     Tracker tkr(Tracker::release);
4033     res = os::Linux::release_memory_special_impl(base, bytes);
4034     if (res) {
4035       tkr.record((address)base, bytes);
4036     }
4037 
4038   } else {
4039     res = os::Linux::release_memory_special_impl(base, bytes);
4040   }
4041   return res;
4042 }
4043 
4044 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
4045   assert(UseLargePages, "only for large pages");
4046   bool res;
4047 
4048   if (UseSHM) {
4049     res = os::Linux::release_memory_special_shm(base, bytes);
4050   } else {
4051     assert(UseHugeTLBFS, "must be");
4052     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
4053   }
4054   return res;
4055 }
4056 
4057 size_t os::large_page_size() {
4058   return _large_page_size;
4059 }
4060 
4061 // With SysV SHM the entire memory region must be allocated as shared
4062 // memory.
4063 // HugeTLBFS allows application to commit large page memory on demand.
4064 // However, when committing memory with HugeTLBFS fails, the region
4065 // that was supposed to be committed will lose the old reservation
4066 // and allow other threads to steal that memory region. Because of this
4067 // behavior we can't commit HugeTLBFS memory.
4068 bool os::can_commit_large_page_memory() {
4069   return UseTransparentHugePages;
4070 }
4071 
4072 bool os::can_execute_large_page_memory() {
4073   return UseTransparentHugePages || UseHugeTLBFS;
4074 }
4075 
4076 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
4077   assert(file_desc >= 0, "file_desc is not valid");
4078   char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
4079   if (result != NULL) {
4080     if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4081       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4082     }
4083   }
4084   return result;
4085 }
4086 
4087 // Reserve memory at an arbitrary address, only if that area is
4088 // available (and not reserved for something else).
4089 
4090 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4091   const int max_tries = 10;
4092   char* base[max_tries];
4093   size_t size[max_tries];
4094   const size_t gap = 0x000000;
4095 
4096   // Assert only that the size is a multiple of the page size, since
4097   // that's all that mmap requires, and since that's all we really know
4098   // about at this low abstraction level.  If we need higher alignment,
4099   // we can either pass an alignment to this method or verify alignment
4100   // in one of the methods further up the call chain.  See bug 5044738.
4101   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4102 
4103   // Repeatedly allocate blocks until the block is allocated at the
4104   // right spot.
4105 
4106   // Linux mmap allows caller to pass an address as hint; give it a try first,
4107   // if kernel honors the hint then we can return immediately.
4108   char * addr = anon_mmap(requested_addr, bytes, false);
4109   if (addr == requested_addr) {
4110     return requested_addr;
4111   }
4112 
4113   if (addr != NULL) {
4114     // mmap() is successful but it fails to reserve at the requested address
4115     anon_munmap(addr, bytes);
4116   }
4117 
4118   int i;
4119   for (i = 0; i < max_tries; ++i) {
4120     base[i] = reserve_memory(bytes);
4121 
4122     if (base[i] != NULL) {
4123       // Is this the block we wanted?
4124       if (base[i] == requested_addr) {
4125         size[i] = bytes;
4126         break;
4127       }
4128 
4129       // Does this overlap the block we wanted? Give back the overlapped
4130       // parts and try again.
4131 
4132       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
4133       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
4134         unmap_memory(base[i], top_overlap);
4135         base[i] += top_overlap;
4136         size[i] = bytes - top_overlap;
4137       } else {
4138         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
4139         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
4140           unmap_memory(requested_addr, bottom_overlap);
4141           size[i] = bytes - bottom_overlap;
4142         } else {
4143           size[i] = bytes;
4144         }
4145       }
4146     }
4147   }
4148 
4149   // Give back the unused reserved pieces.
4150 
4151   for (int j = 0; j < i; ++j) {
4152     if (base[j] != NULL) {
4153       unmap_memory(base[j], size[j]);
4154     }
4155   }
4156 
4157   if (i < max_tries) {
4158     return requested_addr;
4159   } else {
4160     return NULL;
4161   }
4162 }
4163 
4164 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4165 void os::infinite_sleep() {
4166   while (true) {    // sleep forever ...
4167     ::sleep(100);   // ... 100 seconds at a time
4168   }
4169 }
4170 
4171 // Used to convert frequent JVM_Yield() to nops
4172 bool os::dont_yield() {
4173   return DontYieldALot;
4174 }
4175 
4176 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4177 // actually give up the CPU. Since skip buddy (v2.6.28):
4178 //
4179 // * Sets the yielding task as skip buddy for current CPU's run queue.
4180 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4181 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4182 //
4183 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4184 // getting re-scheduled immediately.
4185 //
4186 void os::naked_yield() {
4187   sched_yield();
4188 }
4189 
4190 ////////////////////////////////////////////////////////////////////////////////
4191 // thread priority support
4192 
4193 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4194 // only supports dynamic priority, static priority must be zero. For real-time
4195 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4196 // However, for large multi-threaded applications, SCHED_RR is not only slower
4197 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4198 // of 5 runs - Sep 2005).
4199 //
4200 // The following code actually changes the niceness of kernel-thread/LWP. It
4201 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4202 // not the entire user process, and user level threads are 1:1 mapped to kernel
4203 // threads. It has always been the case, but could change in the future. For
4204 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4205 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4206 // (e.g., root privilege or CAP_SYS_NICE capability).
4207 
4208 int os::java_to_os_priority[CriticalPriority + 1] = {
4209   19,              // 0 Entry should never be used
4210 
4211    4,              // 1 MinPriority
4212    3,              // 2
4213    2,              // 3
4214 
4215    1,              // 4
4216    0,              // 5 NormPriority
4217   -1,              // 6
4218 
4219   -2,              // 7
4220   -3,              // 8
4221   -4,              // 9 NearMaxPriority
4222 
4223   -5,              // 10 MaxPriority
4224 
4225   -5               // 11 CriticalPriority
4226 };
4227 
4228 static int prio_init() {
4229   if (ThreadPriorityPolicy == 1) {
4230     if (geteuid() != 0) {
4231       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4232         warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4233                 "e.g., being the root user. If the necessary permission is not " \
4234                 "possessed, changes to priority will be silently ignored.");
4235       }
4236     }
4237   }
4238   if (UseCriticalJavaThreadPriority) {
4239     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4240   }
4241   return 0;
4242 }
4243 
4244 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4245   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4246 
4247   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4248   return (ret == 0) ? OS_OK : OS_ERR;
4249 }
4250 
4251 OSReturn os::get_native_priority(const Thread* const thread,
4252                                  int *priority_ptr) {
4253   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4254     *priority_ptr = java_to_os_priority[NormPriority];
4255     return OS_OK;
4256   }
4257 
4258   errno = 0;
4259   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4260   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4261 }
4262 
4263 ////////////////////////////////////////////////////////////////////////////////
4264 // suspend/resume support
4265 
4266 //  The low-level signal-based suspend/resume support is a remnant from the
4267 //  old VM-suspension that used to be for java-suspension, safepoints etc,
4268 //  within hotspot. Currently used by JFR's OSThreadSampler
4269 //
4270 //  The remaining code is greatly simplified from the more general suspension
4271 //  code that used to be used.
4272 //
4273 //  The protocol is quite simple:
4274 //  - suspend:
4275 //      - sends a signal to the target thread
4276 //      - polls the suspend state of the osthread using a yield loop
4277 //      - target thread signal handler (SR_handler) sets suspend state
4278 //        and blocks in sigsuspend until continued
4279 //  - resume:
4280 //      - sets target osthread state to continue
4281 //      - sends signal to end the sigsuspend loop in the SR_handler
4282 //
4283 //  Note that the SR_lock plays no role in this suspend/resume protocol,
4284 //  but is checked for NULL in SR_handler as a thread termination indicator.
4285 //  The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4286 //
4287 //  Note that resume_clear_context() and suspend_save_context() are needed
4288 //  by SR_handler(), so that fetch_frame_from_ucontext() works,
4289 //  which in part is used by:
4290 //    - Forte Analyzer: AsyncGetCallTrace()
4291 //    - StackBanging: get_frame_at_stack_banging_point()
4292 
4293 static void resume_clear_context(OSThread *osthread) {
4294   osthread->set_ucontext(NULL);
4295   osthread->set_siginfo(NULL);
4296 }
4297 
4298 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4299                                  ucontext_t* context) {
4300   osthread->set_ucontext(context);
4301   osthread->set_siginfo(siginfo);
4302 }
4303 
4304 // Handler function invoked when a thread's execution is suspended or
4305 // resumed. We have to be careful that only async-safe functions are
4306 // called here (Note: most pthread functions are not async safe and
4307 // should be avoided.)
4308 //
4309 // Note: sigwait() is a more natural fit than sigsuspend() from an
4310 // interface point of view, but sigwait() prevents the signal hander
4311 // from being run. libpthread would get very confused by not having
4312 // its signal handlers run and prevents sigwait()'s use with the
4313 // mutex granting granting signal.
4314 //
4315 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4316 //
4317 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4318   // Save and restore errno to avoid confusing native code with EINTR
4319   // after sigsuspend.
4320   int old_errno = errno;
4321 
4322   Thread* thread = Thread::current_or_null_safe();
4323   assert(thread != NULL, "Missing current thread in SR_handler");
4324 
4325   // On some systems we have seen signal delivery get "stuck" until the signal
4326   // mask is changed as part of thread termination. Check that the current thread
4327   // has not already terminated (via SR_lock()) - else the following assertion
4328   // will fail because the thread is no longer a JavaThread as the ~JavaThread
4329   // destructor has completed.
4330 
4331   if (thread->SR_lock() == NULL) {
4332     return;
4333   }
4334 
4335   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4336 
4337   OSThread* osthread = thread->osthread();
4338 
4339   os::SuspendResume::State current = osthread->sr.state();
4340   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4341     suspend_save_context(osthread, siginfo, context);
4342 
4343     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4344     os::SuspendResume::State state = osthread->sr.suspended();
4345     if (state == os::SuspendResume::SR_SUSPENDED) {
4346       sigset_t suspend_set;  // signals for sigsuspend()
4347       sigemptyset(&suspend_set);
4348       // get current set of blocked signals and unblock resume signal
4349       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4350       sigdelset(&suspend_set, SR_signum);
4351 
4352       sr_semaphore.signal();
4353       // wait here until we are resumed
4354       while (1) {
4355         sigsuspend(&suspend_set);
4356 
4357         os::SuspendResume::State result = osthread->sr.running();
4358         if (result == os::SuspendResume::SR_RUNNING) {
4359           sr_semaphore.signal();
4360           break;
4361         }
4362       }
4363 
4364     } else if (state == os::SuspendResume::SR_RUNNING) {
4365       // request was cancelled, continue
4366     } else {
4367       ShouldNotReachHere();
4368     }
4369 
4370     resume_clear_context(osthread);
4371   } else if (current == os::SuspendResume::SR_RUNNING) {
4372     // request was cancelled, continue
4373   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4374     // ignore
4375   } else {
4376     // ignore
4377   }
4378 
4379   errno = old_errno;
4380 }
4381 
4382 static int SR_initialize() {
4383   struct sigaction act;
4384   char *s;
4385 
4386   // Get signal number to use for suspend/resume
4387   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4388     int sig = ::strtol(s, 0, 10);
4389     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4390         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4391       SR_signum = sig;
4392     } else {
4393       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4394               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4395     }
4396   }
4397 
4398   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4399          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4400 
4401   sigemptyset(&SR_sigset);
4402   sigaddset(&SR_sigset, SR_signum);
4403 
4404   // Set up signal handler for suspend/resume
4405   act.sa_flags = SA_RESTART|SA_SIGINFO;
4406   act.sa_handler = (void (*)(int)) SR_handler;
4407 
4408   // SR_signum is blocked by default.
4409   // 4528190 - We also need to block pthread restart signal (32 on all
4410   // supported Linux platforms). Note that LinuxThreads need to block
4411   // this signal for all threads to work properly. So we don't have
4412   // to use hard-coded signal number when setting up the mask.
4413   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4414 
4415   if (sigaction(SR_signum, &act, 0) == -1) {
4416     return -1;
4417   }
4418 
4419   // Save signal flag
4420   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4421   return 0;
4422 }
4423 
4424 static int sr_notify(OSThread* osthread) {
4425   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4426   assert_status(status == 0, status, "pthread_kill");
4427   return status;
4428 }
4429 
4430 // "Randomly" selected value for how long we want to spin
4431 // before bailing out on suspending a thread, also how often
4432 // we send a signal to a thread we want to resume
4433 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4434 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4435 
4436 // returns true on success and false on error - really an error is fatal
4437 // but this seems the normal response to library errors
4438 static bool do_suspend(OSThread* osthread) {
4439   assert(osthread->sr.is_running(), "thread should be running");
4440   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4441 
4442   // mark as suspended and send signal
4443   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4444     // failed to switch, state wasn't running?
4445     ShouldNotReachHere();
4446     return false;
4447   }
4448 
4449   if (sr_notify(osthread) != 0) {
4450     ShouldNotReachHere();
4451   }
4452 
4453   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4454   while (true) {
4455     if (sr_semaphore.timedwait(2)) {
4456       break;
4457     } else {
4458       // timeout
4459       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4460       if (cancelled == os::SuspendResume::SR_RUNNING) {
4461         return false;
4462       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4463         // make sure that we consume the signal on the semaphore as well
4464         sr_semaphore.wait();
4465         break;
4466       } else {
4467         ShouldNotReachHere();
4468         return false;
4469       }
4470     }
4471   }
4472 
4473   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4474   return true;
4475 }
4476 
4477 static void do_resume(OSThread* osthread) {
4478   assert(osthread->sr.is_suspended(), "thread should be suspended");
4479   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4480 
4481   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4482     // failed to switch to WAKEUP_REQUEST
4483     ShouldNotReachHere();
4484     return;
4485   }
4486 
4487   while (true) {
4488     if (sr_notify(osthread) == 0) {
4489       if (sr_semaphore.timedwait(2)) {
4490         if (osthread->sr.is_running()) {
4491           return;
4492         }
4493       }
4494     } else {
4495       ShouldNotReachHere();
4496     }
4497   }
4498 
4499   guarantee(osthread->sr.is_running(), "Must be running!");
4500 }
4501 
4502 ///////////////////////////////////////////////////////////////////////////////////
4503 // signal handling (except suspend/resume)
4504 
4505 // This routine may be used by user applications as a "hook" to catch signals.
4506 // The user-defined signal handler must pass unrecognized signals to this
4507 // routine, and if it returns true (non-zero), then the signal handler must
4508 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4509 // routine will never retun false (zero), but instead will execute a VM panic
4510 // routine kill the process.
4511 //
4512 // If this routine returns false, it is OK to call it again.  This allows
4513 // the user-defined signal handler to perform checks either before or after
4514 // the VM performs its own checks.  Naturally, the user code would be making
4515 // a serious error if it tried to handle an exception (such as a null check
4516 // or breakpoint) that the VM was generating for its own correct operation.
4517 //
4518 // This routine may recognize any of the following kinds of signals:
4519 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4520 // It should be consulted by handlers for any of those signals.
4521 //
4522 // The caller of this routine must pass in the three arguments supplied
4523 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4524 // field of the structure passed to sigaction().  This routine assumes that
4525 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4526 //
4527 // Note that the VM will print warnings if it detects conflicting signal
4528 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4529 //
4530 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4531                                                  siginfo_t* siginfo,
4532                                                  void* ucontext,
4533                                                  int abort_if_unrecognized);
4534 
4535 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4536   assert(info != NULL && uc != NULL, "it must be old kernel");
4537   int orig_errno = errno;  // Preserve errno value over signal handler.
4538   JVM_handle_linux_signal(sig, info, uc, true);
4539   errno = orig_errno;
4540 }
4541 
4542 
4543 // This boolean allows users to forward their own non-matching signals
4544 // to JVM_handle_linux_signal, harmlessly.
4545 bool os::Linux::signal_handlers_are_installed = false;
4546 
4547 // For signal-chaining
4548 bool os::Linux::libjsig_is_loaded = false;
4549 typedef struct sigaction *(*get_signal_t)(int);
4550 get_signal_t os::Linux::get_signal_action = NULL;
4551 
4552 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4553   struct sigaction *actp = NULL;
4554 
4555   if (libjsig_is_loaded) {
4556     // Retrieve the old signal handler from libjsig
4557     actp = (*get_signal_action)(sig);
4558   }
4559   if (actp == NULL) {
4560     // Retrieve the preinstalled signal handler from jvm
4561     actp = os::Posix::get_preinstalled_handler(sig);
4562   }
4563 
4564   return actp;
4565 }
4566 
4567 static bool call_chained_handler(struct sigaction *actp, int sig,
4568                                  siginfo_t *siginfo, void *context) {
4569   // Call the old signal handler
4570   if (actp->sa_handler == SIG_DFL) {
4571     // It's more reasonable to let jvm treat it as an unexpected exception
4572     // instead of taking the default action.
4573     return false;
4574   } else if (actp->sa_handler != SIG_IGN) {
4575     if ((actp->sa_flags & SA_NODEFER) == 0) {
4576       // automaticlly block the signal
4577       sigaddset(&(actp->sa_mask), sig);
4578     }
4579 
4580     sa_handler_t hand = NULL;
4581     sa_sigaction_t sa = NULL;
4582     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4583     // retrieve the chained handler
4584     if (siginfo_flag_set) {
4585       sa = actp->sa_sigaction;
4586     } else {
4587       hand = actp->sa_handler;
4588     }
4589 
4590     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4591       actp->sa_handler = SIG_DFL;
4592     }
4593 
4594     // try to honor the signal mask
4595     sigset_t oset;
4596     sigemptyset(&oset);
4597     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4598 
4599     // call into the chained handler
4600     if (siginfo_flag_set) {
4601       (*sa)(sig, siginfo, context);
4602     } else {
4603       (*hand)(sig);
4604     }
4605 
4606     // restore the signal mask
4607     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4608   }
4609   // Tell jvm's signal handler the signal is taken care of.
4610   return true;
4611 }
4612 
4613 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4614   bool chained = false;
4615   // signal-chaining
4616   if (UseSignalChaining) {
4617     struct sigaction *actp = get_chained_signal_action(sig);
4618     if (actp != NULL) {
4619       chained = call_chained_handler(actp, sig, siginfo, context);
4620     }
4621   }
4622   return chained;
4623 }
4624 
4625 // for diagnostic
4626 int sigflags[NSIG];
4627 
4628 int os::Linux::get_our_sigflags(int sig) {
4629   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4630   return sigflags[sig];
4631 }
4632 
4633 void os::Linux::set_our_sigflags(int sig, int flags) {
4634   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4635   if (sig > 0 && sig < NSIG) {
4636     sigflags[sig] = flags;
4637   }
4638 }
4639 
4640 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4641   // Check for overwrite.
4642   struct sigaction oldAct;
4643   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4644 
4645   void* oldhand = oldAct.sa_sigaction
4646                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4647                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4648   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4649       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4650       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4651     if (AllowUserSignalHandlers || !set_installed) {
4652       // Do not overwrite; user takes responsibility to forward to us.
4653       return;
4654     } else if (UseSignalChaining) {
4655       // save the old handler in jvm
4656       os::Posix::save_preinstalled_handler(sig, oldAct);
4657       // libjsig also interposes the sigaction() call below and saves the
4658       // old sigaction on it own.
4659     } else {
4660       fatal("Encountered unexpected pre-existing sigaction handler "
4661             "%#lx for signal %d.", (long)oldhand, sig);
4662     }
4663   }
4664 
4665   struct sigaction sigAct;
4666   sigfillset(&(sigAct.sa_mask));
4667   sigAct.sa_handler = SIG_DFL;
4668   if (!set_installed) {
4669     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4670   } else {
4671     sigAct.sa_sigaction = signalHandler;
4672     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4673   }
4674   // Save flags, which are set by ours
4675   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4676   sigflags[sig] = sigAct.sa_flags;
4677 
4678   int ret = sigaction(sig, &sigAct, &oldAct);
4679   assert(ret == 0, "check");
4680 
4681   void* oldhand2  = oldAct.sa_sigaction
4682                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4683                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4684   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4685 }
4686 
4687 // install signal handlers for signals that HotSpot needs to
4688 // handle in order to support Java-level exception handling.
4689 
4690 void os::Linux::install_signal_handlers() {
4691   if (!signal_handlers_are_installed) {
4692     signal_handlers_are_installed = true;
4693 
4694     // signal-chaining
4695     typedef void (*signal_setting_t)();
4696     signal_setting_t begin_signal_setting = NULL;
4697     signal_setting_t end_signal_setting = NULL;
4698     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4699                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4700     if (begin_signal_setting != NULL) {
4701       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4702                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4703       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4704                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4705       libjsig_is_loaded = true;
4706       assert(UseSignalChaining, "should enable signal-chaining");
4707     }
4708     if (libjsig_is_loaded) {
4709       // Tell libjsig jvm is setting signal handlers
4710       (*begin_signal_setting)();
4711     }
4712 
4713     set_signal_handler(SIGSEGV, true);
4714     set_signal_handler(SIGPIPE, true);
4715     set_signal_handler(SIGBUS, true);
4716     set_signal_handler(SIGILL, true);
4717     set_signal_handler(SIGFPE, true);
4718 #if defined(PPC64)
4719     set_signal_handler(SIGTRAP, true);
4720 #endif
4721     set_signal_handler(SIGXFSZ, true);
4722 
4723     if (libjsig_is_loaded) {
4724       // Tell libjsig jvm finishes setting signal handlers
4725       (*end_signal_setting)();
4726     }
4727 
4728     // We don't activate signal checker if libjsig is in place, we trust ourselves
4729     // and if UserSignalHandler is installed all bets are off.
4730     // Log that signal checking is off only if -verbose:jni is specified.
4731     if (CheckJNICalls) {
4732       if (libjsig_is_loaded) {
4733         if (PrintJNIResolving) {
4734           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4735         }
4736         check_signals = false;
4737       }
4738       if (AllowUserSignalHandlers) {
4739         if (PrintJNIResolving) {
4740           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4741         }
4742         check_signals = false;
4743       }
4744     }
4745   }
4746 }
4747 
4748 // This is the fastest way to get thread cpu time on Linux.
4749 // Returns cpu time (user+sys) for any thread, not only for current.
4750 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4751 // It might work on 2.6.10+ with a special kernel/glibc patch.
4752 // For reference, please, see IEEE Std 1003.1-2004:
4753 //   http://www.unix.org/single_unix_specification
4754 
4755 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4756   struct timespec tp;
4757   int rc = os::Posix::clock_gettime(clockid, &tp);
4758   assert(rc == 0, "clock_gettime is expected to return 0 code");
4759 
4760   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4761 }
4762 
4763 void os::Linux::initialize_os_info() {
4764   assert(_os_version == 0, "OS info already initialized");
4765 
4766   struct utsname _uname;
4767 
4768   uint32_t major;
4769   uint32_t minor;
4770   uint32_t fix;
4771 
4772   int rc;
4773 
4774   // Kernel version is unknown if
4775   // verification below fails.
4776   _os_version = 0x01000000;
4777 
4778   rc = uname(&_uname);
4779   if (rc != -1) {
4780 
4781     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4782     if (rc == 3) {
4783 
4784       if (major < 256 && minor < 256 && fix < 256) {
4785         // Kernel version format is as expected,
4786         // set it overriding unknown state.
4787         _os_version = (major << 16) |
4788                       (minor << 8 ) |
4789                       (fix   << 0 ) ;
4790       }
4791     }
4792   }
4793 }
4794 
4795 uint32_t os::Linux::os_version() {
4796   assert(_os_version != 0, "not initialized");
4797   return _os_version & 0x00FFFFFF;
4798 }
4799 
4800 bool os::Linux::os_version_is_known() {
4801   assert(_os_version != 0, "not initialized");
4802   return _os_version & 0x01000000 ? false : true;
4803 }
4804 
4805 /////
4806 // glibc on Linux platform uses non-documented flag
4807 // to indicate, that some special sort of signal
4808 // trampoline is used.
4809 // We will never set this flag, and we should
4810 // ignore this flag in our diagnostic
4811 #ifdef SIGNIFICANT_SIGNAL_MASK
4812   #undef SIGNIFICANT_SIGNAL_MASK
4813 #endif
4814 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4815 
4816 static const char* get_signal_handler_name(address handler,
4817                                            char* buf, int buflen) {
4818   int offset = 0;
4819   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4820   if (found) {
4821     // skip directory names
4822     const char *p1, *p2;
4823     p1 = buf;
4824     size_t len = strlen(os::file_separator());
4825     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4826     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4827   } else {
4828     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4829   }
4830   return buf;
4831 }
4832 
4833 static void print_signal_handler(outputStream* st, int sig,
4834                                  char* buf, size_t buflen) {
4835   struct sigaction sa;
4836 
4837   sigaction(sig, NULL, &sa);
4838 
4839   // See comment for SIGNIFICANT_SIGNAL_MASK define
4840   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4841 
4842   st->print("%s: ", os::exception_name(sig, buf, buflen));
4843 
4844   address handler = (sa.sa_flags & SA_SIGINFO)
4845     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4846     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4847 
4848   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4849     st->print("SIG_DFL");
4850   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4851     st->print("SIG_IGN");
4852   } else {
4853     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4854   }
4855 
4856   st->print(", sa_mask[0]=");
4857   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4858 
4859   address rh = VMError::get_resetted_sighandler(sig);
4860   // May be, handler was resetted by VMError?
4861   if (rh != NULL) {
4862     handler = rh;
4863     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4864   }
4865 
4866   st->print(", sa_flags=");
4867   os::Posix::print_sa_flags(st, sa.sa_flags);
4868 
4869   // Check: is it our handler?
4870   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4871       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4872     // It is our signal handler
4873     // check for flags, reset system-used one!
4874     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4875       st->print(
4876                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4877                 os::Linux::get_our_sigflags(sig));
4878     }
4879   }
4880   st->cr();
4881 }
4882 
4883 
4884 #define DO_SIGNAL_CHECK(sig)                      \
4885   do {                                            \
4886     if (!sigismember(&check_signal_done, sig)) {  \
4887       os::Linux::check_signal_handler(sig);       \
4888     }                                             \
4889   } while (0)
4890 
4891 // This method is a periodic task to check for misbehaving JNI applications
4892 // under CheckJNI, we can add any periodic checks here
4893 
4894 void os::run_periodic_checks() {
4895   if (check_signals == false) return;
4896 
4897   // SEGV and BUS if overridden could potentially prevent
4898   // generation of hs*.log in the event of a crash, debugging
4899   // such a case can be very challenging, so we absolutely
4900   // check the following for a good measure:
4901   DO_SIGNAL_CHECK(SIGSEGV);
4902   DO_SIGNAL_CHECK(SIGILL);
4903   DO_SIGNAL_CHECK(SIGFPE);
4904   DO_SIGNAL_CHECK(SIGBUS);
4905   DO_SIGNAL_CHECK(SIGPIPE);
4906   DO_SIGNAL_CHECK(SIGXFSZ);
4907 #if defined(PPC64)
4908   DO_SIGNAL_CHECK(SIGTRAP);
4909 #endif
4910 
4911   // ReduceSignalUsage allows the user to override these handlers
4912   // see comments at the very top and jvm_md.h
4913   if (!ReduceSignalUsage) {
4914     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4915     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4916     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4917     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4918   }
4919 
4920   DO_SIGNAL_CHECK(SR_signum);
4921 }
4922 
4923 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4924 
4925 static os_sigaction_t os_sigaction = NULL;
4926 
4927 void os::Linux::check_signal_handler(int sig) {
4928   char buf[O_BUFLEN];
4929   address jvmHandler = NULL;
4930 
4931 
4932   struct sigaction act;
4933   if (os_sigaction == NULL) {
4934     // only trust the default sigaction, in case it has been interposed
4935     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4936     if (os_sigaction == NULL) return;
4937   }
4938 
4939   os_sigaction(sig, (struct sigaction*)NULL, &act);
4940 
4941 
4942   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4943 
4944   address thisHandler = (act.sa_flags & SA_SIGINFO)
4945     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4946     : CAST_FROM_FN_PTR(address, act.sa_handler);
4947 
4948 
4949   switch (sig) {
4950   case SIGSEGV:
4951   case SIGBUS:
4952   case SIGFPE:
4953   case SIGPIPE:
4954   case SIGILL:
4955   case SIGXFSZ:
4956     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4957     break;
4958 
4959   case SHUTDOWN1_SIGNAL:
4960   case SHUTDOWN2_SIGNAL:
4961   case SHUTDOWN3_SIGNAL:
4962   case BREAK_SIGNAL:
4963     jvmHandler = (address)user_handler();
4964     break;
4965 
4966   default:
4967     if (sig == SR_signum) {
4968       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4969     } else {
4970       return;
4971     }
4972     break;
4973   }
4974 
4975   if (thisHandler != jvmHandler) {
4976     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4977     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4978     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4979     // No need to check this sig any longer
4980     sigaddset(&check_signal_done, sig);
4981     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4982     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4983       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4984                     exception_name(sig, buf, O_BUFLEN));
4985     }
4986   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4987     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4988     tty->print("expected:");
4989     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4990     tty->cr();
4991     tty->print("  found:");
4992     os::Posix::print_sa_flags(tty, act.sa_flags);
4993     tty->cr();
4994     // No need to check this sig any longer
4995     sigaddset(&check_signal_done, sig);
4996   }
4997 
4998   // Dump all the signal
4999   if (sigismember(&check_signal_done, sig)) {
5000     print_signal_handlers(tty, buf, O_BUFLEN);
5001   }
5002 }
5003 
5004 extern void report_error(char* file_name, int line_no, char* title,
5005                          char* format, ...);
5006 
5007 // this is called _before_ most of the global arguments have been parsed
5008 void os::init(void) {
5009   char dummy;   // used to get a guess on initial stack address
5010 
5011   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5012 
5013   init_random(1234567);
5014 
5015   Linux::set_page_size(sysconf(_SC_PAGESIZE));
5016   if (Linux::page_size() == -1) {
5017     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
5018           os::strerror(errno));
5019   }
5020   init_page_sizes((size_t) Linux::page_size());
5021 
5022   Linux::initialize_system_info();
5023 
5024   Linux::initialize_os_info();
5025 
5026   os::Linux::CPUPerfTicks pticks;
5027   bool res = os::Linux::get_tick_information(&pticks, -1);
5028 
5029   if (res && pticks.has_steal_ticks) {
5030     has_initial_tick_info = true;
5031     initial_total_ticks = pticks.total;
5032     initial_steal_ticks = pticks.steal;
5033   }
5034 
5035   // _main_thread points to the thread that created/loaded the JVM.
5036   Linux::_main_thread = pthread_self();
5037 
5038   // retrieve entry point for pthread_setname_np
5039   Linux::_pthread_setname_np =
5040     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5041 
5042   os::Posix::init();
5043 
5044   initial_time_count = javaTimeNanos();
5045 
5046   // Always warn if no monotonic clock available
5047   if (!os::Posix::supports_monotonic_clock()) {
5048     warning("No monotonic clock was available - timed services may "    \
5049             "be adversely affected if the time-of-day clock changes");
5050   }
5051 }
5052 
5053 // To install functions for atexit system call
5054 extern "C" {
5055   static void perfMemory_exit_helper() {
5056     perfMemory_exit();
5057   }
5058 }
5059 
5060 void os::pd_init_container_support() {
5061   OSContainer::init();
5062 }
5063 
5064 void os::Linux::numa_init() {
5065 
5066   // Java can be invoked as
5067   // 1. Without numactl and heap will be allocated/configured on all nodes as
5068   //    per the system policy.
5069   // 2. With numactl --interleave:
5070   //      Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
5071   //      API for membind case bitmask is reset.
5072   //      Interleave is only hint and Kernel can fallback to other nodes if
5073   //      no memory is available on the target nodes.
5074   // 3. With numactl --membind:
5075   //      Use numa_get_membind(v2) API to get nodes bitmask. The same API for
5076   //      interleave case returns bitmask of all nodes.
5077   // numa_all_nodes_ptr holds bitmask of all nodes.
5078   // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
5079   // bitmask when externally configured to run on all or fewer nodes.
5080 
5081   if (!Linux::libnuma_init()) {
5082     UseNUMA = false;
5083   } else {
5084     if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
5085       // If there's only one node (they start from 0) or if the process
5086       // is bound explicitly to a single node using membind, disable NUMA.
5087       UseNUMA = false;
5088     } else {
5089 
5090       LogTarget(Info,os) log;
5091       LogStream ls(log);
5092 
5093       Linux::set_configured_numa_policy(Linux::identify_numa_policy());
5094 
5095       struct bitmask* bmp = Linux::_numa_membind_bitmask;
5096       const char* numa_mode = "membind";
5097 
5098       if (Linux::is_running_in_interleave_mode()) {
5099         bmp = Linux::_numa_interleave_bitmask;
5100         numa_mode = "interleave";
5101       }
5102 
5103       ls.print("UseNUMA is enabled and invoked in '%s' mode."
5104                " Heap will be configured using NUMA memory nodes:", numa_mode);
5105 
5106       for (int node = 0; node <= Linux::numa_max_node(); node++) {
5107         if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5108           ls.print(" %d", node);
5109         }
5110       }
5111     }
5112   }
5113 
5114   if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5115     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5116     // we can make the adaptive lgrp chunk resizing work. If the user specified both
5117     // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5118     // and disable adaptive resizing.
5119     if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5120       warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5121               "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5122       UseAdaptiveSizePolicy = false;
5123       UseAdaptiveNUMAChunkSizing = false;
5124     }
5125   }
5126 
5127   if (!UseNUMA && ForceNUMA) {
5128     UseNUMA = true;
5129   }
5130 }
5131 
5132 // this is called _after_ the global arguments have been parsed
5133 jint os::init_2(void) {
5134 
5135   // This could be set after os::Posix::init() but all platforms
5136   // have to set it the same so we have to mirror Solaris.
5137   DEBUG_ONLY(os::set_mutex_init_done();)
5138 
5139   os::Posix::init_2();
5140 
5141   Linux::fast_thread_clock_init();
5142 
5143   // initialize suspend/resume support - must do this before signal_sets_init()
5144   if (SR_initialize() != 0) {
5145     perror("SR_initialize failed");
5146     return JNI_ERR;
5147   }
5148 
5149   Linux::signal_sets_init();
5150   Linux::install_signal_handlers();
5151   // Initialize data for jdk.internal.misc.Signal
5152   if (!ReduceSignalUsage) {
5153     jdk_misc_signal_init();
5154   }
5155 
5156   // Check and sets minimum stack sizes against command line options
5157   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5158     return JNI_ERR;
5159   }
5160 
5161 #if defined(IA32)
5162   // Need to ensure we've determined the process's initial stack to
5163   // perform the workaround
5164   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5165   workaround_expand_exec_shield_cs_limit();
5166 #else
5167   suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5168   if (!suppress_primordial_thread_resolution) {
5169     Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5170   }
5171 #endif
5172 
5173   Linux::libpthread_init();
5174   Linux::sched_getcpu_init();
5175   log_info(os)("HotSpot is running with %s, %s",
5176                Linux::glibc_version(), Linux::libpthread_version());
5177 
5178   if (UseNUMA) {
5179     Linux::numa_init();
5180   }
5181 
5182   if (MaxFDLimit) {
5183     // set the number of file descriptors to max. print out error
5184     // if getrlimit/setrlimit fails but continue regardless.
5185     struct rlimit nbr_files;
5186     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5187     if (status != 0) {
5188       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5189     } else {
5190       nbr_files.rlim_cur = nbr_files.rlim_max;
5191       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5192       if (status != 0) {
5193         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5194       }
5195     }
5196   }
5197 
5198   // Initialize lock used to serialize thread creation (see os::create_thread)
5199   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5200 
5201   // at-exit methods are called in the reverse order of their registration.
5202   // atexit functions are called on return from main or as a result of a
5203   // call to exit(3C). There can be only 32 of these functions registered
5204   // and atexit() does not set errno.
5205 
5206   if (PerfAllowAtExitRegistration) {
5207     // only register atexit functions if PerfAllowAtExitRegistration is set.
5208     // atexit functions can be delayed until process exit time, which
5209     // can be problematic for embedded VM situations. Embedded VMs should
5210     // call DestroyJavaVM() to assure that VM resources are released.
5211 
5212     // note: perfMemory_exit_helper atexit function may be removed in
5213     // the future if the appropriate cleanup code can be added to the
5214     // VM_Exit VMOperation's doit method.
5215     if (atexit(perfMemory_exit_helper) != 0) {
5216       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5217     }
5218   }
5219 
5220   // initialize thread priority policy
5221   prio_init();
5222 
5223   if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) {
5224     set_coredump_filter(DAX_SHARED_BIT);
5225   }
5226 
5227   if (DumpPrivateMappingsInCore) {
5228     set_coredump_filter(FILE_BACKED_PVT_BIT);
5229   }
5230 
5231   if (DumpSharedMappingsInCore) {
5232     set_coredump_filter(FILE_BACKED_SHARED_BIT);
5233   }
5234 
5235   return JNI_OK;
5236 }
5237 
5238 // Mark the polling page as unreadable
5239 void os::make_polling_page_unreadable(void) {
5240   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5241     fatal("Could not disable polling page");
5242   }
5243 }
5244 
5245 // Mark the polling page as readable
5246 void os::make_polling_page_readable(void) {
5247   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5248     fatal("Could not enable polling page");
5249   }
5250 }
5251 
5252 // older glibc versions don't have this macro (which expands to
5253 // an optimized bit-counting function) so we have to roll our own
5254 #ifndef CPU_COUNT
5255 
5256 static int _cpu_count(const cpu_set_t* cpus) {
5257   int count = 0;
5258   // only look up to the number of configured processors
5259   for (int i = 0; i < os::processor_count(); i++) {
5260     if (CPU_ISSET(i, cpus)) {
5261       count++;
5262     }
5263   }
5264   return count;
5265 }
5266 
5267 #define CPU_COUNT(cpus) _cpu_count(cpus)
5268 
5269 #endif // CPU_COUNT
5270 
5271 // Get the current number of available processors for this process.
5272 // This value can change at any time during a process's lifetime.
5273 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5274 // If it appears there may be more than 1024 processors then we do a
5275 // dynamic check - see 6515172 for details.
5276 // If anything goes wrong we fallback to returning the number of online
5277 // processors - which can be greater than the number available to the process.
5278 int os::Linux::active_processor_count() {
5279   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
5280   cpu_set_t* cpus_p = &cpus;
5281   int cpus_size = sizeof(cpu_set_t);
5282 
5283   int configured_cpus = os::processor_count();  // upper bound on available cpus
5284   int cpu_count = 0;
5285 
5286 // old build platforms may not support dynamic cpu sets
5287 #ifdef CPU_ALLOC
5288 
5289   // To enable easy testing of the dynamic path on different platforms we
5290   // introduce a diagnostic flag: UseCpuAllocPath
5291   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5292     // kernel may use a mask bigger than cpu_set_t
5293     log_trace(os)("active_processor_count: using dynamic path %s"
5294                   "- configured processors: %d",
5295                   UseCpuAllocPath ? "(forced) " : "",
5296                   configured_cpus);
5297     cpus_p = CPU_ALLOC(configured_cpus);
5298     if (cpus_p != NULL) {
5299       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5300       // zero it just to be safe
5301       CPU_ZERO_S(cpus_size, cpus_p);
5302     }
5303     else {
5304        // failed to allocate so fallback to online cpus
5305        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5306        log_trace(os)("active_processor_count: "
5307                      "CPU_ALLOC failed (%s) - using "
5308                      "online processor count: %d",
5309                      os::strerror(errno), online_cpus);
5310        return online_cpus;
5311     }
5312   }
5313   else {
5314     log_trace(os)("active_processor_count: using static path - configured processors: %d",
5315                   configured_cpus);
5316   }
5317 #else // CPU_ALLOC
5318 // these stubs won't be executed
5319 #define CPU_COUNT_S(size, cpus) -1
5320 #define CPU_FREE(cpus)
5321 
5322   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5323                 configured_cpus);
5324 #endif // CPU_ALLOC
5325 
5326   // pid 0 means the current thread - which we have to assume represents the process
5327   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5328     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5329       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5330     }
5331     else {
5332       cpu_count = CPU_COUNT(cpus_p);
5333     }
5334     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5335   }
5336   else {
5337     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5338     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5339             "which may exceed available processors", os::strerror(errno), cpu_count);
5340   }
5341 
5342   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5343     CPU_FREE(cpus_p);
5344   }
5345 
5346   assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5347   return cpu_count;
5348 }
5349 
5350 // Determine the active processor count from one of
5351 // three different sources:
5352 //
5353 // 1. User option -XX:ActiveProcessorCount
5354 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5355 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5356 //
5357 // Option 1, if specified, will always override.
5358 // If the cgroup subsystem is active and configured, we
5359 // will return the min of the cgroup and option 2 results.
5360 // This is required since tools, such as numactl, that
5361 // alter cpu affinity do not update cgroup subsystem
5362 // cpuset configuration files.
5363 int os::active_processor_count() {
5364   // User has overridden the number of active processors
5365   if (ActiveProcessorCount > 0) {
5366     log_trace(os)("active_processor_count: "
5367                   "active processor count set by user : %d",
5368                   ActiveProcessorCount);
5369     return ActiveProcessorCount;
5370   }
5371 
5372   int active_cpus;
5373   if (OSContainer::is_containerized()) {
5374     active_cpus = OSContainer::active_processor_count();
5375     log_trace(os)("active_processor_count: determined by OSContainer: %d",
5376                    active_cpus);
5377   } else {
5378     active_cpus = os::Linux::active_processor_count();
5379   }
5380 
5381   return active_cpus;
5382 }
5383 
5384 uint os::processor_id() {
5385   const int id = Linux::sched_getcpu();
5386   assert(id >= 0 && id < _processor_count, "Invalid processor id");
5387   return (uint)id;
5388 }
5389 
5390 void os::set_native_thread_name(const char *name) {
5391   if (Linux::_pthread_setname_np) {
5392     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5393     snprintf(buf, sizeof(buf), "%s", name);
5394     buf[sizeof(buf) - 1] = '\0';
5395     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5396     // ERANGE should not happen; all other errors should just be ignored.
5397     assert(rc != ERANGE, "pthread_setname_np failed");
5398   }
5399 }
5400 
5401 bool os::distribute_processes(uint length, uint* distribution) {
5402   // Not yet implemented.
5403   return false;
5404 }
5405 
5406 bool os::bind_to_processor(uint processor_id) {
5407   // Not yet implemented.
5408   return false;
5409 }
5410 
5411 ///
5412 
5413 void os::SuspendedThreadTask::internal_do_task() {
5414   if (do_suspend(_thread->osthread())) {
5415     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5416     do_task(context);
5417     do_resume(_thread->osthread());
5418   }
5419 }
5420 
5421 ////////////////////////////////////////////////////////////////////////////////
5422 // debug support
5423 
5424 bool os::find(address addr, outputStream* st) {
5425   Dl_info dlinfo;
5426   memset(&dlinfo, 0, sizeof(dlinfo));
5427   if (dladdr(addr, &dlinfo) != 0) {
5428     st->print(PTR_FORMAT ": ", p2i(addr));
5429     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5430       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5431                 p2i(addr) - p2i(dlinfo.dli_saddr));
5432     } else if (dlinfo.dli_fbase != NULL) {
5433       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5434     } else {
5435       st->print("<absolute address>");
5436     }
5437     if (dlinfo.dli_fname != NULL) {
5438       st->print(" in %s", dlinfo.dli_fname);
5439     }
5440     if (dlinfo.dli_fbase != NULL) {
5441       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5442     }
5443     st->cr();
5444 
5445     if (Verbose) {
5446       // decode some bytes around the PC
5447       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5448       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5449       address       lowest = (address) dlinfo.dli_sname;
5450       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5451       if (begin < lowest)  begin = lowest;
5452       Dl_info dlinfo2;
5453       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5454           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5455         end = (address) dlinfo2.dli_saddr;
5456       }
5457       Disassembler::decode(begin, end, st);
5458     }
5459     return true;
5460   }
5461   return false;
5462 }
5463 
5464 ////////////////////////////////////////////////////////////////////////////////
5465 // misc
5466 
5467 // This does not do anything on Linux. This is basically a hook for being
5468 // able to use structured exception handling (thread-local exception filters)
5469 // on, e.g., Win32.
5470 void
5471 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5472                          JavaCallArguments* args, Thread* thread) {
5473   f(value, method, args, thread);
5474 }
5475 
5476 void os::print_statistics() {
5477 }
5478 
5479 bool os::message_box(const char* title, const char* message) {
5480   int i;
5481   fdStream err(defaultStream::error_fd());
5482   for (i = 0; i < 78; i++) err.print_raw("=");
5483   err.cr();
5484   err.print_raw_cr(title);
5485   for (i = 0; i < 78; i++) err.print_raw("-");
5486   err.cr();
5487   err.print_raw_cr(message);
5488   for (i = 0; i < 78; i++) err.print_raw("=");
5489   err.cr();
5490 
5491   char buf[16];
5492   // Prevent process from exiting upon "read error" without consuming all CPU
5493   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5494 
5495   return buf[0] == 'y' || buf[0] == 'Y';
5496 }
5497 
5498 // Is a (classpath) directory empty?
5499 bool os::dir_is_empty(const char* path) {
5500   DIR *dir = NULL;
5501   struct dirent *ptr;
5502 
5503   dir = opendir(path);
5504   if (dir == NULL) return true;
5505 
5506   // Scan the directory
5507   bool result = true;
5508   while (result && (ptr = readdir(dir)) != NULL) {
5509     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5510       result = false;
5511     }
5512   }
5513   closedir(dir);
5514   return result;
5515 }
5516 
5517 // This code originates from JDK's sysOpen and open64_w
5518 // from src/solaris/hpi/src/system_md.c
5519 
5520 int os::open(const char *path, int oflag, int mode) {
5521   if (strlen(path) > MAX_PATH - 1) {
5522     errno = ENAMETOOLONG;
5523     return -1;
5524   }
5525 
5526   // All file descriptors that are opened in the Java process and not
5527   // specifically destined for a subprocess should have the close-on-exec
5528   // flag set.  If we don't set it, then careless 3rd party native code
5529   // might fork and exec without closing all appropriate file descriptors
5530   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5531   // turn might:
5532   //
5533   // - cause end-of-file to fail to be detected on some file
5534   //   descriptors, resulting in mysterious hangs, or
5535   //
5536   // - might cause an fopen in the subprocess to fail on a system
5537   //   suffering from bug 1085341.
5538   //
5539   // (Yes, the default setting of the close-on-exec flag is a Unix
5540   // design flaw)
5541   //
5542   // See:
5543   // 1085341: 32-bit stdio routines should support file descriptors >255
5544   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5545   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5546   //
5547   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5548   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5549   // because it saves a system call and removes a small window where the flag
5550   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5551   // and we fall back to using FD_CLOEXEC (see below).
5552 #ifdef O_CLOEXEC
5553   oflag |= O_CLOEXEC;
5554 #endif
5555 
5556   int fd = ::open64(path, oflag, mode);
5557   if (fd == -1) return -1;
5558 
5559   //If the open succeeded, the file might still be a directory
5560   {
5561     struct stat64 buf64;
5562     int ret = ::fstat64(fd, &buf64);
5563     int st_mode = buf64.st_mode;
5564 
5565     if (ret != -1) {
5566       if ((st_mode & S_IFMT) == S_IFDIR) {
5567         errno = EISDIR;
5568         ::close(fd);
5569         return -1;
5570       }
5571     } else {
5572       ::close(fd);
5573       return -1;
5574     }
5575   }
5576 
5577 #ifdef FD_CLOEXEC
5578   // Validate that the use of the O_CLOEXEC flag on open above worked.
5579   // With recent kernels, we will perform this check exactly once.
5580   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5581   if (!O_CLOEXEC_is_known_to_work) {
5582     int flags = ::fcntl(fd, F_GETFD);
5583     if (flags != -1) {
5584       if ((flags & FD_CLOEXEC) != 0)
5585         O_CLOEXEC_is_known_to_work = 1;
5586       else
5587         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5588     }
5589   }
5590 #endif
5591 
5592   return fd;
5593 }
5594 
5595 
5596 // create binary file, rewriting existing file if required
5597 int os::create_binary_file(const char* path, bool rewrite_existing) {
5598   int oflags = O_WRONLY | O_CREAT;
5599   if (!rewrite_existing) {
5600     oflags |= O_EXCL;
5601   }
5602   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5603 }
5604 
5605 // return current position of file pointer
5606 jlong os::current_file_offset(int fd) {
5607   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5608 }
5609 
5610 // move file pointer to the specified offset
5611 jlong os::seek_to_file_offset(int fd, jlong offset) {
5612   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5613 }
5614 
5615 // This code originates from JDK's sysAvailable
5616 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5617 
5618 int os::available(int fd, jlong *bytes) {
5619   jlong cur, end;
5620   int mode;
5621   struct stat64 buf64;
5622 
5623   if (::fstat64(fd, &buf64) >= 0) {
5624     mode = buf64.st_mode;
5625     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5626       int n;
5627       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5628         *bytes = n;
5629         return 1;
5630       }
5631     }
5632   }
5633   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5634     return 0;
5635   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5636     return 0;
5637   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5638     return 0;
5639   }
5640   *bytes = end - cur;
5641   return 1;
5642 }
5643 
5644 // Map a block of memory.
5645 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5646                         char *addr, size_t bytes, bool read_only,
5647                         bool allow_exec) {
5648   int prot;
5649   int flags = MAP_PRIVATE;
5650 
5651   if (read_only) {
5652     prot = PROT_READ;
5653   } else {
5654     prot = PROT_READ | PROT_WRITE;
5655   }
5656 
5657   if (allow_exec) {
5658     prot |= PROT_EXEC;
5659   }
5660 
5661   if (addr != NULL) {
5662     flags |= MAP_FIXED;
5663   }
5664 
5665   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5666                                      fd, file_offset);
5667   if (mapped_address == MAP_FAILED) {
5668     return NULL;
5669   }
5670   return mapped_address;
5671 }
5672 
5673 
5674 // Remap a block of memory.
5675 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5676                           char *addr, size_t bytes, bool read_only,
5677                           bool allow_exec) {
5678   // same as map_memory() on this OS
5679   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5680                         allow_exec);
5681 }
5682 
5683 
5684 // Unmap a block of memory.
5685 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5686   return munmap(addr, bytes) == 0;
5687 }
5688 
5689 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5690 
5691 static jlong fast_cpu_time(Thread *thread) {
5692     clockid_t clockid;
5693     int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5694                                               &clockid);
5695     if (rc == 0) {
5696       return os::Linux::fast_thread_cpu_time(clockid);
5697     } else {
5698       // It's possible to encounter a terminated native thread that failed
5699       // to detach itself from the VM - which should result in ESRCH.
5700       assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5701       return -1;
5702     }
5703 }
5704 
5705 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5706 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5707 // of a thread.
5708 //
5709 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5710 // the fast estimate available on the platform.
5711 
5712 jlong os::current_thread_cpu_time() {
5713   if (os::Linux::supports_fast_thread_cpu_time()) {
5714     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5715   } else {
5716     // return user + sys since the cost is the same
5717     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5718   }
5719 }
5720 
5721 jlong os::thread_cpu_time(Thread* thread) {
5722   // consistent with what current_thread_cpu_time() returns
5723   if (os::Linux::supports_fast_thread_cpu_time()) {
5724     return fast_cpu_time(thread);
5725   } else {
5726     return slow_thread_cpu_time(thread, true /* user + sys */);
5727   }
5728 }
5729 
5730 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5731   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5732     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5733   } else {
5734     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5735   }
5736 }
5737 
5738 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5739   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5740     return fast_cpu_time(thread);
5741   } else {
5742     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5743   }
5744 }
5745 
5746 //  -1 on error.
5747 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5748   pid_t  tid = thread->osthread()->thread_id();
5749   char *s;
5750   char stat[2048];
5751   int statlen;
5752   char proc_name[64];
5753   int count;
5754   long sys_time, user_time;
5755   char cdummy;
5756   int idummy;
5757   long ldummy;
5758   FILE *fp;
5759 
5760   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5761   fp = fopen(proc_name, "r");
5762   if (fp == NULL) return -1;
5763   statlen = fread(stat, 1, 2047, fp);
5764   stat[statlen] = '\0';
5765   fclose(fp);
5766 
5767   // Skip pid and the command string. Note that we could be dealing with
5768   // weird command names, e.g. user could decide to rename java launcher
5769   // to "java 1.4.2 :)", then the stat file would look like
5770   //                1234 (java 1.4.2 :)) R ... ...
5771   // We don't really need to know the command string, just find the last
5772   // occurrence of ")" and then start parsing from there. See bug 4726580.
5773   s = strrchr(stat, ')');
5774   if (s == NULL) return -1;
5775 
5776   // Skip blank chars
5777   do { s++; } while (s && isspace(*s));
5778 
5779   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5780                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5781                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5782                  &user_time, &sys_time);
5783   if (count != 13) return -1;
5784   if (user_sys_cpu_time) {
5785     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5786   } else {
5787     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5788   }
5789 }
5790 
5791 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5792   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5793   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5794   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5795   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5796 }
5797 
5798 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5799   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5800   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5801   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5802   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5803 }
5804 
5805 bool os::is_thread_cpu_time_supported() {
5806   return true;
5807 }
5808 
5809 // System loadavg support.  Returns -1 if load average cannot be obtained.
5810 // Linux doesn't yet have a (official) notion of processor sets,
5811 // so just return the system wide load average.
5812 int os::loadavg(double loadavg[], int nelem) {
5813   return ::getloadavg(loadavg, nelem);
5814 }
5815 
5816 void os::pause() {
5817   char filename[MAX_PATH];
5818   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5819     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5820   } else {
5821     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5822   }
5823 
5824   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5825   if (fd != -1) {
5826     struct stat buf;
5827     ::close(fd);
5828     while (::stat(filename, &buf) == 0) {
5829       (void)::poll(NULL, 0, 100);
5830     }
5831   } else {
5832     jio_fprintf(stderr,
5833                 "Could not open pause file '%s', continuing immediately.\n", filename);
5834   }
5835 }
5836 
5837 extern char** environ;
5838 
5839 // Run the specified command in a separate process. Return its exit value,
5840 // or -1 on failure (e.g. can't fork a new process).
5841 // Unlike system(), this function can be called from signal handler. It
5842 // doesn't block SIGINT et al.
5843 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5844   const char * argv[4] = {"sh", "-c", cmd, NULL};
5845 
5846   pid_t pid ;
5847 
5848   if (use_vfork_if_available) {
5849     pid = vfork();
5850   } else {
5851     pid = fork();
5852   }
5853 
5854   if (pid < 0) {
5855     // fork failed
5856     return -1;
5857 
5858   } else if (pid == 0) {
5859     // child process
5860 
5861     execve("/bin/sh", (char* const*)argv, environ);
5862 
5863     // execve failed
5864     _exit(-1);
5865 
5866   } else  {
5867     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5868     // care about the actual exit code, for now.
5869 
5870     int status;
5871 
5872     // Wait for the child process to exit.  This returns immediately if
5873     // the child has already exited. */
5874     while (waitpid(pid, &status, 0) < 0) {
5875       switch (errno) {
5876       case ECHILD: return 0;
5877       case EINTR: break;
5878       default: return -1;
5879       }
5880     }
5881 
5882     if (WIFEXITED(status)) {
5883       // The child exited normally; get its exit code.
5884       return WEXITSTATUS(status);
5885     } else if (WIFSIGNALED(status)) {
5886       // The child exited because of a signal
5887       // The best value to return is 0x80 + signal number,
5888       // because that is what all Unix shells do, and because
5889       // it allows callers to distinguish between process exit and
5890       // process death by signal.
5891       return 0x80 + WTERMSIG(status);
5892     } else {
5893       // Unknown exit code; pass it through
5894       return status;
5895     }
5896   }
5897 }
5898 
5899 // Get the default path to the core file
5900 // Returns the length of the string
5901 int os::get_core_path(char* buffer, size_t bufferSize) {
5902   /*
5903    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5904    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5905    */
5906   const int core_pattern_len = 129;
5907   char core_pattern[core_pattern_len] = {0};
5908 
5909   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5910   if (core_pattern_file == -1) {
5911     return -1;
5912   }
5913 
5914   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5915   ::close(core_pattern_file);
5916   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5917     return -1;
5918   }
5919   if (core_pattern[ret-1] == '\n') {
5920     core_pattern[ret-1] = '\0';
5921   } else {
5922     core_pattern[ret] = '\0';
5923   }
5924 
5925   // Replace the %p in the core pattern with the process id. NOTE: we do this
5926   // only if the pattern doesn't start with "|", and we support only one %p in
5927   // the pattern.
5928   char *pid_pos = strstr(core_pattern, "%p");
5929   const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : "";  // skip over the "%p"
5930   int written;
5931 
5932   if (core_pattern[0] == '/') {
5933     if (pid_pos != NULL) {
5934       *pid_pos = '\0';
5935       written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5936                              current_process_id(), tail);
5937     } else {
5938       written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5939     }
5940   } else {
5941     char cwd[PATH_MAX];
5942 
5943     const char* p = get_current_directory(cwd, PATH_MAX);
5944     if (p == NULL) {
5945       return -1;
5946     }
5947 
5948     if (core_pattern[0] == '|') {
5949       written = jio_snprintf(buffer, bufferSize,
5950                              "\"%s\" (or dumping to %s/core.%d)",
5951                              &core_pattern[1], p, current_process_id());
5952     } else if (pid_pos != NULL) {
5953       *pid_pos = '\0';
5954       written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5955                              current_process_id(), tail);
5956     } else {
5957       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5958     }
5959   }
5960 
5961   if (written < 0) {
5962     return -1;
5963   }
5964 
5965   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5966     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5967 
5968     if (core_uses_pid_file != -1) {
5969       char core_uses_pid = 0;
5970       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5971       ::close(core_uses_pid_file);
5972 
5973       if (core_uses_pid == '1') {
5974         jio_snprintf(buffer + written, bufferSize - written,
5975                                           ".%d", current_process_id());
5976       }
5977     }
5978   }
5979 
5980   return strlen(buffer);
5981 }
5982 
5983 bool os::start_debugging(char *buf, int buflen) {
5984   int len = (int)strlen(buf);
5985   char *p = &buf[len];
5986 
5987   jio_snprintf(p, buflen-len,
5988                "\n\n"
5989                "Do you want to debug the problem?\n\n"
5990                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5991                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5992                "Otherwise, press RETURN to abort...",
5993                os::current_process_id(), os::current_process_id(),
5994                os::current_thread_id(), os::current_thread_id());
5995 
5996   bool yes = os::message_box("Unexpected Error", buf);
5997 
5998   if (yes) {
5999     // yes, user asked VM to launch debugger
6000     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6001                  os::current_process_id(), os::current_process_id());
6002 
6003     os::fork_and_exec(buf);
6004     yes = false;
6005   }
6006   return yes;
6007 }
6008 
6009 
6010 // Java/Compiler thread:
6011 //
6012 //   Low memory addresses
6013 // P0 +------------------------+
6014 //    |                        |\  Java thread created by VM does not have glibc
6015 //    |    glibc guard page    | - guard page, attached Java thread usually has
6016 //    |                        |/  1 glibc guard page.
6017 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6018 //    |                        |\
6019 //    |  HotSpot Guard Pages   | - red, yellow and reserved pages
6020 //    |                        |/
6021 //    +------------------------+ JavaThread::stack_reserved_zone_base()
6022 //    |                        |\
6023 //    |      Normal Stack      | -
6024 //    |                        |/
6025 // P2 +------------------------+ Thread::stack_base()
6026 //
6027 // Non-Java thread:
6028 //
6029 //   Low memory addresses
6030 // P0 +------------------------+
6031 //    |                        |\
6032 //    |  glibc guard page      | - usually 1 page
6033 //    |                        |/
6034 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6035 //    |                        |\
6036 //    |      Normal Stack      | -
6037 //    |                        |/
6038 // P2 +------------------------+ Thread::stack_base()
6039 //
6040 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6041 //    returned from pthread_attr_getstack().
6042 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6043 //    of the stack size given in pthread_attr. We work around this for
6044 //    threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6045 //
6046 #ifndef ZERO
6047 static void current_stack_region(address * bottom, size_t * size) {
6048   if (os::is_primordial_thread()) {
6049     // primordial thread needs special handling because pthread_getattr_np()
6050     // may return bogus value.
6051     *bottom = os::Linux::initial_thread_stack_bottom();
6052     *size   = os::Linux::initial_thread_stack_size();
6053   } else {
6054     pthread_attr_t attr;
6055 
6056     int rslt = pthread_getattr_np(pthread_self(), &attr);
6057 
6058     // JVM needs to know exact stack location, abort if it fails
6059     if (rslt != 0) {
6060       if (rslt == ENOMEM) {
6061         vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6062       } else {
6063         fatal("pthread_getattr_np failed with error = %d", rslt);
6064       }
6065     }
6066 
6067     if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6068       fatal("Cannot locate current stack attributes!");
6069     }
6070 
6071     // Work around NPTL stack guard error.
6072     size_t guard_size = 0;
6073     rslt = pthread_attr_getguardsize(&attr, &guard_size);
6074     if (rslt != 0) {
6075       fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6076     }
6077     *bottom += guard_size;
6078     *size   -= guard_size;
6079 
6080     pthread_attr_destroy(&attr);
6081 
6082   }
6083   assert(os::current_stack_pointer() >= *bottom &&
6084          os::current_stack_pointer() < *bottom + *size, "just checking");
6085 }
6086 
6087 address os::current_stack_base() {
6088   address bottom;
6089   size_t size;
6090   current_stack_region(&bottom, &size);
6091   return (bottom + size);
6092 }
6093 
6094 size_t os::current_stack_size() {
6095   // This stack size includes the usable stack and HotSpot guard pages
6096   // (for the threads that have Hotspot guard pages).
6097   address bottom;
6098   size_t size;
6099   current_stack_region(&bottom, &size);
6100   return size;
6101 }
6102 #endif
6103 
6104 static inline struct timespec get_mtime(const char* filename) {
6105   struct stat st;
6106   int ret = os::stat(filename, &st);
6107   assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
6108   return st.st_mtim;
6109 }
6110 
6111 int os::compare_file_modified_times(const char* file1, const char* file2) {
6112   struct timespec filetime1 = get_mtime(file1);
6113   struct timespec filetime2 = get_mtime(file2);
6114   int diff = filetime1.tv_sec - filetime2.tv_sec;
6115   if (diff == 0) {
6116     return filetime1.tv_nsec - filetime2.tv_nsec;
6117   }
6118   return diff;
6119 }
6120 
6121 /////////////// Unit tests ///////////////
6122 
6123 #ifndef PRODUCT
6124 
6125 class TestReserveMemorySpecial : AllStatic {
6126  public:
6127   static void small_page_write(void* addr, size_t size) {
6128     size_t page_size = os::vm_page_size();
6129 
6130     char* end = (char*)addr + size;
6131     for (char* p = (char*)addr; p < end; p += page_size) {
6132       *p = 1;
6133     }
6134   }
6135 
6136   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6137     if (!UseHugeTLBFS) {
6138       return;
6139     }
6140 
6141     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6142 
6143     if (addr != NULL) {
6144       small_page_write(addr, size);
6145 
6146       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6147     }
6148   }
6149 
6150   static void test_reserve_memory_special_huge_tlbfs_only() {
6151     if (!UseHugeTLBFS) {
6152       return;
6153     }
6154 
6155     size_t lp = os::large_page_size();
6156 
6157     for (size_t size = lp; size <= lp * 10; size += lp) {
6158       test_reserve_memory_special_huge_tlbfs_only(size);
6159     }
6160   }
6161 
6162   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6163     size_t lp = os::large_page_size();
6164     size_t ag = os::vm_allocation_granularity();
6165 
6166     // sizes to test
6167     const size_t sizes[] = {
6168       lp, lp + ag, lp + lp / 2, lp * 2,
6169       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6170       lp * 10, lp * 10 + lp / 2
6171     };
6172     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6173 
6174     // For each size/alignment combination, we test three scenarios:
6175     // 1) with req_addr == NULL
6176     // 2) with a non-null req_addr at which we expect to successfully allocate
6177     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6178     //    expect the allocation to either fail or to ignore req_addr
6179 
6180     // Pre-allocate two areas; they shall be as large as the largest allocation
6181     //  and aligned to the largest alignment we will be testing.
6182     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6183     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6184       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6185       -1, 0);
6186     assert(mapping1 != MAP_FAILED, "should work");
6187 
6188     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6189       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6190       -1, 0);
6191     assert(mapping2 != MAP_FAILED, "should work");
6192 
6193     // Unmap the first mapping, but leave the second mapping intact: the first
6194     // mapping will serve as a value for a "good" req_addr (case 2). The second
6195     // mapping, still intact, as "bad" req_addr (case 3).
6196     ::munmap(mapping1, mapping_size);
6197 
6198     // Case 1
6199     for (int i = 0; i < num_sizes; i++) {
6200       const size_t size = sizes[i];
6201       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6202         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6203         if (p != NULL) {
6204           assert(is_aligned(p, alignment), "must be");
6205           small_page_write(p, size);
6206           os::Linux::release_memory_special_huge_tlbfs(p, size);
6207         }
6208       }
6209     }
6210 
6211     // Case 2
6212     for (int i = 0; i < num_sizes; i++) {
6213       const size_t size = sizes[i];
6214       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6215         char* const req_addr = align_up(mapping1, alignment);
6216         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6217         if (p != NULL) {
6218           assert(p == req_addr, "must be");
6219           small_page_write(p, size);
6220           os::Linux::release_memory_special_huge_tlbfs(p, size);
6221         }
6222       }
6223     }
6224 
6225     // Case 3
6226     for (int i = 0; i < num_sizes; i++) {
6227       const size_t size = sizes[i];
6228       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6229         char* const req_addr = align_up(mapping2, alignment);
6230         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6231         // as the area around req_addr contains already existing mappings, the API should always
6232         // return NULL (as per contract, it cannot return another address)
6233         assert(p == NULL, "must be");
6234       }
6235     }
6236 
6237     ::munmap(mapping2, mapping_size);
6238 
6239   }
6240 
6241   static void test_reserve_memory_special_huge_tlbfs() {
6242     if (!UseHugeTLBFS) {
6243       return;
6244     }
6245 
6246     test_reserve_memory_special_huge_tlbfs_only();
6247     test_reserve_memory_special_huge_tlbfs_mixed();
6248   }
6249 
6250   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6251     if (!UseSHM) {
6252       return;
6253     }
6254 
6255     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6256 
6257     if (addr != NULL) {
6258       assert(is_aligned(addr, alignment), "Check");
6259       assert(is_aligned(addr, os::large_page_size()), "Check");
6260 
6261       small_page_write(addr, size);
6262 
6263       os::Linux::release_memory_special_shm(addr, size);
6264     }
6265   }
6266 
6267   static void test_reserve_memory_special_shm() {
6268     size_t lp = os::large_page_size();
6269     size_t ag = os::vm_allocation_granularity();
6270 
6271     for (size_t size = ag; size < lp * 3; size += ag) {
6272       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6273         test_reserve_memory_special_shm(size, alignment);
6274       }
6275     }
6276   }
6277 
6278   static void test() {
6279     test_reserve_memory_special_huge_tlbfs();
6280     test_reserve_memory_special_shm();
6281   }
6282 };
6283 
6284 void TestReserveMemorySpecial_test() {
6285   TestReserveMemorySpecial::test();
6286 }
6287 
6288 #endif