rev 56448 : imported patch 8220310.mut.0
rev 56449 : imported patch 8220310.mut.1
rev 56451 : imported patch 8220310.mut.3
rev 56452 : [mq]: 8220310.mut.4
rev 56453 : [mq]: 8220310.mut.5

   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 <endian.h>
  88 # include <errno.h>
  89 # include <dlfcn.h>
  90 # include <stdio.h>
  91 # include <unistd.h>
  92 # include <sys/resource.h>
  93 # include <pthread.h>
  94 # include <sys/stat.h>
  95 # include <sys/time.h>
  96 # include <sys/times.h>
  97 # include <sys/utsname.h>
  98 # include <sys/socket.h>
  99 # include <sys/wait.h>
 100 # include <pwd.h>
 101 # include <poll.h>
 102 # include <fcntl.h>
 103 # include <string.h>
 104 # include <syscall.h>
 105 # include <sys/sysinfo.h>
 106 # include <gnu/libc-version.h>
 107 # include <sys/ipc.h>
 108 # include <sys/shm.h>
 109 # include <link.h>
 110 # include <stdint.h>
 111 # include <inttypes.h>
 112 # include <sys/ioctl.h>
 113 
 114 #ifndef _GNU_SOURCE
 115   #define _GNU_SOURCE
 116   #include <sched.h>
 117   #undef _GNU_SOURCE
 118 #else
 119   #include <sched.h>
 120 #endif
 121 
 122 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 123 // getrusage() is prepared to handle the associated failure.
 124 #ifndef RUSAGE_THREAD
 125   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 126 #endif
 127 
 128 #define MAX_PATH    (2 * K)
 129 
 130 #define MAX_SECS 100000000
 131 
 132 // for timer info max values which include all bits
 133 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 134 
 135 enum CoredumpFilterBit {
 136   FILE_BACKED_PVT_BIT = 1 << 2,
 137   FILE_BACKED_SHARED_BIT = 1 << 3,
 138   LARGEPAGES_BIT = 1 << 6,
 139   DAX_SHARED_BIT = 1 << 8
 140 };
 141 
 142 ////////////////////////////////////////////////////////////////////////////////
 143 // global variables
 144 julong os::Linux::_physical_memory = 0;
 145 
 146 address   os::Linux::_initial_thread_stack_bottom = NULL;
 147 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 148 
 149 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 150 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 151 Mutex* os::Linux::_createThread_lock = NULL;
 152 pthread_t os::Linux::_main_thread;
 153 int os::Linux::_page_size = -1;
 154 bool os::Linux::_supports_fast_thread_cpu_time = false;
 155 uint32_t os::Linux::_os_version = 0;
 156 const char * os::Linux::_glibc_version = NULL;
 157 const char * os::Linux::_libpthread_version = NULL;
 158 
 159 static jlong initial_time_count=0;
 160 
 161 static int clock_tics_per_sec = 100;
 162 
 163 // If the VM might have been created on the primordial thread, we need to resolve the
 164 // primordial thread stack bounds and check if the current thread might be the
 165 // primordial thread in places. If we know that the primordial thread is never used,
 166 // such as when the VM was created by one of the standard java launchers, we can
 167 // avoid this
 168 static bool suppress_primordial_thread_resolution = false;
 169 
 170 // For diagnostics to print a message once. see run_periodic_checks
 171 static sigset_t check_signal_done;
 172 static bool check_signals = true;
 173 
 174 // Signal number used to suspend/resume a thread
 175 
 176 // do not use any signal number less than SIGSEGV, see 4355769
 177 static int SR_signum = SIGUSR2;
 178 sigset_t SR_sigset;
 179 
 180 // utility functions
 181 
 182 static int SR_initialize();
 183 
 184 julong os::available_memory() {
 185   return Linux::available_memory();
 186 }
 187 
 188 julong os::Linux::available_memory() {
 189   // values in struct sysinfo are "unsigned long"
 190   struct sysinfo si;
 191   julong avail_mem;
 192 
 193   if (OSContainer::is_containerized()) {
 194     jlong mem_limit, mem_usage;
 195     if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
 196       log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
 197                              mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
 198     }
 199     if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
 200       log_debug(os, container)("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
 201     }
 202     if (mem_limit > 0 && mem_usage > 0 ) {
 203       avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
 204       log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
 205       return avail_mem;
 206     }
 207   }
 208 
 209   sysinfo(&si);
 210   avail_mem = (julong)si.freeram * si.mem_unit;
 211   log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
 212   return avail_mem;
 213 }
 214 
 215 julong os::physical_memory() {
 216   jlong phys_mem = 0;
 217   if (OSContainer::is_containerized()) {
 218     jlong mem_limit;
 219     if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
 220       log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
 221       return mem_limit;
 222     }
 223     log_debug(os, container)("container memory limit %s: " JLONG_FORMAT ", using host value",
 224                             mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
 225   }
 226 
 227   phys_mem = Linux::physical_memory();
 228   log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
 229   return phys_mem;
 230 }
 231 
 232 static uint64_t initial_total_ticks = 0;
 233 static uint64_t initial_steal_ticks = 0;
 234 static bool     has_initial_tick_info = false;
 235 
 236 static void next_line(FILE *f) {
 237   int c;
 238   do {
 239     c = fgetc(f);
 240   } while (c != '\n' && c != EOF);
 241 }
 242 
 243 bool os::Linux::get_tick_information(CPUPerfTicks* pticks, int which_logical_cpu) {
 244   FILE*         fh;
 245   uint64_t      userTicks, niceTicks, systemTicks, idleTicks;
 246   // since at least kernel 2.6 : iowait: time waiting for I/O to complete
 247   // irq: time  servicing interrupts; softirq: time servicing softirqs
 248   uint64_t      iowTicks = 0, irqTicks = 0, sirqTicks= 0;
 249   // steal (since kernel 2.6.11): time spent in other OS when running in a virtualized environment
 250   uint64_t      stealTicks = 0;
 251   // guest (since kernel 2.6.24): time spent running a virtual CPU for guest OS under the
 252   // control of the Linux kernel
 253   uint64_t      guestNiceTicks = 0;
 254   int           logical_cpu = -1;
 255   const int     required_tickinfo_count = (which_logical_cpu == -1) ? 4 : 5;
 256   int           n;
 257 
 258   memset(pticks, 0, sizeof(CPUPerfTicks));
 259 
 260   if ((fh = fopen("/proc/stat", "r")) == NULL) {
 261     return false;
 262   }
 263 
 264   if (which_logical_cpu == -1) {
 265     n = fscanf(fh, "cpu " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
 266             UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
 267             UINT64_FORMAT " " UINT64_FORMAT " ",
 268             &userTicks, &niceTicks, &systemTicks, &idleTicks,
 269             &iowTicks, &irqTicks, &sirqTicks,
 270             &stealTicks, &guestNiceTicks);
 271   } else {
 272     // Move to next line
 273     next_line(fh);
 274 
 275     // find the line for requested cpu faster to just iterate linefeeds?
 276     for (int i = 0; i < which_logical_cpu; i++) {
 277       next_line(fh);
 278     }
 279 
 280     n = fscanf(fh, "cpu%u " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
 281                UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " " UINT64_FORMAT " "
 282                UINT64_FORMAT " " UINT64_FORMAT " ",
 283                &logical_cpu, &userTicks, &niceTicks,
 284                &systemTicks, &idleTicks, &iowTicks, &irqTicks, &sirqTicks,
 285                &stealTicks, &guestNiceTicks);
 286   }
 287 
 288   fclose(fh);
 289   if (n < required_tickinfo_count || logical_cpu != which_logical_cpu) {
 290     return false;
 291   }
 292   pticks->used       = userTicks + niceTicks;
 293   pticks->usedKernel = systemTicks + irqTicks + sirqTicks;
 294   pticks->total      = userTicks + niceTicks + systemTicks + idleTicks +
 295                        iowTicks + irqTicks + sirqTicks + stealTicks + guestNiceTicks;
 296 
 297   if (n > required_tickinfo_count + 3) {
 298     pticks->steal = stealTicks;
 299     pticks->has_steal_ticks = true;
 300   } else {
 301     pticks->steal = 0;
 302     pticks->has_steal_ticks = false;
 303   }
 304 
 305   return true;
 306 }
 307 
 308 // Return true if user is running as root.
 309 
 310 bool os::have_special_privileges() {
 311   static bool init = false;
 312   static bool privileges = false;
 313   if (!init) {
 314     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 315     init = true;
 316   }
 317   return privileges;
 318 }
 319 
 320 
 321 #ifndef SYS_gettid
 322 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 323   #ifdef __ia64__
 324     #define SYS_gettid 1105
 325   #else
 326     #ifdef __i386__
 327       #define SYS_gettid 224
 328     #else
 329       #ifdef __amd64__
 330         #define SYS_gettid 186
 331       #else
 332         #ifdef __sparc__
 333           #define SYS_gettid 143
 334         #else
 335           #error define gettid for the arch
 336         #endif
 337       #endif
 338     #endif
 339   #endif
 340 #endif
 341 
 342 
 343 // pid_t gettid()
 344 //
 345 // Returns the kernel thread id of the currently running thread. Kernel
 346 // thread id is used to access /proc.
 347 pid_t os::Linux::gettid() {
 348   int rslt = syscall(SYS_gettid);
 349   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 350   return (pid_t)rslt;
 351 }
 352 
 353 // Most versions of linux have a bug where the number of processors are
 354 // determined by looking at the /proc file system.  In a chroot environment,
 355 // the system call returns 1.
 356 static bool unsafe_chroot_detected = false;
 357 static const char *unstable_chroot_error = "/proc file system not found.\n"
 358                      "Java may be unstable running multithreaded in a chroot "
 359                      "environment on Linux when /proc filesystem is not mounted.";
 360 
 361 void os::Linux::initialize_system_info() {
 362   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 363   if (processor_count() == 1) {
 364     pid_t pid = os::Linux::gettid();
 365     char fname[32];
 366     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 367     FILE *fp = fopen(fname, "r");
 368     if (fp == NULL) {
 369       unsafe_chroot_detected = true;
 370     } else {
 371       fclose(fp);
 372     }
 373   }
 374   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 375   assert(processor_count() > 0, "linux error");
 376 }
 377 
 378 void os::init_system_properties_values() {
 379   // The next steps are taken in the product version:
 380   //
 381   // Obtain the JAVA_HOME value from the location of libjvm.so.
 382   // This library should be located at:
 383   // <JAVA_HOME>/lib/{client|server}/libjvm.so.
 384   //
 385   // If "/jre/lib/" appears at the right place in the path, then we
 386   // assume libjvm.so is installed in a JDK and we use this path.
 387   //
 388   // Otherwise exit with message: "Could not create the Java virtual machine."
 389   //
 390   // The following extra steps are taken in the debugging version:
 391   //
 392   // If "/jre/lib/" does NOT appear at the right place in the path
 393   // instead of exit check for $JAVA_HOME environment variable.
 394   //
 395   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 396   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 397   // it looks like libjvm.so is installed there
 398   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 399   //
 400   // Otherwise exit.
 401   //
 402   // Important note: if the location of libjvm.so changes this
 403   // code needs to be changed accordingly.
 404 
 405   // See ld(1):
 406   //      The linker uses the following search paths to locate required
 407   //      shared libraries:
 408   //        1: ...
 409   //        ...
 410   //        7: The default directories, normally /lib and /usr/lib.
 411 #ifndef OVERRIDE_LIBPATH
 412   #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
 413     #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 414   #else
 415     #define DEFAULT_LIBPATH "/lib:/usr/lib"
 416   #endif
 417 #else
 418   #define DEFAULT_LIBPATH OVERRIDE_LIBPATH
 419 #endif
 420 
 421 // Base path of extensions installed on the system.
 422 #define SYS_EXT_DIR     "/usr/java/packages"
 423 #define EXTENSIONS_DIR  "/lib/ext"
 424 
 425   // Buffer that fits several sprintfs.
 426   // Note that the space for the colon and the trailing null are provided
 427   // by the nulls included by the sizeof operator.
 428   const size_t bufsize =
 429     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 430          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 431   char *buf = NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 432 
 433   // sysclasspath, java_home, dll_dir
 434   {
 435     char *pslash;
 436     os::jvm_path(buf, bufsize);
 437 
 438     // Found the full path to libjvm.so.
 439     // Now cut the path to <java_home>/jre if we can.
 440     pslash = strrchr(buf, '/');
 441     if (pslash != NULL) {
 442       *pslash = '\0';            // Get rid of /libjvm.so.
 443     }
 444     pslash = strrchr(buf, '/');
 445     if (pslash != NULL) {
 446       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 447     }
 448     Arguments::set_dll_dir(buf);
 449 
 450     if (pslash != NULL) {
 451       pslash = strrchr(buf, '/');
 452       if (pslash != NULL) {
 453         *pslash = '\0';        // Get rid of /lib.
 454       }
 455     }
 456     Arguments::set_java_home(buf);
 457     if (!set_boot_path('/', ':')) {
 458       vm_exit_during_initialization("Failed setting boot class path.", NULL);
 459     }
 460   }
 461 
 462   // Where to look for native libraries.
 463   //
 464   // Note: Due to a legacy implementation, most of the library path
 465   // is set in the launcher. This was to accomodate linking restrictions
 466   // on legacy Linux implementations (which are no longer supported).
 467   // Eventually, all the library path setting will be done here.
 468   //
 469   // However, to prevent the proliferation of improperly built native
 470   // libraries, the new path component /usr/java/packages is added here.
 471   // Eventually, all the library path setting will be done here.
 472   {
 473     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 474     // should always exist (until the legacy problem cited above is
 475     // addressed).
 476     const char *v = ::getenv("LD_LIBRARY_PATH");
 477     const char *v_colon = ":";
 478     if (v == NULL) { v = ""; v_colon = ""; }
 479     // That's +1 for the colon and +1 for the trailing '\0'.
 480     char *ld_library_path = NEW_C_HEAP_ARRAY(char,
 481                                              strlen(v) + 1 +
 482                                              sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
 483                                              mtInternal);
 484     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
 485     Arguments::set_library_path(ld_library_path);
 486     FREE_C_HEAP_ARRAY(char, ld_library_path);
 487   }
 488 
 489   // Extensions directories.
 490   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 491   Arguments::set_ext_dirs(buf);
 492 
 493   FREE_C_HEAP_ARRAY(char, buf);
 494 
 495 #undef DEFAULT_LIBPATH
 496 #undef SYS_EXT_DIR
 497 #undef EXTENSIONS_DIR
 498 }
 499 
 500 ////////////////////////////////////////////////////////////////////////////////
 501 // breakpoint support
 502 
 503 void os::breakpoint() {
 504   BREAKPOINT;
 505 }
 506 
 507 extern "C" void breakpoint() {
 508   // use debugger to set breakpoint here
 509 }
 510 
 511 ////////////////////////////////////////////////////////////////////////////////
 512 // signal support
 513 
 514 debug_only(static bool signal_sets_initialized = false);
 515 static sigset_t unblocked_sigs, vm_sigs;
 516 
 517 void os::Linux::signal_sets_init() {
 518   // Should also have an assertion stating we are still single-threaded.
 519   assert(!signal_sets_initialized, "Already initialized");
 520   // Fill in signals that are necessarily unblocked for all threads in
 521   // the VM. Currently, we unblock the following signals:
 522   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 523   //                         by -Xrs (=ReduceSignalUsage));
 524   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 525   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 526   // the dispositions or masks wrt these signals.
 527   // Programs embedding the VM that want to use the above signals for their
 528   // own purposes must, at this time, use the "-Xrs" option to prevent
 529   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 530   // (See bug 4345157, and other related bugs).
 531   // In reality, though, unblocking these signals is really a nop, since
 532   // these signals are not blocked by default.
 533   sigemptyset(&unblocked_sigs);
 534   sigaddset(&unblocked_sigs, SIGILL);
 535   sigaddset(&unblocked_sigs, SIGSEGV);
 536   sigaddset(&unblocked_sigs, SIGBUS);
 537   sigaddset(&unblocked_sigs, SIGFPE);
 538 #if defined(PPC64)
 539   sigaddset(&unblocked_sigs, SIGTRAP);
 540 #endif
 541   sigaddset(&unblocked_sigs, SR_signum);
 542 
 543   if (!ReduceSignalUsage) {
 544     if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 545       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 546     }
 547     if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 548       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 549     }
 550     if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 551       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 552     }
 553   }
 554   // Fill in signals that are blocked by all but the VM thread.
 555   sigemptyset(&vm_sigs);
 556   if (!ReduceSignalUsage) {
 557     sigaddset(&vm_sigs, BREAK_SIGNAL);
 558   }
 559   debug_only(signal_sets_initialized = true);
 560 
 561 }
 562 
 563 // These are signals that are unblocked while a thread is running Java.
 564 // (For some reason, they get blocked by default.)
 565 sigset_t* os::Linux::unblocked_signals() {
 566   assert(signal_sets_initialized, "Not initialized");
 567   return &unblocked_sigs;
 568 }
 569 
 570 // These are the signals that are blocked while a (non-VM) thread is
 571 // running Java. Only the VM thread handles these signals.
 572 sigset_t* os::Linux::vm_signals() {
 573   assert(signal_sets_initialized, "Not initialized");
 574   return &vm_sigs;
 575 }
 576 
 577 void os::Linux::hotspot_sigmask(Thread* thread) {
 578 
 579   //Save caller's signal mask before setting VM signal mask
 580   sigset_t caller_sigmask;
 581   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 582 
 583   OSThread* osthread = thread->osthread();
 584   osthread->set_caller_sigmask(caller_sigmask);
 585 
 586   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 587 
 588   if (!ReduceSignalUsage) {
 589     if (thread->is_VM_thread()) {
 590       // Only the VM thread handles BREAK_SIGNAL ...
 591       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 592     } else {
 593       // ... all other threads block BREAK_SIGNAL
 594       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 595     }
 596   }
 597 }
 598 
 599 //////////////////////////////////////////////////////////////////////////////
 600 // detecting pthread library
 601 
 602 void os::Linux::libpthread_init() {
 603   // Save glibc and pthread version strings.
 604 #if !defined(_CS_GNU_LIBC_VERSION) || \
 605     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 606   #error "glibc too old (< 2.3.2)"
 607 #endif
 608 
 609   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 610   assert(n > 0, "cannot retrieve glibc version");
 611   char *str = (char *)malloc(n, mtInternal);
 612   confstr(_CS_GNU_LIBC_VERSION, str, n);
 613   os::Linux::set_glibc_version(str);
 614 
 615   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 616   assert(n > 0, "cannot retrieve pthread version");
 617   str = (char *)malloc(n, mtInternal);
 618   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 619   os::Linux::set_libpthread_version(str);
 620 }
 621 
 622 /////////////////////////////////////////////////////////////////////////////
 623 // thread stack expansion
 624 
 625 // os::Linux::manually_expand_stack() takes care of expanding the thread
 626 // stack. Note that this is normally not needed: pthread stacks allocate
 627 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 628 // committed. Therefore it is not necessary to expand the stack manually.
 629 //
 630 // Manually expanding the stack was historically needed on LinuxThreads
 631 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 632 // it is kept to deal with very rare corner cases:
 633 //
 634 // For one, user may run the VM on an own implementation of threads
 635 // whose stacks are - like the old LinuxThreads - implemented using
 636 // mmap(MAP_GROWSDOWN).
 637 //
 638 // Also, this coding may be needed if the VM is running on the primordial
 639 // thread. Normally we avoid running on the primordial thread; however,
 640 // user may still invoke the VM on the primordial thread.
 641 //
 642 // The following historical comment describes the details about running
 643 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 644 
 645 
 646 // Force Linux kernel to expand current thread stack. If "bottom" is close
 647 // to the stack guard, caller should block all signals.
 648 //
 649 // MAP_GROWSDOWN:
 650 //   A special mmap() flag that is used to implement thread stacks. It tells
 651 //   kernel that the memory region should extend downwards when needed. This
 652 //   allows early versions of LinuxThreads to only mmap the first few pages
 653 //   when creating a new thread. Linux kernel will automatically expand thread
 654 //   stack as needed (on page faults).
 655 //
 656 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 657 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 658 //   region, it's hard to tell if the fault is due to a legitimate stack
 659 //   access or because of reading/writing non-exist memory (e.g. buffer
 660 //   overrun). As a rule, if the fault happens below current stack pointer,
 661 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 662 //   application (see Linux kernel fault.c).
 663 //
 664 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 665 //   stack overflow detection.
 666 //
 667 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 668 //   not use MAP_GROWSDOWN.
 669 //
 670 // To get around the problem and allow stack banging on Linux, we need to
 671 // manually expand thread stack after receiving the SIGSEGV.
 672 //
 673 // There are two ways to expand thread stack to address "bottom", we used
 674 // both of them in JVM before 1.5:
 675 //   1. adjust stack pointer first so that it is below "bottom", and then
 676 //      touch "bottom"
 677 //   2. mmap() the page in question
 678 //
 679 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 680 // if current sp is already near the lower end of page 101, and we need to
 681 // call mmap() to map page 100, it is possible that part of the mmap() frame
 682 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 683 // That will destroy the mmap() frame and cause VM to crash.
 684 //
 685 // The following code works by adjusting sp first, then accessing the "bottom"
 686 // page to force a page fault. Linux kernel will then automatically expand the
 687 // stack mapping.
 688 //
 689 // _expand_stack_to() assumes its frame size is less than page size, which
 690 // should always be true if the function is not inlined.
 691 
 692 static void NOINLINE _expand_stack_to(address bottom) {
 693   address sp;
 694   size_t size;
 695   volatile char *p;
 696 
 697   // Adjust bottom to point to the largest address within the same page, it
 698   // gives us a one-page buffer if alloca() allocates slightly more memory.
 699   bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
 700   bottom += os::Linux::page_size() - 1;
 701 
 702   // sp might be slightly above current stack pointer; if that's the case, we
 703   // will alloca() a little more space than necessary, which is OK. Don't use
 704   // os::current_stack_pointer(), as its result can be slightly below current
 705   // stack pointer, causing us to not alloca enough to reach "bottom".
 706   sp = (address)&sp;
 707 
 708   if (sp > bottom) {
 709     size = sp - bottom;
 710     p = (volatile char *)alloca(size);
 711     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 712     p[0] = '\0';
 713   }
 714 }
 715 
 716 void os::Linux::expand_stack_to(address bottom) {
 717   _expand_stack_to(bottom);
 718 }
 719 
 720 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 721   assert(t!=NULL, "just checking");
 722   assert(t->osthread()->expanding_stack(), "expand should be set");
 723   assert(t->stack_base() != NULL, "stack_base was not initialized");
 724 
 725   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 726     sigset_t mask_all, old_sigset;
 727     sigfillset(&mask_all);
 728     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 729     _expand_stack_to(addr);
 730     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 731     return true;
 732   }
 733   return false;
 734 }
 735 
 736 //////////////////////////////////////////////////////////////////////////////
 737 // create new thread
 738 
 739 // Thread start routine for all newly created threads
 740 static void *thread_native_entry(Thread *thread) {
 741 
 742   thread->record_stack_base_and_size();
 743 
 744   // Try to randomize the cache line index of hot stack frames.
 745   // This helps when threads of the same stack traces evict each other's
 746   // cache lines. The threads can be either from the same JVM instance, or
 747   // from different JVM instances. The benefit is especially true for
 748   // processors with hyperthreading technology.
 749   static int counter = 0;
 750   int pid = os::current_process_id();
 751   alloca(((pid ^ counter++) & 7) * 128);
 752 
 753   thread->initialize_thread_current();
 754 
 755   OSThread* osthread = thread->osthread();
 756   Monitor* sync = osthread->startThread_lock();
 757 
 758   osthread->set_thread_id(os::current_thread_id());
 759 
 760   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 761     os::current_thread_id(), (uintx) pthread_self());
 762 
 763   if (UseNUMA) {
 764     int lgrp_id = os::numa_get_group_id();
 765     if (lgrp_id != -1) {
 766       thread->set_lgrp_id(lgrp_id);
 767     }
 768   }
 769   // initialize signal mask for this thread
 770   os::Linux::hotspot_sigmask(thread);
 771 
 772   // initialize floating point control register
 773   os::Linux::init_thread_fpu_state();
 774 
 775   // handshaking with parent thread
 776   {
 777     MutexLocker ml(sync, Mutex::_no_safepoint_check_flag);
 778 
 779     // notify parent thread
 780     osthread->set_state(INITIALIZED);
 781     sync->notify_all();
 782 
 783     // wait until os::start_thread()
 784     while (osthread->get_state() == INITIALIZED) {
 785       sync->wait_without_safepoint_check();
 786     }
 787   }
 788 
 789   assert(osthread->pthread_id() != 0, "pthread_id was not set as expected");
 790 
 791   // call one more level start routine
 792   thread->call_run();
 793 
 794   // Note: at this point the thread object may already have deleted itself.
 795   // Prevent dereferencing it from here on out.
 796   thread = NULL;
 797 
 798   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 799     os::current_thread_id(), (uintx) pthread_self());
 800 
 801   return 0;
 802 }
 803 
 804 // On Linux, glibc places static TLS blocks (for __thread variables) on
 805 // the thread stack. This decreases the stack size actually available
 806 // to threads.
 807 //
 808 // For large static TLS sizes, this may cause threads to malfunction due
 809 // to insufficient stack space. This is a well-known issue in glibc:
 810 // http://sourceware.org/bugzilla/show_bug.cgi?id=11787.
 811 //
 812 // As a workaround, we call a private but assumed-stable glibc function,
 813 // __pthread_get_minstack() to obtain the minstack size and derive the
 814 // static TLS size from it. We then increase the user requested stack
 815 // size by this TLS size.
 816 //
 817 // Due to compatibility concerns, this size adjustment is opt-in and
 818 // controlled via AdjustStackSizeForTLS.
 819 typedef size_t (*GetMinStack)(const pthread_attr_t *attr);
 820 
 821 GetMinStack _get_minstack_func = NULL;
 822 
 823 static void get_minstack_init() {
 824   _get_minstack_func =
 825         (GetMinStack)dlsym(RTLD_DEFAULT, "__pthread_get_minstack");
 826   log_info(os, thread)("Lookup of __pthread_get_minstack %s",
 827                        _get_minstack_func == NULL ? "failed" : "succeeded");
 828 }
 829 
 830 // Returns the size of the static TLS area glibc puts on thread stacks.
 831 // The value is cached on first use, which occurs when the first thread
 832 // is created during VM initialization.
 833 static size_t get_static_tls_area_size(const pthread_attr_t *attr) {
 834   size_t tls_size = 0;
 835   if (_get_minstack_func != NULL) {
 836     // Obtain the pthread minstack size by calling __pthread_get_minstack.
 837     size_t minstack_size = _get_minstack_func(attr);
 838 
 839     // Remove non-TLS area size included in minstack size returned
 840     // by __pthread_get_minstack() to get the static TLS size.
 841     // In glibc before 2.27, minstack size includes guard_size.
 842     // In glibc 2.27 and later, guard_size is automatically added
 843     // to the stack size by pthread_create and is no longer included
 844     // in minstack size. In both cases, the guard_size is taken into
 845     // account, so there is no need to adjust the result for that.
 846     //
 847     // Although __pthread_get_minstack() is a private glibc function,
 848     // it is expected to have a stable behavior across future glibc
 849     // versions while glibc still allocates the static TLS blocks off
 850     // the stack. Following is glibc 2.28 __pthread_get_minstack():
 851     //
 852     // size_t
 853     // __pthread_get_minstack (const pthread_attr_t *attr)
 854     // {
 855     //   return GLRO(dl_pagesize) + __static_tls_size + PTHREAD_STACK_MIN;
 856     // }
 857     //
 858     //
 859     // The following 'minstack_size > os::vm_page_size() + PTHREAD_STACK_MIN'
 860     // if check is done for precaution.
 861     if (minstack_size > (size_t)os::vm_page_size() + PTHREAD_STACK_MIN) {
 862       tls_size = minstack_size - os::vm_page_size() - PTHREAD_STACK_MIN;
 863     }
 864   }
 865 
 866   log_info(os, thread)("Stack size adjustment for TLS is " SIZE_FORMAT,
 867                        tls_size);
 868   return tls_size;
 869 }
 870 
 871 bool os::create_thread(Thread* thread, ThreadType thr_type,
 872                        size_t req_stack_size) {
 873   assert(thread->osthread() == NULL, "caller responsible");
 874 
 875   // Allocate the OSThread object
 876   OSThread* osthread = new OSThread(NULL, NULL);
 877   if (osthread == NULL) {
 878     return false;
 879   }
 880 
 881   // set the correct thread state
 882   osthread->set_thread_type(thr_type);
 883 
 884   // Initial state is ALLOCATED but not INITIALIZED
 885   osthread->set_state(ALLOCATED);
 886 
 887   thread->set_osthread(osthread);
 888 
 889   // init thread attributes
 890   pthread_attr_t attr;
 891   pthread_attr_init(&attr);
 892   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 893 
 894   // Calculate stack size if it's not specified by caller.
 895   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 896   // In glibc versions prior to 2.7 the guard size mechanism
 897   // is not implemented properly. The posix standard requires adding
 898   // the size of the guard pages to the stack size, instead Linux
 899   // takes the space out of 'stacksize'. Thus we adapt the requested
 900   // stack_size by the size of the guard pages to mimick proper
 901   // behaviour. However, be careful not to end up with a size
 902   // of zero due to overflow. Don't add the guard page in that case.
 903   size_t guard_size = os::Linux::default_guard_size(thr_type);
 904   // Configure glibc guard page. Must happen before calling
 905   // get_static_tls_area_size(), which uses the guard_size.
 906   pthread_attr_setguardsize(&attr, guard_size);
 907 
 908   size_t stack_adjust_size = 0;
 909   if (AdjustStackSizeForTLS) {
 910     // Adjust the stack_size for on-stack TLS - see get_static_tls_area_size().
 911     stack_adjust_size += get_static_tls_area_size(&attr);
 912   } else {
 913     stack_adjust_size += guard_size;
 914   }
 915 
 916   stack_adjust_size = align_up(stack_adjust_size, os::vm_page_size());
 917   if (stack_size <= SIZE_MAX - stack_adjust_size) {
 918     stack_size += stack_adjust_size;
 919   }
 920   assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
 921 
 922   int status = pthread_attr_setstacksize(&attr, stack_size);
 923   assert_status(status == 0, status, "pthread_attr_setstacksize");
 924 
 925   ThreadState state;
 926 
 927   {
 928     pthread_t tid;
 929     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 930 
 931     char buf[64];
 932     if (ret == 0) {
 933       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 934         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 935     } else {
 936       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 937         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 938       // Log some OS information which might explain why creating the thread failed.
 939       log_info(os, thread)("Number of threads approx. running in the VM: %d", Threads::number_of_threads());
 940       LogStream st(Log(os, thread)::info());
 941       os::Posix::print_rlimit_info(&st);
 942       os::print_memory_info(&st);
 943       os::Linux::print_proc_sys_info(&st);
 944       os::Linux::print_container_info(&st);
 945     }
 946 
 947     pthread_attr_destroy(&attr);
 948 
 949     if (ret != 0) {
 950       // Need to clean up stuff we've allocated so far
 951       thread->set_osthread(NULL);
 952       delete osthread;
 953       return false;
 954     }
 955 
 956     // Store pthread info into the OSThread
 957     osthread->set_pthread_id(tid);
 958 
 959     // Wait until child thread is either initialized or aborted
 960     {
 961       Monitor* sync_with_child = osthread->startThread_lock();
 962       MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 963       while ((state = osthread->get_state()) == ALLOCATED) {
 964         sync_with_child->wait_without_safepoint_check();
 965       }
 966     }
 967   }
 968 
 969   // Aborted due to thread limit being reached
 970   if (state == ZOMBIE) {
 971     thread->set_osthread(NULL);
 972     delete osthread;
 973     return false;
 974   }
 975 
 976   // The thread is returned suspended (in state INITIALIZED),
 977   // and is started higher up in the call chain
 978   assert(state == INITIALIZED, "race condition");
 979   return true;
 980 }
 981 
 982 /////////////////////////////////////////////////////////////////////////////
 983 // attach existing thread
 984 
 985 // bootstrap the main thread
 986 bool os::create_main_thread(JavaThread* thread) {
 987   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 988   return create_attached_thread(thread);
 989 }
 990 
 991 bool os::create_attached_thread(JavaThread* thread) {
 992 #ifdef ASSERT
 993   thread->verify_not_published();
 994 #endif
 995 
 996   // Allocate the OSThread object
 997   OSThread* osthread = new OSThread(NULL, NULL);
 998 
 999   if (osthread == NULL) {
1000     return false;
1001   }
1002 
1003   // Store pthread info into the OSThread
1004   osthread->set_thread_id(os::Linux::gettid());
1005   osthread->set_pthread_id(::pthread_self());
1006 
1007   // initialize floating point control register
1008   os::Linux::init_thread_fpu_state();
1009 
1010   // Initial thread state is RUNNABLE
1011   osthread->set_state(RUNNABLE);
1012 
1013   thread->set_osthread(osthread);
1014 
1015   if (UseNUMA) {
1016     int lgrp_id = os::numa_get_group_id();
1017     if (lgrp_id != -1) {
1018       thread->set_lgrp_id(lgrp_id);
1019     }
1020   }
1021 
1022   if (os::is_primordial_thread()) {
1023     // If current thread is primordial thread, its stack is mapped on demand,
1024     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1025     // the entire stack region to avoid SEGV in stack banging.
1026     // It is also useful to get around the heap-stack-gap problem on SuSE
1027     // kernel (see 4821821 for details). We first expand stack to the top
1028     // of yellow zone, then enable stack yellow zone (order is significant,
1029     // enabling yellow zone first will crash JVM on SuSE Linux), so there
1030     // is no gap between the last two virtual memory regions.
1031 
1032     JavaThread *jt = (JavaThread *)thread;
1033     address addr = jt->stack_reserved_zone_base();
1034     assert(addr != NULL, "initialization problem?");
1035     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1036 
1037     osthread->set_expanding_stack();
1038     os::Linux::manually_expand_stack(jt, addr);
1039     osthread->clear_expanding_stack();
1040   }
1041 
1042   // initialize signal mask for this thread
1043   // and save the caller's signal mask
1044   os::Linux::hotspot_sigmask(thread);
1045 
1046   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
1047     os::current_thread_id(), (uintx) pthread_self());
1048 
1049   return true;
1050 }
1051 
1052 void os::pd_start_thread(Thread* thread) {
1053   OSThread * osthread = thread->osthread();
1054   assert(osthread->get_state() != INITIALIZED, "just checking");
1055   Monitor* sync_with_child = osthread->startThread_lock();
1056   MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1057   sync_with_child->notify();
1058 }
1059 
1060 // Free Linux resources related to the OSThread
1061 void os::free_thread(OSThread* osthread) {
1062   assert(osthread != NULL, "osthread not set");
1063 
1064   // We are told to free resources of the argument thread,
1065   // but we can only really operate on the current thread.
1066   assert(Thread::current()->osthread() == osthread,
1067          "os::free_thread but not current thread");
1068 
1069 #ifdef ASSERT
1070   sigset_t current;
1071   sigemptyset(&current);
1072   pthread_sigmask(SIG_SETMASK, NULL, &current);
1073   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
1074 #endif
1075 
1076   // Restore caller's signal mask
1077   sigset_t sigmask = osthread->caller_sigmask();
1078   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1079 
1080   delete osthread;
1081 }
1082 
1083 //////////////////////////////////////////////////////////////////////////////
1084 // primordial thread
1085 
1086 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1087 bool os::is_primordial_thread(void) {
1088   if (suppress_primordial_thread_resolution) {
1089     return false;
1090   }
1091   char dummy;
1092   // If called before init complete, thread stack bottom will be null.
1093   // Can be called if fatal error occurs before initialization.
1094   if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1095   assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1096          os::Linux::initial_thread_stack_size()   != 0,
1097          "os::init did not locate primordial thread's stack region");
1098   if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1099       (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1100                         os::Linux::initial_thread_stack_size()) {
1101     return true;
1102   } else {
1103     return false;
1104   }
1105 }
1106 
1107 // Find the virtual memory area that contains addr
1108 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1109   FILE *fp = fopen("/proc/self/maps", "r");
1110   if (fp) {
1111     address low, high;
1112     while (!feof(fp)) {
1113       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1114         if (low <= addr && addr < high) {
1115           if (vma_low)  *vma_low  = low;
1116           if (vma_high) *vma_high = high;
1117           fclose(fp);
1118           return true;
1119         }
1120       }
1121       for (;;) {
1122         int ch = fgetc(fp);
1123         if (ch == EOF || ch == (int)'\n') break;
1124       }
1125     }
1126     fclose(fp);
1127   }
1128   return false;
1129 }
1130 
1131 // Locate primordial thread stack. This special handling of primordial thread stack
1132 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1133 // bogus value for the primordial process thread. While the launcher has created
1134 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1135 // JNI invocation API from a primordial thread.
1136 void os::Linux::capture_initial_stack(size_t max_size) {
1137 
1138   // max_size is either 0 (which means accept OS default for thread stacks) or
1139   // a user-specified value known to be at least the minimum needed. If we
1140   // are actually on the primordial thread we can make it appear that we have a
1141   // smaller max_size stack by inserting the guard pages at that location. But we
1142   // cannot do anything to emulate a larger stack than what has been provided by
1143   // the OS or threading library. In fact if we try to use a stack greater than
1144   // what is set by rlimit then we will crash the hosting process.
1145 
1146   // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1147   // If this is "unlimited" then it will be a huge value.
1148   struct rlimit rlim;
1149   getrlimit(RLIMIT_STACK, &rlim);
1150   size_t stack_size = rlim.rlim_cur;
1151 
1152   // 6308388: a bug in ld.so will relocate its own .data section to the
1153   //   lower end of primordial stack; reduce ulimit -s value a little bit
1154   //   so we won't install guard page on ld.so's data section.
1155   //   But ensure we don't underflow the stack size - allow 1 page spare
1156   if (stack_size >= (size_t)(3 * page_size())) {
1157     stack_size -= 2 * page_size();
1158   }
1159 
1160   // Try to figure out where the stack base (top) is. This is harder.
1161   //
1162   // When an application is started, glibc saves the initial stack pointer in
1163   // a global variable "__libc_stack_end", which is then used by system
1164   // libraries. __libc_stack_end should be pretty close to stack top. The
1165   // variable is available since the very early days. However, because it is
1166   // a private interface, it could disappear in the future.
1167   //
1168   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1169   // to __libc_stack_end, it is very close to stack top, but isn't the real
1170   // stack top. Note that /proc may not exist if VM is running as a chroot
1171   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1172   // /proc/<pid>/stat could change in the future (though unlikely).
1173   //
1174   // We try __libc_stack_end first. If that doesn't work, look for
1175   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1176   // as a hint, which should work well in most cases.
1177 
1178   uintptr_t stack_start;
1179 
1180   // try __libc_stack_end first
1181   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1182   if (p && *p) {
1183     stack_start = *p;
1184   } else {
1185     // see if we can get the start_stack field from /proc/self/stat
1186     FILE *fp;
1187     int pid;
1188     char state;
1189     int ppid;
1190     int pgrp;
1191     int session;
1192     int nr;
1193     int tpgrp;
1194     unsigned long flags;
1195     unsigned long minflt;
1196     unsigned long cminflt;
1197     unsigned long majflt;
1198     unsigned long cmajflt;
1199     unsigned long utime;
1200     unsigned long stime;
1201     long cutime;
1202     long cstime;
1203     long prio;
1204     long nice;
1205     long junk;
1206     long it_real;
1207     uintptr_t start;
1208     uintptr_t vsize;
1209     intptr_t rss;
1210     uintptr_t rsslim;
1211     uintptr_t scodes;
1212     uintptr_t ecode;
1213     int i;
1214 
1215     // Figure what the primordial thread stack base is. Code is inspired
1216     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1217     // followed by command name surrounded by parentheses, state, etc.
1218     char stat[2048];
1219     int statlen;
1220 
1221     fp = fopen("/proc/self/stat", "r");
1222     if (fp) {
1223       statlen = fread(stat, 1, 2047, fp);
1224       stat[statlen] = '\0';
1225       fclose(fp);
1226 
1227       // Skip pid and the command string. Note that we could be dealing with
1228       // weird command names, e.g. user could decide to rename java launcher
1229       // to "java 1.4.2 :)", then the stat file would look like
1230       //                1234 (java 1.4.2 :)) R ... ...
1231       // We don't really need to know the command string, just find the last
1232       // occurrence of ")" and then start parsing from there. See bug 4726580.
1233       char * s = strrchr(stat, ')');
1234 
1235       i = 0;
1236       if (s) {
1237         // Skip blank chars
1238         do { s++; } while (s && isspace(*s));
1239 
1240 #define _UFM UINTX_FORMAT
1241 #define _DFM INTX_FORMAT
1242 
1243         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1244         //              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
1245         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,
1246                    &state,          // 3  %c
1247                    &ppid,           // 4  %d
1248                    &pgrp,           // 5  %d
1249                    &session,        // 6  %d
1250                    &nr,             // 7  %d
1251                    &tpgrp,          // 8  %d
1252                    &flags,          // 9  %lu
1253                    &minflt,         // 10 %lu
1254                    &cminflt,        // 11 %lu
1255                    &majflt,         // 12 %lu
1256                    &cmajflt,        // 13 %lu
1257                    &utime,          // 14 %lu
1258                    &stime,          // 15 %lu
1259                    &cutime,         // 16 %ld
1260                    &cstime,         // 17 %ld
1261                    &prio,           // 18 %ld
1262                    &nice,           // 19 %ld
1263                    &junk,           // 20 %ld
1264                    &it_real,        // 21 %ld
1265                    &start,          // 22 UINTX_FORMAT
1266                    &vsize,          // 23 UINTX_FORMAT
1267                    &rss,            // 24 INTX_FORMAT
1268                    &rsslim,         // 25 UINTX_FORMAT
1269                    &scodes,         // 26 UINTX_FORMAT
1270                    &ecode,          // 27 UINTX_FORMAT
1271                    &stack_start);   // 28 UINTX_FORMAT
1272       }
1273 
1274 #undef _UFM
1275 #undef _DFM
1276 
1277       if (i != 28 - 2) {
1278         assert(false, "Bad conversion from /proc/self/stat");
1279         // product mode - assume we are the primordial thread, good luck in the
1280         // embedded case.
1281         warning("Can't detect primordial thread stack location - bad conversion");
1282         stack_start = (uintptr_t) &rlim;
1283       }
1284     } else {
1285       // For some reason we can't open /proc/self/stat (for example, running on
1286       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1287       // most cases, so don't abort:
1288       warning("Can't detect primordial thread stack location - no /proc/self/stat");
1289       stack_start = (uintptr_t) &rlim;
1290     }
1291   }
1292 
1293   // Now we have a pointer (stack_start) very close to the stack top, the
1294   // next thing to do is to figure out the exact location of stack top. We
1295   // can find out the virtual memory area that contains stack_start by
1296   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1297   // and its upper limit is the real stack top. (again, this would fail if
1298   // running inside chroot, because /proc may not exist.)
1299 
1300   uintptr_t stack_top;
1301   address low, high;
1302   if (find_vma((address)stack_start, &low, &high)) {
1303     // success, "high" is the true stack top. (ignore "low", because initial
1304     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1305     stack_top = (uintptr_t)high;
1306   } else {
1307     // failed, likely because /proc/self/maps does not exist
1308     warning("Can't detect primordial thread stack location - find_vma failed");
1309     // best effort: stack_start is normally within a few pages below the real
1310     // stack top, use it as stack top, and reduce stack size so we won't put
1311     // guard page outside stack.
1312     stack_top = stack_start;
1313     stack_size -= 16 * page_size();
1314   }
1315 
1316   // stack_top could be partially down the page so align it
1317   stack_top = align_up(stack_top, page_size());
1318 
1319   // Allowed stack value is minimum of max_size and what we derived from rlimit
1320   if (max_size > 0) {
1321     _initial_thread_stack_size = MIN2(max_size, stack_size);
1322   } else {
1323     // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1324     // clamp it at 8MB as we do on Solaris
1325     _initial_thread_stack_size = MIN2(stack_size, 8*M);
1326   }
1327   _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1328   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1329 
1330   assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1331 
1332   if (log_is_enabled(Info, os, thread)) {
1333     // See if we seem to be on primordial process thread
1334     bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1335                       uintptr_t(&rlim) < stack_top;
1336 
1337     log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1338                          SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1339                          primordial ? "primordial" : "user", max_size / K,  _initial_thread_stack_size / K,
1340                          stack_top, intptr_t(_initial_thread_stack_bottom));
1341   }
1342 }
1343 
1344 ////////////////////////////////////////////////////////////////////////////////
1345 // time support
1346 
1347 #ifndef SUPPORTS_CLOCK_MONOTONIC
1348 #error "Build platform doesn't support clock_gettime and related functionality"
1349 #endif
1350 
1351 // Time since start-up in seconds to a fine granularity.
1352 // Used by VMSelfDestructTimer and the MemProfiler.
1353 double os::elapsedTime() {
1354 
1355   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1356 }
1357 
1358 jlong os::elapsed_counter() {
1359   return javaTimeNanos() - initial_time_count;
1360 }
1361 
1362 jlong os::elapsed_frequency() {
1363   return NANOSECS_PER_SEC; // nanosecond resolution
1364 }
1365 
1366 bool os::supports_vtime() { return true; }
1367 bool os::enable_vtime()   { return false; }
1368 bool os::vtime_enabled()  { return false; }
1369 
1370 double os::elapsedVTime() {
1371   struct rusage usage;
1372   int retval = getrusage(RUSAGE_THREAD, &usage);
1373   if (retval == 0) {
1374     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);
1375   } else {
1376     // better than nothing, but not much
1377     return elapsedTime();
1378   }
1379 }
1380 
1381 jlong os::javaTimeMillis() {
1382   timeval time;
1383   int status = gettimeofday(&time, NULL);
1384   assert(status != -1, "linux error");
1385   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1386 }
1387 
1388 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1389   timeval time;
1390   int status = gettimeofday(&time, NULL);
1391   assert(status != -1, "linux error");
1392   seconds = jlong(time.tv_sec);
1393   nanos = jlong(time.tv_usec) * 1000;
1394 }
1395 
1396 void os::Linux::fast_thread_clock_init() {
1397   if (!UseLinuxPosixThreadCPUClocks) {
1398     return;
1399   }
1400   clockid_t clockid;
1401   struct timespec tp;
1402   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1403       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1404 
1405   // Switch to using fast clocks for thread cpu time if
1406   // the clock_getres() returns 0 error code.
1407   // Note, that some kernels may support the current thread
1408   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1409   // returned by the pthread_getcpuclockid().
1410   // If the fast Posix clocks are supported then the clock_getres()
1411   // must return at least tp.tv_sec == 0 which means a resolution
1412   // better than 1 sec. This is extra check for reliability.
1413 
1414   if (pthread_getcpuclockid_func &&
1415       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1416       os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1417     _supports_fast_thread_cpu_time = true;
1418     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1419   }
1420 }
1421 
1422 jlong os::javaTimeNanos() {
1423   if (os::supports_monotonic_clock()) {
1424     struct timespec tp;
1425     int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1426     assert(status == 0, "gettime error");
1427     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1428     return result;
1429   } else {
1430     timeval time;
1431     int status = gettimeofday(&time, NULL);
1432     assert(status != -1, "linux error");
1433     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1434     return 1000 * usecs;
1435   }
1436 }
1437 
1438 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1439   if (os::supports_monotonic_clock()) {
1440     info_ptr->max_value = ALL_64_BITS;
1441 
1442     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1443     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1444     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1445   } else {
1446     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1447     info_ptr->max_value = ALL_64_BITS;
1448 
1449     // gettimeofday is a real time clock so it skips
1450     info_ptr->may_skip_backward = true;
1451     info_ptr->may_skip_forward = true;
1452   }
1453 
1454   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1455 }
1456 
1457 // Return the real, user, and system times in seconds from an
1458 // arbitrary fixed point in the past.
1459 bool os::getTimesSecs(double* process_real_time,
1460                       double* process_user_time,
1461                       double* process_system_time) {
1462   struct tms ticks;
1463   clock_t real_ticks = times(&ticks);
1464 
1465   if (real_ticks == (clock_t) (-1)) {
1466     return false;
1467   } else {
1468     double ticks_per_second = (double) clock_tics_per_sec;
1469     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1470     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1471     *process_real_time = ((double) real_ticks) / ticks_per_second;
1472 
1473     return true;
1474   }
1475 }
1476 
1477 
1478 char * os::local_time_string(char *buf, size_t buflen) {
1479   struct tm t;
1480   time_t long_time;
1481   time(&long_time);
1482   localtime_r(&long_time, &t);
1483   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1484                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1485                t.tm_hour, t.tm_min, t.tm_sec);
1486   return buf;
1487 }
1488 
1489 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1490   return localtime_r(clock, res);
1491 }
1492 
1493 ////////////////////////////////////////////////////////////////////////////////
1494 // runtime exit support
1495 
1496 // Note: os::shutdown() might be called very early during initialization, or
1497 // called from signal handler. Before adding something to os::shutdown(), make
1498 // sure it is async-safe and can handle partially initialized VM.
1499 void os::shutdown() {
1500 
1501   // allow PerfMemory to attempt cleanup of any persistent resources
1502   perfMemory_exit();
1503 
1504   // needs to remove object in file system
1505   AttachListener::abort();
1506 
1507   // flush buffered output, finish log files
1508   ostream_abort();
1509 
1510   // Check for abort hook
1511   abort_hook_t abort_hook = Arguments::abort_hook();
1512   if (abort_hook != NULL) {
1513     abort_hook();
1514   }
1515 
1516 }
1517 
1518 // Note: os::abort() might be called very early during initialization, or
1519 // called from signal handler. Before adding something to os::abort(), make
1520 // sure it is async-safe and can handle partially initialized VM.
1521 void os::abort(bool dump_core, void* siginfo, const void* context) {
1522   os::shutdown();
1523   if (dump_core) {
1524     if (DumpPrivateMappingsInCore) {
1525       ClassLoader::close_jrt_image();
1526     }
1527 #ifndef PRODUCT
1528     fdStream out(defaultStream::output_fd());
1529     out.print_raw("Current thread is ");
1530     char buf[16];
1531     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1532     out.print_raw_cr(buf);
1533     out.print_raw_cr("Dumping core ...");
1534 #endif
1535     ::abort(); // dump core
1536   }
1537 
1538   ::exit(1);
1539 }
1540 
1541 // Die immediately, no exit hook, no abort hook, no cleanup.
1542 // Dump a core file, if possible, for debugging.
1543 void os::die() {
1544   if (TestUnresponsiveErrorHandler && !CreateCoredumpOnCrash) {
1545     // For TimeoutInErrorHandlingTest.java, we just kill the VM
1546     // and don't take the time to generate a core file.
1547     os::signal_raise(SIGKILL);
1548   } else {
1549     ::abort();
1550   }
1551 }
1552 
1553 // thread_id is kernel thread id (similar to Solaris LWP id)
1554 intx os::current_thread_id() { return os::Linux::gettid(); }
1555 int os::current_process_id() {
1556   return ::getpid();
1557 }
1558 
1559 // DLL functions
1560 
1561 const char* os::dll_file_extension() { return ".so"; }
1562 
1563 // This must be hard coded because it's the system's temporary
1564 // directory not the java application's temp directory, ala java.io.tmpdir.
1565 const char* os::get_temp_directory() { return "/tmp"; }
1566 
1567 static bool file_exists(const char* filename) {
1568   struct stat statbuf;
1569   if (filename == NULL || strlen(filename) == 0) {
1570     return false;
1571   }
1572   return os::stat(filename, &statbuf) == 0;
1573 }
1574 
1575 // check if addr is inside libjvm.so
1576 bool os::address_is_in_vm(address addr) {
1577   static address libjvm_base_addr;
1578   Dl_info dlinfo;
1579 
1580   if (libjvm_base_addr == NULL) {
1581     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1582       libjvm_base_addr = (address)dlinfo.dli_fbase;
1583     }
1584     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1585   }
1586 
1587   if (dladdr((void *)addr, &dlinfo) != 0) {
1588     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1589   }
1590 
1591   return false;
1592 }
1593 
1594 bool os::dll_address_to_function_name(address addr, char *buf,
1595                                       int buflen, int *offset,
1596                                       bool demangle) {
1597   // buf is not optional, but offset is optional
1598   assert(buf != NULL, "sanity check");
1599 
1600   Dl_info dlinfo;
1601 
1602   if (dladdr((void*)addr, &dlinfo) != 0) {
1603     // see if we have a matching symbol
1604     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1605       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1606         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1607       }
1608       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1609       return true;
1610     }
1611     // no matching symbol so try for just file info
1612     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1613       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1614                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1615         return true;
1616       }
1617     }
1618   }
1619 
1620   buf[0] = '\0';
1621   if (offset != NULL) *offset = -1;
1622   return false;
1623 }
1624 
1625 struct _address_to_library_name {
1626   address addr;          // input : memory address
1627   size_t  buflen;        //         size of fname
1628   char*   fname;         // output: library name
1629   address base;          //         library base addr
1630 };
1631 
1632 static int address_to_library_name_callback(struct dl_phdr_info *info,
1633                                             size_t size, void *data) {
1634   int i;
1635   bool found = false;
1636   address libbase = NULL;
1637   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1638 
1639   // iterate through all loadable segments
1640   for (i = 0; i < info->dlpi_phnum; i++) {
1641     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1642     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1643       // base address of a library is the lowest address of its loaded
1644       // segments.
1645       if (libbase == NULL || libbase > segbase) {
1646         libbase = segbase;
1647       }
1648       // see if 'addr' is within current segment
1649       if (segbase <= d->addr &&
1650           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1651         found = true;
1652       }
1653     }
1654   }
1655 
1656   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1657   // so dll_address_to_library_name() can fall through to use dladdr() which
1658   // can figure out executable name from argv[0].
1659   if (found && info->dlpi_name && info->dlpi_name[0]) {
1660     d->base = libbase;
1661     if (d->fname) {
1662       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1663     }
1664     return 1;
1665   }
1666   return 0;
1667 }
1668 
1669 bool os::dll_address_to_library_name(address addr, char* buf,
1670                                      int buflen, int* offset) {
1671   // buf is not optional, but offset is optional
1672   assert(buf != NULL, "sanity check");
1673 
1674   Dl_info dlinfo;
1675   struct _address_to_library_name data;
1676 
1677   // There is a bug in old glibc dladdr() implementation that it could resolve
1678   // to wrong library name if the .so file has a base address != NULL. Here
1679   // we iterate through the program headers of all loaded libraries to find
1680   // out which library 'addr' really belongs to. This workaround can be
1681   // removed once the minimum requirement for glibc is moved to 2.3.x.
1682   data.addr = addr;
1683   data.fname = buf;
1684   data.buflen = buflen;
1685   data.base = NULL;
1686   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1687 
1688   if (rslt) {
1689     // buf already contains library name
1690     if (offset) *offset = addr - data.base;
1691     return true;
1692   }
1693   if (dladdr((void*)addr, &dlinfo) != 0) {
1694     if (dlinfo.dli_fname != NULL) {
1695       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1696     }
1697     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1698       *offset = addr - (address)dlinfo.dli_fbase;
1699     }
1700     return true;
1701   }
1702 
1703   buf[0] = '\0';
1704   if (offset) *offset = -1;
1705   return false;
1706 }
1707 
1708 // Loads .dll/.so and
1709 // in case of error it checks if .dll/.so was built for the
1710 // same architecture as Hotspot is running on
1711 
1712 
1713 // Remember the stack's state. The Linux dynamic linker will change
1714 // the stack to 'executable' at most once, so we must safepoint only once.
1715 bool os::Linux::_stack_is_executable = false;
1716 
1717 // VM operation that loads a library.  This is necessary if stack protection
1718 // of the Java stacks can be lost during loading the library.  If we
1719 // do not stop the Java threads, they can stack overflow before the stacks
1720 // are protected again.
1721 class VM_LinuxDllLoad: public VM_Operation {
1722  private:
1723   const char *_filename;
1724   char *_ebuf;
1725   int _ebuflen;
1726   void *_lib;
1727  public:
1728   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1729     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1730   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1731   void doit() {
1732     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1733     os::Linux::_stack_is_executable = true;
1734   }
1735   void* loaded_library() { return _lib; }
1736 };
1737 
1738 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1739   void * result = NULL;
1740   bool load_attempted = false;
1741 
1742   log_info(os)("attempting shared library load of %s", filename);
1743 
1744   // Check whether the library to load might change execution rights
1745   // of the stack. If they are changed, the protection of the stack
1746   // guard pages will be lost. We need a safepoint to fix this.
1747   //
1748   // See Linux man page execstack(8) for more info.
1749   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1750     if (!ElfFile::specifies_noexecstack(filename)) {
1751       if (!is_init_completed()) {
1752         os::Linux::_stack_is_executable = true;
1753         // This is OK - No Java threads have been created yet, and hence no
1754         // stack guard pages to fix.
1755         //
1756         // Dynamic loader will make all stacks executable after
1757         // this function returns, and will not do that again.
1758         assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1759       } else {
1760         warning("You have loaded library %s which might have disabled stack guard. "
1761                 "The VM will try to fix the stack guard now.\n"
1762                 "It's highly recommended that you fix the library with "
1763                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1764                 filename);
1765 
1766         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1767         JavaThread *jt = JavaThread::current();
1768         if (jt->thread_state() != _thread_in_native) {
1769           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1770           // that requires ExecStack. Cannot enter safe point. Let's give up.
1771           warning("Unable to fix stack guard. Giving up.");
1772         } else {
1773           if (!LoadExecStackDllInVMThread) {
1774             // This is for the case where the DLL has an static
1775             // constructor function that executes JNI code. We cannot
1776             // load such DLLs in the VMThread.
1777             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1778           }
1779 
1780           ThreadInVMfromNative tiv(jt);
1781           debug_only(VMNativeEntryWrapper vew;)
1782 
1783           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1784           VMThread::execute(&op);
1785           if (LoadExecStackDllInVMThread) {
1786             result = op.loaded_library();
1787           }
1788           load_attempted = true;
1789         }
1790       }
1791     }
1792   }
1793 
1794   if (!load_attempted) {
1795     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1796   }
1797 
1798   if (result != NULL) {
1799     // Successful loading
1800     return result;
1801   }
1802 
1803   Elf32_Ehdr elf_head;
1804   int diag_msg_max_length=ebuflen-strlen(ebuf);
1805   char* diag_msg_buf=ebuf+strlen(ebuf);
1806 
1807   if (diag_msg_max_length==0) {
1808     // No more space in ebuf for additional diagnostics message
1809     return NULL;
1810   }
1811 
1812 
1813   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1814 
1815   if (file_descriptor < 0) {
1816     // Can't open library, report dlerror() message
1817     return NULL;
1818   }
1819 
1820   bool failed_to_read_elf_head=
1821     (sizeof(elf_head)!=
1822      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1823 
1824   ::close(file_descriptor);
1825   if (failed_to_read_elf_head) {
1826     // file i/o error - report dlerror() msg
1827     return NULL;
1828   }
1829 
1830   if (elf_head.e_ident[EI_DATA] != LITTLE_ENDIAN_ONLY(ELFDATA2LSB) BIG_ENDIAN_ONLY(ELFDATA2MSB)) {
1831     // handle invalid/out of range endianness values
1832     if (elf_head.e_ident[EI_DATA] == 0 || elf_head.e_ident[EI_DATA] > 2) {
1833       return NULL;
1834     }
1835 
1836 #if defined(VM_LITTLE_ENDIAN)
1837     // VM is LE, shared object BE
1838     elf_head.e_machine = be16toh(elf_head.e_machine);
1839 #else
1840     // VM is BE, shared object LE
1841     elf_head.e_machine = le16toh(elf_head.e_machine);
1842 #endif
1843   }
1844 
1845   typedef struct {
1846     Elf32_Half    code;         // Actual value as defined in elf.h
1847     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1848     unsigned char elf_class;    // 32 or 64 bit
1849     unsigned char endianness;   // MSB or LSB
1850     char*         name;         // String representation
1851   } arch_t;
1852 
1853 #ifndef EM_486
1854   #define EM_486          6               /* Intel 80486 */
1855 #endif
1856 #ifndef EM_AARCH64
1857   #define EM_AARCH64    183               /* ARM AARCH64 */
1858 #endif
1859 
1860   static const arch_t arch_array[]={
1861     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1862     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1863     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1864     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1865     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1866     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1867     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1868     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1869 #if defined(VM_LITTLE_ENDIAN)
1870     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1871     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1872 #else
1873     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1874     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1875 #endif
1876     {EM_ARM,         EM_ARM,     ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1877     // we only support 64 bit z architecture
1878     {EM_S390,        EM_S390,    ELFCLASS64, ELFDATA2MSB, (char*)"IBM System/390"},
1879     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1880     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1881     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1882     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1883     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1884     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1885   };
1886 
1887 #if  (defined IA32)
1888   static  Elf32_Half running_arch_code=EM_386;
1889 #elif   (defined AMD64) || (defined X32)
1890   static  Elf32_Half running_arch_code=EM_X86_64;
1891 #elif  (defined IA64)
1892   static  Elf32_Half running_arch_code=EM_IA_64;
1893 #elif  (defined __sparc) && (defined _LP64)
1894   static  Elf32_Half running_arch_code=EM_SPARCV9;
1895 #elif  (defined __sparc) && (!defined _LP64)
1896   static  Elf32_Half running_arch_code=EM_SPARC;
1897 #elif  (defined __powerpc64__)
1898   static  Elf32_Half running_arch_code=EM_PPC64;
1899 #elif  (defined __powerpc__)
1900   static  Elf32_Half running_arch_code=EM_PPC;
1901 #elif  (defined AARCH64)
1902   static  Elf32_Half running_arch_code=EM_AARCH64;
1903 #elif  (defined ARM)
1904   static  Elf32_Half running_arch_code=EM_ARM;
1905 #elif  (defined S390)
1906   static  Elf32_Half running_arch_code=EM_S390;
1907 #elif  (defined ALPHA)
1908   static  Elf32_Half running_arch_code=EM_ALPHA;
1909 #elif  (defined MIPSEL)
1910   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1911 #elif  (defined PARISC)
1912   static  Elf32_Half running_arch_code=EM_PARISC;
1913 #elif  (defined MIPS)
1914   static  Elf32_Half running_arch_code=EM_MIPS;
1915 #elif  (defined M68K)
1916   static  Elf32_Half running_arch_code=EM_68K;
1917 #elif  (defined SH)
1918   static  Elf32_Half running_arch_code=EM_SH;
1919 #else
1920     #error Method os::dll_load requires that one of following is defined:\
1921         AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1922 #endif
1923 
1924   // Identify compatibility class for VM's architecture and library's architecture
1925   // Obtain string descriptions for architectures
1926 
1927   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1928   int running_arch_index=-1;
1929 
1930   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1931     if (running_arch_code == arch_array[i].code) {
1932       running_arch_index    = i;
1933     }
1934     if (lib_arch.code == arch_array[i].code) {
1935       lib_arch.compat_class = arch_array[i].compat_class;
1936       lib_arch.name         = arch_array[i].name;
1937     }
1938   }
1939 
1940   assert(running_arch_index != -1,
1941          "Didn't find running architecture code (running_arch_code) in arch_array");
1942   if (running_arch_index == -1) {
1943     // Even though running architecture detection failed
1944     // we may still continue with reporting dlerror() message
1945     return NULL;
1946   }
1947 
1948   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1949     if (lib_arch.name != NULL) {
1950       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1951                  " (Possible cause: can't load %s .so on a %s platform)",
1952                  lib_arch.name, arch_array[running_arch_index].name);
1953     } else {
1954       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1955                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s platform)",
1956                  lib_arch.code, arch_array[running_arch_index].name);
1957     }
1958     return NULL;
1959   }
1960 
1961   if (lib_arch.endianness != arch_array[running_arch_index].endianness) {
1962     ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: endianness mismatch)");
1963     return NULL;
1964   }
1965 
1966   // ELF file class/capacity : 0 - invalid, 1 - 32bit, 2 - 64bit
1967   if (lib_arch.elf_class > 2 || lib_arch.elf_class < 1) {
1968     ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: invalid ELF file class)");
1969     return NULL;
1970   }
1971 
1972   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1973     ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1974                " (Possible cause: architecture word width mismatch, can't load %d-bit .so on a %d-bit platform)",
1975                (int) lib_arch.elf_class * 32, arch_array[running_arch_index].elf_class * 32);
1976     return NULL;
1977   }
1978 
1979   return NULL;
1980 }
1981 
1982 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1983                                 int ebuflen) {
1984   void * result = ::dlopen(filename, RTLD_LAZY);
1985   if (result == NULL) {
1986     const char* error_report = ::dlerror();
1987     if (error_report == NULL) {
1988       error_report = "dlerror returned no error description";
1989     }
1990     if (ebuf != NULL && ebuflen > 0) {
1991       ::strncpy(ebuf, error_report, ebuflen-1);
1992       ebuf[ebuflen-1]='\0';
1993     }
1994     Events::log(NULL, "Loading shared library %s failed, %s", filename, error_report);
1995     log_info(os)("shared library load of %s failed, %s", filename, error_report);
1996   } else {
1997     Events::log(NULL, "Loaded shared library %s", filename);
1998     log_info(os)("shared library load of %s was successful", filename);
1999   }
2000   return result;
2001 }
2002 
2003 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2004                                        int ebuflen) {
2005   void * result = NULL;
2006   if (LoadExecStackDllInVMThread) {
2007     result = dlopen_helper(filename, ebuf, ebuflen);
2008   }
2009 
2010   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2011   // library that requires an executable stack, or which does not have this
2012   // stack attribute set, dlopen changes the stack attribute to executable. The
2013   // read protection of the guard pages gets lost.
2014   //
2015   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2016   // may have been queued at the same time.
2017 
2018   if (!_stack_is_executable) {
2019     for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
2020       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
2021           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
2022         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
2023           warning("Attempt to reguard stack yellow zone failed.");
2024         }
2025       }
2026     }
2027   }
2028 
2029   return result;
2030 }
2031 
2032 void* os::dll_lookup(void* handle, const char* name) {
2033   void* res = dlsym(handle, name);
2034   return res;
2035 }
2036 
2037 void* os::get_default_process_handle() {
2038   return (void*)::dlopen(NULL, RTLD_LAZY);
2039 }
2040 
2041 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
2042   int fd = ::open(filename, O_RDONLY);
2043   if (fd == -1) {
2044     return false;
2045   }
2046 
2047   if (hdr != NULL) {
2048     st->print_cr("%s", hdr);
2049   }
2050 
2051   char buf[33];
2052   int bytes;
2053   buf[32] = '\0';
2054   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
2055     st->print_raw(buf, bytes);
2056   }
2057 
2058   ::close(fd);
2059 
2060   return true;
2061 }
2062 
2063 void os::print_dll_info(outputStream *st) {
2064   st->print_cr("Dynamic libraries:");
2065 
2066   char fname[32];
2067   pid_t pid = os::Linux::gettid();
2068 
2069   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2070 
2071   if (!_print_ascii_file(fname, st)) {
2072     st->print("Can not get library information for pid = %d\n", pid);
2073   }
2074 }
2075 
2076 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2077   FILE *procmapsFile = NULL;
2078 
2079   // Open the procfs maps file for the current process
2080   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2081     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2082     char line[PATH_MAX + 100];
2083 
2084     // Read line by line from 'file'
2085     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2086       u8 base, top, offset, inode;
2087       char permissions[5];
2088       char device[6];
2089       char name[PATH_MAX + 1];
2090 
2091       // Parse fields from line
2092       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s",
2093              &base, &top, permissions, &offset, device, &inode, name);
2094 
2095       // Filter by device id '00:00' so that we only get file system mapped files.
2096       if (strcmp(device, "00:00") != 0) {
2097 
2098         // Call callback with the fields of interest
2099         if(callback(name, (address)base, (address)top, param)) {
2100           // Oops abort, callback aborted
2101           fclose(procmapsFile);
2102           return 1;
2103         }
2104       }
2105     }
2106     fclose(procmapsFile);
2107   }
2108   return 0;
2109 }
2110 
2111 void os::print_os_info_brief(outputStream* st) {
2112   os::Linux::print_distro_info(st);
2113 
2114   os::Posix::print_uname_info(st);
2115 
2116   os::Linux::print_libversion_info(st);
2117 
2118 }
2119 
2120 void os::print_os_info(outputStream* st) {
2121   st->print("OS:");
2122 
2123   os::Linux::print_distro_info(st);
2124 
2125   os::Posix::print_uname_info(st);
2126 
2127   // Print warning if unsafe chroot environment detected
2128   if (unsafe_chroot_detected) {
2129     st->print("WARNING!! ");
2130     st->print_cr("%s", unstable_chroot_error);
2131   }
2132 
2133   os::Linux::print_libversion_info(st);
2134 
2135   os::Posix::print_rlimit_info(st);
2136 
2137   os::Posix::print_load_average(st);
2138 
2139   os::Linux::print_full_memory_info(st);
2140 
2141   os::Linux::print_proc_sys_info(st);
2142 
2143   os::Linux::print_ld_preload_file(st);
2144 
2145   os::Linux::print_container_info(st);
2146 
2147   VM_Version::print_platform_virtualization_info(st);
2148 
2149   os::Linux::print_steal_info(st);
2150 }
2151 
2152 // Try to identify popular distros.
2153 // Most Linux distributions have a /etc/XXX-release file, which contains
2154 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2155 // file that also contains the OS version string. Some have more than one
2156 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2157 // /etc/redhat-release.), so the order is important.
2158 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2159 // their own specific XXX-release file as well as a redhat-release file.
2160 // Because of this the XXX-release file needs to be searched for before the
2161 // redhat-release file.
2162 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2163 // search for redhat-release / SuSE-release needs to be before lsb-release.
2164 // Since the lsb-release file is the new standard it needs to be searched
2165 // before the older style release files.
2166 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2167 // next to last resort.  The os-release file is a new standard that contains
2168 // distribution information and the system-release file seems to be an old
2169 // standard that has been replaced by the lsb-release and os-release files.
2170 // Searching for the debian_version file is the last resort.  It contains
2171 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2172 // "Debian " is printed before the contents of the debian_version file.
2173 
2174 const char* distro_files[] = {
2175   "/etc/oracle-release",
2176   "/etc/mandriva-release",
2177   "/etc/mandrake-release",
2178   "/etc/sun-release",
2179   "/etc/redhat-release",
2180   "/etc/SuSE-release",
2181   "/etc/lsb-release",
2182   "/etc/turbolinux-release",
2183   "/etc/gentoo-release",
2184   "/etc/ltib-release",
2185   "/etc/angstrom-version",
2186   "/etc/system-release",
2187   "/etc/os-release",
2188   NULL };
2189 
2190 void os::Linux::print_distro_info(outputStream* st) {
2191   for (int i = 0;; i++) {
2192     const char* file = distro_files[i];
2193     if (file == NULL) {
2194       break;  // done
2195     }
2196     // If file prints, we found it.
2197     if (_print_ascii_file(file, st)) {
2198       return;
2199     }
2200   }
2201 
2202   if (file_exists("/etc/debian_version")) {
2203     st->print("Debian ");
2204     _print_ascii_file("/etc/debian_version", st);
2205   } else {
2206     st->print("Linux");
2207   }
2208   st->cr();
2209 }
2210 
2211 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2212   char buf[256];
2213   while (fgets(buf, sizeof(buf), fp)) {
2214     // Edit out extra stuff in expected format
2215     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2216       char* ptr = strstr(buf, "\"");  // the name is in quotes
2217       if (ptr != NULL) {
2218         ptr++; // go beyond first quote
2219         char* nl = strchr(ptr, '\"');
2220         if (nl != NULL) *nl = '\0';
2221         strncpy(distro, ptr, length);
2222       } else {
2223         ptr = strstr(buf, "=");
2224         ptr++; // go beyond equals then
2225         char* nl = strchr(ptr, '\n');
2226         if (nl != NULL) *nl = '\0';
2227         strncpy(distro, ptr, length);
2228       }
2229       return;
2230     } else if (get_first_line) {
2231       char* nl = strchr(buf, '\n');
2232       if (nl != NULL) *nl = '\0';
2233       strncpy(distro, buf, length);
2234       return;
2235     }
2236   }
2237   // print last line and close
2238   char* nl = strchr(buf, '\n');
2239   if (nl != NULL) *nl = '\0';
2240   strncpy(distro, buf, length);
2241 }
2242 
2243 static void parse_os_info(char* distro, size_t length, const char* file) {
2244   FILE* fp = fopen(file, "r");
2245   if (fp != NULL) {
2246     // if suse format, print out first line
2247     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2248     parse_os_info_helper(fp, distro, length, get_first_line);
2249     fclose(fp);
2250   }
2251 }
2252 
2253 void os::get_summary_os_info(char* buf, size_t buflen) {
2254   for (int i = 0;; i++) {
2255     const char* file = distro_files[i];
2256     if (file == NULL) {
2257       break; // ran out of distro_files
2258     }
2259     if (file_exists(file)) {
2260       parse_os_info(buf, buflen, file);
2261       return;
2262     }
2263   }
2264   // special case for debian
2265   if (file_exists("/etc/debian_version")) {
2266     strncpy(buf, "Debian ", buflen);
2267     if (buflen > 7) {
2268       parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2269     }
2270   } else {
2271     strncpy(buf, "Linux", buflen);
2272   }
2273 }
2274 
2275 void os::Linux::print_libversion_info(outputStream* st) {
2276   // libc, pthread
2277   st->print("libc:");
2278   st->print("%s ", os::Linux::glibc_version());
2279   st->print("%s ", os::Linux::libpthread_version());
2280   st->cr();
2281 }
2282 
2283 void os::Linux::print_proc_sys_info(outputStream* st) {
2284   st->cr();
2285   st->print_cr("/proc/sys/kernel/threads-max (system-wide limit on the number of threads):");
2286   _print_ascii_file("/proc/sys/kernel/threads-max", st);
2287   st->cr();
2288   st->cr();
2289 
2290   st->print_cr("/proc/sys/vm/max_map_count (maximum number of memory map areas a process may have):");
2291   _print_ascii_file("/proc/sys/vm/max_map_count", st);
2292   st->cr();
2293   st->cr();
2294 
2295   st->print_cr("/proc/sys/kernel/pid_max (system-wide limit on number of process identifiers):");
2296   _print_ascii_file("/proc/sys/kernel/pid_max", st);
2297   st->cr();
2298   st->cr();
2299 }
2300 
2301 void os::Linux::print_full_memory_info(outputStream* st) {
2302   st->print("\n/proc/meminfo:\n");
2303   _print_ascii_file("/proc/meminfo", st);
2304   st->cr();
2305 }
2306 
2307 void os::Linux::print_ld_preload_file(outputStream* st) {
2308   _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2309   st->cr();
2310 }
2311 
2312 void os::Linux::print_container_info(outputStream* st) {
2313   if (!OSContainer::is_containerized()) {
2314     return;
2315   }
2316 
2317   st->print("container (cgroup) information:\n");
2318 
2319   const char *p_ct = OSContainer::container_type();
2320   st->print("container_type: %s\n", p_ct != NULL ? p_ct : "not supported");
2321 
2322   char *p = OSContainer::cpu_cpuset_cpus();
2323   st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "not supported");
2324   free(p);
2325 
2326   p = OSContainer::cpu_cpuset_memory_nodes();
2327   st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "not supported");
2328   free(p);
2329 
2330   int i = OSContainer::active_processor_count();
2331   st->print("active_processor_count: ");
2332   if (i > 0) {
2333     st->print("%d\n", i);
2334   } else {
2335     st->print("not supported\n");
2336   }
2337 
2338   i = OSContainer::cpu_quota();
2339   st->print("cpu_quota: ");
2340   if (i > 0) {
2341     st->print("%d\n", i);
2342   } else {
2343     st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no quota");
2344   }
2345 
2346   i = OSContainer::cpu_period();
2347   st->print("cpu_period: ");
2348   if (i > 0) {
2349     st->print("%d\n", i);
2350   } else {
2351     st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no period");
2352   }
2353 
2354   i = OSContainer::cpu_shares();
2355   st->print("cpu_shares: ");
2356   if (i > 0) {
2357     st->print("%d\n", i);
2358   } else {
2359     st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no shares");
2360   }
2361 
2362   jlong j = OSContainer::memory_limit_in_bytes();
2363   st->print("memory_limit_in_bytes: ");
2364   if (j > 0) {
2365     st->print(JLONG_FORMAT "\n", j);
2366   } else {
2367     st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2368   }
2369 
2370   j = OSContainer::memory_and_swap_limit_in_bytes();
2371   st->print("memory_and_swap_limit_in_bytes: ");
2372   if (j > 0) {
2373     st->print(JLONG_FORMAT "\n", j);
2374   } else {
2375     st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2376   }
2377 
2378   j = OSContainer::memory_soft_limit_in_bytes();
2379   st->print("memory_soft_limit_in_bytes: ");
2380   if (j > 0) {
2381     st->print(JLONG_FORMAT "\n", j);
2382   } else {
2383     st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2384   }
2385 
2386   j = OSContainer::OSContainer::memory_usage_in_bytes();
2387   st->print("memory_usage_in_bytes: ");
2388   if (j > 0) {
2389     st->print(JLONG_FORMAT "\n", j);
2390   } else {
2391     st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2392   }
2393 
2394   j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2395   st->print("memory_max_usage_in_bytes: ");
2396   if (j > 0) {
2397     st->print(JLONG_FORMAT "\n", j);
2398   } else {
2399     st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2400   }
2401   st->cr();
2402 }
2403 
2404 void os::Linux::print_steal_info(outputStream* st) {
2405   if (has_initial_tick_info) {
2406     CPUPerfTicks pticks;
2407     bool res = os::Linux::get_tick_information(&pticks, -1);
2408 
2409     if (res && pticks.has_steal_ticks) {
2410       uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks;
2411       uint64_t total_ticks_difference = pticks.total - initial_total_ticks;
2412       double steal_ticks_perc = 0.0;
2413       if (total_ticks_difference != 0) {
2414         steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference;
2415       }
2416       st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference);
2417       st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc);
2418     }
2419   }
2420 }
2421 
2422 void os::print_memory_info(outputStream* st) {
2423 
2424   st->print("Memory:");
2425   st->print(" %dk page", os::vm_page_size()>>10);
2426 
2427   // values in struct sysinfo are "unsigned long"
2428   struct sysinfo si;
2429   sysinfo(&si);
2430 
2431   st->print(", physical " UINT64_FORMAT "k",
2432             os::physical_memory() >> 10);
2433   st->print("(" UINT64_FORMAT "k free)",
2434             os::available_memory() >> 10);
2435   st->print(", swap " UINT64_FORMAT "k",
2436             ((jlong)si.totalswap * si.mem_unit) >> 10);
2437   st->print("(" UINT64_FORMAT "k free)",
2438             ((jlong)si.freeswap * si.mem_unit) >> 10);
2439   st->cr();
2440 }
2441 
2442 // Print the first "model name" line and the first "flags" line
2443 // that we find and nothing more. We assume "model name" comes
2444 // before "flags" so if we find a second "model name", then the
2445 // "flags" field is considered missing.
2446 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2447 #if defined(IA32) || defined(AMD64)
2448   // Other platforms have less repetitive cpuinfo files
2449   FILE *fp = fopen("/proc/cpuinfo", "r");
2450   if (fp) {
2451     while (!feof(fp)) {
2452       if (fgets(buf, buflen, fp)) {
2453         // Assume model name comes before flags
2454         bool model_name_printed = false;
2455         if (strstr(buf, "model name") != NULL) {
2456           if (!model_name_printed) {
2457             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2458             st->print_raw(buf);
2459             model_name_printed = true;
2460           } else {
2461             // model name printed but not flags?  Odd, just return
2462             fclose(fp);
2463             return true;
2464           }
2465         }
2466         // print the flags line too
2467         if (strstr(buf, "flags") != NULL) {
2468           st->print_raw(buf);
2469           fclose(fp);
2470           return true;
2471         }
2472       }
2473     }
2474     fclose(fp);
2475   }
2476 #endif // x86 platforms
2477   return false;
2478 }
2479 
2480 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2481   // Only print the model name if the platform provides this as a summary
2482   if (!print_model_name_and_flags(st, buf, buflen)) {
2483     st->print("\n/proc/cpuinfo:\n");
2484     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2485       st->print_cr("  <Not Available>");
2486     }
2487   }
2488 }
2489 
2490 #if defined(AMD64) || defined(IA32) || defined(X32)
2491 const char* search_string = "model name";
2492 #elif defined(M68K)
2493 const char* search_string = "CPU";
2494 #elif defined(PPC64)
2495 const char* search_string = "cpu";
2496 #elif defined(S390)
2497 const char* search_string = "machine =";
2498 #elif defined(SPARC)
2499 const char* search_string = "cpu";
2500 #else
2501 const char* search_string = "Processor";
2502 #endif
2503 
2504 // Parses the cpuinfo file for string representing the model name.
2505 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2506   FILE* fp = fopen("/proc/cpuinfo", "r");
2507   if (fp != NULL) {
2508     while (!feof(fp)) {
2509       char buf[256];
2510       if (fgets(buf, sizeof(buf), fp)) {
2511         char* start = strstr(buf, search_string);
2512         if (start != NULL) {
2513           char *ptr = start + strlen(search_string);
2514           char *end = buf + strlen(buf);
2515           while (ptr != end) {
2516              // skip whitespace and colon for the rest of the name.
2517              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2518                break;
2519              }
2520              ptr++;
2521           }
2522           if (ptr != end) {
2523             // reasonable string, get rid of newline and keep the rest
2524             char* nl = strchr(buf, '\n');
2525             if (nl != NULL) *nl = '\0';
2526             strncpy(cpuinfo, ptr, length);
2527             fclose(fp);
2528             return;
2529           }
2530         }
2531       }
2532     }
2533     fclose(fp);
2534   }
2535   // cpuinfo not found or parsing failed, just print generic string.  The entire
2536   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2537 #if   defined(AARCH64)
2538   strncpy(cpuinfo, "AArch64", length);
2539 #elif defined(AMD64)
2540   strncpy(cpuinfo, "x86_64", length);
2541 #elif defined(ARM)  // Order wrt. AARCH64 is relevant!
2542   strncpy(cpuinfo, "ARM", length);
2543 #elif defined(IA32)
2544   strncpy(cpuinfo, "x86_32", length);
2545 #elif defined(IA64)
2546   strncpy(cpuinfo, "IA64", length);
2547 #elif defined(PPC)
2548   strncpy(cpuinfo, "PPC64", length);
2549 #elif defined(S390)
2550   strncpy(cpuinfo, "S390", length);
2551 #elif defined(SPARC)
2552   strncpy(cpuinfo, "sparcv9", length);
2553 #elif defined(ZERO_LIBARCH)
2554   strncpy(cpuinfo, ZERO_LIBARCH, length);
2555 #else
2556   strncpy(cpuinfo, "unknown", length);
2557 #endif
2558 }
2559 
2560 static void print_signal_handler(outputStream* st, int sig,
2561                                  char* buf, size_t buflen);
2562 
2563 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2564   st->print_cr("Signal Handlers:");
2565   print_signal_handler(st, SIGSEGV, buf, buflen);
2566   print_signal_handler(st, SIGBUS , buf, buflen);
2567   print_signal_handler(st, SIGFPE , buf, buflen);
2568   print_signal_handler(st, SIGPIPE, buf, buflen);
2569   print_signal_handler(st, SIGXFSZ, buf, buflen);
2570   print_signal_handler(st, SIGILL , buf, buflen);
2571   print_signal_handler(st, SR_signum, buf, buflen);
2572   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2573   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2574   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2575   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2576 #if defined(PPC64)
2577   print_signal_handler(st, SIGTRAP, buf, buflen);
2578 #endif
2579 }
2580 
2581 static char saved_jvm_path[MAXPATHLEN] = {0};
2582 
2583 // Find the full path to the current module, libjvm.so
2584 void os::jvm_path(char *buf, jint buflen) {
2585   // Error checking.
2586   if (buflen < MAXPATHLEN) {
2587     assert(false, "must use a large-enough buffer");
2588     buf[0] = '\0';
2589     return;
2590   }
2591   // Lazy resolve the path to current module.
2592   if (saved_jvm_path[0] != 0) {
2593     strcpy(buf, saved_jvm_path);
2594     return;
2595   }
2596 
2597   char dli_fname[MAXPATHLEN];
2598   bool ret = dll_address_to_library_name(
2599                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2600                                          dli_fname, sizeof(dli_fname), NULL);
2601   assert(ret, "cannot locate libjvm");
2602   char *rp = NULL;
2603   if (ret && dli_fname[0] != '\0') {
2604     rp = os::Posix::realpath(dli_fname, buf, buflen);
2605   }
2606   if (rp == NULL) {
2607     return;
2608   }
2609 
2610   if (Arguments::sun_java_launcher_is_altjvm()) {
2611     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2612     // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2613     // If "/jre/lib/" appears at the right place in the string, then
2614     // assume we are installed in a JDK and we're done. Otherwise, check
2615     // for a JAVA_HOME environment variable and fix up the path so it
2616     // looks like libjvm.so is installed there (append a fake suffix
2617     // hotspot/libjvm.so).
2618     const char *p = buf + strlen(buf) - 1;
2619     for (int count = 0; p > buf && count < 5; ++count) {
2620       for (--p; p > buf && *p != '/'; --p)
2621         /* empty */ ;
2622     }
2623 
2624     if (strncmp(p, "/jre/lib/", 9) != 0) {
2625       // Look for JAVA_HOME in the environment.
2626       char* java_home_var = ::getenv("JAVA_HOME");
2627       if (java_home_var != NULL && java_home_var[0] != 0) {
2628         char* jrelib_p;
2629         int len;
2630 
2631         // Check the current module name "libjvm.so".
2632         p = strrchr(buf, '/');
2633         if (p == NULL) {
2634           return;
2635         }
2636         assert(strstr(p, "/libjvm") == p, "invalid library name");
2637 
2638         rp = os::Posix::realpath(java_home_var, buf, buflen);
2639         if (rp == NULL) {
2640           return;
2641         }
2642 
2643         // determine if this is a legacy image or modules image
2644         // modules image doesn't have "jre" subdirectory
2645         len = strlen(buf);
2646         assert(len < buflen, "Ran out of buffer room");
2647         jrelib_p = buf + len;
2648         snprintf(jrelib_p, buflen-len, "/jre/lib");
2649         if (0 != access(buf, F_OK)) {
2650           snprintf(jrelib_p, buflen-len, "/lib");
2651         }
2652 
2653         if (0 == access(buf, F_OK)) {
2654           // Use current module name "libjvm.so"
2655           len = strlen(buf);
2656           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2657         } else {
2658           // Go back to path of .so
2659           rp = os::Posix::realpath(dli_fname, buf, buflen);
2660           if (rp == NULL) {
2661             return;
2662           }
2663         }
2664       }
2665     }
2666   }
2667 
2668   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2669   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2670 }
2671 
2672 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2673   // no prefix required, not even "_"
2674 }
2675 
2676 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2677   // no suffix required
2678 }
2679 
2680 ////////////////////////////////////////////////////////////////////////////////
2681 // sun.misc.Signal support
2682 
2683 static void UserHandler(int sig, void *siginfo, void *context) {
2684   // Ctrl-C is pressed during error reporting, likely because the error
2685   // handler fails to abort. Let VM die immediately.
2686   if (sig == SIGINT && VMError::is_error_reported()) {
2687     os::die();
2688   }
2689 
2690   os::signal_notify(sig);
2691 }
2692 
2693 void* os::user_handler() {
2694   return CAST_FROM_FN_PTR(void*, UserHandler);
2695 }
2696 
2697 extern "C" {
2698   typedef void (*sa_handler_t)(int);
2699   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2700 }
2701 
2702 void* os::signal(int signal_number, void* handler) {
2703   struct sigaction sigAct, oldSigAct;
2704 
2705   sigfillset(&(sigAct.sa_mask));
2706   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2707   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2708 
2709   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2710     // -1 means registration failed
2711     return (void *)-1;
2712   }
2713 
2714   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2715 }
2716 
2717 void os::signal_raise(int signal_number) {
2718   ::raise(signal_number);
2719 }
2720 
2721 // The following code is moved from os.cpp for making this
2722 // code platform specific, which it is by its very nature.
2723 
2724 // Will be modified when max signal is changed to be dynamic
2725 int os::sigexitnum_pd() {
2726   return NSIG;
2727 }
2728 
2729 // a counter for each possible signal value
2730 static volatile jint pending_signals[NSIG+1] = { 0 };
2731 
2732 // Linux(POSIX) specific hand shaking semaphore.
2733 static Semaphore* sig_sem = NULL;
2734 static PosixSemaphore sr_semaphore;
2735 
2736 static void jdk_misc_signal_init() {
2737   // Initialize signal structures
2738   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2739 
2740   // Initialize signal semaphore
2741   sig_sem = new Semaphore();
2742 }
2743 
2744 void os::signal_notify(int sig) {
2745   if (sig_sem != NULL) {
2746     Atomic::inc(&pending_signals[sig]);
2747     sig_sem->signal();
2748   } else {
2749     // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2750     // initialization isn't called.
2751     assert(ReduceSignalUsage, "signal semaphore should be created");
2752   }
2753 }
2754 
2755 static int check_pending_signals() {
2756   for (;;) {
2757     for (int i = 0; i < NSIG + 1; i++) {
2758       jint n = pending_signals[i];
2759       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2760         return i;
2761       }
2762     }
2763     JavaThread *thread = JavaThread::current();
2764     ThreadBlockInVM tbivm(thread);
2765 
2766     bool threadIsSuspended;
2767     do {
2768       thread->set_suspend_equivalent();
2769       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2770       sig_sem->wait();
2771 
2772       // were we externally suspended while we were waiting?
2773       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2774       if (threadIsSuspended) {
2775         // The semaphore has been incremented, but while we were waiting
2776         // another thread suspended us. We don't want to continue running
2777         // while suspended because that would surprise the thread that
2778         // suspended us.
2779         sig_sem->signal();
2780 
2781         thread->java_suspend_self();
2782       }
2783     } while (threadIsSuspended);
2784   }
2785 }
2786 
2787 int os::signal_wait() {
2788   return check_pending_signals();
2789 }
2790 
2791 ////////////////////////////////////////////////////////////////////////////////
2792 // Virtual Memory
2793 
2794 int os::vm_page_size() {
2795   // Seems redundant as all get out
2796   assert(os::Linux::page_size() != -1, "must call os::init");
2797   return os::Linux::page_size();
2798 }
2799 
2800 // Solaris allocates memory by pages.
2801 int os::vm_allocation_granularity() {
2802   assert(os::Linux::page_size() != -1, "must call os::init");
2803   return os::Linux::page_size();
2804 }
2805 
2806 // Rationale behind this function:
2807 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2808 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2809 //  samples for JITted code. Here we create private executable mapping over the code cache
2810 //  and then we can use standard (well, almost, as mapping can change) way to provide
2811 //  info for the reporting script by storing timestamp and location of symbol
2812 void linux_wrap_code(char* base, size_t size) {
2813   static volatile jint cnt = 0;
2814 
2815   if (!UseOprofile) {
2816     return;
2817   }
2818 
2819   char buf[PATH_MAX+1];
2820   int num = Atomic::add(1, &cnt);
2821 
2822   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2823            os::get_temp_directory(), os::current_process_id(), num);
2824   unlink(buf);
2825 
2826   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2827 
2828   if (fd != -1) {
2829     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2830     if (rv != (off_t)-1) {
2831       if (::write(fd, "", 1) == 1) {
2832         mmap(base, size,
2833              PROT_READ|PROT_WRITE|PROT_EXEC,
2834              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2835       }
2836     }
2837     ::close(fd);
2838     unlink(buf);
2839   }
2840 }
2841 
2842 static bool recoverable_mmap_error(int err) {
2843   // See if the error is one we can let the caller handle. This
2844   // list of errno values comes from JBS-6843484. I can't find a
2845   // Linux man page that documents this specific set of errno
2846   // values so while this list currently matches Solaris, it may
2847   // change as we gain experience with this failure mode.
2848   switch (err) {
2849   case EBADF:
2850   case EINVAL:
2851   case ENOTSUP:
2852     // let the caller deal with these errors
2853     return true;
2854 
2855   default:
2856     // Any remaining errors on this OS can cause our reserved mapping
2857     // to be lost. That can cause confusion where different data
2858     // structures think they have the same memory mapped. The worst
2859     // scenario is if both the VM and a library think they have the
2860     // same memory mapped.
2861     return false;
2862   }
2863 }
2864 
2865 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2866                                     int err) {
2867   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2868           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2869           os::strerror(err), err);
2870 }
2871 
2872 static void warn_fail_commit_memory(char* addr, size_t size,
2873                                     size_t alignment_hint, bool exec,
2874                                     int err) {
2875   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2876           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2877           alignment_hint, exec, os::strerror(err), err);
2878 }
2879 
2880 // NOTE: Linux kernel does not really reserve the pages for us.
2881 //       All it does is to check if there are enough free pages
2882 //       left at the time of mmap(). This could be a potential
2883 //       problem.
2884 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2885   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2886   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2887                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2888   if (res != (uintptr_t) MAP_FAILED) {
2889     if (UseNUMAInterleaving) {
2890       numa_make_global(addr, size);
2891     }
2892     return 0;
2893   }
2894 
2895   int err = errno;  // save errno from mmap() call above
2896 
2897   if (!recoverable_mmap_error(err)) {
2898     warn_fail_commit_memory(addr, size, exec, err);
2899     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2900   }
2901 
2902   return err;
2903 }
2904 
2905 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2906   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2907 }
2908 
2909 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2910                                   const char* mesg) {
2911   assert(mesg != NULL, "mesg must be specified");
2912   int err = os::Linux::commit_memory_impl(addr, size, exec);
2913   if (err != 0) {
2914     // the caller wants all commit errors to exit with the specified mesg:
2915     warn_fail_commit_memory(addr, size, exec, err);
2916     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2917   }
2918 }
2919 
2920 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2921 #ifndef MAP_HUGETLB
2922   #define MAP_HUGETLB 0x40000
2923 #endif
2924 
2925 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2926 #ifndef MADV_HUGEPAGE
2927   #define MADV_HUGEPAGE 14
2928 #endif
2929 
2930 int os::Linux::commit_memory_impl(char* addr, size_t size,
2931                                   size_t alignment_hint, bool exec) {
2932   int err = os::Linux::commit_memory_impl(addr, size, exec);
2933   if (err == 0) {
2934     realign_memory(addr, size, alignment_hint);
2935   }
2936   return err;
2937 }
2938 
2939 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2940                           bool exec) {
2941   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2942 }
2943 
2944 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2945                                   size_t alignment_hint, bool exec,
2946                                   const char* mesg) {
2947   assert(mesg != NULL, "mesg must be specified");
2948   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2949   if (err != 0) {
2950     // the caller wants all commit errors to exit with the specified mesg:
2951     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2952     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2953   }
2954 }
2955 
2956 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2957   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2958     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2959     // be supported or the memory may already be backed by huge pages.
2960     ::madvise(addr, bytes, MADV_HUGEPAGE);
2961   }
2962 }
2963 
2964 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2965   // This method works by doing an mmap over an existing mmaping and effectively discarding
2966   // the existing pages. However it won't work for SHM-based large pages that cannot be
2967   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2968   // small pages on top of the SHM segment. This method always works for small pages, so we
2969   // allow that in any case.
2970   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2971     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2972   }
2973 }
2974 
2975 void os::numa_make_global(char *addr, size_t bytes) {
2976   Linux::numa_interleave_memory(addr, bytes);
2977 }
2978 
2979 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2980 // bind policy to MPOL_PREFERRED for the current thread.
2981 #define USE_MPOL_PREFERRED 0
2982 
2983 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2984   // To make NUMA and large pages more robust when both enabled, we need to ease
2985   // the requirements on where the memory should be allocated. MPOL_BIND is the
2986   // default policy and it will force memory to be allocated on the specified
2987   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2988   // the specified node, but will not force it. Using this policy will prevent
2989   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2990   // free large pages.
2991   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2992   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2993 }
2994 
2995 bool os::numa_topology_changed() { return false; }
2996 
2997 size_t os::numa_get_groups_num() {
2998   // Return just the number of nodes in which it's possible to allocate memory
2999   // (in numa terminology, configured nodes).
3000   return Linux::numa_num_configured_nodes();
3001 }
3002 
3003 int os::numa_get_group_id() {
3004   int cpu_id = Linux::sched_getcpu();
3005   if (cpu_id != -1) {
3006     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
3007     if (lgrp_id != -1) {
3008       return lgrp_id;
3009     }
3010   }
3011   return 0;
3012 }
3013 
3014 int os::numa_get_address_id(void* address) {
3015 #ifndef MPOL_F_NODE
3016 #define MPOL_F_NODE     (1<<0)  // Return next IL mode instead of node mask
3017 #endif
3018 
3019 #ifndef MPOL_F_ADDR
3020 #define MPOL_F_ADDR     (1<<1)  // Look up VMA using address
3021 #endif
3022   
3023   int id = InvalidNUMAId;
3024 
3025   if (syscall(SYS_get_mempolicy, &id, NULL, 0, address, MPOL_F_NODE | MPOL_F_ADDR) == -1) {
3026     warning("Failed to get numa id at " PTR_FORMAT " with errno=%d", p2i(address), errno);
3027     return InvalidNUMAId;
3028   }
3029   return id;
3030 }
3031 
3032 int os::Linux::get_existing_num_nodes() {
3033   int node;
3034   int highest_node_number = Linux::numa_max_node();
3035   int num_nodes = 0;
3036 
3037   // Get the total number of nodes in the system including nodes without memory.
3038   for (node = 0; node <= highest_node_number; node++) {
3039     if (is_node_in_existing_nodes(node)) {
3040       num_nodes++;
3041     }
3042   }
3043   return num_nodes;
3044 }
3045 
3046 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
3047   int highest_node_number = Linux::numa_max_node();
3048   size_t i = 0;
3049 
3050   // Map all node ids in which it is possible to allocate memory. Also nodes are
3051   // not always consecutively available, i.e. available from 0 to the highest
3052   // node number. If the nodes have been bound explicitly using numactl membind,
3053   // then allocate memory from those nodes only.
3054   for (int node = 0; node <= highest_node_number; node++) {
3055     if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
3056       ids[i++] = node;
3057     }
3058   }
3059   return i;
3060 }
3061 
3062 bool os::get_page_info(char *start, page_info* info) {
3063   return false;
3064 }
3065 
3066 char *os::scan_pages(char *start, char* end, page_info* page_expected,
3067                      page_info* page_found) {
3068   return end;
3069 }
3070 
3071 
3072 int os::Linux::sched_getcpu_syscall(void) {
3073   unsigned int cpu = 0;
3074   int retval = -1;
3075 
3076 #if defined(IA32)
3077   #ifndef SYS_getcpu
3078     #define SYS_getcpu 318
3079   #endif
3080   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
3081 #elif defined(AMD64)
3082 // Unfortunately we have to bring all these macros here from vsyscall.h
3083 // to be able to compile on old linuxes.
3084   #define __NR_vgetcpu 2
3085   #define VSYSCALL_START (-10UL << 20)
3086   #define VSYSCALL_SIZE 1024
3087   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
3088   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
3089   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
3090   retval = vgetcpu(&cpu, NULL, NULL);
3091 #endif
3092 
3093   return (retval == -1) ? retval : cpu;
3094 }
3095 
3096 void os::Linux::sched_getcpu_init() {
3097   // sched_getcpu() should be in libc.
3098   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3099                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
3100 
3101   // If it's not, try a direct syscall.
3102   if (sched_getcpu() == -1) {
3103     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3104                                     (void*)&sched_getcpu_syscall));
3105   }
3106 
3107   if (sched_getcpu() == -1) {
3108     vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
3109   }
3110 }
3111 
3112 // Something to do with the numa-aware allocator needs these symbols
3113 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
3114 extern "C" JNIEXPORT void numa_error(char *where) { }
3115 
3116 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
3117 // load symbol from base version instead.
3118 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
3119   void *f = dlvsym(handle, name, "libnuma_1.1");
3120   if (f == NULL) {
3121     f = dlsym(handle, name);
3122   }
3123   return f;
3124 }
3125 
3126 // Handle request to load libnuma symbol version 1.2 (API v2) only.
3127 // Return NULL if the symbol is not defined in this particular version.
3128 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
3129   return dlvsym(handle, name, "libnuma_1.2");
3130 }
3131 
3132 bool os::Linux::libnuma_init() {
3133   if (sched_getcpu() != -1) { // Requires sched_getcpu() support
3134     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3135     if (handle != NULL) {
3136       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3137                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
3138       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3139                                        libnuma_dlsym(handle, "numa_max_node")));
3140       set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3141                                                    libnuma_dlsym(handle, "numa_num_configured_nodes")));
3142       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3143                                         libnuma_dlsym(handle, "numa_available")));
3144       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3145                                             libnuma_dlsym(handle, "numa_tonode_memory")));
3146       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3147                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
3148       set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3149                                                 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3150       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3151                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
3152       set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3153                                                libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3154       set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3155                                        libnuma_dlsym(handle, "numa_distance")));
3156       set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
3157                                           libnuma_v2_dlsym(handle, "numa_get_membind")));
3158       set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
3159                                                   libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
3160 
3161       if (numa_available() != -1) {
3162         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3163         set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3164         set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3165         set_numa_interleave_bitmask(_numa_get_interleave_mask());
3166         set_numa_membind_bitmask(_numa_get_membind());
3167         // Create an index -> node mapping, since nodes are not always consecutive
3168         _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3169         rebuild_nindex_to_node_map();
3170         // Create a cpu -> node mapping
3171         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3172         rebuild_cpu_to_node_map();
3173         return true;
3174       }
3175     }
3176   }
3177   return false;
3178 }
3179 
3180 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
3181   // Creating guard page is very expensive. Java thread has HotSpot
3182   // guard pages, only enable glibc guard page for non-Java threads.
3183   // (Remember: compiler thread is a Java thread, too!)
3184   return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
3185 }
3186 
3187 void os::Linux::rebuild_nindex_to_node_map() {
3188   int highest_node_number = Linux::numa_max_node();
3189 
3190   nindex_to_node()->clear();
3191   for (int node = 0; node <= highest_node_number; node++) {
3192     if (Linux::is_node_in_existing_nodes(node)) {
3193       nindex_to_node()->append(node);
3194     }
3195   }
3196 }
3197 
3198 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3199 // The table is later used in get_node_by_cpu().
3200 void os::Linux::rebuild_cpu_to_node_map() {
3201   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3202                               // in libnuma (possible values are starting from 16,
3203                               // and continuing up with every other power of 2, but less
3204                               // than the maximum number of CPUs supported by kernel), and
3205                               // is a subject to change (in libnuma version 2 the requirements
3206                               // are more reasonable) we'll just hardcode the number they use
3207                               // in the library.
3208   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3209 
3210   size_t cpu_num = processor_count();
3211   size_t cpu_map_size = NCPUS / BitsPerCLong;
3212   size_t cpu_map_valid_size =
3213     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3214 
3215   cpu_to_node()->clear();
3216   cpu_to_node()->at_grow(cpu_num - 1);
3217 
3218   size_t node_num = get_existing_num_nodes();
3219 
3220   int distance = 0;
3221   int closest_distance = INT_MAX;
3222   int closest_node = 0;
3223   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3224   for (size_t i = 0; i < node_num; i++) {
3225     // Check if node is configured (not a memory-less node). If it is not, find
3226     // the closest configured node. Check also if node is bound, i.e. it's allowed
3227     // to allocate memory from the node. If it's not allowed, map cpus in that node
3228     // to the closest node from which memory allocation is allowed.
3229     if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
3230         !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
3231       closest_distance = INT_MAX;
3232       // Check distance from all remaining nodes in the system. Ignore distance
3233       // from itself, from another non-configured node, and from another non-bound
3234       // node.
3235       for (size_t m = 0; m < node_num; m++) {
3236         if (m != i &&
3237             is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
3238             is_node_in_bound_nodes(nindex_to_node()->at(m))) {
3239           distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3240           // If a closest node is found, update. There is always at least one
3241           // configured and bound node in the system so there is always at least
3242           // one node close.
3243           if (distance != 0 && distance < closest_distance) {
3244             closest_distance = distance;
3245             closest_node = nindex_to_node()->at(m);
3246           }
3247         }
3248       }
3249      } else {
3250        // Current node is already a configured node.
3251        closest_node = nindex_to_node()->at(i);
3252      }
3253 
3254     // Get cpus from the original node and map them to the closest node. If node
3255     // is a configured node (not a memory-less node), then original node and
3256     // closest node are the same.
3257     if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3258       for (size_t j = 0; j < cpu_map_valid_size; j++) {
3259         if (cpu_map[j] != 0) {
3260           for (size_t k = 0; k < BitsPerCLong; k++) {
3261             if (cpu_map[j] & (1UL << k)) {
3262               cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3263             }
3264           }
3265         }
3266       }
3267     }
3268   }
3269   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3270 }
3271 
3272 int os::Linux::get_node_by_cpu(int cpu_id) {
3273   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3274     return cpu_to_node()->at(cpu_id);
3275   }
3276   return -1;
3277 }
3278 
3279 GrowableArray<int>* os::Linux::_cpu_to_node;
3280 GrowableArray<int>* os::Linux::_nindex_to_node;
3281 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3282 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3283 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3284 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3285 os::Linux::numa_available_func_t os::Linux::_numa_available;
3286 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3287 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3288 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3289 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3290 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3291 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3292 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3293 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3294 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy;
3295 unsigned long* os::Linux::_numa_all_nodes;
3296 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3297 struct bitmask* os::Linux::_numa_nodes_ptr;
3298 struct bitmask* os::Linux::_numa_interleave_bitmask;
3299 struct bitmask* os::Linux::_numa_membind_bitmask;
3300 
3301 bool os::pd_uncommit_memory(char* addr, size_t size) {
3302   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3303                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3304   return res  != (uintptr_t) MAP_FAILED;
3305 }
3306 
3307 static address get_stack_commited_bottom(address bottom, size_t size) {
3308   address nbot = bottom;
3309   address ntop = bottom + size;
3310 
3311   size_t page_sz = os::vm_page_size();
3312   unsigned pages = size / page_sz;
3313 
3314   unsigned char vec[1];
3315   unsigned imin = 1, imax = pages + 1, imid;
3316   int mincore_return_value = 0;
3317 
3318   assert(imin <= imax, "Unexpected page size");
3319 
3320   while (imin < imax) {
3321     imid = (imax + imin) / 2;
3322     nbot = ntop - (imid * page_sz);
3323 
3324     // Use a trick with mincore to check whether the page is mapped or not.
3325     // mincore sets vec to 1 if page resides in memory and to 0 if page
3326     // is swapped output but if page we are asking for is unmapped
3327     // it returns -1,ENOMEM
3328     mincore_return_value = mincore(nbot, page_sz, vec);
3329 
3330     if (mincore_return_value == -1) {
3331       // Page is not mapped go up
3332       // to find first mapped page
3333       if (errno != EAGAIN) {
3334         assert(errno == ENOMEM, "Unexpected mincore errno");
3335         imax = imid;
3336       }
3337     } else {
3338       // Page is mapped go down
3339       // to find first not mapped page
3340       imin = imid + 1;
3341     }
3342   }
3343 
3344   nbot = nbot + page_sz;
3345 
3346   // Adjust stack bottom one page up if last checked page is not mapped
3347   if (mincore_return_value == -1) {
3348     nbot = nbot + page_sz;
3349   }
3350 
3351   return nbot;
3352 }
3353 
3354 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3355   int mincore_return_value;
3356   const size_t stripe = 1024;  // query this many pages each time
3357   unsigned char vec[stripe + 1];
3358   // set a guard
3359   vec[stripe] = 'X';
3360 
3361   const size_t page_sz = os::vm_page_size();
3362   size_t pages = size / page_sz;
3363 
3364   assert(is_aligned(start, page_sz), "Start address must be page aligned");
3365   assert(is_aligned(size, page_sz), "Size must be page aligned");
3366 
3367   committed_start = NULL;
3368 
3369   int loops = (pages + stripe - 1) / stripe;
3370   int committed_pages = 0;
3371   address loop_base = start;
3372   bool found_range = false;
3373 
3374   for (int index = 0; index < loops && !found_range; index ++) {
3375     assert(pages > 0, "Nothing to do");
3376     int pages_to_query = (pages >= stripe) ? stripe : pages;
3377     pages -= pages_to_query;
3378 
3379     // Get stable read
3380     while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3381 
3382     // During shutdown, some memory goes away without properly notifying NMT,
3383     // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3384     // Bailout and return as not committed for now.
3385     if (mincore_return_value == -1 && errno == ENOMEM) {
3386       return false;
3387     }
3388 
3389     assert(vec[stripe] == 'X', "overflow guard");
3390     assert(mincore_return_value == 0, "Range must be valid");
3391     // Process this stripe
3392     for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3393       if ((vec[vecIdx] & 0x01) == 0) { // not committed
3394         // End of current contiguous region
3395         if (committed_start != NULL) {
3396           found_range = true;
3397           break;
3398         }
3399       } else { // committed
3400         // Start of region
3401         if (committed_start == NULL) {
3402           committed_start = loop_base + page_sz * vecIdx;
3403         }
3404         committed_pages ++;
3405       }
3406     }
3407 
3408     loop_base += pages_to_query * page_sz;
3409   }
3410 
3411   if (committed_start != NULL) {
3412     assert(committed_pages > 0, "Must have committed region");
3413     assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3414     assert(committed_start >= start && committed_start < start + size, "Out of range");
3415     committed_size = page_sz * committed_pages;
3416     return true;
3417   } else {
3418     assert(committed_pages == 0, "Should not have committed region");
3419     return false;
3420   }
3421 }
3422 
3423 
3424 // Linux uses a growable mapping for the stack, and if the mapping for
3425 // the stack guard pages is not removed when we detach a thread the
3426 // stack cannot grow beyond the pages where the stack guard was
3427 // mapped.  If at some point later in the process the stack expands to
3428 // that point, the Linux kernel cannot expand the stack any further
3429 // because the guard pages are in the way, and a segfault occurs.
3430 //
3431 // However, it's essential not to split the stack region by unmapping
3432 // a region (leaving a hole) that's already part of the stack mapping,
3433 // so if the stack mapping has already grown beyond the guard pages at
3434 // the time we create them, we have to truncate the stack mapping.
3435 // So, we need to know the extent of the stack mapping when
3436 // create_stack_guard_pages() is called.
3437 
3438 // We only need this for stacks that are growable: at the time of
3439 // writing thread stacks don't use growable mappings (i.e. those
3440 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3441 // only applies to the main thread.
3442 
3443 // If the (growable) stack mapping already extends beyond the point
3444 // where we're going to put our guard pages, truncate the mapping at
3445 // that point by munmap()ping it.  This ensures that when we later
3446 // munmap() the guard pages we don't leave a hole in the stack
3447 // mapping. This only affects the main/primordial thread
3448 
3449 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3450   if (os::is_primordial_thread()) {
3451     // As we manually grow stack up to bottom inside create_attached_thread(),
3452     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3453     // we don't need to do anything special.
3454     // Check it first, before calling heavy function.
3455     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3456     unsigned char vec[1];
3457 
3458     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3459       // Fallback to slow path on all errors, including EAGAIN
3460       stack_extent = (uintptr_t) get_stack_commited_bottom(
3461                                                            os::Linux::initial_thread_stack_bottom(),
3462                                                            (size_t)addr - stack_extent);
3463     }
3464 
3465     if (stack_extent < (uintptr_t)addr) {
3466       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3467     }
3468   }
3469 
3470   return os::commit_memory(addr, size, !ExecMem);
3471 }
3472 
3473 // If this is a growable mapping, remove the guard pages entirely by
3474 // munmap()ping them.  If not, just call uncommit_memory(). This only
3475 // affects the main/primordial thread, but guard against future OS changes.
3476 // It's safe to always unmap guard pages for primordial thread because we
3477 // always place it right after end of the mapped region.
3478 
3479 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3480   uintptr_t stack_extent, stack_base;
3481 
3482   if (os::is_primordial_thread()) {
3483     return ::munmap(addr, size) == 0;
3484   }
3485 
3486   return os::uncommit_memory(addr, size);
3487 }
3488 
3489 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3490 // at 'requested_addr'. If there are existing memory mappings at the same
3491 // location, however, they will be overwritten. If 'fixed' is false,
3492 // 'requested_addr' is only treated as a hint, the return value may or
3493 // may not start from the requested address. Unlike Linux mmap(), this
3494 // function returns NULL to indicate failure.
3495 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3496   char * addr;
3497   int flags;
3498 
3499   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3500   if (fixed) {
3501     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3502     flags |= MAP_FIXED;
3503   }
3504 
3505   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3506   // touch an uncommitted page. Otherwise, the read/write might
3507   // succeed if we have enough swap space to back the physical page.
3508   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3509                        flags, -1, 0);
3510 
3511   return addr == MAP_FAILED ? NULL : addr;
3512 }
3513 
3514 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3515 //   (req_addr != NULL) or with a given alignment.
3516 //  - bytes shall be a multiple of alignment.
3517 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3518 //  - alignment sets the alignment at which memory shall be allocated.
3519 //     It must be a multiple of allocation granularity.
3520 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3521 //  req_addr or NULL.
3522 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3523 
3524   size_t extra_size = bytes;
3525   if (req_addr == NULL && alignment > 0) {
3526     extra_size += alignment;
3527   }
3528 
3529   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3530     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3531     -1, 0);
3532   if (start == MAP_FAILED) {
3533     start = NULL;
3534   } else {
3535     if (req_addr != NULL) {
3536       if (start != req_addr) {
3537         ::munmap(start, extra_size);
3538         start = NULL;
3539       }
3540     } else {
3541       char* const start_aligned = align_up(start, alignment);
3542       char* const end_aligned = start_aligned + bytes;
3543       char* const end = start + extra_size;
3544       if (start_aligned > start) {
3545         ::munmap(start, start_aligned - start);
3546       }
3547       if (end_aligned < end) {
3548         ::munmap(end_aligned, end - end_aligned);
3549       }
3550       start = start_aligned;
3551     }
3552   }
3553   return start;
3554 }
3555 
3556 static int anon_munmap(char * addr, size_t size) {
3557   return ::munmap(addr, size) == 0;
3558 }
3559 
3560 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3561                             size_t alignment_hint) {
3562   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3563 }
3564 
3565 bool os::pd_release_memory(char* addr, size_t size) {
3566   return anon_munmap(addr, size);
3567 }
3568 
3569 static bool linux_mprotect(char* addr, size_t size, int prot) {
3570   // Linux wants the mprotect address argument to be page aligned.
3571   char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3572 
3573   // According to SUSv3, mprotect() should only be used with mappings
3574   // established by mmap(), and mmap() always maps whole pages. Unaligned
3575   // 'addr' likely indicates problem in the VM (e.g. trying to change
3576   // protection of malloc'ed or statically allocated memory). Check the
3577   // caller if you hit this assert.
3578   assert(addr == bottom, "sanity check");
3579 
3580   size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3581   Events::log(NULL, "Protecting memory [" INTPTR_FORMAT "," INTPTR_FORMAT "] with protection modes %x", p2i(bottom), p2i(bottom+size), prot);
3582   return ::mprotect(bottom, size, prot) == 0;
3583 }
3584 
3585 // Set protections specified
3586 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3587                         bool is_committed) {
3588   unsigned int p = 0;
3589   switch (prot) {
3590   case MEM_PROT_NONE: p = PROT_NONE; break;
3591   case MEM_PROT_READ: p = PROT_READ; break;
3592   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3593   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3594   default:
3595     ShouldNotReachHere();
3596   }
3597   // is_committed is unused.
3598   return linux_mprotect(addr, bytes, p);
3599 }
3600 
3601 bool os::guard_memory(char* addr, size_t size) {
3602   return linux_mprotect(addr, size, PROT_NONE);
3603 }
3604 
3605 bool os::unguard_memory(char* addr, size_t size) {
3606   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3607 }
3608 
3609 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3610                                                     size_t page_size) {
3611   bool result = false;
3612   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3613                  MAP_ANONYMOUS|MAP_PRIVATE,
3614                  -1, 0);
3615   if (p != MAP_FAILED) {
3616     void *aligned_p = align_up(p, page_size);
3617 
3618     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3619 
3620     munmap(p, page_size * 2);
3621   }
3622 
3623   if (warn && !result) {
3624     warning("TransparentHugePages is not supported by the operating system.");
3625   }
3626 
3627   return result;
3628 }
3629 
3630 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3631   bool result = false;
3632   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3633                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3634                  -1, 0);
3635 
3636   if (p != MAP_FAILED) {
3637     // We don't know if this really is a huge page or not.
3638     FILE *fp = fopen("/proc/self/maps", "r");
3639     if (fp) {
3640       while (!feof(fp)) {
3641         char chars[257];
3642         long x = 0;
3643         if (fgets(chars, sizeof(chars), fp)) {
3644           if (sscanf(chars, "%lx-%*x", &x) == 1
3645               && x == (long)p) {
3646             if (strstr (chars, "hugepage")) {
3647               result = true;
3648               break;
3649             }
3650           }
3651         }
3652       }
3653       fclose(fp);
3654     }
3655     munmap(p, page_size);
3656   }
3657 
3658   if (warn && !result) {
3659     warning("HugeTLBFS is not supported by the operating system.");
3660   }
3661 
3662   return result;
3663 }
3664 
3665 // From the coredump_filter documentation:
3666 //
3667 // - (bit 0) anonymous private memory
3668 // - (bit 1) anonymous shared memory
3669 // - (bit 2) file-backed private memory
3670 // - (bit 3) file-backed shared memory
3671 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3672 //           effective only if the bit 2 is cleared)
3673 // - (bit 5) hugetlb private memory
3674 // - (bit 6) hugetlb shared memory
3675 // - (bit 7) dax private memory
3676 // - (bit 8) dax shared memory
3677 //
3678 static void set_coredump_filter(CoredumpFilterBit bit) {
3679   FILE *f;
3680   long cdm;
3681 
3682   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3683     return;
3684   }
3685 
3686   if (fscanf(f, "%lx", &cdm) != 1) {
3687     fclose(f);
3688     return;
3689   }
3690 
3691   long saved_cdm = cdm;
3692   rewind(f);
3693   cdm |= bit;
3694 
3695   if (cdm != saved_cdm) {
3696     fprintf(f, "%#lx", cdm);
3697   }
3698 
3699   fclose(f);
3700 }
3701 
3702 // Large page support
3703 
3704 static size_t _large_page_size = 0;
3705 
3706 size_t os::Linux::find_large_page_size() {
3707   size_t large_page_size = 0;
3708 
3709   // large_page_size on Linux is used to round up heap size. x86 uses either
3710   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3711   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3712   // page as large as 256M.
3713   //
3714   // Here we try to figure out page size by parsing /proc/meminfo and looking
3715   // for a line with the following format:
3716   //    Hugepagesize:     2048 kB
3717   //
3718   // If we can't determine the value (e.g. /proc is not mounted, or the text
3719   // format has been changed), we'll use the largest page size supported by
3720   // the processor.
3721 
3722 #ifndef ZERO
3723   large_page_size =
3724     AARCH64_ONLY(2 * M)
3725     AMD64_ONLY(2 * M)
3726     ARM32_ONLY(2 * M)
3727     IA32_ONLY(4 * M)
3728     IA64_ONLY(256 * M)
3729     PPC_ONLY(4 * M)
3730     S390_ONLY(1 * M)
3731     SPARC_ONLY(4 * M);
3732 #endif // ZERO
3733 
3734   FILE *fp = fopen("/proc/meminfo", "r");
3735   if (fp) {
3736     while (!feof(fp)) {
3737       int x = 0;
3738       char buf[16];
3739       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3740         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3741           large_page_size = x * K;
3742           break;
3743         }
3744       } else {
3745         // skip to next line
3746         for (;;) {
3747           int ch = fgetc(fp);
3748           if (ch == EOF || ch == (int)'\n') break;
3749         }
3750       }
3751     }
3752     fclose(fp);
3753   }
3754 
3755   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3756     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3757             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3758             proper_unit_for_byte_size(large_page_size));
3759   }
3760 
3761   return large_page_size;
3762 }
3763 
3764 size_t os::Linux::setup_large_page_size() {
3765   _large_page_size = Linux::find_large_page_size();
3766   const size_t default_page_size = (size_t)Linux::page_size();
3767   if (_large_page_size > default_page_size) {
3768     _page_sizes[0] = _large_page_size;
3769     _page_sizes[1] = default_page_size;
3770     _page_sizes[2] = 0;
3771   }
3772 
3773   return _large_page_size;
3774 }
3775 
3776 bool os::Linux::setup_large_page_type(size_t page_size) {
3777   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3778       FLAG_IS_DEFAULT(UseSHM) &&
3779       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3780 
3781     // The type of large pages has not been specified by the user.
3782 
3783     // Try UseHugeTLBFS and then UseSHM.
3784     UseHugeTLBFS = UseSHM = true;
3785 
3786     // Don't try UseTransparentHugePages since there are known
3787     // performance issues with it turned on. This might change in the future.
3788     UseTransparentHugePages = false;
3789   }
3790 
3791   if (UseTransparentHugePages) {
3792     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3793     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3794       UseHugeTLBFS = false;
3795       UseSHM = false;
3796       return true;
3797     }
3798     UseTransparentHugePages = false;
3799   }
3800 
3801   if (UseHugeTLBFS) {
3802     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3803     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3804       UseSHM = false;
3805       return true;
3806     }
3807     UseHugeTLBFS = false;
3808   }
3809 
3810   return UseSHM;
3811 }
3812 
3813 void os::large_page_init() {
3814   if (!UseLargePages &&
3815       !UseTransparentHugePages &&
3816       !UseHugeTLBFS &&
3817       !UseSHM) {
3818     // Not using large pages.
3819     return;
3820   }
3821 
3822   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3823     // The user explicitly turned off large pages.
3824     // Ignore the rest of the large pages flags.
3825     UseTransparentHugePages = false;
3826     UseHugeTLBFS = false;
3827     UseSHM = false;
3828     return;
3829   }
3830 
3831   size_t large_page_size = Linux::setup_large_page_size();
3832   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3833 
3834   set_coredump_filter(LARGEPAGES_BIT);
3835 }
3836 
3837 #ifndef SHM_HUGETLB
3838   #define SHM_HUGETLB 04000
3839 #endif
3840 
3841 #define shm_warning_format(format, ...)              \
3842   do {                                               \
3843     if (UseLargePages &&                             \
3844         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3845          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3846          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3847       warning(format, __VA_ARGS__);                  \
3848     }                                                \
3849   } while (0)
3850 
3851 #define shm_warning(str) shm_warning_format("%s", str)
3852 
3853 #define shm_warning_with_errno(str)                \
3854   do {                                             \
3855     int err = errno;                               \
3856     shm_warning_format(str " (error = %d)", err);  \
3857   } while (0)
3858 
3859 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3860   assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3861 
3862   if (!is_aligned(alignment, SHMLBA)) {
3863     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3864     return NULL;
3865   }
3866 
3867   // To ensure that we get 'alignment' aligned memory from shmat,
3868   // we pre-reserve aligned virtual memory and then attach to that.
3869 
3870   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3871   if (pre_reserved_addr == NULL) {
3872     // Couldn't pre-reserve aligned memory.
3873     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3874     return NULL;
3875   }
3876 
3877   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3878   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3879 
3880   if ((intptr_t)addr == -1) {
3881     int err = errno;
3882     shm_warning_with_errno("Failed to attach shared memory.");
3883 
3884     assert(err != EACCES, "Unexpected error");
3885     assert(err != EIDRM,  "Unexpected error");
3886     assert(err != EINVAL, "Unexpected error");
3887 
3888     // Since we don't know if the kernel unmapped the pre-reserved memory area
3889     // we can't unmap it, since that would potentially unmap memory that was
3890     // mapped from other threads.
3891     return NULL;
3892   }
3893 
3894   return addr;
3895 }
3896 
3897 static char* shmat_at_address(int shmid, char* req_addr) {
3898   if (!is_aligned(req_addr, SHMLBA)) {
3899     assert(false, "Requested address needs to be SHMLBA aligned");
3900     return NULL;
3901   }
3902 
3903   char* addr = (char*)shmat(shmid, req_addr, 0);
3904 
3905   if ((intptr_t)addr == -1) {
3906     shm_warning_with_errno("Failed to attach shared memory.");
3907     return NULL;
3908   }
3909 
3910   return addr;
3911 }
3912 
3913 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3914   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3915   if (req_addr != NULL) {
3916     assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3917     assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3918     return shmat_at_address(shmid, req_addr);
3919   }
3920 
3921   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3922   // return large page size aligned memory addresses when req_addr == NULL.
3923   // However, if the alignment is larger than the large page size, we have
3924   // to manually ensure that the memory returned is 'alignment' aligned.
3925   if (alignment > os::large_page_size()) {
3926     assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3927     return shmat_with_alignment(shmid, bytes, alignment);
3928   } else {
3929     return shmat_at_address(shmid, NULL);
3930   }
3931 }
3932 
3933 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3934                                             char* req_addr, bool exec) {
3935   // "exec" is passed in but not used.  Creating the shared image for
3936   // the code cache doesn't have an SHM_X executable permission to check.
3937   assert(UseLargePages && UseSHM, "only for SHM large pages");
3938   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3939   assert(is_aligned(req_addr, alignment), "Unaligned address");
3940 
3941   if (!is_aligned(bytes, os::large_page_size())) {
3942     return NULL; // Fallback to small pages.
3943   }
3944 
3945   // Create a large shared memory region to attach to based on size.
3946   // Currently, size is the total size of the heap.
3947   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3948   if (shmid == -1) {
3949     // Possible reasons for shmget failure:
3950     // 1. shmmax is too small for Java heap.
3951     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3952     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3953     // 2. not enough large page memory.
3954     //    > check available large pages: cat /proc/meminfo
3955     //    > increase amount of large pages:
3956     //          echo new_value > /proc/sys/vm/nr_hugepages
3957     //      Note 1: different Linux may use different name for this property,
3958     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3959     //      Note 2: it's possible there's enough physical memory available but
3960     //            they are so fragmented after a long run that they can't
3961     //            coalesce into large pages. Try to reserve large pages when
3962     //            the system is still "fresh".
3963     shm_warning_with_errno("Failed to reserve shared memory.");
3964     return NULL;
3965   }
3966 
3967   // Attach to the region.
3968   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3969 
3970   // Remove shmid. If shmat() is successful, the actual shared memory segment
3971   // will be deleted when it's detached by shmdt() or when the process
3972   // terminates. If shmat() is not successful this will remove the shared
3973   // segment immediately.
3974   shmctl(shmid, IPC_RMID, NULL);
3975 
3976   return addr;
3977 }
3978 
3979 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3980                                         int error) {
3981   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3982 
3983   bool warn_on_failure = UseLargePages &&
3984       (!FLAG_IS_DEFAULT(UseLargePages) ||
3985        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3986        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3987 
3988   if (warn_on_failure) {
3989     char msg[128];
3990     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3991                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3992     warning("%s", msg);
3993   }
3994 }
3995 
3996 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3997                                                         char* req_addr,
3998                                                         bool exec) {
3999   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4000   assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
4001   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
4002 
4003   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4004   char* addr = (char*)::mmap(req_addr, bytes, prot,
4005                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
4006                              -1, 0);
4007 
4008   if (addr == MAP_FAILED) {
4009     warn_on_large_pages_failure(req_addr, bytes, errno);
4010     return NULL;
4011   }
4012 
4013   assert(is_aligned(addr, os::large_page_size()), "Must be");
4014 
4015   return addr;
4016 }
4017 
4018 // Reserve memory using mmap(MAP_HUGETLB).
4019 //  - bytes shall be a multiple of alignment.
4020 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
4021 //  - alignment sets the alignment at which memory shall be allocated.
4022 //     It must be a multiple of allocation granularity.
4023 // Returns address of memory or NULL. If req_addr was not NULL, will only return
4024 //  req_addr or NULL.
4025 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
4026                                                          size_t alignment,
4027                                                          char* req_addr,
4028                                                          bool exec) {
4029   size_t large_page_size = os::large_page_size();
4030   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
4031 
4032   assert(is_aligned(req_addr, alignment), "Must be");
4033   assert(is_aligned(bytes, alignment), "Must be");
4034 
4035   // First reserve - but not commit - the address range in small pages.
4036   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
4037 
4038   if (start == NULL) {
4039     return NULL;
4040   }
4041 
4042   assert(is_aligned(start, alignment), "Must be");
4043 
4044   char* end = start + bytes;
4045 
4046   // Find the regions of the allocated chunk that can be promoted to large pages.
4047   char* lp_start = align_up(start, large_page_size);
4048   char* lp_end   = align_down(end, large_page_size);
4049 
4050   size_t lp_bytes = lp_end - lp_start;
4051 
4052   assert(is_aligned(lp_bytes, large_page_size), "Must be");
4053 
4054   if (lp_bytes == 0) {
4055     // The mapped region doesn't even span the start and the end of a large page.
4056     // Fall back to allocate a non-special area.
4057     ::munmap(start, end - start);
4058     return NULL;
4059   }
4060 
4061   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4062 
4063   void* result;
4064 
4065   // Commit small-paged leading area.
4066   if (start != lp_start) {
4067     result = ::mmap(start, lp_start - start, prot,
4068                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4069                     -1, 0);
4070     if (result == MAP_FAILED) {
4071       ::munmap(lp_start, end - lp_start);
4072       return NULL;
4073     }
4074   }
4075 
4076   // Commit large-paged area.
4077   result = ::mmap(lp_start, lp_bytes, prot,
4078                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
4079                   -1, 0);
4080   if (result == MAP_FAILED) {
4081     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
4082     // If the mmap above fails, the large pages region will be unmapped and we
4083     // have regions before and after with small pages. Release these regions.
4084     //
4085     // |  mapped  |  unmapped  |  mapped  |
4086     // ^          ^            ^          ^
4087     // start      lp_start     lp_end     end
4088     //
4089     ::munmap(start, lp_start - start);
4090     ::munmap(lp_end, end - lp_end);
4091     return NULL;
4092   }
4093 
4094   // Commit small-paged trailing area.
4095   if (lp_end != end) {
4096     result = ::mmap(lp_end, end - lp_end, prot,
4097                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4098                     -1, 0);
4099     if (result == MAP_FAILED) {
4100       ::munmap(start, lp_end - start);
4101       return NULL;
4102     }
4103   }
4104 
4105   return start;
4106 }
4107 
4108 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
4109                                                    size_t alignment,
4110                                                    char* req_addr,
4111                                                    bool exec) {
4112   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4113   assert(is_aligned(req_addr, alignment), "Must be");
4114   assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
4115   assert(is_power_of_2(os::large_page_size()), "Must be");
4116   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
4117 
4118   if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
4119     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
4120   } else {
4121     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
4122   }
4123 }
4124 
4125 char* os::reserve_memory_special(size_t bytes, size_t alignment,
4126                                  char* req_addr, bool exec) {
4127   assert(UseLargePages, "only for large pages");
4128 
4129   char* addr;
4130   if (UseSHM) {
4131     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
4132   } else {
4133     assert(UseHugeTLBFS, "must be");
4134     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
4135   }
4136 
4137   if (addr != NULL) {
4138     if (UseNUMAInterleaving) {
4139       numa_make_global(addr, bytes);
4140     }
4141 
4142     // The memory is committed
4143     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
4144   }
4145 
4146   return addr;
4147 }
4148 
4149 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
4150   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
4151   return shmdt(base) == 0;
4152 }
4153 
4154 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
4155   return pd_release_memory(base, bytes);
4156 }
4157 
4158 bool os::release_memory_special(char* base, size_t bytes) {
4159   bool res;
4160   if (MemTracker::tracking_level() > NMT_minimal) {
4161     Tracker tkr(Tracker::release);
4162     res = os::Linux::release_memory_special_impl(base, bytes);
4163     if (res) {
4164       tkr.record((address)base, bytes);
4165     }
4166 
4167   } else {
4168     res = os::Linux::release_memory_special_impl(base, bytes);
4169   }
4170   return res;
4171 }
4172 
4173 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
4174   assert(UseLargePages, "only for large pages");
4175   bool res;
4176 
4177   if (UseSHM) {
4178     res = os::Linux::release_memory_special_shm(base, bytes);
4179   } else {
4180     assert(UseHugeTLBFS, "must be");
4181     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
4182   }
4183   return res;
4184 }
4185 
4186 size_t os::large_page_size() {
4187   return _large_page_size;
4188 }
4189 
4190 // With SysV SHM the entire memory region must be allocated as shared
4191 // memory.
4192 // HugeTLBFS allows application to commit large page memory on demand.
4193 // However, when committing memory with HugeTLBFS fails, the region
4194 // that was supposed to be committed will lose the old reservation
4195 // and allow other threads to steal that memory region. Because of this
4196 // behavior we can't commit HugeTLBFS memory.
4197 bool os::can_commit_large_page_memory() {
4198   return UseTransparentHugePages;
4199 }
4200 
4201 bool os::can_execute_large_page_memory() {
4202   return UseTransparentHugePages || UseHugeTLBFS;
4203 }
4204 
4205 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
4206   assert(file_desc >= 0, "file_desc is not valid");
4207   char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
4208   if (result != NULL) {
4209     if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4210       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4211     }
4212   }
4213   return result;
4214 }
4215 
4216 // Reserve memory at an arbitrary address, only if that area is
4217 // available (and not reserved for something else).
4218 
4219 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4220   // Assert only that the size is a multiple of the page size, since
4221   // that's all that mmap requires, and since that's all we really know
4222   // about at this low abstraction level.  If we need higher alignment,
4223   // we can either pass an alignment to this method or verify alignment
4224   // in one of the methods further up the call chain.  See bug 5044738.
4225   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4226 
4227   // Repeatedly allocate blocks until the block is allocated at the
4228   // right spot.
4229 
4230   // Linux mmap allows caller to pass an address as hint; give it a try first,
4231   // if kernel honors the hint then we can return immediately.
4232   char * addr = anon_mmap(requested_addr, bytes, false);
4233   if (addr == requested_addr) {
4234     return requested_addr;
4235   }
4236 
4237   if (addr != NULL) {
4238     // mmap() is successful but it fails to reserve at the requested address
4239     anon_munmap(addr, bytes);
4240   }
4241 
4242   return NULL;
4243 }
4244 
4245 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4246 void os::infinite_sleep() {
4247   while (true) {    // sleep forever ...
4248     ::sleep(100);   // ... 100 seconds at a time
4249   }
4250 }
4251 
4252 // Used to convert frequent JVM_Yield() to nops
4253 bool os::dont_yield() {
4254   return DontYieldALot;
4255 }
4256 
4257 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4258 // actually give up the CPU. Since skip buddy (v2.6.28):
4259 //
4260 // * Sets the yielding task as skip buddy for current CPU's run queue.
4261 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4262 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4263 //
4264 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4265 // getting re-scheduled immediately.
4266 //
4267 void os::naked_yield() {
4268   sched_yield();
4269 }
4270 
4271 ////////////////////////////////////////////////////////////////////////////////
4272 // thread priority support
4273 
4274 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4275 // only supports dynamic priority, static priority must be zero. For real-time
4276 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4277 // However, for large multi-threaded applications, SCHED_RR is not only slower
4278 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4279 // of 5 runs - Sep 2005).
4280 //
4281 // The following code actually changes the niceness of kernel-thread/LWP. It
4282 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4283 // not the entire user process, and user level threads are 1:1 mapped to kernel
4284 // threads. It has always been the case, but could change in the future. For
4285 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4286 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4287 // (e.g., root privilege or CAP_SYS_NICE capability).
4288 
4289 int os::java_to_os_priority[CriticalPriority + 1] = {
4290   19,              // 0 Entry should never be used
4291 
4292    4,              // 1 MinPriority
4293    3,              // 2
4294    2,              // 3
4295 
4296    1,              // 4
4297    0,              // 5 NormPriority
4298   -1,              // 6
4299 
4300   -2,              // 7
4301   -3,              // 8
4302   -4,              // 9 NearMaxPriority
4303 
4304   -5,              // 10 MaxPriority
4305 
4306   -5               // 11 CriticalPriority
4307 };
4308 
4309 static int prio_init() {
4310   if (ThreadPriorityPolicy == 1) {
4311     if (geteuid() != 0) {
4312       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4313         warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4314                 "e.g., being the root user. If the necessary permission is not " \
4315                 "possessed, changes to priority will be silently ignored.");
4316       }
4317     }
4318   }
4319   if (UseCriticalJavaThreadPriority) {
4320     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4321   }
4322   return 0;
4323 }
4324 
4325 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4326   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4327 
4328   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4329   return (ret == 0) ? OS_OK : OS_ERR;
4330 }
4331 
4332 OSReturn os::get_native_priority(const Thread* const thread,
4333                                  int *priority_ptr) {
4334   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4335     *priority_ptr = java_to_os_priority[NormPriority];
4336     return OS_OK;
4337   }
4338 
4339   errno = 0;
4340   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4341   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4342 }
4343 
4344 ////////////////////////////////////////////////////////////////////////////////
4345 // suspend/resume support
4346 
4347 //  The low-level signal-based suspend/resume support is a remnant from the
4348 //  old VM-suspension that used to be for java-suspension, safepoints etc,
4349 //  within hotspot. Currently used by JFR's OSThreadSampler
4350 //
4351 //  The remaining code is greatly simplified from the more general suspension
4352 //  code that used to be used.
4353 //
4354 //  The protocol is quite simple:
4355 //  - suspend:
4356 //      - sends a signal to the target thread
4357 //      - polls the suspend state of the osthread using a yield loop
4358 //      - target thread signal handler (SR_handler) sets suspend state
4359 //        and blocks in sigsuspend until continued
4360 //  - resume:
4361 //      - sets target osthread state to continue
4362 //      - sends signal to end the sigsuspend loop in the SR_handler
4363 //
4364 //  Note that the SR_lock plays no role in this suspend/resume protocol,
4365 //  but is checked for NULL in SR_handler as a thread termination indicator.
4366 //  The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4367 //
4368 //  Note that resume_clear_context() and suspend_save_context() are needed
4369 //  by SR_handler(), so that fetch_frame_from_ucontext() works,
4370 //  which in part is used by:
4371 //    - Forte Analyzer: AsyncGetCallTrace()
4372 //    - StackBanging: get_frame_at_stack_banging_point()
4373 
4374 static void resume_clear_context(OSThread *osthread) {
4375   osthread->set_ucontext(NULL);
4376   osthread->set_siginfo(NULL);
4377 }
4378 
4379 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4380                                  ucontext_t* context) {
4381   osthread->set_ucontext(context);
4382   osthread->set_siginfo(siginfo);
4383 }
4384 
4385 // Handler function invoked when a thread's execution is suspended or
4386 // resumed. We have to be careful that only async-safe functions are
4387 // called here (Note: most pthread functions are not async safe and
4388 // should be avoided.)
4389 //
4390 // Note: sigwait() is a more natural fit than sigsuspend() from an
4391 // interface point of view, but sigwait() prevents the signal hander
4392 // from being run. libpthread would get very confused by not having
4393 // its signal handlers run and prevents sigwait()'s use with the
4394 // mutex granting granting signal.
4395 //
4396 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4397 //
4398 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4399   // Save and restore errno to avoid confusing native code with EINTR
4400   // after sigsuspend.
4401   int old_errno = errno;
4402 
4403   Thread* thread = Thread::current_or_null_safe();
4404   assert(thread != NULL, "Missing current thread in SR_handler");
4405 
4406   // On some systems we have seen signal delivery get "stuck" until the signal
4407   // mask is changed as part of thread termination. Check that the current thread
4408   // has not already terminated (via SR_lock()) - else the following assertion
4409   // will fail because the thread is no longer a JavaThread as the ~JavaThread
4410   // destructor has completed.
4411 
4412   if (thread->SR_lock() == NULL) {
4413     return;
4414   }
4415 
4416   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4417 
4418   OSThread* osthread = thread->osthread();
4419 
4420   os::SuspendResume::State current = osthread->sr.state();
4421   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4422     suspend_save_context(osthread, siginfo, context);
4423 
4424     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4425     os::SuspendResume::State state = osthread->sr.suspended();
4426     if (state == os::SuspendResume::SR_SUSPENDED) {
4427       sigset_t suspend_set;  // signals for sigsuspend()
4428       sigemptyset(&suspend_set);
4429       // get current set of blocked signals and unblock resume signal
4430       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4431       sigdelset(&suspend_set, SR_signum);
4432 
4433       sr_semaphore.signal();
4434       // wait here until we are resumed
4435       while (1) {
4436         sigsuspend(&suspend_set);
4437 
4438         os::SuspendResume::State result = osthread->sr.running();
4439         if (result == os::SuspendResume::SR_RUNNING) {
4440           sr_semaphore.signal();
4441           break;
4442         }
4443       }
4444 
4445     } else if (state == os::SuspendResume::SR_RUNNING) {
4446       // request was cancelled, continue
4447     } else {
4448       ShouldNotReachHere();
4449     }
4450 
4451     resume_clear_context(osthread);
4452   } else if (current == os::SuspendResume::SR_RUNNING) {
4453     // request was cancelled, continue
4454   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4455     // ignore
4456   } else {
4457     // ignore
4458   }
4459 
4460   errno = old_errno;
4461 }
4462 
4463 static int SR_initialize() {
4464   struct sigaction act;
4465   char *s;
4466 
4467   // Get signal number to use for suspend/resume
4468   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4469     int sig = ::strtol(s, 0, 10);
4470     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4471         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4472       SR_signum = sig;
4473     } else {
4474       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4475               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4476     }
4477   }
4478 
4479   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4480          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4481 
4482   sigemptyset(&SR_sigset);
4483   sigaddset(&SR_sigset, SR_signum);
4484 
4485   // Set up signal handler for suspend/resume
4486   act.sa_flags = SA_RESTART|SA_SIGINFO;
4487   act.sa_handler = (void (*)(int)) SR_handler;
4488 
4489   // SR_signum is blocked by default.
4490   // 4528190 - We also need to block pthread restart signal (32 on all
4491   // supported Linux platforms). Note that LinuxThreads need to block
4492   // this signal for all threads to work properly. So we don't have
4493   // to use hard-coded signal number when setting up the mask.
4494   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4495 
4496   if (sigaction(SR_signum, &act, 0) == -1) {
4497     return -1;
4498   }
4499 
4500   // Save signal flag
4501   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4502   return 0;
4503 }
4504 
4505 static int sr_notify(OSThread* osthread) {
4506   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4507   assert_status(status == 0, status, "pthread_kill");
4508   return status;
4509 }
4510 
4511 // "Randomly" selected value for how long we want to spin
4512 // before bailing out on suspending a thread, also how often
4513 // we send a signal to a thread we want to resume
4514 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4515 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4516 
4517 // returns true on success and false on error - really an error is fatal
4518 // but this seems the normal response to library errors
4519 static bool do_suspend(OSThread* osthread) {
4520   assert(osthread->sr.is_running(), "thread should be running");
4521   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4522 
4523   // mark as suspended and send signal
4524   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4525     // failed to switch, state wasn't running?
4526     ShouldNotReachHere();
4527     return false;
4528   }
4529 
4530   if (sr_notify(osthread) != 0) {
4531     ShouldNotReachHere();
4532   }
4533 
4534   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4535   while (true) {
4536     if (sr_semaphore.timedwait(2)) {
4537       break;
4538     } else {
4539       // timeout
4540       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4541       if (cancelled == os::SuspendResume::SR_RUNNING) {
4542         return false;
4543       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4544         // make sure that we consume the signal on the semaphore as well
4545         sr_semaphore.wait();
4546         break;
4547       } else {
4548         ShouldNotReachHere();
4549         return false;
4550       }
4551     }
4552   }
4553 
4554   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4555   return true;
4556 }
4557 
4558 static void do_resume(OSThread* osthread) {
4559   assert(osthread->sr.is_suspended(), "thread should be suspended");
4560   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4561 
4562   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4563     // failed to switch to WAKEUP_REQUEST
4564     ShouldNotReachHere();
4565     return;
4566   }
4567 
4568   while (true) {
4569     if (sr_notify(osthread) == 0) {
4570       if (sr_semaphore.timedwait(2)) {
4571         if (osthread->sr.is_running()) {
4572           return;
4573         }
4574       }
4575     } else {
4576       ShouldNotReachHere();
4577     }
4578   }
4579 
4580   guarantee(osthread->sr.is_running(), "Must be running!");
4581 }
4582 
4583 ///////////////////////////////////////////////////////////////////////////////////
4584 // signal handling (except suspend/resume)
4585 
4586 // This routine may be used by user applications as a "hook" to catch signals.
4587 // The user-defined signal handler must pass unrecognized signals to this
4588 // routine, and if it returns true (non-zero), then the signal handler must
4589 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4590 // routine will never retun false (zero), but instead will execute a VM panic
4591 // routine kill the process.
4592 //
4593 // If this routine returns false, it is OK to call it again.  This allows
4594 // the user-defined signal handler to perform checks either before or after
4595 // the VM performs its own checks.  Naturally, the user code would be making
4596 // a serious error if it tried to handle an exception (such as a null check
4597 // or breakpoint) that the VM was generating for its own correct operation.
4598 //
4599 // This routine may recognize any of the following kinds of signals:
4600 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4601 // It should be consulted by handlers for any of those signals.
4602 //
4603 // The caller of this routine must pass in the three arguments supplied
4604 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4605 // field of the structure passed to sigaction().  This routine assumes that
4606 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4607 //
4608 // Note that the VM will print warnings if it detects conflicting signal
4609 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4610 //
4611 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4612                                                  siginfo_t* siginfo,
4613                                                  void* ucontext,
4614                                                  int abort_if_unrecognized);
4615 
4616 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4617   assert(info != NULL && uc != NULL, "it must be old kernel");
4618   int orig_errno = errno;  // Preserve errno value over signal handler.
4619   JVM_handle_linux_signal(sig, info, uc, true);
4620   errno = orig_errno;
4621 }
4622 
4623 
4624 // This boolean allows users to forward their own non-matching signals
4625 // to JVM_handle_linux_signal, harmlessly.
4626 bool os::Linux::signal_handlers_are_installed = false;
4627 
4628 // For signal-chaining
4629 bool os::Linux::libjsig_is_loaded = false;
4630 typedef struct sigaction *(*get_signal_t)(int);
4631 get_signal_t os::Linux::get_signal_action = NULL;
4632 
4633 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4634   struct sigaction *actp = NULL;
4635 
4636   if (libjsig_is_loaded) {
4637     // Retrieve the old signal handler from libjsig
4638     actp = (*get_signal_action)(sig);
4639   }
4640   if (actp == NULL) {
4641     // Retrieve the preinstalled signal handler from jvm
4642     actp = os::Posix::get_preinstalled_handler(sig);
4643   }
4644 
4645   return actp;
4646 }
4647 
4648 static bool call_chained_handler(struct sigaction *actp, int sig,
4649                                  siginfo_t *siginfo, void *context) {
4650   // Call the old signal handler
4651   if (actp->sa_handler == SIG_DFL) {
4652     // It's more reasonable to let jvm treat it as an unexpected exception
4653     // instead of taking the default action.
4654     return false;
4655   } else if (actp->sa_handler != SIG_IGN) {
4656     if ((actp->sa_flags & SA_NODEFER) == 0) {
4657       // automaticlly block the signal
4658       sigaddset(&(actp->sa_mask), sig);
4659     }
4660 
4661     sa_handler_t hand = NULL;
4662     sa_sigaction_t sa = NULL;
4663     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4664     // retrieve the chained handler
4665     if (siginfo_flag_set) {
4666       sa = actp->sa_sigaction;
4667     } else {
4668       hand = actp->sa_handler;
4669     }
4670 
4671     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4672       actp->sa_handler = SIG_DFL;
4673     }
4674 
4675     // try to honor the signal mask
4676     sigset_t oset;
4677     sigemptyset(&oset);
4678     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4679 
4680     // call into the chained handler
4681     if (siginfo_flag_set) {
4682       (*sa)(sig, siginfo, context);
4683     } else {
4684       (*hand)(sig);
4685     }
4686 
4687     // restore the signal mask
4688     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4689   }
4690   // Tell jvm's signal handler the signal is taken care of.
4691   return true;
4692 }
4693 
4694 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4695   bool chained = false;
4696   // signal-chaining
4697   if (UseSignalChaining) {
4698     struct sigaction *actp = get_chained_signal_action(sig);
4699     if (actp != NULL) {
4700       chained = call_chained_handler(actp, sig, siginfo, context);
4701     }
4702   }
4703   return chained;
4704 }
4705 
4706 // for diagnostic
4707 int sigflags[NSIG];
4708 
4709 int os::Linux::get_our_sigflags(int sig) {
4710   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4711   return sigflags[sig];
4712 }
4713 
4714 void os::Linux::set_our_sigflags(int sig, int flags) {
4715   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4716   if (sig > 0 && sig < NSIG) {
4717     sigflags[sig] = flags;
4718   }
4719 }
4720 
4721 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4722   // Check for overwrite.
4723   struct sigaction oldAct;
4724   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4725 
4726   void* oldhand = oldAct.sa_sigaction
4727                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4728                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4729   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4730       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4731       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4732     if (AllowUserSignalHandlers || !set_installed) {
4733       // Do not overwrite; user takes responsibility to forward to us.
4734       return;
4735     } else if (UseSignalChaining) {
4736       // save the old handler in jvm
4737       os::Posix::save_preinstalled_handler(sig, oldAct);
4738       // libjsig also interposes the sigaction() call below and saves the
4739       // old sigaction on it own.
4740     } else {
4741       fatal("Encountered unexpected pre-existing sigaction handler "
4742             "%#lx for signal %d.", (long)oldhand, sig);
4743     }
4744   }
4745 
4746   struct sigaction sigAct;
4747   sigfillset(&(sigAct.sa_mask));
4748   sigAct.sa_handler = SIG_DFL;
4749   if (!set_installed) {
4750     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4751   } else {
4752     sigAct.sa_sigaction = signalHandler;
4753     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4754   }
4755   // Save flags, which are set by ours
4756   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4757   sigflags[sig] = sigAct.sa_flags;
4758 
4759   int ret = sigaction(sig, &sigAct, &oldAct);
4760   assert(ret == 0, "check");
4761 
4762   void* oldhand2  = oldAct.sa_sigaction
4763                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4764                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4765   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4766 }
4767 
4768 // install signal handlers for signals that HotSpot needs to
4769 // handle in order to support Java-level exception handling.
4770 
4771 void os::Linux::install_signal_handlers() {
4772   if (!signal_handlers_are_installed) {
4773     signal_handlers_are_installed = true;
4774 
4775     // signal-chaining
4776     typedef void (*signal_setting_t)();
4777     signal_setting_t begin_signal_setting = NULL;
4778     signal_setting_t end_signal_setting = NULL;
4779     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4780                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4781     if (begin_signal_setting != NULL) {
4782       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4783                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4784       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4785                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4786       libjsig_is_loaded = true;
4787       assert(UseSignalChaining, "should enable signal-chaining");
4788     }
4789     if (libjsig_is_loaded) {
4790       // Tell libjsig jvm is setting signal handlers
4791       (*begin_signal_setting)();
4792     }
4793 
4794     set_signal_handler(SIGSEGV, true);
4795     set_signal_handler(SIGPIPE, true);
4796     set_signal_handler(SIGBUS, true);
4797     set_signal_handler(SIGILL, true);
4798     set_signal_handler(SIGFPE, true);
4799 #if defined(PPC64)
4800     set_signal_handler(SIGTRAP, true);
4801 #endif
4802     set_signal_handler(SIGXFSZ, true);
4803 
4804     if (libjsig_is_loaded) {
4805       // Tell libjsig jvm finishes setting signal handlers
4806       (*end_signal_setting)();
4807     }
4808 
4809     // We don't activate signal checker if libjsig is in place, we trust ourselves
4810     // and if UserSignalHandler is installed all bets are off.
4811     // Log that signal checking is off only if -verbose:jni is specified.
4812     if (CheckJNICalls) {
4813       if (libjsig_is_loaded) {
4814         if (PrintJNIResolving) {
4815           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4816         }
4817         check_signals = false;
4818       }
4819       if (AllowUserSignalHandlers) {
4820         if (PrintJNIResolving) {
4821           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4822         }
4823         check_signals = false;
4824       }
4825     }
4826   }
4827 }
4828 
4829 // This is the fastest way to get thread cpu time on Linux.
4830 // Returns cpu time (user+sys) for any thread, not only for current.
4831 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4832 // It might work on 2.6.10+ with a special kernel/glibc patch.
4833 // For reference, please, see IEEE Std 1003.1-2004:
4834 //   http://www.unix.org/single_unix_specification
4835 
4836 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4837   struct timespec tp;
4838   int rc = os::Posix::clock_gettime(clockid, &tp);
4839   assert(rc == 0, "clock_gettime is expected to return 0 code");
4840 
4841   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4842 }
4843 
4844 void os::Linux::initialize_os_info() {
4845   assert(_os_version == 0, "OS info already initialized");
4846 
4847   struct utsname _uname;
4848 
4849   uint32_t major;
4850   uint32_t minor;
4851   uint32_t fix;
4852 
4853   int rc;
4854 
4855   // Kernel version is unknown if
4856   // verification below fails.
4857   _os_version = 0x01000000;
4858 
4859   rc = uname(&_uname);
4860   if (rc != -1) {
4861 
4862     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4863     if (rc == 3) {
4864 
4865       if (major < 256 && minor < 256 && fix < 256) {
4866         // Kernel version format is as expected,
4867         // set it overriding unknown state.
4868         _os_version = (major << 16) |
4869                       (minor << 8 ) |
4870                       (fix   << 0 ) ;
4871       }
4872     }
4873   }
4874 }
4875 
4876 uint32_t os::Linux::os_version() {
4877   assert(_os_version != 0, "not initialized");
4878   return _os_version & 0x00FFFFFF;
4879 }
4880 
4881 bool os::Linux::os_version_is_known() {
4882   assert(_os_version != 0, "not initialized");
4883   return _os_version & 0x01000000 ? false : true;
4884 }
4885 
4886 /////
4887 // glibc on Linux platform uses non-documented flag
4888 // to indicate, that some special sort of signal
4889 // trampoline is used.
4890 // We will never set this flag, and we should
4891 // ignore this flag in our diagnostic
4892 #ifdef SIGNIFICANT_SIGNAL_MASK
4893   #undef SIGNIFICANT_SIGNAL_MASK
4894 #endif
4895 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4896 
4897 static const char* get_signal_handler_name(address handler,
4898                                            char* buf, int buflen) {
4899   int offset = 0;
4900   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4901   if (found) {
4902     // skip directory names
4903     const char *p1, *p2;
4904     p1 = buf;
4905     size_t len = strlen(os::file_separator());
4906     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4907     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4908   } else {
4909     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4910   }
4911   return buf;
4912 }
4913 
4914 static void print_signal_handler(outputStream* st, int sig,
4915                                  char* buf, size_t buflen) {
4916   struct sigaction sa;
4917 
4918   sigaction(sig, NULL, &sa);
4919 
4920   // See comment for SIGNIFICANT_SIGNAL_MASK define
4921   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4922 
4923   st->print("%s: ", os::exception_name(sig, buf, buflen));
4924 
4925   address handler = (sa.sa_flags & SA_SIGINFO)
4926     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4927     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4928 
4929   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4930     st->print("SIG_DFL");
4931   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4932     st->print("SIG_IGN");
4933   } else {
4934     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4935   }
4936 
4937   st->print(", sa_mask[0]=");
4938   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4939 
4940   address rh = VMError::get_resetted_sighandler(sig);
4941   // May be, handler was resetted by VMError?
4942   if (rh != NULL) {
4943     handler = rh;
4944     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4945   }
4946 
4947   st->print(", sa_flags=");
4948   os::Posix::print_sa_flags(st, sa.sa_flags);
4949 
4950   // Check: is it our handler?
4951   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4952       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4953     // It is our signal handler
4954     // check for flags, reset system-used one!
4955     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4956       st->print(
4957                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4958                 os::Linux::get_our_sigflags(sig));
4959     }
4960   }
4961   st->cr();
4962 }
4963 
4964 
4965 #define DO_SIGNAL_CHECK(sig)                      \
4966   do {                                            \
4967     if (!sigismember(&check_signal_done, sig)) {  \
4968       os::Linux::check_signal_handler(sig);       \
4969     }                                             \
4970   } while (0)
4971 
4972 // This method is a periodic task to check for misbehaving JNI applications
4973 // under CheckJNI, we can add any periodic checks here
4974 
4975 void os::run_periodic_checks() {
4976   if (check_signals == false) return;
4977 
4978   // SEGV and BUS if overridden could potentially prevent
4979   // generation of hs*.log in the event of a crash, debugging
4980   // such a case can be very challenging, so we absolutely
4981   // check the following for a good measure:
4982   DO_SIGNAL_CHECK(SIGSEGV);
4983   DO_SIGNAL_CHECK(SIGILL);
4984   DO_SIGNAL_CHECK(SIGFPE);
4985   DO_SIGNAL_CHECK(SIGBUS);
4986   DO_SIGNAL_CHECK(SIGPIPE);
4987   DO_SIGNAL_CHECK(SIGXFSZ);
4988 #if defined(PPC64)
4989   DO_SIGNAL_CHECK(SIGTRAP);
4990 #endif
4991 
4992   // ReduceSignalUsage allows the user to override these handlers
4993   // see comments at the very top and jvm_md.h
4994   if (!ReduceSignalUsage) {
4995     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4996     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4997     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4998     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4999   }
5000 
5001   DO_SIGNAL_CHECK(SR_signum);
5002 }
5003 
5004 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
5005 
5006 static os_sigaction_t os_sigaction = NULL;
5007 
5008 void os::Linux::check_signal_handler(int sig) {
5009   char buf[O_BUFLEN];
5010   address jvmHandler = NULL;
5011 
5012 
5013   struct sigaction act;
5014   if (os_sigaction == NULL) {
5015     // only trust the default sigaction, in case it has been interposed
5016     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
5017     if (os_sigaction == NULL) return;
5018   }
5019 
5020   os_sigaction(sig, (struct sigaction*)NULL, &act);
5021 
5022 
5023   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
5024 
5025   address thisHandler = (act.sa_flags & SA_SIGINFO)
5026     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
5027     : CAST_FROM_FN_PTR(address, act.sa_handler);
5028 
5029 
5030   switch (sig) {
5031   case SIGSEGV:
5032   case SIGBUS:
5033   case SIGFPE:
5034   case SIGPIPE:
5035   case SIGILL:
5036   case SIGXFSZ:
5037     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
5038     break;
5039 
5040   case SHUTDOWN1_SIGNAL:
5041   case SHUTDOWN2_SIGNAL:
5042   case SHUTDOWN3_SIGNAL:
5043   case BREAK_SIGNAL:
5044     jvmHandler = (address)user_handler();
5045     break;
5046 
5047   default:
5048     if (sig == SR_signum) {
5049       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
5050     } else {
5051       return;
5052     }
5053     break;
5054   }
5055 
5056   if (thisHandler != jvmHandler) {
5057     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
5058     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
5059     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
5060     // No need to check this sig any longer
5061     sigaddset(&check_signal_done, sig);
5062     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
5063     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
5064       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
5065                     exception_name(sig, buf, O_BUFLEN));
5066     }
5067   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
5068     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
5069     tty->print("expected:");
5070     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
5071     tty->cr();
5072     tty->print("  found:");
5073     os::Posix::print_sa_flags(tty, act.sa_flags);
5074     tty->cr();
5075     // No need to check this sig any longer
5076     sigaddset(&check_signal_done, sig);
5077   }
5078 
5079   // Dump all the signal
5080   if (sigismember(&check_signal_done, sig)) {
5081     print_signal_handlers(tty, buf, O_BUFLEN);
5082   }
5083 }
5084 
5085 extern void report_error(char* file_name, int line_no, char* title,
5086                          char* format, ...);
5087 
5088 // this is called _before_ most of the global arguments have been parsed
5089 void os::init(void) {
5090   char dummy;   // used to get a guess on initial stack address
5091 
5092   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5093 
5094   init_random(1234567);
5095 
5096   Linux::set_page_size(sysconf(_SC_PAGESIZE));
5097   if (Linux::page_size() == -1) {
5098     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
5099           os::strerror(errno));
5100   }
5101   init_page_sizes((size_t) Linux::page_size());
5102 
5103   Linux::initialize_system_info();
5104 
5105   Linux::initialize_os_info();
5106 
5107   os::Linux::CPUPerfTicks pticks;
5108   bool res = os::Linux::get_tick_information(&pticks, -1);
5109 
5110   if (res && pticks.has_steal_ticks) {
5111     has_initial_tick_info = true;
5112     initial_total_ticks = pticks.total;
5113     initial_steal_ticks = pticks.steal;
5114   }
5115 
5116   // _main_thread points to the thread that created/loaded the JVM.
5117   Linux::_main_thread = pthread_self();
5118 
5119   // retrieve entry point for pthread_setname_np
5120   Linux::_pthread_setname_np =
5121     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5122 
5123   os::Posix::init();
5124 
5125   initial_time_count = javaTimeNanos();
5126 
5127   // Always warn if no monotonic clock available
5128   if (!os::Posix::supports_monotonic_clock()) {
5129     warning("No monotonic clock was available - timed services may "    \
5130             "be adversely affected if the time-of-day clock changes");
5131   }
5132 }
5133 
5134 // To install functions for atexit system call
5135 extern "C" {
5136   static void perfMemory_exit_helper() {
5137     perfMemory_exit();
5138   }
5139 }
5140 
5141 void os::pd_init_container_support() {
5142   OSContainer::init();
5143 }
5144 
5145 void os::Linux::numa_init() {
5146 
5147   // Java can be invoked as
5148   // 1. Without numactl and heap will be allocated/configured on all nodes as
5149   //    per the system policy.
5150   // 2. With numactl --interleave:
5151   //      Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
5152   //      API for membind case bitmask is reset.
5153   //      Interleave is only hint and Kernel can fallback to other nodes if
5154   //      no memory is available on the target nodes.
5155   // 3. With numactl --membind:
5156   //      Use numa_get_membind(v2) API to get nodes bitmask. The same API for
5157   //      interleave case returns bitmask of all nodes.
5158   // numa_all_nodes_ptr holds bitmask of all nodes.
5159   // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
5160   // bitmask when externally configured to run on all or fewer nodes.
5161 
5162   if (!Linux::libnuma_init()) {
5163     UseNUMA = false;
5164   } else {
5165     if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
5166       // If there's only one node (they start from 0) or if the process
5167       // is bound explicitly to a single node using membind, disable NUMA.
5168       UseNUMA = false;
5169     } else {
5170 
5171       LogTarget(Info,os) log;
5172       LogStream ls(log);
5173 
5174       Linux::set_configured_numa_policy(Linux::identify_numa_policy());
5175 
5176       struct bitmask* bmp = Linux::_numa_membind_bitmask;
5177       const char* numa_mode = "membind";
5178 
5179       if (Linux::is_running_in_interleave_mode()) {
5180         bmp = Linux::_numa_interleave_bitmask;
5181         numa_mode = "interleave";
5182       }
5183 
5184       ls.print("UseNUMA is enabled and invoked in '%s' mode."
5185                " Heap will be configured using NUMA memory nodes:", numa_mode);
5186 
5187       for (int node = 0; node <= Linux::numa_max_node(); node++) {
5188         if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5189           ls.print(" %d", node);
5190         }
5191       }
5192     }
5193   }
5194 
5195   if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5196     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5197     // we can make the adaptive lgrp chunk resizing work. If the user specified both
5198     // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5199     // and disable adaptive resizing.
5200     if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5201       warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5202               "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5203       UseAdaptiveSizePolicy = false;
5204       UseAdaptiveNUMAChunkSizing = false;
5205     }
5206   }
5207 
5208   if (!UseNUMA && ForceNUMA) {
5209     UseNUMA = true;
5210   }
5211 }
5212 
5213 // this is called _after_ the global arguments have been parsed
5214 jint os::init_2(void) {
5215 
5216   // This could be set after os::Posix::init() but all platforms
5217   // have to set it the same so we have to mirror Solaris.
5218   DEBUG_ONLY(os::set_mutex_init_done();)
5219 
5220   os::Posix::init_2();
5221 
5222   Linux::fast_thread_clock_init();
5223 
5224   // initialize suspend/resume support - must do this before signal_sets_init()
5225   if (SR_initialize() != 0) {
5226     perror("SR_initialize failed");
5227     return JNI_ERR;
5228   }
5229 
5230   Linux::signal_sets_init();
5231   Linux::install_signal_handlers();
5232   // Initialize data for jdk.internal.misc.Signal
5233   if (!ReduceSignalUsage) {
5234     jdk_misc_signal_init();
5235   }
5236 
5237   if (AdjustStackSizeForTLS) {
5238     get_minstack_init();
5239   }
5240 
5241   // Check and sets minimum stack sizes against command line options
5242   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5243     return JNI_ERR;
5244   }
5245 
5246 #if defined(IA32)
5247   // Need to ensure we've determined the process's initial stack to
5248   // perform the workaround
5249   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5250   workaround_expand_exec_shield_cs_limit();
5251 #else
5252   suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5253   if (!suppress_primordial_thread_resolution) {
5254     Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5255   }
5256 #endif
5257 
5258   Linux::libpthread_init();
5259   Linux::sched_getcpu_init();
5260   log_info(os)("HotSpot is running with %s, %s",
5261                Linux::glibc_version(), Linux::libpthread_version());
5262 
5263   if (UseNUMA) {
5264     Linux::numa_init();
5265   }
5266 
5267   if (MaxFDLimit) {
5268     // set the number of file descriptors to max. print out error
5269     // if getrlimit/setrlimit fails but continue regardless.
5270     struct rlimit nbr_files;
5271     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5272     if (status != 0) {
5273       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5274     } else {
5275       nbr_files.rlim_cur = nbr_files.rlim_max;
5276       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5277       if (status != 0) {
5278         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5279       }
5280     }
5281   }
5282 
5283   // Initialize lock used to serialize thread creation (see os::create_thread)
5284   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5285 
5286   // at-exit methods are called in the reverse order of their registration.
5287   // atexit functions are called on return from main or as a result of a
5288   // call to exit(3C). There can be only 32 of these functions registered
5289   // and atexit() does not set errno.
5290 
5291   if (PerfAllowAtExitRegistration) {
5292     // only register atexit functions if PerfAllowAtExitRegistration is set.
5293     // atexit functions can be delayed until process exit time, which
5294     // can be problematic for embedded VM situations. Embedded VMs should
5295     // call DestroyJavaVM() to assure that VM resources are released.
5296 
5297     // note: perfMemory_exit_helper atexit function may be removed in
5298     // the future if the appropriate cleanup code can be added to the
5299     // VM_Exit VMOperation's doit method.
5300     if (atexit(perfMemory_exit_helper) != 0) {
5301       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5302     }
5303   }
5304 
5305   // initialize thread priority policy
5306   prio_init();
5307 
5308   if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) {
5309     set_coredump_filter(DAX_SHARED_BIT);
5310   }
5311 
5312   if (DumpPrivateMappingsInCore) {
5313     set_coredump_filter(FILE_BACKED_PVT_BIT);
5314   }
5315 
5316   if (DumpSharedMappingsInCore) {
5317     set_coredump_filter(FILE_BACKED_SHARED_BIT);
5318   }
5319 
5320   return JNI_OK;
5321 }
5322 
5323 // Mark the polling page as unreadable
5324 void os::make_polling_page_unreadable(void) {
5325   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5326     fatal("Could not disable polling page");
5327   }
5328 }
5329 
5330 // Mark the polling page as readable
5331 void os::make_polling_page_readable(void) {
5332   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5333     fatal("Could not enable polling page");
5334   }
5335 }
5336 
5337 // older glibc versions don't have this macro (which expands to
5338 // an optimized bit-counting function) so we have to roll our own
5339 #ifndef CPU_COUNT
5340 
5341 static int _cpu_count(const cpu_set_t* cpus) {
5342   int count = 0;
5343   // only look up to the number of configured processors
5344   for (int i = 0; i < os::processor_count(); i++) {
5345     if (CPU_ISSET(i, cpus)) {
5346       count++;
5347     }
5348   }
5349   return count;
5350 }
5351 
5352 #define CPU_COUNT(cpus) _cpu_count(cpus)
5353 
5354 #endif // CPU_COUNT
5355 
5356 // Get the current number of available processors for this process.
5357 // This value can change at any time during a process's lifetime.
5358 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5359 // If it appears there may be more than 1024 processors then we do a
5360 // dynamic check - see 6515172 for details.
5361 // If anything goes wrong we fallback to returning the number of online
5362 // processors - which can be greater than the number available to the process.
5363 int os::Linux::active_processor_count() {
5364   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
5365   cpu_set_t* cpus_p = &cpus;
5366   int cpus_size = sizeof(cpu_set_t);
5367 
5368   int configured_cpus = os::processor_count();  // upper bound on available cpus
5369   int cpu_count = 0;
5370 
5371 // old build platforms may not support dynamic cpu sets
5372 #ifdef CPU_ALLOC
5373 
5374   // To enable easy testing of the dynamic path on different platforms we
5375   // introduce a diagnostic flag: UseCpuAllocPath
5376   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5377     // kernel may use a mask bigger than cpu_set_t
5378     log_trace(os)("active_processor_count: using dynamic path %s"
5379                   "- configured processors: %d",
5380                   UseCpuAllocPath ? "(forced) " : "",
5381                   configured_cpus);
5382     cpus_p = CPU_ALLOC(configured_cpus);
5383     if (cpus_p != NULL) {
5384       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5385       // zero it just to be safe
5386       CPU_ZERO_S(cpus_size, cpus_p);
5387     }
5388     else {
5389        // failed to allocate so fallback to online cpus
5390        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5391        log_trace(os)("active_processor_count: "
5392                      "CPU_ALLOC failed (%s) - using "
5393                      "online processor count: %d",
5394                      os::strerror(errno), online_cpus);
5395        return online_cpus;
5396     }
5397   }
5398   else {
5399     log_trace(os)("active_processor_count: using static path - configured processors: %d",
5400                   configured_cpus);
5401   }
5402 #else // CPU_ALLOC
5403 // these stubs won't be executed
5404 #define CPU_COUNT_S(size, cpus) -1
5405 #define CPU_FREE(cpus)
5406 
5407   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5408                 configured_cpus);
5409 #endif // CPU_ALLOC
5410 
5411   // pid 0 means the current thread - which we have to assume represents the process
5412   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5413     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5414       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5415     }
5416     else {
5417       cpu_count = CPU_COUNT(cpus_p);
5418     }
5419     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5420   }
5421   else {
5422     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5423     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5424             "which may exceed available processors", os::strerror(errno), cpu_count);
5425   }
5426 
5427   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5428     CPU_FREE(cpus_p);
5429   }
5430 
5431   assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5432   return cpu_count;
5433 }
5434 
5435 // Determine the active processor count from one of
5436 // three different sources:
5437 //
5438 // 1. User option -XX:ActiveProcessorCount
5439 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5440 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5441 //
5442 // Option 1, if specified, will always override.
5443 // If the cgroup subsystem is active and configured, we
5444 // will return the min of the cgroup and option 2 results.
5445 // This is required since tools, such as numactl, that
5446 // alter cpu affinity do not update cgroup subsystem
5447 // cpuset configuration files.
5448 int os::active_processor_count() {
5449   // User has overridden the number of active processors
5450   if (ActiveProcessorCount > 0) {
5451     log_trace(os)("active_processor_count: "
5452                   "active processor count set by user : %d",
5453                   ActiveProcessorCount);
5454     return ActiveProcessorCount;
5455   }
5456 
5457   int active_cpus;
5458   if (OSContainer::is_containerized()) {
5459     active_cpus = OSContainer::active_processor_count();
5460     log_trace(os)("active_processor_count: determined by OSContainer: %d",
5461                    active_cpus);
5462   } else {
5463     active_cpus = os::Linux::active_processor_count();
5464   }
5465 
5466   return active_cpus;
5467 }
5468 
5469 uint os::processor_id() {
5470   const int id = Linux::sched_getcpu();
5471   assert(id >= 0 && id < _processor_count, "Invalid processor id");
5472   return (uint)id;
5473 }
5474 
5475 void os::set_native_thread_name(const char *name) {
5476   if (Linux::_pthread_setname_np) {
5477     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5478     snprintf(buf, sizeof(buf), "%s", name);
5479     buf[sizeof(buf) - 1] = '\0';
5480     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5481     // ERANGE should not happen; all other errors should just be ignored.
5482     assert(rc != ERANGE, "pthread_setname_np failed");
5483   }
5484 }
5485 
5486 bool os::distribute_processes(uint length, uint* distribution) {
5487   // Not yet implemented.
5488   return false;
5489 }
5490 
5491 bool os::bind_to_processor(uint processor_id) {
5492   // Not yet implemented.
5493   return false;
5494 }
5495 
5496 ///
5497 
5498 void os::SuspendedThreadTask::internal_do_task() {
5499   if (do_suspend(_thread->osthread())) {
5500     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5501     do_task(context);
5502     do_resume(_thread->osthread());
5503   }
5504 }
5505 
5506 ////////////////////////////////////////////////////////////////////////////////
5507 // debug support
5508 
5509 bool os::find(address addr, outputStream* st) {
5510   Dl_info dlinfo;
5511   memset(&dlinfo, 0, sizeof(dlinfo));
5512   if (dladdr(addr, &dlinfo) != 0) {
5513     st->print(PTR_FORMAT ": ", p2i(addr));
5514     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5515       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5516                 p2i(addr) - p2i(dlinfo.dli_saddr));
5517     } else if (dlinfo.dli_fbase != NULL) {
5518       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5519     } else {
5520       st->print("<absolute address>");
5521     }
5522     if (dlinfo.dli_fname != NULL) {
5523       st->print(" in %s", dlinfo.dli_fname);
5524     }
5525     if (dlinfo.dli_fbase != NULL) {
5526       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5527     }
5528     st->cr();
5529 
5530     if (Verbose) {
5531       // decode some bytes around the PC
5532       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5533       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5534       address       lowest = (address) dlinfo.dli_sname;
5535       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5536       if (begin < lowest)  begin = lowest;
5537       Dl_info dlinfo2;
5538       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5539           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5540         end = (address) dlinfo2.dli_saddr;
5541       }
5542       Disassembler::decode(begin, end, st);
5543     }
5544     return true;
5545   }
5546   return false;
5547 }
5548 
5549 ////////////////////////////////////////////////////////////////////////////////
5550 // misc
5551 
5552 // This does not do anything on Linux. This is basically a hook for being
5553 // able to use structured exception handling (thread-local exception filters)
5554 // on, e.g., Win32.
5555 void
5556 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5557                          JavaCallArguments* args, Thread* thread) {
5558   f(value, method, args, thread);
5559 }
5560 
5561 void os::print_statistics() {
5562 }
5563 
5564 bool os::message_box(const char* title, const char* message) {
5565   int i;
5566   fdStream err(defaultStream::error_fd());
5567   for (i = 0; i < 78; i++) err.print_raw("=");
5568   err.cr();
5569   err.print_raw_cr(title);
5570   for (i = 0; i < 78; i++) err.print_raw("-");
5571   err.cr();
5572   err.print_raw_cr(message);
5573   for (i = 0; i < 78; i++) err.print_raw("=");
5574   err.cr();
5575 
5576   char buf[16];
5577   // Prevent process from exiting upon "read error" without consuming all CPU
5578   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5579 
5580   return buf[0] == 'y' || buf[0] == 'Y';
5581 }
5582 
5583 // Is a (classpath) directory empty?
5584 bool os::dir_is_empty(const char* path) {
5585   DIR *dir = NULL;
5586   struct dirent *ptr;
5587 
5588   dir = opendir(path);
5589   if (dir == NULL) return true;
5590 
5591   // Scan the directory
5592   bool result = true;
5593   while (result && (ptr = readdir(dir)) != NULL) {
5594     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5595       result = false;
5596     }
5597   }
5598   closedir(dir);
5599   return result;
5600 }
5601 
5602 // This code originates from JDK's sysOpen and open64_w
5603 // from src/solaris/hpi/src/system_md.c
5604 
5605 int os::open(const char *path, int oflag, int mode) {
5606   if (strlen(path) > MAX_PATH - 1) {
5607     errno = ENAMETOOLONG;
5608     return -1;
5609   }
5610 
5611   // All file descriptors that are opened in the Java process and not
5612   // specifically destined for a subprocess should have the close-on-exec
5613   // flag set.  If we don't set it, then careless 3rd party native code
5614   // might fork and exec without closing all appropriate file descriptors
5615   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5616   // turn might:
5617   //
5618   // - cause end-of-file to fail to be detected on some file
5619   //   descriptors, resulting in mysterious hangs, or
5620   //
5621   // - might cause an fopen in the subprocess to fail on a system
5622   //   suffering from bug 1085341.
5623   //
5624   // (Yes, the default setting of the close-on-exec flag is a Unix
5625   // design flaw)
5626   //
5627   // See:
5628   // 1085341: 32-bit stdio routines should support file descriptors >255
5629   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5630   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5631   //
5632   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5633   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5634   // because it saves a system call and removes a small window where the flag
5635   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5636   // and we fall back to using FD_CLOEXEC (see below).
5637 #ifdef O_CLOEXEC
5638   oflag |= O_CLOEXEC;
5639 #endif
5640 
5641   int fd = ::open64(path, oflag, mode);
5642   if (fd == -1) return -1;
5643 
5644   //If the open succeeded, the file might still be a directory
5645   {
5646     struct stat64 buf64;
5647     int ret = ::fstat64(fd, &buf64);
5648     int st_mode = buf64.st_mode;
5649 
5650     if (ret != -1) {
5651       if ((st_mode & S_IFMT) == S_IFDIR) {
5652         errno = EISDIR;
5653         ::close(fd);
5654         return -1;
5655       }
5656     } else {
5657       ::close(fd);
5658       return -1;
5659     }
5660   }
5661 
5662 #ifdef FD_CLOEXEC
5663   // Validate that the use of the O_CLOEXEC flag on open above worked.
5664   // With recent kernels, we will perform this check exactly once.
5665   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5666   if (!O_CLOEXEC_is_known_to_work) {
5667     int flags = ::fcntl(fd, F_GETFD);
5668     if (flags != -1) {
5669       if ((flags & FD_CLOEXEC) != 0)
5670         O_CLOEXEC_is_known_to_work = 1;
5671       else
5672         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5673     }
5674   }
5675 #endif
5676 
5677   return fd;
5678 }
5679 
5680 
5681 // create binary file, rewriting existing file if required
5682 int os::create_binary_file(const char* path, bool rewrite_existing) {
5683   int oflags = O_WRONLY | O_CREAT;
5684   if (!rewrite_existing) {
5685     oflags |= O_EXCL;
5686   }
5687   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5688 }
5689 
5690 // return current position of file pointer
5691 jlong os::current_file_offset(int fd) {
5692   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5693 }
5694 
5695 // move file pointer to the specified offset
5696 jlong os::seek_to_file_offset(int fd, jlong offset) {
5697   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5698 }
5699 
5700 // This code originates from JDK's sysAvailable
5701 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5702 
5703 int os::available(int fd, jlong *bytes) {
5704   jlong cur, end;
5705   int mode;
5706   struct stat64 buf64;
5707 
5708   if (::fstat64(fd, &buf64) >= 0) {
5709     mode = buf64.st_mode;
5710     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5711       int n;
5712       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5713         *bytes = n;
5714         return 1;
5715       }
5716     }
5717   }
5718   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5719     return 0;
5720   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5721     return 0;
5722   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5723     return 0;
5724   }
5725   *bytes = end - cur;
5726   return 1;
5727 }
5728 
5729 // Map a block of memory.
5730 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5731                         char *addr, size_t bytes, bool read_only,
5732                         bool allow_exec) {
5733   int prot;
5734   int flags = MAP_PRIVATE;
5735 
5736   if (read_only) {
5737     prot = PROT_READ;
5738   } else {
5739     prot = PROT_READ | PROT_WRITE;
5740   }
5741 
5742   if (allow_exec) {
5743     prot |= PROT_EXEC;
5744   }
5745 
5746   if (addr != NULL) {
5747     flags |= MAP_FIXED;
5748   }
5749 
5750   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5751                                      fd, file_offset);
5752   if (mapped_address == MAP_FAILED) {
5753     return NULL;
5754   }
5755   return mapped_address;
5756 }
5757 
5758 
5759 // Remap a block of memory.
5760 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5761                           char *addr, size_t bytes, bool read_only,
5762                           bool allow_exec) {
5763   // same as map_memory() on this OS
5764   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5765                         allow_exec);
5766 }
5767 
5768 
5769 // Unmap a block of memory.
5770 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5771   return munmap(addr, bytes) == 0;
5772 }
5773 
5774 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5775 
5776 static jlong fast_cpu_time(Thread *thread) {
5777     clockid_t clockid;
5778     int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5779                                               &clockid);
5780     if (rc == 0) {
5781       return os::Linux::fast_thread_cpu_time(clockid);
5782     } else {
5783       // It's possible to encounter a terminated native thread that failed
5784       // to detach itself from the VM - which should result in ESRCH.
5785       assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5786       return -1;
5787     }
5788 }
5789 
5790 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5791 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5792 // of a thread.
5793 //
5794 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5795 // the fast estimate available on the platform.
5796 
5797 jlong os::current_thread_cpu_time() {
5798   if (os::Linux::supports_fast_thread_cpu_time()) {
5799     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5800   } else {
5801     // return user + sys since the cost is the same
5802     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5803   }
5804 }
5805 
5806 jlong os::thread_cpu_time(Thread* thread) {
5807   // consistent with what current_thread_cpu_time() returns
5808   if (os::Linux::supports_fast_thread_cpu_time()) {
5809     return fast_cpu_time(thread);
5810   } else {
5811     return slow_thread_cpu_time(thread, true /* user + sys */);
5812   }
5813 }
5814 
5815 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5816   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5817     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5818   } else {
5819     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5820   }
5821 }
5822 
5823 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5824   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5825     return fast_cpu_time(thread);
5826   } else {
5827     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5828   }
5829 }
5830 
5831 //  -1 on error.
5832 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5833   pid_t  tid = thread->osthread()->thread_id();
5834   char *s;
5835   char stat[2048];
5836   int statlen;
5837   char proc_name[64];
5838   int count;
5839   long sys_time, user_time;
5840   char cdummy;
5841   int idummy;
5842   long ldummy;
5843   FILE *fp;
5844 
5845   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5846   fp = fopen(proc_name, "r");
5847   if (fp == NULL) return -1;
5848   statlen = fread(stat, 1, 2047, fp);
5849   stat[statlen] = '\0';
5850   fclose(fp);
5851 
5852   // Skip pid and the command string. Note that we could be dealing with
5853   // weird command names, e.g. user could decide to rename java launcher
5854   // to "java 1.4.2 :)", then the stat file would look like
5855   //                1234 (java 1.4.2 :)) R ... ...
5856   // We don't really need to know the command string, just find the last
5857   // occurrence of ")" and then start parsing from there. See bug 4726580.
5858   s = strrchr(stat, ')');
5859   if (s == NULL) return -1;
5860 
5861   // Skip blank chars
5862   do { s++; } while (s && isspace(*s));
5863 
5864   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5865                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5866                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5867                  &user_time, &sys_time);
5868   if (count != 13) return -1;
5869   if (user_sys_cpu_time) {
5870     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5871   } else {
5872     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5873   }
5874 }
5875 
5876 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5877   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5878   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5879   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5880   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5881 }
5882 
5883 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5884   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5885   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5886   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5887   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5888 }
5889 
5890 bool os::is_thread_cpu_time_supported() {
5891   return true;
5892 }
5893 
5894 // System loadavg support.  Returns -1 if load average cannot be obtained.
5895 // Linux doesn't yet have a (official) notion of processor sets,
5896 // so just return the system wide load average.
5897 int os::loadavg(double loadavg[], int nelem) {
5898   return ::getloadavg(loadavg, nelem);
5899 }
5900 
5901 void os::pause() {
5902   char filename[MAX_PATH];
5903   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5904     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5905   } else {
5906     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5907   }
5908 
5909   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5910   if (fd != -1) {
5911     struct stat buf;
5912     ::close(fd);
5913     while (::stat(filename, &buf) == 0) {
5914       (void)::poll(NULL, 0, 100);
5915     }
5916   } else {
5917     jio_fprintf(stderr,
5918                 "Could not open pause file '%s', continuing immediately.\n", filename);
5919   }
5920 }
5921 
5922 extern char** environ;
5923 
5924 // Run the specified command in a separate process. Return its exit value,
5925 // or -1 on failure (e.g. can't fork a new process).
5926 // Unlike system(), this function can be called from signal handler. It
5927 // doesn't block SIGINT et al.
5928 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5929   const char * argv[4] = {"sh", "-c", cmd, NULL};
5930 
5931   pid_t pid ;
5932 
5933   if (use_vfork_if_available) {
5934     pid = vfork();
5935   } else {
5936     pid = fork();
5937   }
5938 
5939   if (pid < 0) {
5940     // fork failed
5941     return -1;
5942 
5943   } else if (pid == 0) {
5944     // child process
5945 
5946     execve("/bin/sh", (char* const*)argv, environ);
5947 
5948     // execve failed
5949     _exit(-1);
5950 
5951   } else  {
5952     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5953     // care about the actual exit code, for now.
5954 
5955     int status;
5956 
5957     // Wait for the child process to exit.  This returns immediately if
5958     // the child has already exited. */
5959     while (waitpid(pid, &status, 0) < 0) {
5960       switch (errno) {
5961       case ECHILD: return 0;
5962       case EINTR: break;
5963       default: return -1;
5964       }
5965     }
5966 
5967     if (WIFEXITED(status)) {
5968       // The child exited normally; get its exit code.
5969       return WEXITSTATUS(status);
5970     } else if (WIFSIGNALED(status)) {
5971       // The child exited because of a signal
5972       // The best value to return is 0x80 + signal number,
5973       // because that is what all Unix shells do, and because
5974       // it allows callers to distinguish between process exit and
5975       // process death by signal.
5976       return 0x80 + WTERMSIG(status);
5977     } else {
5978       // Unknown exit code; pass it through
5979       return status;
5980     }
5981   }
5982 }
5983 
5984 // Get the default path to the core file
5985 // Returns the length of the string
5986 int os::get_core_path(char* buffer, size_t bufferSize) {
5987   /*
5988    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5989    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5990    */
5991   const int core_pattern_len = 129;
5992   char core_pattern[core_pattern_len] = {0};
5993 
5994   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5995   if (core_pattern_file == -1) {
5996     return -1;
5997   }
5998 
5999   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6000   ::close(core_pattern_file);
6001   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
6002     return -1;
6003   }
6004   if (core_pattern[ret-1] == '\n') {
6005     core_pattern[ret-1] = '\0';
6006   } else {
6007     core_pattern[ret] = '\0';
6008   }
6009 
6010   // Replace the %p in the core pattern with the process id. NOTE: we do this
6011   // only if the pattern doesn't start with "|", and we support only one %p in
6012   // the pattern.
6013   char *pid_pos = strstr(core_pattern, "%p");
6014   const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : "";  // skip over the "%p"
6015   int written;
6016 
6017   if (core_pattern[0] == '/') {
6018     if (pid_pos != NULL) {
6019       *pid_pos = '\0';
6020       written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
6021                              current_process_id(), tail);
6022     } else {
6023       written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
6024     }
6025   } else {
6026     char cwd[PATH_MAX];
6027 
6028     const char* p = get_current_directory(cwd, PATH_MAX);
6029     if (p == NULL) {
6030       return -1;
6031     }
6032 
6033     if (core_pattern[0] == '|') {
6034       written = jio_snprintf(buffer, bufferSize,
6035                              "\"%s\" (or dumping to %s/core.%d)",
6036                              &core_pattern[1], p, current_process_id());
6037     } else if (pid_pos != NULL) {
6038       *pid_pos = '\0';
6039       written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
6040                              current_process_id(), tail);
6041     } else {
6042       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6043     }
6044   }
6045 
6046   if (written < 0) {
6047     return -1;
6048   }
6049 
6050   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6051     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6052 
6053     if (core_uses_pid_file != -1) {
6054       char core_uses_pid = 0;
6055       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6056       ::close(core_uses_pid_file);
6057 
6058       if (core_uses_pid == '1') {
6059         jio_snprintf(buffer + written, bufferSize - written,
6060                                           ".%d", current_process_id());
6061       }
6062     }
6063   }
6064 
6065   return strlen(buffer);
6066 }
6067 
6068 bool os::start_debugging(char *buf, int buflen) {
6069   int len = (int)strlen(buf);
6070   char *p = &buf[len];
6071 
6072   jio_snprintf(p, buflen-len,
6073                "\n\n"
6074                "Do you want to debug the problem?\n\n"
6075                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6076                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6077                "Otherwise, press RETURN to abort...",
6078                os::current_process_id(), os::current_process_id(),
6079                os::current_thread_id(), os::current_thread_id());
6080 
6081   bool yes = os::message_box("Unexpected Error", buf);
6082 
6083   if (yes) {
6084     // yes, user asked VM to launch debugger
6085     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6086                  os::current_process_id(), os::current_process_id());
6087 
6088     os::fork_and_exec(buf);
6089     yes = false;
6090   }
6091   return yes;
6092 }
6093 
6094 
6095 // Java/Compiler thread:
6096 //
6097 //   Low memory addresses
6098 // P0 +------------------------+
6099 //    |                        |\  Java thread created by VM does not have glibc
6100 //    |    glibc guard page    | - guard page, attached Java thread usually has
6101 //    |                        |/  1 glibc guard page.
6102 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6103 //    |                        |\
6104 //    |  HotSpot Guard Pages   | - red, yellow and reserved pages
6105 //    |                        |/
6106 //    +------------------------+ JavaThread::stack_reserved_zone_base()
6107 //    |                        |\
6108 //    |      Normal Stack      | -
6109 //    |                        |/
6110 // P2 +------------------------+ Thread::stack_base()
6111 //
6112 // Non-Java thread:
6113 //
6114 //   Low memory addresses
6115 // P0 +------------------------+
6116 //    |                        |\
6117 //    |  glibc guard page      | - usually 1 page
6118 //    |                        |/
6119 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6120 //    |                        |\
6121 //    |      Normal Stack      | -
6122 //    |                        |/
6123 // P2 +------------------------+ Thread::stack_base()
6124 //
6125 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6126 //    returned from pthread_attr_getstack().
6127 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6128 //    of the stack size given in pthread_attr. We work around this for
6129 //    threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6130 //
6131 #ifndef ZERO
6132 static void current_stack_region(address * bottom, size_t * size) {
6133   if (os::is_primordial_thread()) {
6134     // primordial thread needs special handling because pthread_getattr_np()
6135     // may return bogus value.
6136     *bottom = os::Linux::initial_thread_stack_bottom();
6137     *size   = os::Linux::initial_thread_stack_size();
6138   } else {
6139     pthread_attr_t attr;
6140 
6141     int rslt = pthread_getattr_np(pthread_self(), &attr);
6142 
6143     // JVM needs to know exact stack location, abort if it fails
6144     if (rslt != 0) {
6145       if (rslt == ENOMEM) {
6146         vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6147       } else {
6148         fatal("pthread_getattr_np failed with error = %d", rslt);
6149       }
6150     }
6151 
6152     if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6153       fatal("Cannot locate current stack attributes!");
6154     }
6155 
6156     // Work around NPTL stack guard error.
6157     size_t guard_size = 0;
6158     rslt = pthread_attr_getguardsize(&attr, &guard_size);
6159     if (rslt != 0) {
6160       fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6161     }
6162     *bottom += guard_size;
6163     *size   -= guard_size;
6164 
6165     pthread_attr_destroy(&attr);
6166 
6167   }
6168   assert(os::current_stack_pointer() >= *bottom &&
6169          os::current_stack_pointer() < *bottom + *size, "just checking");
6170 }
6171 
6172 address os::current_stack_base() {
6173   address bottom;
6174   size_t size;
6175   current_stack_region(&bottom, &size);
6176   return (bottom + size);
6177 }
6178 
6179 size_t os::current_stack_size() {
6180   // This stack size includes the usable stack and HotSpot guard pages
6181   // (for the threads that have Hotspot guard pages).
6182   address bottom;
6183   size_t size;
6184   current_stack_region(&bottom, &size);
6185   return size;
6186 }
6187 #endif
6188 
6189 static inline struct timespec get_mtime(const char* filename) {
6190   struct stat st;
6191   int ret = os::stat(filename, &st);
6192   assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
6193   return st.st_mtim;
6194 }
6195 
6196 int os::compare_file_modified_times(const char* file1, const char* file2) {
6197   struct timespec filetime1 = get_mtime(file1);
6198   struct timespec filetime2 = get_mtime(file2);
6199   int diff = filetime1.tv_sec - filetime2.tv_sec;
6200   if (diff == 0) {
6201     return filetime1.tv_nsec - filetime2.tv_nsec;
6202   }
6203   return diff;
6204 }
6205 
6206 bool os::supports_map_sync() {
6207   return true;
6208 }
6209 
6210 /////////////// Unit tests ///////////////
6211 
6212 #ifndef PRODUCT
6213 
6214 class TestReserveMemorySpecial : AllStatic {
6215  public:
6216   static void small_page_write(void* addr, size_t size) {
6217     size_t page_size = os::vm_page_size();
6218 
6219     char* end = (char*)addr + size;
6220     for (char* p = (char*)addr; p < end; p += page_size) {
6221       *p = 1;
6222     }
6223   }
6224 
6225   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6226     if (!UseHugeTLBFS) {
6227       return;
6228     }
6229 
6230     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6231 
6232     if (addr != NULL) {
6233       small_page_write(addr, size);
6234 
6235       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6236     }
6237   }
6238 
6239   static void test_reserve_memory_special_huge_tlbfs_only() {
6240     if (!UseHugeTLBFS) {
6241       return;
6242     }
6243 
6244     size_t lp = os::large_page_size();
6245 
6246     for (size_t size = lp; size <= lp * 10; size += lp) {
6247       test_reserve_memory_special_huge_tlbfs_only(size);
6248     }
6249   }
6250 
6251   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6252     size_t lp = os::large_page_size();
6253     size_t ag = os::vm_allocation_granularity();
6254 
6255     // sizes to test
6256     const size_t sizes[] = {
6257       lp, lp + ag, lp + lp / 2, lp * 2,
6258       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6259       lp * 10, lp * 10 + lp / 2
6260     };
6261     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6262 
6263     // For each size/alignment combination, we test three scenarios:
6264     // 1) with req_addr == NULL
6265     // 2) with a non-null req_addr at which we expect to successfully allocate
6266     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6267     //    expect the allocation to either fail or to ignore req_addr
6268 
6269     // Pre-allocate two areas; they shall be as large as the largest allocation
6270     //  and aligned to the largest alignment we will be testing.
6271     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6272     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6273       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6274       -1, 0);
6275     assert(mapping1 != MAP_FAILED, "should work");
6276 
6277     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6278       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6279       -1, 0);
6280     assert(mapping2 != MAP_FAILED, "should work");
6281 
6282     // Unmap the first mapping, but leave the second mapping intact: the first
6283     // mapping will serve as a value for a "good" req_addr (case 2). The second
6284     // mapping, still intact, as "bad" req_addr (case 3).
6285     ::munmap(mapping1, mapping_size);
6286 
6287     // Case 1
6288     for (int i = 0; i < num_sizes; i++) {
6289       const size_t size = sizes[i];
6290       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6291         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6292         if (p != NULL) {
6293           assert(is_aligned(p, alignment), "must be");
6294           small_page_write(p, size);
6295           os::Linux::release_memory_special_huge_tlbfs(p, size);
6296         }
6297       }
6298     }
6299 
6300     // Case 2
6301     for (int i = 0; i < num_sizes; i++) {
6302       const size_t size = sizes[i];
6303       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6304         char* const req_addr = align_up(mapping1, alignment);
6305         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6306         if (p != NULL) {
6307           assert(p == req_addr, "must be");
6308           small_page_write(p, size);
6309           os::Linux::release_memory_special_huge_tlbfs(p, size);
6310         }
6311       }
6312     }
6313 
6314     // Case 3
6315     for (int i = 0; i < num_sizes; i++) {
6316       const size_t size = sizes[i];
6317       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6318         char* const req_addr = align_up(mapping2, alignment);
6319         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6320         // as the area around req_addr contains already existing mappings, the API should always
6321         // return NULL (as per contract, it cannot return another address)
6322         assert(p == NULL, "must be");
6323       }
6324     }
6325 
6326     ::munmap(mapping2, mapping_size);
6327 
6328   }
6329 
6330   static void test_reserve_memory_special_huge_tlbfs() {
6331     if (!UseHugeTLBFS) {
6332       return;
6333     }
6334 
6335     test_reserve_memory_special_huge_tlbfs_only();
6336     test_reserve_memory_special_huge_tlbfs_mixed();
6337   }
6338 
6339   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6340     if (!UseSHM) {
6341       return;
6342     }
6343 
6344     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6345 
6346     if (addr != NULL) {
6347       assert(is_aligned(addr, alignment), "Check");
6348       assert(is_aligned(addr, os::large_page_size()), "Check");
6349 
6350       small_page_write(addr, size);
6351 
6352       os::Linux::release_memory_special_shm(addr, size);
6353     }
6354   }
6355 
6356   static void test_reserve_memory_special_shm() {
6357     size_t lp = os::large_page_size();
6358     size_t ag = os::vm_allocation_granularity();
6359 
6360     for (size_t size = ag; size < lp * 3; size += ag) {
6361       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6362         test_reserve_memory_special_shm(size, alignment);
6363       }
6364     }
6365   }
6366 
6367   static void test() {
6368     test_reserve_memory_special_huge_tlbfs();
6369     test_reserve_memory_special_shm();
6370   }
6371 };
6372 
6373 void TestReserveMemorySpecial_test() {
6374   TestReserveMemorySpecial::test();
6375 }
6376 
6377 #endif
--- EOF ---