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