1 /*
   2  * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "jvm.h"
  27 #include "classfile/classLoader.hpp"
  28 #include "classfile/systemDictionary.hpp"
  29 #include "classfile/vmSymbols.hpp"
  30 #include "code/icBuffer.hpp"
  31 #include "code/vtableStubs.hpp"
  32 #include "compiler/compileBroker.hpp"
  33 #include "compiler/disassembler.hpp"
  34 #include "interpreter/interpreter.hpp"
  35 #include "logging/log.hpp"
  36 #include "logging/logStream.hpp"
  37 #include "memory/allocation.inline.hpp"
  38 #include "memory/filemap.hpp"
  39 #include "oops/oop.inline.hpp"
  40 #include "os_linux.inline.hpp"
  41 #include "os_posix.inline.hpp"
  42 #include "os_share_linux.hpp"
  43 #include "osContainer_linux.hpp"
  44 #include "prims/jniFastGetField.hpp"
  45 #include "prims/jvm_misc.hpp"
  46 #include "runtime/arguments.hpp"
  47 #include "runtime/atomic.hpp"
  48 #include "runtime/extendedPC.hpp"
  49 #include "runtime/globals.hpp"
  50 #include "runtime/interfaceSupport.inline.hpp"
  51 #include "runtime/init.hpp"
  52 #include "runtime/java.hpp"
  53 #include "runtime/javaCalls.hpp"
  54 #include "runtime/mutexLocker.hpp"
  55 #include "runtime/objectMonitor.hpp"
  56 #include "runtime/orderAccess.hpp"
  57 #include "runtime/osThread.hpp"
  58 #include "runtime/perfMemory.hpp"
  59 #include "runtime/sharedRuntime.hpp"
  60 #include "runtime/statSampler.hpp"
  61 #include "runtime/stubRoutines.hpp"
  62 #include "runtime/thread.inline.hpp"
  63 #include "runtime/threadCritical.hpp"
  64 #include "runtime/threadSMR.hpp"
  65 #include "runtime/timer.hpp"
  66 #include "runtime/vm_version.hpp"
  67 #include "semaphore_posix.hpp"
  68 #include "services/attachListener.hpp"
  69 #include "services/memTracker.hpp"
  70 #include "services/runtimeService.hpp"
  71 #include "utilities/align.hpp"
  72 #include "utilities/decoder.hpp"
  73 #include "utilities/defaultStream.hpp"
  74 #include "utilities/events.hpp"
  75 #include "utilities/elfFile.hpp"
  76 #include "utilities/growableArray.hpp"
  77 #include "utilities/macros.hpp"
  78 #include "utilities/vmError.hpp"
  79 
  80 // put OS-includes here
  81 # include <sys/types.h>
  82 # include <sys/mman.h>
  83 # include <sys/stat.h>
  84 # include <sys/select.h>
  85 # include <pthread.h>
  86 # include <signal.h>
  87 # include <endian.h>
  88 # include <errno.h>
  89 # include <dlfcn.h>
  90 # include <stdio.h>
  91 # include <unistd.h>
  92 # include <sys/resource.h>
  93 # include <pthread.h>
  94 # include <sys/stat.h>
  95 # include <sys/time.h>
  96 # include <sys/times.h>
  97 # include <sys/utsname.h>
  98 # include <sys/socket.h>
  99 # include <sys/wait.h>
 100 # include <pwd.h>
 101 # include <poll.h>
 102 # include <fcntl.h>
 103 # include <string.h>
 104 # include <syscall.h>
 105 # include <sys/sysinfo.h>
 106 # include <gnu/libc-version.h>
 107 # include <sys/ipc.h>
 108 # include <sys/shm.h>
 109 # include <link.h>
 110 # include <stdint.h>
 111 # include <inttypes.h>
 112 # include <sys/ioctl.h>
 113 
 114 #ifndef _GNU_SOURCE
 115   #define _GNU_SOURCE
 116   #include <sched.h>
 117   #undef _GNU_SOURCE
 118 #else
 119   #include <sched.h>
 120 #endif
 121 
 122 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 123 // getrusage() is prepared to handle the associated failure.
 124 #ifndef RUSAGE_THREAD
 125   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 126 #endif
 127 
 128 #define MAX_PATH    (2 * K)
 129 
 130 #define MAX_SECS 100000000
 131 
 132 // for timer info max values which include all bits
 133 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 134 
 135 enum CoredumpFilterBit {
 136   FILE_BACKED_PVT_BIT = 1 << 2,
 137   FILE_BACKED_SHARED_BIT = 1 << 3,
 138   LARGEPAGES_BIT = 1 << 6,
 139   DAX_SHARED_BIT = 1 << 8
 140 };
 141 
 142 ////////////////////////////////////////////////////////////////////////////////
 143 // global variables
 144 julong os::Linux::_physical_memory = 0;
 145 
 146 address   os::Linux::_initial_thread_stack_bottom = NULL;
 147 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 148 
 149 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 150 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 151 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   assert(t->stack_base() != NULL, "stack_base was not initialized");
 722 
 723   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 724     sigset_t mask_all, old_sigset;
 725     sigfillset(&mask_all);
 726     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 727     _expand_stack_to(addr);
 728     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 729     return true;
 730   }
 731   return false;
 732 }
 733 
 734 //////////////////////////////////////////////////////////////////////////////
 735 // create new thread
 736 
 737 // Thread start routine for all newly created threads
 738 static void *thread_native_entry(Thread *thread) {
 739 
 740   thread->record_stack_base_and_size();
 741 
 742   // Try to randomize the cache line index of hot stack frames.
 743   // This helps when threads of the same stack traces evict each other's
 744   // cache lines. The threads can be either from the same JVM instance, or
 745   // from different JVM instances. The benefit is especially true for
 746   // processors with hyperthreading technology.
 747   static int counter = 0;
 748   int pid = os::current_process_id();
 749   alloca(((pid ^ counter++) & 7) * 128);
 750 
 751   thread->initialize_thread_current();
 752 
 753   OSThread* osthread = thread->osthread();
 754   Monitor* sync = osthread->startThread_lock();
 755 
 756   osthread->set_thread_id(os::current_thread_id());
 757 
 758   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 759     os::current_thread_id(), (uintx) pthread_self());
 760 
 761   if (UseNUMA) {
 762     int lgrp_id = os::numa_get_group_id();
 763     if (lgrp_id != -1) {
 764       thread->set_lgrp_id(lgrp_id);
 765     }
 766   }
 767   // initialize signal mask for this thread
 768   os::Linux::hotspot_sigmask(thread);
 769 
 770   // initialize floating point control register
 771   os::Linux::init_thread_fpu_state();
 772 
 773   // handshaking with parent thread
 774   {
 775     MutexLocker ml(sync, Mutex::_no_safepoint_check_flag);
 776 
 777     // notify parent thread
 778     osthread->set_state(INITIALIZED);
 779     sync->notify_all();
 780 
 781     // wait until os::start_thread()
 782     while (osthread->get_state() == INITIALIZED) {
 783       sync->wait_without_safepoint_check();
 784     }
 785   }
 786 
 787   assert(osthread->pthread_id() != 0, "pthread_id was not set as expected");
 788 
 789   // call one more level start routine
 790   thread->call_run();
 791 
 792   // Note: at this point the thread object may already have deleted itself.
 793   // Prevent dereferencing it from here on out.
 794   thread = NULL;
 795 
 796   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 797     os::current_thread_id(), (uintx) pthread_self());
 798 
 799   return 0;
 800 }
 801 
 802 // On Linux, glibc places static TLS blocks (for __thread variables) on
 803 // the thread stack. This decreases the stack size actually available
 804 // to threads.
 805 //
 806 // For large static TLS sizes, this may cause threads to malfunction due
 807 // to insufficient stack space. This is a well-known issue in glibc:
 808 // http://sourceware.org/bugzilla/show_bug.cgi?id=11787.
 809 //
 810 // As a workaround, we call a private but assumed-stable glibc function,
 811 // __pthread_get_minstack() to obtain the minstack size and derive the
 812 // static TLS size from it. We then increase the user requested stack
 813 // size by this TLS size.
 814 //
 815 // Due to compatibility concerns, this size adjustment is opt-in and
 816 // controlled via AdjustStackSizeForTLS.
 817 typedef size_t (*GetMinStack)(const pthread_attr_t *attr);
 818 
 819 GetMinStack _get_minstack_func = NULL;
 820 
 821 static void get_minstack_init() {
 822   _get_minstack_func =
 823         (GetMinStack)dlsym(RTLD_DEFAULT, "__pthread_get_minstack");
 824   log_info(os, thread)("Lookup of __pthread_get_minstack %s",
 825                        _get_minstack_func == NULL ? "failed" : "succeeded");
 826 }
 827 
 828 // Returns the size of the static TLS area glibc puts on thread stacks.
 829 // The value is cached on first use, which occurs when the first thread
 830 // is created during VM initialization.
 831 static size_t get_static_tls_area_size(const pthread_attr_t *attr) {
 832   size_t tls_size = 0;
 833   if (_get_minstack_func != NULL) {
 834     // Obtain the pthread minstack size by calling __pthread_get_minstack.
 835     size_t minstack_size = _get_minstack_func(attr);
 836 
 837     // Remove non-TLS area size included in minstack size returned
 838     // by __pthread_get_minstack() to get the static TLS size.
 839     // In glibc before 2.27, minstack size includes guard_size.
 840     // In glibc 2.27 and later, guard_size is automatically added
 841     // to the stack size by pthread_create and is no longer included
 842     // in minstack size. In both cases, the guard_size is taken into
 843     // account, so there is no need to adjust the result for that.
 844     //
 845     // Although __pthread_get_minstack() is a private glibc function,
 846     // it is expected to have a stable behavior across future glibc
 847     // versions while glibc still allocates the static TLS blocks off
 848     // the stack. Following is glibc 2.28 __pthread_get_minstack():
 849     //
 850     // size_t
 851     // __pthread_get_minstack (const pthread_attr_t *attr)
 852     // {
 853     //   return GLRO(dl_pagesize) + __static_tls_size + PTHREAD_STACK_MIN;
 854     // }
 855     //
 856     //
 857     // The following 'minstack_size > os::vm_page_size() + PTHREAD_STACK_MIN'
 858     // if check is done for precaution.
 859     if (minstack_size > (size_t)os::vm_page_size() + PTHREAD_STACK_MIN) {
 860       tls_size = minstack_size - os::vm_page_size() - PTHREAD_STACK_MIN;
 861     }
 862   }
 863 
 864   log_info(os, thread)("Stack size adjustment for TLS is " SIZE_FORMAT,
 865                        tls_size);
 866   return tls_size;
 867 }
 868 
 869 bool os::create_thread(Thread* thread, ThreadType thr_type,
 870                        size_t req_stack_size) {
 871   assert(thread->osthread() == NULL, "caller responsible");
 872 
 873   // Allocate the OSThread object
 874   OSThread* osthread = new OSThread(NULL, NULL);
 875   if (osthread == NULL) {
 876     return false;
 877   }
 878 
 879   // set the correct thread state
 880   osthread->set_thread_type(thr_type);
 881 
 882   // Initial state is ALLOCATED but not INITIALIZED
 883   osthread->set_state(ALLOCATED);
 884 
 885   thread->set_osthread(osthread);
 886 
 887   // init thread attributes
 888   pthread_attr_t attr;
 889   pthread_attr_init(&attr);
 890   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 891 
 892   // Calculate stack size if it's not specified by caller.
 893   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 894   // In glibc versions prior to 2.7 the guard size mechanism
 895   // is not implemented properly. The posix standard requires adding
 896   // the size of the guard pages to the stack size, instead Linux
 897   // takes the space out of 'stacksize'. Thus we adapt the requested
 898   // stack_size by the size of the guard pages to mimick proper
 899   // behaviour. However, be careful not to end up with a size
 900   // of zero due to overflow. Don't add the guard page in that case.
 901   size_t guard_size = os::Linux::default_guard_size(thr_type);
 902   // Configure glibc guard page. Must happen before calling
 903   // get_static_tls_area_size(), which uses the guard_size.
 904   pthread_attr_setguardsize(&attr, guard_size);
 905 
 906   size_t stack_adjust_size = 0;
 907   if (AdjustStackSizeForTLS) {
 908     // Adjust the stack_size for on-stack TLS - see get_static_tls_area_size().
 909     stack_adjust_size += get_static_tls_area_size(&attr);
 910   } else {
 911     stack_adjust_size += guard_size;
 912   }
 913 
 914   stack_adjust_size = align_up(stack_adjust_size, os::vm_page_size());
 915   if (stack_size <= SIZE_MAX - stack_adjust_size) {
 916     stack_size += stack_adjust_size;
 917   }
 918   assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
 919 
 920   int status = pthread_attr_setstacksize(&attr, stack_size);
 921   assert_status(status == 0, status, "pthread_attr_setstacksize");
 922 
 923   ThreadState state;
 924 
 925   {
 926     pthread_t tid;
 927     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 928 
 929     char buf[64];
 930     if (ret == 0) {
 931       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 932         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 933     } else {
 934       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 935         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 936       // Log some OS information which might explain why creating the thread failed.
 937       log_info(os, thread)("Number of threads approx. running in the VM: %d", Threads::number_of_threads());
 938       LogStream st(Log(os, thread)::info());
 939       os::Posix::print_rlimit_info(&st);
 940       os::print_memory_info(&st);
 941       os::Linux::print_proc_sys_info(&st);
 942       os::Linux::print_container_info(&st);
 943     }
 944 
 945     pthread_attr_destroy(&attr);
 946 
 947     if (ret != 0) {
 948       // Need to clean up stuff we've allocated so far
 949       thread->set_osthread(NULL);
 950       delete osthread;
 951       return false;
 952     }
 953 
 954     // Store pthread info into the OSThread
 955     osthread->set_pthread_id(tid);
 956 
 957     // Wait until child thread is either initialized or aborted
 958     {
 959       Monitor* sync_with_child = osthread->startThread_lock();
 960       MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 961       while ((state = osthread->get_state()) == ALLOCATED) {
 962         sync_with_child->wait_without_safepoint_check();
 963       }
 964     }
 965   }
 966 
 967   // Aborted due to thread limit being reached
 968   if (state == ZOMBIE) {
 969     thread->set_osthread(NULL);
 970     delete osthread;
 971     return false;
 972   }
 973 
 974   // The thread is returned suspended (in state INITIALIZED),
 975   // and is started higher up in the call chain
 976   assert(state == INITIALIZED, "race condition");
 977   return true;
 978 }
 979 
 980 /////////////////////////////////////////////////////////////////////////////
 981 // attach existing thread
 982 
 983 // bootstrap the main thread
 984 bool os::create_main_thread(JavaThread* thread) {
 985   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 986   return create_attached_thread(thread);
 987 }
 988 
 989 bool os::create_attached_thread(JavaThread* thread) {
 990 #ifdef ASSERT
 991   thread->verify_not_published();
 992 #endif
 993 
 994   // Allocate the OSThread object
 995   OSThread* osthread = new OSThread(NULL, NULL);
 996 
 997   if (osthread == NULL) {
 998     return false;
 999   }
1000 
1001   // Store pthread info into the OSThread
1002   osthread->set_thread_id(os::Linux::gettid());
1003   osthread->set_pthread_id(::pthread_self());
1004 
1005   // initialize floating point control register
1006   os::Linux::init_thread_fpu_state();
1007 
1008   // Initial thread state is RUNNABLE
1009   osthread->set_state(RUNNABLE);
1010 
1011   thread->set_osthread(osthread);
1012 
1013   if (UseNUMA) {
1014     int lgrp_id = os::numa_get_group_id();
1015     if (lgrp_id != -1) {
1016       thread->set_lgrp_id(lgrp_id);
1017     }
1018   }
1019 
1020   if (os::is_primordial_thread()) {
1021     // If current thread is primordial thread, its stack is mapped on demand,
1022     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1023     // the entire stack region to avoid SEGV in stack banging.
1024     // It is also useful to get around the heap-stack-gap problem on SuSE
1025     // kernel (see 4821821 for details). We first expand stack to the top
1026     // of yellow zone, then enable stack yellow zone (order is significant,
1027     // enabling yellow zone first will crash JVM on SuSE Linux), so there
1028     // is no gap between the last two virtual memory regions.
1029 
1030     JavaThread *jt = (JavaThread *)thread;
1031     address addr = jt->stack_reserved_zone_base();
1032     assert(addr != NULL, "initialization problem?");
1033     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1034 
1035     osthread->set_expanding_stack();
1036     os::Linux::manually_expand_stack(jt, addr);
1037     osthread->clear_expanding_stack();
1038   }
1039 
1040   // initialize signal mask for this thread
1041   // and save the caller's signal mask
1042   os::Linux::hotspot_sigmask(thread);
1043 
1044   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
1045     os::current_thread_id(), (uintx) pthread_self());
1046 
1047   return true;
1048 }
1049 
1050 void os::pd_start_thread(Thread* thread) {
1051   OSThread * osthread = thread->osthread();
1052   assert(osthread->get_state() != INITIALIZED, "just checking");
1053   Monitor* sync_with_child = osthread->startThread_lock();
1054   MutexLocker ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1055   sync_with_child->notify();
1056 }
1057 
1058 // Free Linux resources related to the OSThread
1059 void os::free_thread(OSThread* osthread) {
1060   assert(osthread != NULL, "osthread not set");
1061 
1062   // We are told to free resources of the argument thread,
1063   // but we can only really operate on the current thread.
1064   assert(Thread::current()->osthread() == osthread,
1065          "os::free_thread but not current thread");
1066 
1067 #ifdef ASSERT
1068   sigset_t current;
1069   sigemptyset(&current);
1070   pthread_sigmask(SIG_SETMASK, NULL, &current);
1071   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
1072 #endif
1073 
1074   // Restore caller's signal mask
1075   sigset_t sigmask = osthread->caller_sigmask();
1076   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1077 
1078   delete osthread;
1079 }
1080 
1081 //////////////////////////////////////////////////////////////////////////////
1082 // primordial thread
1083 
1084 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1085 bool os::is_primordial_thread(void) {
1086   if (suppress_primordial_thread_resolution) {
1087     return false;
1088   }
1089   char dummy;
1090   // If called before init complete, thread stack bottom will be null.
1091   // Can be called if fatal error occurs before initialization.
1092   if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1093   assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1094          os::Linux::initial_thread_stack_size()   != 0,
1095          "os::init did not locate primordial thread's stack region");
1096   if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1097       (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1098                         os::Linux::initial_thread_stack_size()) {
1099     return true;
1100   } else {
1101     return false;
1102   }
1103 }
1104 
1105 // Find the virtual memory area that contains addr
1106 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1107   FILE *fp = fopen("/proc/self/maps", "r");
1108   if (fp) {
1109     address low, high;
1110     while (!feof(fp)) {
1111       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1112         if (low <= addr && addr < high) {
1113           if (vma_low)  *vma_low  = low;
1114           if (vma_high) *vma_high = high;
1115           fclose(fp);
1116           return true;
1117         }
1118       }
1119       for (;;) {
1120         int ch = fgetc(fp);
1121         if (ch == EOF || ch == (int)'\n') break;
1122       }
1123     }
1124     fclose(fp);
1125   }
1126   return false;
1127 }
1128 
1129 // Locate primordial thread stack. This special handling of primordial thread stack
1130 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1131 // bogus value for the primordial process thread. While the launcher has created
1132 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1133 // JNI invocation API from a primordial thread.
1134 void os::Linux::capture_initial_stack(size_t max_size) {
1135 
1136   // max_size is either 0 (which means accept OS default for thread stacks) or
1137   // a user-specified value known to be at least the minimum needed. If we
1138   // are actually on the primordial thread we can make it appear that we have a
1139   // smaller max_size stack by inserting the guard pages at that location. But we
1140   // cannot do anything to emulate a larger stack than what has been provided by
1141   // the OS or threading library. In fact if we try to use a stack greater than
1142   // what is set by rlimit then we will crash the hosting process.
1143 
1144   // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1145   // If this is "unlimited" then it will be a huge value.
1146   struct rlimit rlim;
1147   getrlimit(RLIMIT_STACK, &rlim);
1148   size_t stack_size = rlim.rlim_cur;
1149 
1150   // 6308388: a bug in ld.so will relocate its own .data section to the
1151   //   lower end of primordial stack; reduce ulimit -s value a little bit
1152   //   so we won't install guard page on ld.so's data section.
1153   //   But ensure we don't underflow the stack size - allow 1 page spare
1154   if (stack_size >= (size_t)(3 * page_size())) {
1155     stack_size -= 2 * page_size();
1156   }
1157 
1158   // Try to figure out where the stack base (top) is. This is harder.
1159   //
1160   // When an application is started, glibc saves the initial stack pointer in
1161   // a global variable "__libc_stack_end", which is then used by system
1162   // libraries. __libc_stack_end should be pretty close to stack top. The
1163   // variable is available since the very early days. However, because it is
1164   // a private interface, it could disappear in the future.
1165   //
1166   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1167   // to __libc_stack_end, it is very close to stack top, but isn't the real
1168   // stack top. Note that /proc may not exist if VM is running as a chroot
1169   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1170   // /proc/<pid>/stat could change in the future (though unlikely).
1171   //
1172   // We try __libc_stack_end first. If that doesn't work, look for
1173   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1174   // as a hint, which should work well in most cases.
1175 
1176   uintptr_t stack_start;
1177 
1178   // try __libc_stack_end first
1179   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1180   if (p && *p) {
1181     stack_start = *p;
1182   } else {
1183     // see if we can get the start_stack field from /proc/self/stat
1184     FILE *fp;
1185     int pid;
1186     char state;
1187     int ppid;
1188     int pgrp;
1189     int session;
1190     int nr;
1191     int tpgrp;
1192     unsigned long flags;
1193     unsigned long minflt;
1194     unsigned long cminflt;
1195     unsigned long majflt;
1196     unsigned long cmajflt;
1197     unsigned long utime;
1198     unsigned long stime;
1199     long cutime;
1200     long cstime;
1201     long prio;
1202     long nice;
1203     long junk;
1204     long it_real;
1205     uintptr_t start;
1206     uintptr_t vsize;
1207     intptr_t rss;
1208     uintptr_t rsslim;
1209     uintptr_t scodes;
1210     uintptr_t ecode;
1211     int i;
1212 
1213     // Figure what the primordial thread stack base is. Code is inspired
1214     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1215     // followed by command name surrounded by parentheses, state, etc.
1216     char stat[2048];
1217     int statlen;
1218 
1219     fp = fopen("/proc/self/stat", "r");
1220     if (fp) {
1221       statlen = fread(stat, 1, 2047, fp);
1222       stat[statlen] = '\0';
1223       fclose(fp);
1224 
1225       // Skip pid and the command string. Note that we could be dealing with
1226       // weird command names, e.g. user could decide to rename java launcher
1227       // to "java 1.4.2 :)", then the stat file would look like
1228       //                1234 (java 1.4.2 :)) R ... ...
1229       // We don't really need to know the command string, just find the last
1230       // occurrence of ")" and then start parsing from there. See bug 4726580.
1231       char * s = strrchr(stat, ')');
1232 
1233       i = 0;
1234       if (s) {
1235         // Skip blank chars
1236         do { s++; } while (s && isspace(*s));
1237 
1238 #define _UFM UINTX_FORMAT
1239 #define _DFM INTX_FORMAT
1240 
1241         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1242         //              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
1243         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,
1244                    &state,          // 3  %c
1245                    &ppid,           // 4  %d
1246                    &pgrp,           // 5  %d
1247                    &session,        // 6  %d
1248                    &nr,             // 7  %d
1249                    &tpgrp,          // 8  %d
1250                    &flags,          // 9  %lu
1251                    &minflt,         // 10 %lu
1252                    &cminflt,        // 11 %lu
1253                    &majflt,         // 12 %lu
1254                    &cmajflt,        // 13 %lu
1255                    &utime,          // 14 %lu
1256                    &stime,          // 15 %lu
1257                    &cutime,         // 16 %ld
1258                    &cstime,         // 17 %ld
1259                    &prio,           // 18 %ld
1260                    &nice,           // 19 %ld
1261                    &junk,           // 20 %ld
1262                    &it_real,        // 21 %ld
1263                    &start,          // 22 UINTX_FORMAT
1264                    &vsize,          // 23 UINTX_FORMAT
1265                    &rss,            // 24 INTX_FORMAT
1266                    &rsslim,         // 25 UINTX_FORMAT
1267                    &scodes,         // 26 UINTX_FORMAT
1268                    &ecode,          // 27 UINTX_FORMAT
1269                    &stack_start);   // 28 UINTX_FORMAT
1270       }
1271 
1272 #undef _UFM
1273 #undef _DFM
1274 
1275       if (i != 28 - 2) {
1276         assert(false, "Bad conversion from /proc/self/stat");
1277         // product mode - assume we are the primordial thread, good luck in the
1278         // embedded case.
1279         warning("Can't detect primordial thread stack location - bad conversion");
1280         stack_start = (uintptr_t) &rlim;
1281       }
1282     } else {
1283       // For some reason we can't open /proc/self/stat (for example, running on
1284       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1285       // most cases, so don't abort:
1286       warning("Can't detect primordial thread stack location - no /proc/self/stat");
1287       stack_start = (uintptr_t) &rlim;
1288     }
1289   }
1290 
1291   // Now we have a pointer (stack_start) very close to the stack top, the
1292   // next thing to do is to figure out the exact location of stack top. We
1293   // can find out the virtual memory area that contains stack_start by
1294   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1295   // and its upper limit is the real stack top. (again, this would fail if
1296   // running inside chroot, because /proc may not exist.)
1297 
1298   uintptr_t stack_top;
1299   address low, high;
1300   if (find_vma((address)stack_start, &low, &high)) {
1301     // success, "high" is the true stack top. (ignore "low", because initial
1302     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1303     stack_top = (uintptr_t)high;
1304   } else {
1305     // failed, likely because /proc/self/maps does not exist
1306     warning("Can't detect primordial thread stack location - find_vma failed");
1307     // best effort: stack_start is normally within a few pages below the real
1308     // stack top, use it as stack top, and reduce stack size so we won't put
1309     // guard page outside stack.
1310     stack_top = stack_start;
1311     stack_size -= 16 * page_size();
1312   }
1313 
1314   // stack_top could be partially down the page so align it
1315   stack_top = align_up(stack_top, page_size());
1316 
1317   // Allowed stack value is minimum of max_size and what we derived from rlimit
1318   if (max_size > 0) {
1319     _initial_thread_stack_size = MIN2(max_size, stack_size);
1320   } else {
1321     // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1322     // clamp it at 8MB as we do on Solaris
1323     _initial_thread_stack_size = MIN2(stack_size, 8*M);
1324   }
1325   _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1326   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1327 
1328   assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1329 
1330   if (log_is_enabled(Info, os, thread)) {
1331     // See if we seem to be on primordial process thread
1332     bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1333                       uintptr_t(&rlim) < stack_top;
1334 
1335     log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1336                          SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1337                          primordial ? "primordial" : "user", max_size / K,  _initial_thread_stack_size / K,
1338                          stack_top, intptr_t(_initial_thread_stack_bottom));
1339   }
1340 }
1341 
1342 ////////////////////////////////////////////////////////////////////////////////
1343 // time support
1344 
1345 #ifndef SUPPORTS_CLOCK_MONOTONIC
1346 #error "Build platform doesn't support clock_gettime and related functionality"
1347 #endif
1348 
1349 // Time since start-up in seconds to a fine granularity.
1350 // Used by VMSelfDestructTimer and the MemProfiler.
1351 double os::elapsedTime() {
1352 
1353   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1354 }
1355 
1356 jlong os::elapsed_counter() {
1357   return javaTimeNanos() - initial_time_count;
1358 }
1359 
1360 jlong os::elapsed_frequency() {
1361   return NANOSECS_PER_SEC; // nanosecond resolution
1362 }
1363 
1364 bool os::supports_vtime() { return true; }
1365 
1366 double os::elapsedVTime() {
1367   struct rusage usage;
1368   int retval = getrusage(RUSAGE_THREAD, &usage);
1369   if (retval == 0) {
1370     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);
1371   } else {
1372     // better than nothing, but not much
1373     return elapsedTime();
1374   }
1375 }
1376 
1377 jlong os::javaTimeMillis() {
1378   timeval time;
1379   int status = gettimeofday(&time, NULL);
1380   assert(status != -1, "linux error");
1381   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1382 }
1383 
1384 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1385   timeval time;
1386   int status = gettimeofday(&time, NULL);
1387   assert(status != -1, "linux error");
1388   seconds = jlong(time.tv_sec);
1389   nanos = jlong(time.tv_usec) * 1000;
1390 }
1391 
1392 void os::Linux::fast_thread_clock_init() {
1393   if (!UseLinuxPosixThreadCPUClocks) {
1394     return;
1395   }
1396   clockid_t clockid;
1397   struct timespec tp;
1398   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1399       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1400 
1401   // Switch to using fast clocks for thread cpu time if
1402   // the clock_getres() returns 0 error code.
1403   // Note, that some kernels may support the current thread
1404   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1405   // returned by the pthread_getcpuclockid().
1406   // If the fast Posix clocks are supported then the clock_getres()
1407   // must return at least tp.tv_sec == 0 which means a resolution
1408   // better than 1 sec. This is extra check for reliability.
1409 
1410   if (pthread_getcpuclockid_func &&
1411       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1412       os::Posix::clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1413     _supports_fast_thread_cpu_time = true;
1414     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1415   }
1416 }
1417 
1418 jlong os::javaTimeNanos() {
1419   if (os::supports_monotonic_clock()) {
1420     struct timespec tp;
1421     int status = os::Posix::clock_gettime(CLOCK_MONOTONIC, &tp);
1422     assert(status == 0, "gettime error");
1423     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1424     return result;
1425   } else {
1426     timeval time;
1427     int status = gettimeofday(&time, NULL);
1428     assert(status != -1, "linux error");
1429     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1430     return 1000 * usecs;
1431   }
1432 }
1433 
1434 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1435   if (os::supports_monotonic_clock()) {
1436     info_ptr->max_value = ALL_64_BITS;
1437 
1438     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1439     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1440     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1441   } else {
1442     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1443     info_ptr->max_value = ALL_64_BITS;
1444 
1445     // gettimeofday is a real time clock so it skips
1446     info_ptr->may_skip_backward = true;
1447     info_ptr->may_skip_forward = true;
1448   }
1449 
1450   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1451 }
1452 
1453 // Return the real, user, and system times in seconds from an
1454 // arbitrary fixed point in the past.
1455 bool os::getTimesSecs(double* process_real_time,
1456                       double* process_user_time,
1457                       double* process_system_time) {
1458   struct tms ticks;
1459   clock_t real_ticks = times(&ticks);
1460 
1461   if (real_ticks == (clock_t) (-1)) {
1462     return false;
1463   } else {
1464     double ticks_per_second = (double) clock_tics_per_sec;
1465     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1466     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1467     *process_real_time = ((double) real_ticks) / ticks_per_second;
1468 
1469     return true;
1470   }
1471 }
1472 
1473 
1474 char * os::local_time_string(char *buf, size_t buflen) {
1475   struct tm t;
1476   time_t long_time;
1477   time(&long_time);
1478   localtime_r(&long_time, &t);
1479   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1480                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1481                t.tm_hour, t.tm_min, t.tm_sec);
1482   return buf;
1483 }
1484 
1485 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1486   return localtime_r(clock, res);
1487 }
1488 
1489 ////////////////////////////////////////////////////////////////////////////////
1490 // runtime exit support
1491 
1492 // Note: os::shutdown() might be called very early during initialization, or
1493 // called from signal handler. Before adding something to os::shutdown(), make
1494 // sure it is async-safe and can handle partially initialized VM.
1495 void os::shutdown() {
1496 
1497   // allow PerfMemory to attempt cleanup of any persistent resources
1498   perfMemory_exit();
1499 
1500   // needs to remove object in file system
1501   AttachListener::abort();
1502 
1503   // flush buffered output, finish log files
1504   ostream_abort();
1505 
1506   // Check for abort hook
1507   abort_hook_t abort_hook = Arguments::abort_hook();
1508   if (abort_hook != NULL) {
1509     abort_hook();
1510   }
1511 
1512 }
1513 
1514 // Note: os::abort() might be called very early during initialization, or
1515 // called from signal handler. Before adding something to os::abort(), make
1516 // sure it is async-safe and can handle partially initialized VM.
1517 void os::abort(bool dump_core, void* siginfo, const void* context) {
1518   os::shutdown();
1519   if (dump_core) {
1520     if (DumpPrivateMappingsInCore) {
1521       ClassLoader::close_jrt_image();
1522     }
1523 #ifndef PRODUCT
1524     fdStream out(defaultStream::output_fd());
1525     out.print_raw("Current thread is ");
1526     char buf[16];
1527     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1528     out.print_raw_cr(buf);
1529     out.print_raw_cr("Dumping core ...");
1530 #endif
1531     ::abort(); // dump core
1532   }
1533 
1534   ::exit(1);
1535 }
1536 
1537 // Die immediately, no exit hook, no abort hook, no cleanup.
1538 // Dump a core file, if possible, for debugging.
1539 void os::die() {
1540   if (TestUnresponsiveErrorHandler && !CreateCoredumpOnCrash) {
1541     // For TimeoutInErrorHandlingTest.java, we just kill the VM
1542     // and don't take the time to generate a core file.
1543     os::signal_raise(SIGKILL);
1544   } else {
1545     ::abort();
1546   }
1547 }
1548 
1549 // thread_id is kernel thread id (similar to Solaris LWP id)
1550 intx os::current_thread_id() { return os::Linux::gettid(); }
1551 int os::current_process_id() {
1552   return ::getpid();
1553 }
1554 
1555 // DLL functions
1556 
1557 const char* os::dll_file_extension() { return ".so"; }
1558 
1559 // This must be hard coded because it's the system's temporary
1560 // directory not the java application's temp directory, ala java.io.tmpdir.
1561 const char* os::get_temp_directory() { return "/tmp"; }
1562 
1563 static bool file_exists(const char* filename) {
1564   struct stat statbuf;
1565   if (filename == NULL || strlen(filename) == 0) {
1566     return false;
1567   }
1568   return os::stat(filename, &statbuf) == 0;
1569 }
1570 
1571 // check if addr is inside libjvm.so
1572 bool os::address_is_in_vm(address addr) {
1573   static address libjvm_base_addr;
1574   Dl_info dlinfo;
1575 
1576   if (libjvm_base_addr == NULL) {
1577     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1578       libjvm_base_addr = (address)dlinfo.dli_fbase;
1579     }
1580     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1581   }
1582 
1583   if (dladdr((void *)addr, &dlinfo) != 0) {
1584     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1585   }
1586 
1587   return false;
1588 }
1589 
1590 bool os::dll_address_to_function_name(address addr, char *buf,
1591                                       int buflen, int *offset,
1592                                       bool demangle) {
1593   // buf is not optional, but offset is optional
1594   assert(buf != NULL, "sanity check");
1595 
1596   Dl_info dlinfo;
1597 
1598   if (dladdr((void*)addr, &dlinfo) != 0) {
1599     // see if we have a matching symbol
1600     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1601       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1602         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1603       }
1604       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1605       return true;
1606     }
1607     // no matching symbol so try for just file info
1608     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1609       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1610                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1611         return true;
1612       }
1613     }
1614   }
1615 
1616   buf[0] = '\0';
1617   if (offset != NULL) *offset = -1;
1618   return false;
1619 }
1620 
1621 struct _address_to_library_name {
1622   address addr;          // input : memory address
1623   size_t  buflen;        //         size of fname
1624   char*   fname;         // output: library name
1625   address base;          //         library base addr
1626 };
1627 
1628 static int address_to_library_name_callback(struct dl_phdr_info *info,
1629                                             size_t size, void *data) {
1630   int i;
1631   bool found = false;
1632   address libbase = NULL;
1633   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1634 
1635   // iterate through all loadable segments
1636   for (i = 0; i < info->dlpi_phnum; i++) {
1637     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1638     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1639       // base address of a library is the lowest address of its loaded
1640       // segments.
1641       if (libbase == NULL || libbase > segbase) {
1642         libbase = segbase;
1643       }
1644       // see if 'addr' is within current segment
1645       if (segbase <= d->addr &&
1646           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1647         found = true;
1648       }
1649     }
1650   }
1651 
1652   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1653   // so dll_address_to_library_name() can fall through to use dladdr() which
1654   // can figure out executable name from argv[0].
1655   if (found && info->dlpi_name && info->dlpi_name[0]) {
1656     d->base = libbase;
1657     if (d->fname) {
1658       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1659     }
1660     return 1;
1661   }
1662   return 0;
1663 }
1664 
1665 bool os::dll_address_to_library_name(address addr, char* buf,
1666                                      int buflen, int* offset) {
1667   // buf is not optional, but offset is optional
1668   assert(buf != NULL, "sanity check");
1669 
1670   Dl_info dlinfo;
1671   struct _address_to_library_name data;
1672 
1673   // There is a bug in old glibc dladdr() implementation that it could resolve
1674   // to wrong library name if the .so file has a base address != NULL. Here
1675   // we iterate through the program headers of all loaded libraries to find
1676   // out which library 'addr' really belongs to. This workaround can be
1677   // removed once the minimum requirement for glibc is moved to 2.3.x.
1678   data.addr = addr;
1679   data.fname = buf;
1680   data.buflen = buflen;
1681   data.base = NULL;
1682   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1683 
1684   if (rslt) {
1685     // buf already contains library name
1686     if (offset) *offset = addr - data.base;
1687     return true;
1688   }
1689   if (dladdr((void*)addr, &dlinfo) != 0) {
1690     if (dlinfo.dli_fname != NULL) {
1691       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1692     }
1693     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1694       *offset = addr - (address)dlinfo.dli_fbase;
1695     }
1696     return true;
1697   }
1698 
1699   buf[0] = '\0';
1700   if (offset) *offset = -1;
1701   return false;
1702 }
1703 
1704 // Loads .dll/.so and
1705 // in case of error it checks if .dll/.so was built for the
1706 // same architecture as Hotspot is running on
1707 
1708 
1709 // Remember the stack's state. The Linux dynamic linker will change
1710 // the stack to 'executable' at most once, so we must safepoint only once.
1711 bool os::Linux::_stack_is_executable = false;
1712 
1713 // VM operation that loads a library.  This is necessary if stack protection
1714 // of the Java stacks can be lost during loading the library.  If we
1715 // do not stop the Java threads, they can stack overflow before the stacks
1716 // are protected again.
1717 class VM_LinuxDllLoad: public VM_Operation {
1718  private:
1719   const char *_filename;
1720   char *_ebuf;
1721   int _ebuflen;
1722   void *_lib;
1723  public:
1724   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1725     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1726   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1727   void doit() {
1728     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1729     os::Linux::_stack_is_executable = true;
1730   }
1731   void* loaded_library() { return _lib; }
1732 };
1733 
1734 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1735   void * result = NULL;
1736   bool load_attempted = false;
1737 
1738   log_info(os)("attempting shared library load of %s", filename);
1739 
1740   // Check whether the library to load might change execution rights
1741   // of the stack. If they are changed, the protection of the stack
1742   // guard pages will be lost. We need a safepoint to fix this.
1743   //
1744   // See Linux man page execstack(8) for more info.
1745   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1746     if (!ElfFile::specifies_noexecstack(filename)) {
1747       if (!is_init_completed()) {
1748         os::Linux::_stack_is_executable = true;
1749         // This is OK - No Java threads have been created yet, and hence no
1750         // stack guard pages to fix.
1751         //
1752         // Dynamic loader will make all stacks executable after
1753         // this function returns, and will not do that again.
1754         assert(Threads::number_of_threads() == 0, "no Java threads should exist yet.");
1755       } else {
1756         warning("You have loaded library %s which might have disabled stack guard. "
1757                 "The VM will try to fix the stack guard now.\n"
1758                 "It's highly recommended that you fix the library with "
1759                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1760                 filename);
1761 
1762         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1763         JavaThread *jt = JavaThread::current();
1764         if (jt->thread_state() != _thread_in_native) {
1765           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1766           // that requires ExecStack. Cannot enter safe point. Let's give up.
1767           warning("Unable to fix stack guard. Giving up.");
1768         } else {
1769           if (!LoadExecStackDllInVMThread) {
1770             // This is for the case where the DLL has an static
1771             // constructor function that executes JNI code. We cannot
1772             // load such DLLs in the VMThread.
1773             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1774           }
1775 
1776           ThreadInVMfromNative tiv(jt);
1777           debug_only(VMNativeEntryWrapper vew;)
1778 
1779           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1780           VMThread::execute(&op);
1781           if (LoadExecStackDllInVMThread) {
1782             result = op.loaded_library();
1783           }
1784           load_attempted = true;
1785         }
1786       }
1787     }
1788   }
1789 
1790   if (!load_attempted) {
1791     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1792   }
1793 
1794   if (result != NULL) {
1795     // Successful loading
1796     return result;
1797   }
1798 
1799   Elf32_Ehdr elf_head;
1800   int diag_msg_max_length=ebuflen-strlen(ebuf);
1801   char* diag_msg_buf=ebuf+strlen(ebuf);
1802 
1803   if (diag_msg_max_length==0) {
1804     // No more space in ebuf for additional diagnostics message
1805     return NULL;
1806   }
1807 
1808 
1809   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1810 
1811   if (file_descriptor < 0) {
1812     // Can't open library, report dlerror() message
1813     return NULL;
1814   }
1815 
1816   bool failed_to_read_elf_head=
1817     (sizeof(elf_head)!=
1818      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1819 
1820   ::close(file_descriptor);
1821   if (failed_to_read_elf_head) {
1822     // file i/o error - report dlerror() msg
1823     return NULL;
1824   }
1825 
1826   if (elf_head.e_ident[EI_DATA] != LITTLE_ENDIAN_ONLY(ELFDATA2LSB) BIG_ENDIAN_ONLY(ELFDATA2MSB)) {
1827     // handle invalid/out of range endianness values
1828     if (elf_head.e_ident[EI_DATA] == 0 || elf_head.e_ident[EI_DATA] > 2) {
1829       return NULL;
1830     }
1831 
1832 #if defined(VM_LITTLE_ENDIAN)
1833     // VM is LE, shared object BE
1834     elf_head.e_machine = be16toh(elf_head.e_machine);
1835 #else
1836     // VM is BE, shared object LE
1837     elf_head.e_machine = le16toh(elf_head.e_machine);
1838 #endif
1839   }
1840 
1841   typedef struct {
1842     Elf32_Half    code;         // Actual value as defined in elf.h
1843     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1844     unsigned char elf_class;    // 32 or 64 bit
1845     unsigned char endianness;   // MSB or LSB
1846     char*         name;         // String representation
1847   } arch_t;
1848 
1849 #ifndef EM_486
1850   #define EM_486          6               /* Intel 80486 */
1851 #endif
1852 #ifndef EM_AARCH64
1853   #define EM_AARCH64    183               /* ARM AARCH64 */
1854 #endif
1855 
1856   static const arch_t arch_array[]={
1857     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1858     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1859     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1860     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1861     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1862     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1863     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1864     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1865 #if defined(VM_LITTLE_ENDIAN)
1866     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1867     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1868 #else
1869     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1870     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1871 #endif
1872     {EM_ARM,         EM_ARM,     ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1873     // we only support 64 bit z architecture
1874     {EM_S390,        EM_S390,    ELFCLASS64, ELFDATA2MSB, (char*)"IBM System/390"},
1875     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1876     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1877     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1878     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1879     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1880     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1881   };
1882 
1883 #if  (defined IA32)
1884   static  Elf32_Half running_arch_code=EM_386;
1885 #elif   (defined AMD64) || (defined X32)
1886   static  Elf32_Half running_arch_code=EM_X86_64;
1887 #elif  (defined IA64)
1888   static  Elf32_Half running_arch_code=EM_IA_64;
1889 #elif  (defined __sparc) && (defined _LP64)
1890   static  Elf32_Half running_arch_code=EM_SPARCV9;
1891 #elif  (defined __sparc) && (!defined _LP64)
1892   static  Elf32_Half running_arch_code=EM_SPARC;
1893 #elif  (defined __powerpc64__)
1894   static  Elf32_Half running_arch_code=EM_PPC64;
1895 #elif  (defined __powerpc__)
1896   static  Elf32_Half running_arch_code=EM_PPC;
1897 #elif  (defined AARCH64)
1898   static  Elf32_Half running_arch_code=EM_AARCH64;
1899 #elif  (defined ARM)
1900   static  Elf32_Half running_arch_code=EM_ARM;
1901 #elif  (defined S390)
1902   static  Elf32_Half running_arch_code=EM_S390;
1903 #elif  (defined ALPHA)
1904   static  Elf32_Half running_arch_code=EM_ALPHA;
1905 #elif  (defined MIPSEL)
1906   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1907 #elif  (defined PARISC)
1908   static  Elf32_Half running_arch_code=EM_PARISC;
1909 #elif  (defined MIPS)
1910   static  Elf32_Half running_arch_code=EM_MIPS;
1911 #elif  (defined M68K)
1912   static  Elf32_Half running_arch_code=EM_68K;
1913 #elif  (defined SH)
1914   static  Elf32_Half running_arch_code=EM_SH;
1915 #else
1916     #error Method os::dll_load requires that one of following is defined:\
1917         AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1918 #endif
1919 
1920   // Identify compatibility class for VM's architecture and library's architecture
1921   // Obtain string descriptions for architectures
1922 
1923   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1924   int running_arch_index=-1;
1925 
1926   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1927     if (running_arch_code == arch_array[i].code) {
1928       running_arch_index    = i;
1929     }
1930     if (lib_arch.code == arch_array[i].code) {
1931       lib_arch.compat_class = arch_array[i].compat_class;
1932       lib_arch.name         = arch_array[i].name;
1933     }
1934   }
1935 
1936   assert(running_arch_index != -1,
1937          "Didn't find running architecture code (running_arch_code) in arch_array");
1938   if (running_arch_index == -1) {
1939     // Even though running architecture detection failed
1940     // we may still continue with reporting dlerror() message
1941     return NULL;
1942   }
1943 
1944   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1945     if (lib_arch.name != NULL) {
1946       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1947                  " (Possible cause: can't load %s .so on a %s platform)",
1948                  lib_arch.name, arch_array[running_arch_index].name);
1949     } else {
1950       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1951                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s platform)",
1952                  lib_arch.code, arch_array[running_arch_index].name);
1953     }
1954     return NULL;
1955   }
1956 
1957   if (lib_arch.endianness != arch_array[running_arch_index].endianness) {
1958     ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: endianness mismatch)");
1959     return NULL;
1960   }
1961 
1962   // ELF file class/capacity : 0 - invalid, 1 - 32bit, 2 - 64bit
1963   if (lib_arch.elf_class > 2 || lib_arch.elf_class < 1) {
1964     ::snprintf(diag_msg_buf, diag_msg_max_length-1, " (Possible cause: invalid ELF file class)");
1965     return NULL;
1966   }
1967 
1968   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1969     ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1970                " (Possible cause: architecture word width mismatch, can't load %d-bit .so on a %d-bit platform)",
1971                (int) lib_arch.elf_class * 32, arch_array[running_arch_index].elf_class * 32);
1972     return NULL;
1973   }
1974 
1975   return NULL;
1976 }
1977 
1978 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1979                                 int ebuflen) {
1980   void * result = ::dlopen(filename, RTLD_LAZY);
1981   if (result == NULL) {
1982     const char* error_report = ::dlerror();
1983     if (error_report == NULL) {
1984       error_report = "dlerror returned no error description";
1985     }
1986     if (ebuf != NULL && ebuflen > 0) {
1987       ::strncpy(ebuf, error_report, ebuflen-1);
1988       ebuf[ebuflen-1]='\0';
1989     }
1990     Events::log(NULL, "Loading shared library %s failed, %s", filename, error_report);
1991     log_info(os)("shared library load of %s failed, %s", filename, error_report);
1992   } else {
1993     Events::log(NULL, "Loaded shared library %s", filename);
1994     log_info(os)("shared library load of %s was successful", filename);
1995   }
1996   return result;
1997 }
1998 
1999 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2000                                        int ebuflen) {
2001   void * result = NULL;
2002   if (LoadExecStackDllInVMThread) {
2003     result = dlopen_helper(filename, ebuf, ebuflen);
2004   }
2005 
2006   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2007   // library that requires an executable stack, or which does not have this
2008   // stack attribute set, dlopen changes the stack attribute to executable. The
2009   // read protection of the guard pages gets lost.
2010   //
2011   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2012   // may have been queued at the same time.
2013 
2014   if (!_stack_is_executable) {
2015     for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
2016       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
2017           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
2018         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
2019           warning("Attempt to reguard stack yellow zone failed.");
2020         }
2021       }
2022     }
2023   }
2024 
2025   return result;
2026 }
2027 
2028 void* os::dll_lookup(void* handle, const char* name) {
2029   void* res = dlsym(handle, name);
2030   return res;
2031 }
2032 
2033 void* os::get_default_process_handle() {
2034   return (void*)::dlopen(NULL, RTLD_LAZY);
2035 }
2036 
2037 static bool _print_ascii_file(const char* filename, outputStream* st, const char* hdr = NULL) {
2038   int fd = ::open(filename, O_RDONLY);
2039   if (fd == -1) {
2040     return false;
2041   }
2042 
2043   if (hdr != NULL) {
2044     st->print_cr("%s", hdr);
2045   }
2046 
2047   char buf[33];
2048   int bytes;
2049   buf[32] = '\0';
2050   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
2051     st->print_raw(buf, bytes);
2052   }
2053 
2054   ::close(fd);
2055 
2056   return true;
2057 }
2058 
2059 void os::print_dll_info(outputStream *st) {
2060   st->print_cr("Dynamic libraries:");
2061 
2062   char fname[32];
2063   pid_t pid = os::Linux::gettid();
2064 
2065   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2066 
2067   if (!_print_ascii_file(fname, st)) {
2068     st->print("Can not get library information for pid = %d\n", pid);
2069   }
2070 }
2071 
2072 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2073   FILE *procmapsFile = NULL;
2074 
2075   // Open the procfs maps file for the current process
2076   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2077     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2078     char line[PATH_MAX + 100];
2079 
2080     // Read line by line from 'file'
2081     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2082       u8 base, top, offset, inode;
2083       char permissions[5];
2084       char device[6];
2085       char name[PATH_MAX + 1];
2086 
2087       // Parse fields from line
2088       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s",
2089              &base, &top, permissions, &offset, device, &inode, name);
2090 
2091       // Filter by device id '00:00' so that we only get file system mapped files.
2092       if (strcmp(device, "00:00") != 0) {
2093 
2094         // Call callback with the fields of interest
2095         if(callback(name, (address)base, (address)top, param)) {
2096           // Oops abort, callback aborted
2097           fclose(procmapsFile);
2098           return 1;
2099         }
2100       }
2101     }
2102     fclose(procmapsFile);
2103   }
2104   return 0;
2105 }
2106 
2107 void os::print_os_info_brief(outputStream* st) {
2108   os::Linux::print_distro_info(st);
2109 
2110   os::Posix::print_uname_info(st);
2111 
2112   os::Linux::print_libversion_info(st);
2113 
2114 }
2115 
2116 void os::print_os_info(outputStream* st) {
2117   st->print("OS:");
2118 
2119   os::Linux::print_distro_info(st);
2120 
2121   os::Posix::print_uname_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 
2303 void os::Linux::print_ld_preload_file(outputStream* st) {
2304   _print_ascii_file("/etc/ld.so.preload", st, "\n/etc/ld.so.preload:");
2305   st->cr();
2306 }
2307 
2308 void os::Linux::print_container_info(outputStream* st) {
2309   if (!OSContainer::is_containerized()) {
2310     return;
2311   }
2312 
2313   st->print("container (cgroup) information:\n");
2314 
2315   const char *p_ct = OSContainer::container_type();
2316   st->print("container_type: %s\n", p_ct != NULL ? p_ct : "not supported");
2317 
2318   char *p = OSContainer::cpu_cpuset_cpus();
2319   st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "not supported");
2320   free(p);
2321 
2322   p = OSContainer::cpu_cpuset_memory_nodes();
2323   st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "not supported");
2324   free(p);
2325 
2326   int i = OSContainer::active_processor_count();
2327   st->print("active_processor_count: ");
2328   if (i > 0) {
2329     st->print("%d\n", i);
2330   } else {
2331     st->print("not supported\n");
2332   }
2333 
2334   i = OSContainer::cpu_quota();
2335   st->print("cpu_quota: ");
2336   if (i > 0) {
2337     st->print("%d\n", i);
2338   } else {
2339     st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no quota");
2340   }
2341 
2342   i = OSContainer::cpu_period();
2343   st->print("cpu_period: ");
2344   if (i > 0) {
2345     st->print("%d\n", i);
2346   } else {
2347     st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no period");
2348   }
2349 
2350   i = OSContainer::cpu_shares();
2351   st->print("cpu_shares: ");
2352   if (i > 0) {
2353     st->print("%d\n", i);
2354   } else {
2355     st->print("%s\n", i == OSCONTAINER_ERROR ? "not supported" : "no shares");
2356   }
2357 
2358   jlong j = OSContainer::memory_limit_in_bytes();
2359   st->print("memory_limit_in_bytes: ");
2360   if (j > 0) {
2361     st->print(JLONG_FORMAT "\n", j);
2362   } else {
2363     st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2364   }
2365 
2366   j = OSContainer::memory_and_swap_limit_in_bytes();
2367   st->print("memory_and_swap_limit_in_bytes: ");
2368   if (j > 0) {
2369     st->print(JLONG_FORMAT "\n", j);
2370   } else {
2371     st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2372   }
2373 
2374   j = OSContainer::memory_soft_limit_in_bytes();
2375   st->print("memory_soft_limit_in_bytes: ");
2376   if (j > 0) {
2377     st->print(JLONG_FORMAT "\n", j);
2378   } else {
2379     st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2380   }
2381 
2382   j = OSContainer::OSContainer::memory_usage_in_bytes();
2383   st->print("memory_usage_in_bytes: ");
2384   if (j > 0) {
2385     st->print(JLONG_FORMAT "\n", j);
2386   } else {
2387     st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2388   }
2389 
2390   j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2391   st->print("memory_max_usage_in_bytes: ");
2392   if (j > 0) {
2393     st->print(JLONG_FORMAT "\n", j);
2394   } else {
2395     st->print("%s\n", j == OSCONTAINER_ERROR ? "not supported" : "unlimited");
2396   }
2397   st->cr();
2398 }
2399 
2400 void os::Linux::print_steal_info(outputStream* st) {
2401   if (has_initial_tick_info) {
2402     CPUPerfTicks pticks;
2403     bool res = os::Linux::get_tick_information(&pticks, -1);
2404 
2405     if (res && pticks.has_steal_ticks) {
2406       uint64_t steal_ticks_difference = pticks.steal - initial_steal_ticks;
2407       uint64_t total_ticks_difference = pticks.total - initial_total_ticks;
2408       double steal_ticks_perc = 0.0;
2409       if (total_ticks_difference != 0) {
2410         steal_ticks_perc = (double) steal_ticks_difference / total_ticks_difference;
2411       }
2412       st->print_cr("Steal ticks since vm start: " UINT64_FORMAT, steal_ticks_difference);
2413       st->print_cr("Steal ticks percentage since vm start:%7.3f", steal_ticks_perc);
2414     }
2415   }
2416 }
2417 
2418 void os::print_memory_info(outputStream* st) {
2419 
2420   st->print("Memory:");
2421   st->print(" %dk page", os::vm_page_size()>>10);
2422 
2423   // values in struct sysinfo are "unsigned long"
2424   struct sysinfo si;
2425   sysinfo(&si);
2426 
2427   st->print(", physical " UINT64_FORMAT "k",
2428             os::physical_memory() >> 10);
2429   st->print("(" UINT64_FORMAT "k free)",
2430             os::available_memory() >> 10);
2431   st->print(", swap " UINT64_FORMAT "k",
2432             ((jlong)si.totalswap * si.mem_unit) >> 10);
2433   st->print("(" UINT64_FORMAT "k free)",
2434             ((jlong)si.freeswap * si.mem_unit) >> 10);
2435   st->cr();
2436 }
2437 
2438 // Print the first "model name" line and the first "flags" line
2439 // that we find and nothing more. We assume "model name" comes
2440 // before "flags" so if we find a second "model name", then the
2441 // "flags" field is considered missing.
2442 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2443 #if defined(IA32) || defined(AMD64)
2444   // Other platforms have less repetitive cpuinfo files
2445   FILE *fp = fopen("/proc/cpuinfo", "r");
2446   if (fp) {
2447     while (!feof(fp)) {
2448       if (fgets(buf, buflen, fp)) {
2449         // Assume model name comes before flags
2450         bool model_name_printed = false;
2451         if (strstr(buf, "model name") != NULL) {
2452           if (!model_name_printed) {
2453             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2454             st->print_raw(buf);
2455             model_name_printed = true;
2456           } else {
2457             // model name printed but not flags?  Odd, just return
2458             fclose(fp);
2459             return true;
2460           }
2461         }
2462         // print the flags line too
2463         if (strstr(buf, "flags") != NULL) {
2464           st->print_raw(buf);
2465           fclose(fp);
2466           return true;
2467         }
2468       }
2469     }
2470     fclose(fp);
2471   }
2472 #endif // x86 platforms
2473   return false;
2474 }
2475 
2476 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2477   // Only print the model name if the platform provides this as a summary
2478   if (!print_model_name_and_flags(st, buf, buflen)) {
2479     st->print("\n/proc/cpuinfo:\n");
2480     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2481       st->print_cr("  <Not Available>");
2482     }
2483   }
2484 }
2485 
2486 #if defined(AMD64) || defined(IA32) || defined(X32)
2487 const char* search_string = "model name";
2488 #elif defined(M68K)
2489 const char* search_string = "CPU";
2490 #elif defined(PPC64)
2491 const char* search_string = "cpu";
2492 #elif defined(S390)
2493 const char* search_string = "machine =";
2494 #elif defined(SPARC)
2495 const char* search_string = "cpu";
2496 #else
2497 const char* search_string = "Processor";
2498 #endif
2499 
2500 // Parses the cpuinfo file for string representing the model name.
2501 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2502   FILE* fp = fopen("/proc/cpuinfo", "r");
2503   if (fp != NULL) {
2504     while (!feof(fp)) {
2505       char buf[256];
2506       if (fgets(buf, sizeof(buf), fp)) {
2507         char* start = strstr(buf, search_string);
2508         if (start != NULL) {
2509           char *ptr = start + strlen(search_string);
2510           char *end = buf + strlen(buf);
2511           while (ptr != end) {
2512              // skip whitespace and colon for the rest of the name.
2513              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2514                break;
2515              }
2516              ptr++;
2517           }
2518           if (ptr != end) {
2519             // reasonable string, get rid of newline and keep the rest
2520             char* nl = strchr(buf, '\n');
2521             if (nl != NULL) *nl = '\0';
2522             strncpy(cpuinfo, ptr, length);
2523             fclose(fp);
2524             return;
2525           }
2526         }
2527       }
2528     }
2529     fclose(fp);
2530   }
2531   // cpuinfo not found or parsing failed, just print generic string.  The entire
2532   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2533 #if   defined(AARCH64)
2534   strncpy(cpuinfo, "AArch64", length);
2535 #elif defined(AMD64)
2536   strncpy(cpuinfo, "x86_64", length);
2537 #elif defined(ARM)  // Order wrt. AARCH64 is relevant!
2538   strncpy(cpuinfo, "ARM", length);
2539 #elif defined(IA32)
2540   strncpy(cpuinfo, "x86_32", length);
2541 #elif defined(IA64)
2542   strncpy(cpuinfo, "IA64", length);
2543 #elif defined(PPC)
2544   strncpy(cpuinfo, "PPC64", length);
2545 #elif defined(S390)
2546   strncpy(cpuinfo, "S390", length);
2547 #elif defined(SPARC)
2548   strncpy(cpuinfo, "sparcv9", length);
2549 #elif defined(ZERO_LIBARCH)
2550   strncpy(cpuinfo, ZERO_LIBARCH, length);
2551 #else
2552   strncpy(cpuinfo, "unknown", length);
2553 #endif
2554 }
2555 
2556 static void print_signal_handler(outputStream* st, int sig,
2557                                  char* buf, size_t buflen);
2558 
2559 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2560   st->print_cr("Signal Handlers:");
2561   print_signal_handler(st, SIGSEGV, buf, buflen);
2562   print_signal_handler(st, SIGBUS , buf, buflen);
2563   print_signal_handler(st, SIGFPE , buf, buflen);
2564   print_signal_handler(st, SIGPIPE, buf, buflen);
2565   print_signal_handler(st, SIGXFSZ, buf, buflen);
2566   print_signal_handler(st, SIGILL , buf, buflen);
2567   print_signal_handler(st, SR_signum, buf, buflen);
2568   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2569   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2570   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2571   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2572 #if defined(PPC64)
2573   print_signal_handler(st, SIGTRAP, buf, buflen);
2574 #endif
2575 }
2576 
2577 static char saved_jvm_path[MAXPATHLEN] = {0};
2578 
2579 // Find the full path to the current module, libjvm.so
2580 void os::jvm_path(char *buf, jint buflen) {
2581   // Error checking.
2582   if (buflen < MAXPATHLEN) {
2583     assert(false, "must use a large-enough buffer");
2584     buf[0] = '\0';
2585     return;
2586   }
2587   // Lazy resolve the path to current module.
2588   if (saved_jvm_path[0] != 0) {
2589     strcpy(buf, saved_jvm_path);
2590     return;
2591   }
2592 
2593   char dli_fname[MAXPATHLEN];
2594   bool ret = dll_address_to_library_name(
2595                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2596                                          dli_fname, sizeof(dli_fname), NULL);
2597   assert(ret, "cannot locate libjvm");
2598   char *rp = NULL;
2599   if (ret && dli_fname[0] != '\0') {
2600     rp = os::Posix::realpath(dli_fname, buf, buflen);
2601   }
2602   if (rp == NULL) {
2603     return;
2604   }
2605 
2606   if (Arguments::sun_java_launcher_is_altjvm()) {
2607     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2608     // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2609     // If "/jre/lib/" appears at the right place in the string, then
2610     // assume we are installed in a JDK and we're done. Otherwise, check
2611     // for a JAVA_HOME environment variable and fix up the path so it
2612     // looks like libjvm.so is installed there (append a fake suffix
2613     // hotspot/libjvm.so).
2614     const char *p = buf + strlen(buf) - 1;
2615     for (int count = 0; p > buf && count < 5; ++count) {
2616       for (--p; p > buf && *p != '/'; --p)
2617         /* empty */ ;
2618     }
2619 
2620     if (strncmp(p, "/jre/lib/", 9) != 0) {
2621       // Look for JAVA_HOME in the environment.
2622       char* java_home_var = ::getenv("JAVA_HOME");
2623       if (java_home_var != NULL && java_home_var[0] != 0) {
2624         char* jrelib_p;
2625         int len;
2626 
2627         // Check the current module name "libjvm.so".
2628         p = strrchr(buf, '/');
2629         if (p == NULL) {
2630           return;
2631         }
2632         assert(strstr(p, "/libjvm") == p, "invalid library name");
2633 
2634         rp = os::Posix::realpath(java_home_var, buf, buflen);
2635         if (rp == NULL) {
2636           return;
2637         }
2638 
2639         // determine if this is a legacy image or modules image
2640         // modules image doesn't have "jre" subdirectory
2641         len = strlen(buf);
2642         assert(len < buflen, "Ran out of buffer room");
2643         jrelib_p = buf + len;
2644         snprintf(jrelib_p, buflen-len, "/jre/lib");
2645         if (0 != access(buf, F_OK)) {
2646           snprintf(jrelib_p, buflen-len, "/lib");
2647         }
2648 
2649         if (0 == access(buf, F_OK)) {
2650           // Use current module name "libjvm.so"
2651           len = strlen(buf);
2652           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2653         } else {
2654           // Go back to path of .so
2655           rp = os::Posix::realpath(dli_fname, buf, buflen);
2656           if (rp == NULL) {
2657             return;
2658           }
2659         }
2660       }
2661     }
2662   }
2663 
2664   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2665   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2666 }
2667 
2668 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2669   // no prefix required, not even "_"
2670 }
2671 
2672 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2673   // no suffix required
2674 }
2675 
2676 ////////////////////////////////////////////////////////////////////////////////
2677 // sun.misc.Signal support
2678 
2679 static void UserHandler(int sig, void *siginfo, void *context) {
2680   // Ctrl-C is pressed during error reporting, likely because the error
2681   // handler fails to abort. Let VM die immediately.
2682   if (sig == SIGINT && VMError::is_error_reported()) {
2683     os::die();
2684   }
2685 
2686   os::signal_notify(sig);
2687 }
2688 
2689 void* os::user_handler() {
2690   return CAST_FROM_FN_PTR(void*, UserHandler);
2691 }
2692 
2693 extern "C" {
2694   typedef void (*sa_handler_t)(int);
2695   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2696 }
2697 
2698 void* os::signal(int signal_number, void* handler) {
2699   struct sigaction sigAct, oldSigAct;
2700 
2701   sigfillset(&(sigAct.sa_mask));
2702   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2703   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2704 
2705   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2706     // -1 means registration failed
2707     return (void *)-1;
2708   }
2709 
2710   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2711 }
2712 
2713 void os::signal_raise(int signal_number) {
2714   ::raise(signal_number);
2715 }
2716 
2717 // The following code is moved from os.cpp for making this
2718 // code platform specific, which it is by its very nature.
2719 
2720 // Will be modified when max signal is changed to be dynamic
2721 int os::sigexitnum_pd() {
2722   return NSIG;
2723 }
2724 
2725 // a counter for each possible signal value
2726 static volatile jint pending_signals[NSIG+1] = { 0 };
2727 
2728 // Linux(POSIX) specific hand shaking semaphore.
2729 static Semaphore* sig_sem = NULL;
2730 static PosixSemaphore sr_semaphore;
2731 
2732 static void jdk_misc_signal_init() {
2733   // Initialize signal structures
2734   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2735 
2736   // Initialize signal semaphore
2737   sig_sem = new Semaphore();
2738 }
2739 
2740 void os::signal_notify(int sig) {
2741   if (sig_sem != NULL) {
2742     Atomic::inc(&pending_signals[sig]);
2743     sig_sem->signal();
2744   } else {
2745     // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2746     // initialization isn't called.
2747     assert(ReduceSignalUsage, "signal semaphore should be created");
2748   }
2749 }
2750 
2751 static int check_pending_signals() {
2752   for (;;) {
2753     for (int i = 0; i < NSIG + 1; i++) {
2754       jint n = pending_signals[i];
2755       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2756         return i;
2757       }
2758     }
2759     JavaThread *thread = JavaThread::current();
2760     ThreadBlockInVM tbivm(thread);
2761 
2762     bool threadIsSuspended;
2763     do {
2764       thread->set_suspend_equivalent();
2765       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2766       sig_sem->wait();
2767 
2768       // were we externally suspended while we were waiting?
2769       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2770       if (threadIsSuspended) {
2771         // The semaphore has been incremented, but while we were waiting
2772         // another thread suspended us. We don't want to continue running
2773         // while suspended because that would surprise the thread that
2774         // suspended us.
2775         sig_sem->signal();
2776 
2777         thread->java_suspend_self();
2778       }
2779     } while (threadIsSuspended);
2780   }
2781 }
2782 
2783 int os::signal_wait() {
2784   return check_pending_signals();
2785 }
2786 
2787 ////////////////////////////////////////////////////////////////////////////////
2788 // Virtual Memory
2789 
2790 int os::vm_page_size() {
2791   // Seems redundant as all get out
2792   assert(os::Linux::page_size() != -1, "must call os::init");
2793   return os::Linux::page_size();
2794 }
2795 
2796 // Solaris allocates memory by pages.
2797 int os::vm_allocation_granularity() {
2798   assert(os::Linux::page_size() != -1, "must call os::init");
2799   return os::Linux::page_size();
2800 }
2801 
2802 // Rationale behind this function:
2803 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2804 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2805 //  samples for JITted code. Here we create private executable mapping over the code cache
2806 //  and then we can use standard (well, almost, as mapping can change) way to provide
2807 //  info for the reporting script by storing timestamp and location of symbol
2808 void linux_wrap_code(char* base, size_t size) {
2809   static volatile jint cnt = 0;
2810 
2811   if (!UseOprofile) {
2812     return;
2813   }
2814 
2815   char buf[PATH_MAX+1];
2816   int num = Atomic::add(1, &cnt);
2817 
2818   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2819            os::get_temp_directory(), os::current_process_id(), num);
2820   unlink(buf);
2821 
2822   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2823 
2824   if (fd != -1) {
2825     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2826     if (rv != (off_t)-1) {
2827       if (::write(fd, "", 1) == 1) {
2828         mmap(base, size,
2829              PROT_READ|PROT_WRITE|PROT_EXEC,
2830              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2831       }
2832     }
2833     ::close(fd);
2834     unlink(buf);
2835   }
2836 }
2837 
2838 static bool recoverable_mmap_error(int err) {
2839   // See if the error is one we can let the caller handle. This
2840   // list of errno values comes from JBS-6843484. I can't find a
2841   // Linux man page that documents this specific set of errno
2842   // values so while this list currently matches Solaris, it may
2843   // change as we gain experience with this failure mode.
2844   switch (err) {
2845   case EBADF:
2846   case EINVAL:
2847   case ENOTSUP:
2848     // let the caller deal with these errors
2849     return true;
2850 
2851   default:
2852     // Any remaining errors on this OS can cause our reserved mapping
2853     // to be lost. That can cause confusion where different data
2854     // structures think they have the same memory mapped. The worst
2855     // scenario is if both the VM and a library think they have the
2856     // same memory mapped.
2857     return false;
2858   }
2859 }
2860 
2861 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2862                                     int err) {
2863   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2864           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2865           os::strerror(err), err);
2866 }
2867 
2868 static void warn_fail_commit_memory(char* addr, size_t size,
2869                                     size_t alignment_hint, bool exec,
2870                                     int err) {
2871   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2872           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2873           alignment_hint, exec, os::strerror(err), err);
2874 }
2875 
2876 // NOTE: Linux kernel does not really reserve the pages for us.
2877 //       All it does is to check if there are enough free pages
2878 //       left at the time of mmap(). This could be a potential
2879 //       problem.
2880 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2881   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2882   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2883                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2884   if (res != (uintptr_t) MAP_FAILED) {
2885     if (UseNUMAInterleaving) {
2886       numa_make_global(addr, size);
2887     }
2888     return 0;
2889   }
2890 
2891   int err = errno;  // save errno from mmap() call above
2892 
2893   if (!recoverable_mmap_error(err)) {
2894     warn_fail_commit_memory(addr, size, exec, err);
2895     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2896   }
2897 
2898   return err;
2899 }
2900 
2901 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2902   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2903 }
2904 
2905 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2906                                   const char* mesg) {
2907   assert(mesg != NULL, "mesg must be specified");
2908   int err = os::Linux::commit_memory_impl(addr, size, exec);
2909   if (err != 0) {
2910     // the caller wants all commit errors to exit with the specified mesg:
2911     warn_fail_commit_memory(addr, size, exec, err);
2912     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2913   }
2914 }
2915 
2916 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2917 #ifndef MAP_HUGETLB
2918   #define MAP_HUGETLB 0x40000
2919 #endif
2920 
2921 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2922 #ifndef MADV_HUGEPAGE
2923   #define MADV_HUGEPAGE 14
2924 #endif
2925 
2926 int os::Linux::commit_memory_impl(char* addr, size_t size,
2927                                   size_t alignment_hint, bool exec) {
2928   int err = os::Linux::commit_memory_impl(addr, size, exec);
2929   if (err == 0) {
2930     realign_memory(addr, size, alignment_hint);
2931   }
2932   return err;
2933 }
2934 
2935 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2936                           bool exec) {
2937   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2938 }
2939 
2940 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2941                                   size_t alignment_hint, bool exec,
2942                                   const char* mesg) {
2943   assert(mesg != NULL, "mesg must be specified");
2944   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2945   if (err != 0) {
2946     // the caller wants all commit errors to exit with the specified mesg:
2947     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2948     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2949   }
2950 }
2951 
2952 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2953   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2954     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2955     // be supported or the memory may already be backed by huge pages.
2956     ::madvise(addr, bytes, MADV_HUGEPAGE);
2957   }
2958 }
2959 
2960 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2961   // This method works by doing an mmap over an existing mmaping and effectively discarding
2962   // the existing pages. However it won't work for SHM-based large pages that cannot be
2963   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2964   // small pages on top of the SHM segment. This method always works for small pages, so we
2965   // allow that in any case.
2966   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2967     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2968   }
2969 }
2970 
2971 void os::numa_make_global(char *addr, size_t bytes) {
2972   Linux::numa_interleave_memory(addr, bytes);
2973 }
2974 
2975 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2976 // bind policy to MPOL_PREFERRED for the current thread.
2977 #define USE_MPOL_PREFERRED 0
2978 
2979 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2980   // To make NUMA and large pages more robust when both enabled, we need to ease
2981   // the requirements on where the memory should be allocated. MPOL_BIND is the
2982   // default policy and it will force memory to be allocated on the specified
2983   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2984   // the specified node, but will not force it. Using this policy will prevent
2985   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2986   // free large pages.
2987   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2988   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2989 }
2990 
2991 bool os::numa_topology_changed() { return false; }
2992 
2993 size_t os::numa_get_groups_num() {
2994   // Return just the number of nodes in which it's possible to allocate memory
2995   // (in numa terminology, configured nodes).
2996   return Linux::numa_num_configured_nodes();
2997 }
2998 
2999 int os::numa_get_group_id() {
3000   int cpu_id = Linux::sched_getcpu();
3001   if (cpu_id != -1) {
3002     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
3003     if (lgrp_id != -1) {
3004       return lgrp_id;
3005     }
3006   }
3007   return 0;
3008 }
3009 
3010 int os::numa_get_group_id_for_address(const void* address) {
3011 #ifndef MPOL_F_NODE
3012 #define MPOL_F_NODE     (1<<0)  // Return next IL mode instead of node mask
3013 #endif
3014 
3015 #ifndef MPOL_F_ADDR
3016 #define MPOL_F_ADDR     (1<<1)  // Look up VMA using address
3017 #endif
3018 
3019   int id = 0;
3020 
3021   if (syscall(SYS_get_mempolicy, &id, NULL, 0, const_cast<void*>(address), MPOL_F_NODE | MPOL_F_ADDR) == -1) {
3022     return -1;
3023   }
3024   return id;
3025 }
3026 
3027 int os::Linux::get_existing_num_nodes() {
3028   int node;
3029   int highest_node_number = Linux::numa_max_node();
3030   int num_nodes = 0;
3031 
3032   // Get the total number of nodes in the system including nodes without memory.
3033   for (node = 0; node <= highest_node_number; node++) {
3034     if (is_node_in_existing_nodes(node)) {
3035       num_nodes++;
3036     }
3037   }
3038   return num_nodes;
3039 }
3040 
3041 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
3042   int highest_node_number = Linux::numa_max_node();
3043   size_t i = 0;
3044 
3045   // Map all node ids in which it is possible to allocate memory. Also nodes are
3046   // not always consecutively available, i.e. available from 0 to the highest
3047   // node number. If the nodes have been bound explicitly using numactl membind,
3048   // then allocate memory from those nodes only.
3049   for (int node = 0; node <= highest_node_number; node++) {
3050     if (Linux::is_node_in_bound_nodes((unsigned int)node)) {
3051       ids[i++] = node;
3052     }
3053   }
3054   return i;
3055 }
3056 
3057 bool os::get_page_info(char *start, page_info* info) {
3058   return false;
3059 }
3060 
3061 char *os::scan_pages(char *start, char* end, page_info* page_expected,
3062                      page_info* page_found) {
3063   return end;
3064 }
3065 
3066 
3067 int os::Linux::sched_getcpu_syscall(void) {
3068   unsigned int cpu = 0;
3069   int retval = -1;
3070 
3071 #if defined(IA32)
3072   #ifndef SYS_getcpu
3073     #define SYS_getcpu 318
3074   #endif
3075   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
3076 #elif defined(AMD64)
3077 // Unfortunately we have to bring all these macros here from vsyscall.h
3078 // to be able to compile on old linuxes.
3079   #define __NR_vgetcpu 2
3080   #define VSYSCALL_START (-10UL << 20)
3081   #define VSYSCALL_SIZE 1024
3082   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
3083   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
3084   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
3085   retval = vgetcpu(&cpu, NULL, NULL);
3086 #endif
3087 
3088   return (retval == -1) ? retval : cpu;
3089 }
3090 
3091 void os::Linux::sched_getcpu_init() {
3092   // sched_getcpu() should be in libc.
3093   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3094                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
3095 
3096   // If it's not, try a direct syscall.
3097   if (sched_getcpu() == -1) {
3098     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3099                                     (void*)&sched_getcpu_syscall));
3100   }
3101 
3102   if (sched_getcpu() == -1) {
3103     vm_exit_during_initialization("getcpu(2) system call not supported by kernel");
3104   }
3105 }
3106 
3107 // Something to do with the numa-aware allocator needs these symbols
3108 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
3109 extern "C" JNIEXPORT void numa_error(char *where) { }
3110 
3111 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
3112 // load symbol from base version instead.
3113 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
3114   void *f = dlvsym(handle, name, "libnuma_1.1");
3115   if (f == NULL) {
3116     f = dlsym(handle, name);
3117   }
3118   return f;
3119 }
3120 
3121 // Handle request to load libnuma symbol version 1.2 (API v2) only.
3122 // Return NULL if the symbol is not defined in this particular version.
3123 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
3124   return dlvsym(handle, name, "libnuma_1.2");
3125 }
3126 
3127 bool os::Linux::libnuma_init() {
3128   if (sched_getcpu() != -1) { // Requires sched_getcpu() support
3129     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3130     if (handle != NULL) {
3131       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3132                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
3133       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3134                                        libnuma_dlsym(handle, "numa_max_node")));
3135       set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3136                                                    libnuma_dlsym(handle, "numa_num_configured_nodes")));
3137       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3138                                         libnuma_dlsym(handle, "numa_available")));
3139       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3140                                             libnuma_dlsym(handle, "numa_tonode_memory")));
3141       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3142                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
3143       set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3144                                                 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3145       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3146                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
3147       set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3148                                                libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3149       set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3150                                        libnuma_dlsym(handle, "numa_distance")));
3151       set_numa_get_membind(CAST_TO_FN_PTR(numa_get_membind_func_t,
3152                                           libnuma_v2_dlsym(handle, "numa_get_membind")));
3153       set_numa_get_interleave_mask(CAST_TO_FN_PTR(numa_get_interleave_mask_func_t,
3154                                                   libnuma_v2_dlsym(handle, "numa_get_interleave_mask")));
3155 
3156       if (numa_available() != -1) {
3157         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3158         set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3159         set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3160         set_numa_interleave_bitmask(_numa_get_interleave_mask());
3161         set_numa_membind_bitmask(_numa_get_membind());
3162         // Create an index -> node mapping, since nodes are not always consecutive
3163         _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3164         rebuild_nindex_to_node_map();
3165         // Create a cpu -> node mapping
3166         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3167         rebuild_cpu_to_node_map();
3168         return true;
3169       }
3170     }
3171   }
3172   return false;
3173 }
3174 
3175 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
3176   // Creating guard page is very expensive. Java thread has HotSpot
3177   // guard pages, only enable glibc guard page for non-Java threads.
3178   // (Remember: compiler thread is a Java thread, too!)
3179   return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
3180 }
3181 
3182 void os::Linux::rebuild_nindex_to_node_map() {
3183   int highest_node_number = Linux::numa_max_node();
3184 
3185   nindex_to_node()->clear();
3186   for (int node = 0; node <= highest_node_number; node++) {
3187     if (Linux::is_node_in_existing_nodes(node)) {
3188       nindex_to_node()->append(node);
3189     }
3190   }
3191 }
3192 
3193 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3194 // The table is later used in get_node_by_cpu().
3195 void os::Linux::rebuild_cpu_to_node_map() {
3196   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3197                               // in libnuma (possible values are starting from 16,
3198                               // and continuing up with every other power of 2, but less
3199                               // than the maximum number of CPUs supported by kernel), and
3200                               // is a subject to change (in libnuma version 2 the requirements
3201                               // are more reasonable) we'll just hardcode the number they use
3202                               // in the library.
3203   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3204 
3205   size_t cpu_num = processor_count();
3206   size_t cpu_map_size = NCPUS / BitsPerCLong;
3207   size_t cpu_map_valid_size =
3208     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3209 
3210   cpu_to_node()->clear();
3211   cpu_to_node()->at_grow(cpu_num - 1);
3212 
3213   size_t node_num = get_existing_num_nodes();
3214 
3215   int distance = 0;
3216   int closest_distance = INT_MAX;
3217   int closest_node = 0;
3218   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3219   for (size_t i = 0; i < node_num; i++) {
3220     // Check if node is configured (not a memory-less node). If it is not, find
3221     // the closest configured node. Check also if node is bound, i.e. it's allowed
3222     // to allocate memory from the node. If it's not allowed, map cpus in that node
3223     // to the closest node from which memory allocation is allowed.
3224     if (!is_node_in_configured_nodes(nindex_to_node()->at(i)) ||
3225         !is_node_in_bound_nodes(nindex_to_node()->at(i))) {
3226       closest_distance = INT_MAX;
3227       // Check distance from all remaining nodes in the system. Ignore distance
3228       // from itself, from another non-configured node, and from another non-bound
3229       // node.
3230       for (size_t m = 0; m < node_num; m++) {
3231         if (m != i &&
3232             is_node_in_configured_nodes(nindex_to_node()->at(m)) &&
3233             is_node_in_bound_nodes(nindex_to_node()->at(m))) {
3234           distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3235           // If a closest node is found, update. There is always at least one
3236           // configured and bound node in the system so there is always at least
3237           // one node close.
3238           if (distance != 0 && distance < closest_distance) {
3239             closest_distance = distance;
3240             closest_node = nindex_to_node()->at(m);
3241           }
3242         }
3243       }
3244      } else {
3245        // Current node is already a configured node.
3246        closest_node = nindex_to_node()->at(i);
3247      }
3248 
3249     // Get cpus from the original node and map them to the closest node. If node
3250     // is a configured node (not a memory-less node), then original node and
3251     // closest node are the same.
3252     if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3253       for (size_t j = 0; j < cpu_map_valid_size; j++) {
3254         if (cpu_map[j] != 0) {
3255           for (size_t k = 0; k < BitsPerCLong; k++) {
3256             if (cpu_map[j] & (1UL << k)) {
3257               cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3258             }
3259           }
3260         }
3261       }
3262     }
3263   }
3264   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3265 }
3266 
3267 int os::Linux::get_node_by_cpu(int cpu_id) {
3268   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3269     return cpu_to_node()->at(cpu_id);
3270   }
3271   return -1;
3272 }
3273 
3274 GrowableArray<int>* os::Linux::_cpu_to_node;
3275 GrowableArray<int>* os::Linux::_nindex_to_node;
3276 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3277 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3278 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3279 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3280 os::Linux::numa_available_func_t os::Linux::_numa_available;
3281 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3282 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3283 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3284 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3285 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3286 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3287 os::Linux::numa_get_membind_func_t os::Linux::_numa_get_membind;
3288 os::Linux::numa_get_interleave_mask_func_t os::Linux::_numa_get_interleave_mask;
3289 os::Linux::NumaAllocationPolicy os::Linux::_current_numa_policy;
3290 unsigned long* os::Linux::_numa_all_nodes;
3291 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3292 struct bitmask* os::Linux::_numa_nodes_ptr;
3293 struct bitmask* os::Linux::_numa_interleave_bitmask;
3294 struct bitmask* os::Linux::_numa_membind_bitmask;
3295 
3296 bool os::pd_uncommit_memory(char* addr, size_t size) {
3297   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3298                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3299   return res  != (uintptr_t) MAP_FAILED;
3300 }
3301 
3302 static address get_stack_commited_bottom(address bottom, size_t size) {
3303   address nbot = bottom;
3304   address ntop = bottom + size;
3305 
3306   size_t page_sz = os::vm_page_size();
3307   unsigned pages = size / page_sz;
3308 
3309   unsigned char vec[1];
3310   unsigned imin = 1, imax = pages + 1, imid;
3311   int mincore_return_value = 0;
3312 
3313   assert(imin <= imax, "Unexpected page size");
3314 
3315   while (imin < imax) {
3316     imid = (imax + imin) / 2;
3317     nbot = ntop - (imid * page_sz);
3318 
3319     // Use a trick with mincore to check whether the page is mapped or not.
3320     // mincore sets vec to 1 if page resides in memory and to 0 if page
3321     // is swapped output but if page we are asking for is unmapped
3322     // it returns -1,ENOMEM
3323     mincore_return_value = mincore(nbot, page_sz, vec);
3324 
3325     if (mincore_return_value == -1) {
3326       // Page is not mapped go up
3327       // to find first mapped page
3328       if (errno != EAGAIN) {
3329         assert(errno == ENOMEM, "Unexpected mincore errno");
3330         imax = imid;
3331       }
3332     } else {
3333       // Page is mapped go down
3334       // to find first not mapped page
3335       imin = imid + 1;
3336     }
3337   }
3338 
3339   nbot = nbot + page_sz;
3340 
3341   // Adjust stack bottom one page up if last checked page is not mapped
3342   if (mincore_return_value == -1) {
3343     nbot = nbot + page_sz;
3344   }
3345 
3346   return nbot;
3347 }
3348 
3349 bool os::committed_in_range(address start, size_t size, address& committed_start, size_t& committed_size) {
3350   int mincore_return_value;
3351   const size_t stripe = 1024;  // query this many pages each time
3352   unsigned char vec[stripe + 1];
3353   // set a guard
3354   vec[stripe] = 'X';
3355 
3356   const size_t page_sz = os::vm_page_size();
3357   size_t pages = size / page_sz;
3358 
3359   assert(is_aligned(start, page_sz), "Start address must be page aligned");
3360   assert(is_aligned(size, page_sz), "Size must be page aligned");
3361 
3362   committed_start = NULL;
3363 
3364   int loops = (pages + stripe - 1) / stripe;
3365   int committed_pages = 0;
3366   address loop_base = start;
3367   bool found_range = false;
3368 
3369   for (int index = 0; index < loops && !found_range; index ++) {
3370     assert(pages > 0, "Nothing to do");
3371     int pages_to_query = (pages >= stripe) ? stripe : pages;
3372     pages -= pages_to_query;
3373 
3374     // Get stable read
3375     while ((mincore_return_value = mincore(loop_base, pages_to_query * page_sz, vec)) == -1 && errno == EAGAIN);
3376 
3377     // During shutdown, some memory goes away without properly notifying NMT,
3378     // E.g. ConcurrentGCThread/WatcherThread can exit without deleting thread object.
3379     // Bailout and return as not committed for now.
3380     if (mincore_return_value == -1 && errno == ENOMEM) {
3381       return false;
3382     }
3383 
3384     assert(vec[stripe] == 'X', "overflow guard");
3385     assert(mincore_return_value == 0, "Range must be valid");
3386     // Process this stripe
3387     for (int vecIdx = 0; vecIdx < pages_to_query; vecIdx ++) {
3388       if ((vec[vecIdx] & 0x01) == 0) { // not committed
3389         // End of current contiguous region
3390         if (committed_start != NULL) {
3391           found_range = true;
3392           break;
3393         }
3394       } else { // committed
3395         // Start of region
3396         if (committed_start == NULL) {
3397           committed_start = loop_base + page_sz * vecIdx;
3398         }
3399         committed_pages ++;
3400       }
3401     }
3402 
3403     loop_base += pages_to_query * page_sz;
3404   }
3405 
3406   if (committed_start != NULL) {
3407     assert(committed_pages > 0, "Must have committed region");
3408     assert(committed_pages <= int(size / page_sz), "Can not commit more than it has");
3409     assert(committed_start >= start && committed_start < start + size, "Out of range");
3410     committed_size = page_sz * committed_pages;
3411     return true;
3412   } else {
3413     assert(committed_pages == 0, "Should not have committed region");
3414     return false;
3415   }
3416 }
3417 
3418 
3419 // Linux uses a growable mapping for the stack, and if the mapping for
3420 // the stack guard pages is not removed when we detach a thread the
3421 // stack cannot grow beyond the pages where the stack guard was
3422 // mapped.  If at some point later in the process the stack expands to
3423 // that point, the Linux kernel cannot expand the stack any further
3424 // because the guard pages are in the way, and a segfault occurs.
3425 //
3426 // However, it's essential not to split the stack region by unmapping
3427 // a region (leaving a hole) that's already part of the stack mapping,
3428 // so if the stack mapping has already grown beyond the guard pages at
3429 // the time we create them, we have to truncate the stack mapping.
3430 // So, we need to know the extent of the stack mapping when
3431 // create_stack_guard_pages() is called.
3432 
3433 // We only need this for stacks that are growable: at the time of
3434 // writing thread stacks don't use growable mappings (i.e. those
3435 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3436 // only applies to the main thread.
3437 
3438 // If the (growable) stack mapping already extends beyond the point
3439 // where we're going to put our guard pages, truncate the mapping at
3440 // that point by munmap()ping it.  This ensures that when we later
3441 // munmap() the guard pages we don't leave a hole in the stack
3442 // mapping. This only affects the main/primordial thread
3443 
3444 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3445   if (os::is_primordial_thread()) {
3446     // As we manually grow stack up to bottom inside create_attached_thread(),
3447     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3448     // we don't need to do anything special.
3449     // Check it first, before calling heavy function.
3450     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3451     unsigned char vec[1];
3452 
3453     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3454       // Fallback to slow path on all errors, including EAGAIN
3455       stack_extent = (uintptr_t) get_stack_commited_bottom(
3456                                                            os::Linux::initial_thread_stack_bottom(),
3457                                                            (size_t)addr - stack_extent);
3458     }
3459 
3460     if (stack_extent < (uintptr_t)addr) {
3461       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3462     }
3463   }
3464 
3465   return os::commit_memory(addr, size, !ExecMem);
3466 }
3467 
3468 // If this is a growable mapping, remove the guard pages entirely by
3469 // munmap()ping them.  If not, just call uncommit_memory(). This only
3470 // affects the main/primordial thread, but guard against future OS changes.
3471 // It's safe to always unmap guard pages for primordial thread because we
3472 // always place it right after end of the mapped region.
3473 
3474 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3475   uintptr_t stack_extent, stack_base;
3476 
3477   if (os::is_primordial_thread()) {
3478     return ::munmap(addr, size) == 0;
3479   }
3480 
3481   return os::uncommit_memory(addr, size);
3482 }
3483 
3484 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3485 // at 'requested_addr'. If there are existing memory mappings at the same
3486 // location, however, they will be overwritten. If 'fixed' is false,
3487 // 'requested_addr' is only treated as a hint, the return value may or
3488 // may not start from the requested address. Unlike Linux mmap(), this
3489 // function returns NULL to indicate failure.
3490 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3491   char * addr;
3492   int flags;
3493 
3494   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3495   if (fixed) {
3496     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3497     flags |= MAP_FIXED;
3498   }
3499 
3500   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3501   // touch an uncommitted page. Otherwise, the read/write might
3502   // succeed if we have enough swap space to back the physical page.
3503   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3504                        flags, -1, 0);
3505 
3506   return addr == MAP_FAILED ? NULL : addr;
3507 }
3508 
3509 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3510 //   (req_addr != NULL) or with a given alignment.
3511 //  - bytes shall be a multiple of alignment.
3512 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3513 //  - alignment sets the alignment at which memory shall be allocated.
3514 //     It must be a multiple of allocation granularity.
3515 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3516 //  req_addr or NULL.
3517 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3518 
3519   size_t extra_size = bytes;
3520   if (req_addr == NULL && alignment > 0) {
3521     extra_size += alignment;
3522   }
3523 
3524   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3525     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3526     -1, 0);
3527   if (start == MAP_FAILED) {
3528     start = NULL;
3529   } else {
3530     if (req_addr != NULL) {
3531       if (start != req_addr) {
3532         ::munmap(start, extra_size);
3533         start = NULL;
3534       }
3535     } else {
3536       char* const start_aligned = align_up(start, alignment);
3537       char* const end_aligned = start_aligned + bytes;
3538       char* const end = start + extra_size;
3539       if (start_aligned > start) {
3540         ::munmap(start, start_aligned - start);
3541       }
3542       if (end_aligned < end) {
3543         ::munmap(end_aligned, end - end_aligned);
3544       }
3545       start = start_aligned;
3546     }
3547   }
3548   return start;
3549 }
3550 
3551 static int anon_munmap(char * addr, size_t size) {
3552   return ::munmap(addr, size) == 0;
3553 }
3554 
3555 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3556                             size_t alignment_hint) {
3557   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3558 }
3559 
3560 bool os::pd_release_memory(char* addr, size_t size) {
3561   return anon_munmap(addr, size);
3562 }
3563 
3564 static bool linux_mprotect(char* addr, size_t size, int prot) {
3565   // Linux wants the mprotect address argument to be page aligned.
3566   char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3567 
3568   // According to SUSv3, mprotect() should only be used with mappings
3569   // established by mmap(), and mmap() always maps whole pages. Unaligned
3570   // 'addr' likely indicates problem in the VM (e.g. trying to change
3571   // protection of malloc'ed or statically allocated memory). Check the
3572   // caller if you hit this assert.
3573   assert(addr == bottom, "sanity check");
3574 
3575   size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3576   Events::log(NULL, "Protecting memory [" INTPTR_FORMAT "," INTPTR_FORMAT "] with protection modes %x", p2i(bottom), p2i(bottom+size), prot);
3577   return ::mprotect(bottom, size, prot) == 0;
3578 }
3579 
3580 // Set protections specified
3581 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3582                         bool is_committed) {
3583   unsigned int p = 0;
3584   switch (prot) {
3585   case MEM_PROT_NONE: p = PROT_NONE; break;
3586   case MEM_PROT_READ: p = PROT_READ; break;
3587   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3588   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3589   default:
3590     ShouldNotReachHere();
3591   }
3592   // is_committed is unused.
3593   return linux_mprotect(addr, bytes, p);
3594 }
3595 
3596 bool os::guard_memory(char* addr, size_t size) {
3597   return linux_mprotect(addr, size, PROT_NONE);
3598 }
3599 
3600 bool os::unguard_memory(char* addr, size_t size) {
3601   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3602 }
3603 
3604 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3605                                                     size_t page_size) {
3606   bool result = false;
3607   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3608                  MAP_ANONYMOUS|MAP_PRIVATE,
3609                  -1, 0);
3610   if (p != MAP_FAILED) {
3611     void *aligned_p = align_up(p, page_size);
3612 
3613     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3614 
3615     munmap(p, page_size * 2);
3616   }
3617 
3618   if (warn && !result) {
3619     warning("TransparentHugePages is not supported by the operating system.");
3620   }
3621 
3622   return result;
3623 }
3624 
3625 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3626   bool result = false;
3627   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3628                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3629                  -1, 0);
3630 
3631   if (p != MAP_FAILED) {
3632     // We don't know if this really is a huge page or not.
3633     FILE *fp = fopen("/proc/self/maps", "r");
3634     if (fp) {
3635       while (!feof(fp)) {
3636         char chars[257];
3637         long x = 0;
3638         if (fgets(chars, sizeof(chars), fp)) {
3639           if (sscanf(chars, "%lx-%*x", &x) == 1
3640               && x == (long)p) {
3641             if (strstr (chars, "hugepage")) {
3642               result = true;
3643               break;
3644             }
3645           }
3646         }
3647       }
3648       fclose(fp);
3649     }
3650     munmap(p, page_size);
3651   }
3652 
3653   if (warn && !result) {
3654     warning("HugeTLBFS is not supported by the operating system.");
3655   }
3656 
3657   return result;
3658 }
3659 
3660 // From the coredump_filter documentation:
3661 //
3662 // - (bit 0) anonymous private memory
3663 // - (bit 1) anonymous shared memory
3664 // - (bit 2) file-backed private memory
3665 // - (bit 3) file-backed shared memory
3666 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3667 //           effective only if the bit 2 is cleared)
3668 // - (bit 5) hugetlb private memory
3669 // - (bit 6) hugetlb shared memory
3670 // - (bit 7) dax private memory
3671 // - (bit 8) dax shared memory
3672 //
3673 static void set_coredump_filter(CoredumpFilterBit bit) {
3674   FILE *f;
3675   long cdm;
3676 
3677   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3678     return;
3679   }
3680 
3681   if (fscanf(f, "%lx", &cdm) != 1) {
3682     fclose(f);
3683     return;
3684   }
3685 
3686   long saved_cdm = cdm;
3687   rewind(f);
3688   cdm |= bit;
3689 
3690   if (cdm != saved_cdm) {
3691     fprintf(f, "%#lx", cdm);
3692   }
3693 
3694   fclose(f);
3695 }
3696 
3697 // Large page support
3698 
3699 static size_t _large_page_size = 0;
3700 
3701 size_t os::Linux::find_large_page_size() {
3702   size_t large_page_size = 0;
3703 
3704   // large_page_size on Linux is used to round up heap size. x86 uses either
3705   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3706   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3707   // page as large as 256M.
3708   //
3709   // Here we try to figure out page size by parsing /proc/meminfo and looking
3710   // for a line with the following format:
3711   //    Hugepagesize:     2048 kB
3712   //
3713   // If we can't determine the value (e.g. /proc is not mounted, or the text
3714   // format has been changed), we'll use the largest page size supported by
3715   // the processor.
3716 
3717 #ifndef ZERO
3718   large_page_size =
3719     AARCH64_ONLY(2 * M)
3720     AMD64_ONLY(2 * M)
3721     ARM32_ONLY(2 * M)
3722     IA32_ONLY(4 * M)
3723     IA64_ONLY(256 * M)
3724     PPC_ONLY(4 * M)
3725     S390_ONLY(1 * M)
3726     SPARC_ONLY(4 * M);
3727 #endif // ZERO
3728 
3729   FILE *fp = fopen("/proc/meminfo", "r");
3730   if (fp) {
3731     while (!feof(fp)) {
3732       int x = 0;
3733       char buf[16];
3734       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3735         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3736           large_page_size = x * K;
3737           break;
3738         }
3739       } else {
3740         // skip to next line
3741         for (;;) {
3742           int ch = fgetc(fp);
3743           if (ch == EOF || ch == (int)'\n') break;
3744         }
3745       }
3746     }
3747     fclose(fp);
3748   }
3749 
3750   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3751     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3752             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3753             proper_unit_for_byte_size(large_page_size));
3754   }
3755 
3756   return large_page_size;
3757 }
3758 
3759 size_t os::Linux::setup_large_page_size() {
3760   _large_page_size = Linux::find_large_page_size();
3761   const size_t default_page_size = (size_t)Linux::page_size();
3762   if (_large_page_size > default_page_size) {
3763     _page_sizes[0] = _large_page_size;
3764     _page_sizes[1] = default_page_size;
3765     _page_sizes[2] = 0;
3766   }
3767 
3768   return _large_page_size;
3769 }
3770 
3771 bool os::Linux::setup_large_page_type(size_t page_size) {
3772   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3773       FLAG_IS_DEFAULT(UseSHM) &&
3774       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3775 
3776     // The type of large pages has not been specified by the user.
3777 
3778     // Try UseHugeTLBFS and then UseSHM.
3779     UseHugeTLBFS = UseSHM = true;
3780 
3781     // Don't try UseTransparentHugePages since there are known
3782     // performance issues with it turned on. This might change in the future.
3783     UseTransparentHugePages = false;
3784   }
3785 
3786   if (UseTransparentHugePages) {
3787     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3788     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3789       UseHugeTLBFS = false;
3790       UseSHM = false;
3791       return true;
3792     }
3793     UseTransparentHugePages = false;
3794   }
3795 
3796   if (UseHugeTLBFS) {
3797     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3798     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3799       UseSHM = false;
3800       return true;
3801     }
3802     UseHugeTLBFS = false;
3803   }
3804 
3805   return UseSHM;
3806 }
3807 
3808 void os::large_page_init() {
3809   if (!UseLargePages &&
3810       !UseTransparentHugePages &&
3811       !UseHugeTLBFS &&
3812       !UseSHM) {
3813     // Not using large pages.
3814     return;
3815   }
3816 
3817   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3818     // The user explicitly turned off large pages.
3819     // Ignore the rest of the large pages flags.
3820     UseTransparentHugePages = false;
3821     UseHugeTLBFS = false;
3822     UseSHM = false;
3823     return;
3824   }
3825 
3826   size_t large_page_size = Linux::setup_large_page_size();
3827   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3828 
3829   set_coredump_filter(LARGEPAGES_BIT);
3830 }
3831 
3832 #ifndef SHM_HUGETLB
3833   #define SHM_HUGETLB 04000
3834 #endif
3835 
3836 #define shm_warning_format(format, ...)              \
3837   do {                                               \
3838     if (UseLargePages &&                             \
3839         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3840          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3841          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3842       warning(format, __VA_ARGS__);                  \
3843     }                                                \
3844   } while (0)
3845 
3846 #define shm_warning(str) shm_warning_format("%s", str)
3847 
3848 #define shm_warning_with_errno(str)                \
3849   do {                                             \
3850     int err = errno;                               \
3851     shm_warning_format(str " (error = %d)", err);  \
3852   } while (0)
3853 
3854 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3855   assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3856 
3857   if (!is_aligned(alignment, SHMLBA)) {
3858     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3859     return NULL;
3860   }
3861 
3862   // To ensure that we get 'alignment' aligned memory from shmat,
3863   // we pre-reserve aligned virtual memory and then attach to that.
3864 
3865   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3866   if (pre_reserved_addr == NULL) {
3867     // Couldn't pre-reserve aligned memory.
3868     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3869     return NULL;
3870   }
3871 
3872   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3873   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3874 
3875   if ((intptr_t)addr == -1) {
3876     int err = errno;
3877     shm_warning_with_errno("Failed to attach shared memory.");
3878 
3879     assert(err != EACCES, "Unexpected error");
3880     assert(err != EIDRM,  "Unexpected error");
3881     assert(err != EINVAL, "Unexpected error");
3882 
3883     // Since we don't know if the kernel unmapped the pre-reserved memory area
3884     // we can't unmap it, since that would potentially unmap memory that was
3885     // mapped from other threads.
3886     return NULL;
3887   }
3888 
3889   return addr;
3890 }
3891 
3892 static char* shmat_at_address(int shmid, char* req_addr) {
3893   if (!is_aligned(req_addr, SHMLBA)) {
3894     assert(false, "Requested address needs to be SHMLBA aligned");
3895     return NULL;
3896   }
3897 
3898   char* addr = (char*)shmat(shmid, req_addr, 0);
3899 
3900   if ((intptr_t)addr == -1) {
3901     shm_warning_with_errno("Failed to attach shared memory.");
3902     return NULL;
3903   }
3904 
3905   return addr;
3906 }
3907 
3908 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3909   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3910   if (req_addr != NULL) {
3911     assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3912     assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3913     return shmat_at_address(shmid, req_addr);
3914   }
3915 
3916   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3917   // return large page size aligned memory addresses when req_addr == NULL.
3918   // However, if the alignment is larger than the large page size, we have
3919   // to manually ensure that the memory returned is 'alignment' aligned.
3920   if (alignment > os::large_page_size()) {
3921     assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3922     return shmat_with_alignment(shmid, bytes, alignment);
3923   } else {
3924     return shmat_at_address(shmid, NULL);
3925   }
3926 }
3927 
3928 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3929                                             char* req_addr, bool exec) {
3930   // "exec" is passed in but not used.  Creating the shared image for
3931   // the code cache doesn't have an SHM_X executable permission to check.
3932   assert(UseLargePages && UseSHM, "only for SHM large pages");
3933   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3934   assert(is_aligned(req_addr, alignment), "Unaligned address");
3935 
3936   if (!is_aligned(bytes, os::large_page_size())) {
3937     return NULL; // Fallback to small pages.
3938   }
3939 
3940   // Create a large shared memory region to attach to based on size.
3941   // Currently, size is the total size of the heap.
3942   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3943   if (shmid == -1) {
3944     // Possible reasons for shmget failure:
3945     // 1. shmmax is too small for Java heap.
3946     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3947     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3948     // 2. not enough large page memory.
3949     //    > check available large pages: cat /proc/meminfo
3950     //    > increase amount of large pages:
3951     //          echo new_value > /proc/sys/vm/nr_hugepages
3952     //      Note 1: different Linux may use different name for this property,
3953     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3954     //      Note 2: it's possible there's enough physical memory available but
3955     //            they are so fragmented after a long run that they can't
3956     //            coalesce into large pages. Try to reserve large pages when
3957     //            the system is still "fresh".
3958     shm_warning_with_errno("Failed to reserve shared memory.");
3959     return NULL;
3960   }
3961 
3962   // Attach to the region.
3963   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3964 
3965   // Remove shmid. If shmat() is successful, the actual shared memory segment
3966   // will be deleted when it's detached by shmdt() or when the process
3967   // terminates. If shmat() is not successful this will remove the shared
3968   // segment immediately.
3969   shmctl(shmid, IPC_RMID, NULL);
3970 
3971   return addr;
3972 }
3973 
3974 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3975                                         int error) {
3976   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3977 
3978   bool warn_on_failure = UseLargePages &&
3979       (!FLAG_IS_DEFAULT(UseLargePages) ||
3980        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3981        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3982 
3983   if (warn_on_failure) {
3984     char msg[128];
3985     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3986                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3987     warning("%s", msg);
3988   }
3989 }
3990 
3991 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3992                                                         char* req_addr,
3993                                                         bool exec) {
3994   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3995   assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3996   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3997 
3998   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3999   char* addr = (char*)::mmap(req_addr, bytes, prot,
4000                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
4001                              -1, 0);
4002 
4003   if (addr == MAP_FAILED) {
4004     warn_on_large_pages_failure(req_addr, bytes, errno);
4005     return NULL;
4006   }
4007 
4008   assert(is_aligned(addr, os::large_page_size()), "Must be");
4009 
4010   return addr;
4011 }
4012 
4013 // Reserve memory using mmap(MAP_HUGETLB).
4014 //  - bytes shall be a multiple of alignment.
4015 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
4016 //  - alignment sets the alignment at which memory shall be allocated.
4017 //     It must be a multiple of allocation granularity.
4018 // Returns address of memory or NULL. If req_addr was not NULL, will only return
4019 //  req_addr or NULL.
4020 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
4021                                                          size_t alignment,
4022                                                          char* req_addr,
4023                                                          bool exec) {
4024   size_t large_page_size = os::large_page_size();
4025   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
4026 
4027   assert(is_aligned(req_addr, alignment), "Must be");
4028   assert(is_aligned(bytes, alignment), "Must be");
4029 
4030   // First reserve - but not commit - the address range in small pages.
4031   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
4032 
4033   if (start == NULL) {
4034     return NULL;
4035   }
4036 
4037   assert(is_aligned(start, alignment), "Must be");
4038 
4039   char* end = start + bytes;
4040 
4041   // Find the regions of the allocated chunk that can be promoted to large pages.
4042   char* lp_start = align_up(start, large_page_size);
4043   char* lp_end   = align_down(end, large_page_size);
4044 
4045   size_t lp_bytes = lp_end - lp_start;
4046 
4047   assert(is_aligned(lp_bytes, large_page_size), "Must be");
4048 
4049   if (lp_bytes == 0) {
4050     // The mapped region doesn't even span the start and the end of a large page.
4051     // Fall back to allocate a non-special area.
4052     ::munmap(start, end - start);
4053     return NULL;
4054   }
4055 
4056   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
4057 
4058   void* result;
4059 
4060   // Commit small-paged leading area.
4061   if (start != lp_start) {
4062     result = ::mmap(start, lp_start - start, prot,
4063                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4064                     -1, 0);
4065     if (result == MAP_FAILED) {
4066       ::munmap(lp_start, end - lp_start);
4067       return NULL;
4068     }
4069   }
4070 
4071   // Commit large-paged area.
4072   result = ::mmap(lp_start, lp_bytes, prot,
4073                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
4074                   -1, 0);
4075   if (result == MAP_FAILED) {
4076     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
4077     // If the mmap above fails, the large pages region will be unmapped and we
4078     // have regions before and after with small pages. Release these regions.
4079     //
4080     // |  mapped  |  unmapped  |  mapped  |
4081     // ^          ^            ^          ^
4082     // start      lp_start     lp_end     end
4083     //
4084     ::munmap(start, lp_start - start);
4085     ::munmap(lp_end, end - lp_end);
4086     return NULL;
4087   }
4088 
4089   // Commit small-paged trailing area.
4090   if (lp_end != end) {
4091     result = ::mmap(lp_end, end - lp_end, prot,
4092                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
4093                     -1, 0);
4094     if (result == MAP_FAILED) {
4095       ::munmap(start, lp_end - start);
4096       return NULL;
4097     }
4098   }
4099 
4100   return start;
4101 }
4102 
4103 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
4104                                                    size_t alignment,
4105                                                    char* req_addr,
4106                                                    bool exec) {
4107   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
4108   assert(is_aligned(req_addr, alignment), "Must be");
4109   assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
4110   assert(is_power_of_2(os::large_page_size()), "Must be");
4111   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
4112 
4113   if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
4114     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
4115   } else {
4116     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
4117   }
4118 }
4119 
4120 char* os::reserve_memory_special(size_t bytes, size_t alignment,
4121                                  char* req_addr, bool exec) {
4122   assert(UseLargePages, "only for large pages");
4123 
4124   char* addr;
4125   if (UseSHM) {
4126     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
4127   } else {
4128     assert(UseHugeTLBFS, "must be");
4129     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
4130   }
4131 
4132   if (addr != NULL) {
4133     if (UseNUMAInterleaving) {
4134       numa_make_global(addr, bytes);
4135     }
4136 
4137     // The memory is committed
4138     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
4139   }
4140 
4141   return addr;
4142 }
4143 
4144 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
4145   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
4146   return shmdt(base) == 0;
4147 }
4148 
4149 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
4150   return pd_release_memory(base, bytes);
4151 }
4152 
4153 bool os::release_memory_special(char* base, size_t bytes) {
4154   bool res;
4155   if (MemTracker::tracking_level() > NMT_minimal) {
4156     Tracker tkr(Tracker::release);
4157     res = os::Linux::release_memory_special_impl(base, bytes);
4158     if (res) {
4159       tkr.record((address)base, bytes);
4160     }
4161 
4162   } else {
4163     res = os::Linux::release_memory_special_impl(base, bytes);
4164   }
4165   return res;
4166 }
4167 
4168 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
4169   assert(UseLargePages, "only for large pages");
4170   bool res;
4171 
4172   if (UseSHM) {
4173     res = os::Linux::release_memory_special_shm(base, bytes);
4174   } else {
4175     assert(UseHugeTLBFS, "must be");
4176     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
4177   }
4178   return res;
4179 }
4180 
4181 size_t os::large_page_size() {
4182   return _large_page_size;
4183 }
4184 
4185 // With SysV SHM the entire memory region must be allocated as shared
4186 // memory.
4187 // HugeTLBFS allows application to commit large page memory on demand.
4188 // However, when committing memory with HugeTLBFS fails, the region
4189 // that was supposed to be committed will lose the old reservation
4190 // and allow other threads to steal that memory region. Because of this
4191 // behavior we can't commit HugeTLBFS memory.
4192 bool os::can_commit_large_page_memory() {
4193   return UseTransparentHugePages;
4194 }
4195 
4196 bool os::can_execute_large_page_memory() {
4197   return UseTransparentHugePages || UseHugeTLBFS;
4198 }
4199 
4200 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
4201   assert(file_desc >= 0, "file_desc is not valid");
4202   char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
4203   if (result != NULL) {
4204     if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
4205       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
4206     }
4207   }
4208   return result;
4209 }
4210 
4211 // Reserve memory at an arbitrary address, only if that area is
4212 // available (and not reserved for something else).
4213 
4214 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
4215   // Assert only that the size is a multiple of the page size, since
4216   // that's all that mmap requires, and since that's all we really know
4217   // about at this low abstraction level.  If we need higher alignment,
4218   // we can either pass an alignment to this method or verify alignment
4219   // in one of the methods further up the call chain.  See bug 5044738.
4220   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4221 
4222   // Repeatedly allocate blocks until the block is allocated at the
4223   // right spot.
4224 
4225   // Linux mmap allows caller to pass an address as hint; give it a try first,
4226   // if kernel honors the hint then we can return immediately.
4227   char * addr = anon_mmap(requested_addr, bytes, false);
4228   if (addr == requested_addr) {
4229     return requested_addr;
4230   }
4231 
4232   if (addr != NULL) {
4233     // mmap() is successful but it fails to reserve at the requested address
4234     anon_munmap(addr, bytes);
4235   }
4236 
4237   return NULL;
4238 }
4239 
4240 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4241 void os::infinite_sleep() {
4242   while (true) {    // sleep forever ...
4243     ::sleep(100);   // ... 100 seconds at a time
4244   }
4245 }
4246 
4247 // Used to convert frequent JVM_Yield() to nops
4248 bool os::dont_yield() {
4249   return DontYieldALot;
4250 }
4251 
4252 // Linux CFS scheduler (since 2.6.23) does not guarantee sched_yield(2) will
4253 // actually give up the CPU. Since skip buddy (v2.6.28):
4254 //
4255 // * Sets the yielding task as skip buddy for current CPU's run queue.
4256 // * Picks next from run queue, if empty, picks a skip buddy (can be the yielding task).
4257 // * Clears skip buddies for this run queue (yielding task no longer a skip buddy).
4258 //
4259 // An alternative is calling os::naked_short_nanosleep with a small number to avoid
4260 // getting re-scheduled immediately.
4261 //
4262 void os::naked_yield() {
4263   sched_yield();
4264 }
4265 
4266 ////////////////////////////////////////////////////////////////////////////////
4267 // thread priority support
4268 
4269 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4270 // only supports dynamic priority, static priority must be zero. For real-time
4271 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4272 // However, for large multi-threaded applications, SCHED_RR is not only slower
4273 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4274 // of 5 runs - Sep 2005).
4275 //
4276 // The following code actually changes the niceness of kernel-thread/LWP. It
4277 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4278 // not the entire user process, and user level threads are 1:1 mapped to kernel
4279 // threads. It has always been the case, but could change in the future. For
4280 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4281 // It is only used when ThreadPriorityPolicy=1 and may require system level permission
4282 // (e.g., root privilege or CAP_SYS_NICE capability).
4283 
4284 int os::java_to_os_priority[CriticalPriority + 1] = {
4285   19,              // 0 Entry should never be used
4286 
4287    4,              // 1 MinPriority
4288    3,              // 2
4289    2,              // 3
4290 
4291    1,              // 4
4292    0,              // 5 NormPriority
4293   -1,              // 6
4294 
4295   -2,              // 7
4296   -3,              // 8
4297   -4,              // 9 NearMaxPriority
4298 
4299   -5,              // 10 MaxPriority
4300 
4301   -5               // 11 CriticalPriority
4302 };
4303 
4304 static int prio_init() {
4305   if (ThreadPriorityPolicy == 1) {
4306     if (geteuid() != 0) {
4307       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4308         warning("-XX:ThreadPriorityPolicy=1 may require system level permission, " \
4309                 "e.g., being the root user. If the necessary permission is not " \
4310                 "possessed, changes to priority will be silently ignored.");
4311       }
4312     }
4313   }
4314   if (UseCriticalJavaThreadPriority) {
4315     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4316   }
4317   return 0;
4318 }
4319 
4320 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4321   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4322 
4323   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4324   return (ret == 0) ? OS_OK : OS_ERR;
4325 }
4326 
4327 OSReturn os::get_native_priority(const Thread* const thread,
4328                                  int *priority_ptr) {
4329   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4330     *priority_ptr = java_to_os_priority[NormPriority];
4331     return OS_OK;
4332   }
4333 
4334   errno = 0;
4335   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4336   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4337 }
4338 
4339 ////////////////////////////////////////////////////////////////////////////////
4340 // suspend/resume support
4341 
4342 //  The low-level signal-based suspend/resume support is a remnant from the
4343 //  old VM-suspension that used to be for java-suspension, safepoints etc,
4344 //  within hotspot. Currently used by JFR's OSThreadSampler
4345 //
4346 //  The remaining code is greatly simplified from the more general suspension
4347 //  code that used to be used.
4348 //
4349 //  The protocol is quite simple:
4350 //  - suspend:
4351 //      - sends a signal to the target thread
4352 //      - polls the suspend state of the osthread using a yield loop
4353 //      - target thread signal handler (SR_handler) sets suspend state
4354 //        and blocks in sigsuspend until continued
4355 //  - resume:
4356 //      - sets target osthread state to continue
4357 //      - sends signal to end the sigsuspend loop in the SR_handler
4358 //
4359 //  Note that the SR_lock plays no role in this suspend/resume protocol,
4360 //  but is checked for NULL in SR_handler as a thread termination indicator.
4361 //  The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4362 //
4363 //  Note that resume_clear_context() and suspend_save_context() are needed
4364 //  by SR_handler(), so that fetch_frame_from_ucontext() works,
4365 //  which in part is used by:
4366 //    - Forte Analyzer: AsyncGetCallTrace()
4367 //    - StackBanging: get_frame_at_stack_banging_point()
4368 
4369 static void resume_clear_context(OSThread *osthread) {
4370   osthread->set_ucontext(NULL);
4371   osthread->set_siginfo(NULL);
4372 }
4373 
4374 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4375                                  ucontext_t* context) {
4376   osthread->set_ucontext(context);
4377   osthread->set_siginfo(siginfo);
4378 }
4379 
4380 // Handler function invoked when a thread's execution is suspended or
4381 // resumed. We have to be careful that only async-safe functions are
4382 // called here (Note: most pthread functions are not async safe and
4383 // should be avoided.)
4384 //
4385 // Note: sigwait() is a more natural fit than sigsuspend() from an
4386 // interface point of view, but sigwait() prevents the signal hander
4387 // from being run. libpthread would get very confused by not having
4388 // its signal handlers run and prevents sigwait()'s use with the
4389 // mutex granting granting signal.
4390 //
4391 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4392 //
4393 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4394   // Save and restore errno to avoid confusing native code with EINTR
4395   // after sigsuspend.
4396   int old_errno = errno;
4397 
4398   Thread* thread = Thread::current_or_null_safe();
4399   assert(thread != NULL, "Missing current thread in SR_handler");
4400 
4401   // On some systems we have seen signal delivery get "stuck" until the signal
4402   // mask is changed as part of thread termination. Check that the current thread
4403   // has not already terminated (via SR_lock()) - else the following assertion
4404   // will fail because the thread is no longer a JavaThread as the ~JavaThread
4405   // destructor has completed.
4406 
4407   if (thread->SR_lock() == NULL) {
4408     return;
4409   }
4410 
4411   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4412 
4413   OSThread* osthread = thread->osthread();
4414 
4415   os::SuspendResume::State current = osthread->sr.state();
4416   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4417     suspend_save_context(osthread, siginfo, context);
4418 
4419     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4420     os::SuspendResume::State state = osthread->sr.suspended();
4421     if (state == os::SuspendResume::SR_SUSPENDED) {
4422       sigset_t suspend_set;  // signals for sigsuspend()
4423       sigemptyset(&suspend_set);
4424       // get current set of blocked signals and unblock resume signal
4425       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4426       sigdelset(&suspend_set, SR_signum);
4427 
4428       sr_semaphore.signal();
4429       // wait here until we are resumed
4430       while (1) {
4431         sigsuspend(&suspend_set);
4432 
4433         os::SuspendResume::State result = osthread->sr.running();
4434         if (result == os::SuspendResume::SR_RUNNING) {
4435           sr_semaphore.signal();
4436           break;
4437         }
4438       }
4439 
4440     } else if (state == os::SuspendResume::SR_RUNNING) {
4441       // request was cancelled, continue
4442     } else {
4443       ShouldNotReachHere();
4444     }
4445 
4446     resume_clear_context(osthread);
4447   } else if (current == os::SuspendResume::SR_RUNNING) {
4448     // request was cancelled, continue
4449   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4450     // ignore
4451   } else {
4452     // ignore
4453   }
4454 
4455   errno = old_errno;
4456 }
4457 
4458 static int SR_initialize() {
4459   struct sigaction act;
4460   char *s;
4461 
4462   // Get signal number to use for suspend/resume
4463   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4464     int sig = ::strtol(s, 0, 10);
4465     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4466         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4467       SR_signum = sig;
4468     } else {
4469       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4470               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4471     }
4472   }
4473 
4474   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4475          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4476 
4477   sigemptyset(&SR_sigset);
4478   sigaddset(&SR_sigset, SR_signum);
4479 
4480   // Set up signal handler for suspend/resume
4481   act.sa_flags = SA_RESTART|SA_SIGINFO;
4482   act.sa_handler = (void (*)(int)) SR_handler;
4483 
4484   // SR_signum is blocked by default.
4485   // 4528190 - We also need to block pthread restart signal (32 on all
4486   // supported Linux platforms). Note that LinuxThreads need to block
4487   // this signal for all threads to work properly. So we don't have
4488   // to use hard-coded signal number when setting up the mask.
4489   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4490 
4491   if (sigaction(SR_signum, &act, 0) == -1) {
4492     return -1;
4493   }
4494 
4495   // Save signal flag
4496   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4497   return 0;
4498 }
4499 
4500 static int sr_notify(OSThread* osthread) {
4501   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4502   assert_status(status == 0, status, "pthread_kill");
4503   return status;
4504 }
4505 
4506 // "Randomly" selected value for how long we want to spin
4507 // before bailing out on suspending a thread, also how often
4508 // we send a signal to a thread we want to resume
4509 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4510 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4511 
4512 // returns true on success and false on error - really an error is fatal
4513 // but this seems the normal response to library errors
4514 static bool do_suspend(OSThread* osthread) {
4515   assert(osthread->sr.is_running(), "thread should be running");
4516   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4517 
4518   // mark as suspended and send signal
4519   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4520     // failed to switch, state wasn't running?
4521     ShouldNotReachHere();
4522     return false;
4523   }
4524 
4525   if (sr_notify(osthread) != 0) {
4526     ShouldNotReachHere();
4527   }
4528 
4529   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4530   while (true) {
4531     if (sr_semaphore.timedwait(2)) {
4532       break;
4533     } else {
4534       // timeout
4535       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4536       if (cancelled == os::SuspendResume::SR_RUNNING) {
4537         return false;
4538       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4539         // make sure that we consume the signal on the semaphore as well
4540         sr_semaphore.wait();
4541         break;
4542       } else {
4543         ShouldNotReachHere();
4544         return false;
4545       }
4546     }
4547   }
4548 
4549   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4550   return true;
4551 }
4552 
4553 static void do_resume(OSThread* osthread) {
4554   assert(osthread->sr.is_suspended(), "thread should be suspended");
4555   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4556 
4557   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4558     // failed to switch to WAKEUP_REQUEST
4559     ShouldNotReachHere();
4560     return;
4561   }
4562 
4563   while (true) {
4564     if (sr_notify(osthread) == 0) {
4565       if (sr_semaphore.timedwait(2)) {
4566         if (osthread->sr.is_running()) {
4567           return;
4568         }
4569       }
4570     } else {
4571       ShouldNotReachHere();
4572     }
4573   }
4574 
4575   guarantee(osthread->sr.is_running(), "Must be running!");
4576 }
4577 
4578 ///////////////////////////////////////////////////////////////////////////////////
4579 // signal handling (except suspend/resume)
4580 
4581 // This routine may be used by user applications as a "hook" to catch signals.
4582 // The user-defined signal handler must pass unrecognized signals to this
4583 // routine, and if it returns true (non-zero), then the signal handler must
4584 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4585 // routine will never retun false (zero), but instead will execute a VM panic
4586 // routine kill the process.
4587 //
4588 // If this routine returns false, it is OK to call it again.  This allows
4589 // the user-defined signal handler to perform checks either before or after
4590 // the VM performs its own checks.  Naturally, the user code would be making
4591 // a serious error if it tried to handle an exception (such as a null check
4592 // or breakpoint) that the VM was generating for its own correct operation.
4593 //
4594 // This routine may recognize any of the following kinds of signals:
4595 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4596 // It should be consulted by handlers for any of those signals.
4597 //
4598 // The caller of this routine must pass in the three arguments supplied
4599 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4600 // field of the structure passed to sigaction().  This routine assumes that
4601 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4602 //
4603 // Note that the VM will print warnings if it detects conflicting signal
4604 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4605 //
4606 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4607                                                  siginfo_t* siginfo,
4608                                                  void* ucontext,
4609                                                  int abort_if_unrecognized);
4610 
4611 static void signalHandler(int sig, siginfo_t* info, void* uc) {
4612   assert(info != NULL && uc != NULL, "it must be old kernel");
4613   int orig_errno = errno;  // Preserve errno value over signal handler.
4614   JVM_handle_linux_signal(sig, info, uc, true);
4615   errno = orig_errno;
4616 }
4617 
4618 
4619 // This boolean allows users to forward their own non-matching signals
4620 // to JVM_handle_linux_signal, harmlessly.
4621 bool os::Linux::signal_handlers_are_installed = false;
4622 
4623 // For signal-chaining
4624 bool os::Linux::libjsig_is_loaded = false;
4625 typedef struct sigaction *(*get_signal_t)(int);
4626 get_signal_t os::Linux::get_signal_action = NULL;
4627 
4628 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4629   struct sigaction *actp = NULL;
4630 
4631   if (libjsig_is_loaded) {
4632     // Retrieve the old signal handler from libjsig
4633     actp = (*get_signal_action)(sig);
4634   }
4635   if (actp == NULL) {
4636     // Retrieve the preinstalled signal handler from jvm
4637     actp = os::Posix::get_preinstalled_handler(sig);
4638   }
4639 
4640   return actp;
4641 }
4642 
4643 static bool call_chained_handler(struct sigaction *actp, int sig,
4644                                  siginfo_t *siginfo, void *context) {
4645   // Call the old signal handler
4646   if (actp->sa_handler == SIG_DFL) {
4647     // It's more reasonable to let jvm treat it as an unexpected exception
4648     // instead of taking the default action.
4649     return false;
4650   } else if (actp->sa_handler != SIG_IGN) {
4651     if ((actp->sa_flags & SA_NODEFER) == 0) {
4652       // automaticlly block the signal
4653       sigaddset(&(actp->sa_mask), sig);
4654     }
4655 
4656     sa_handler_t hand = NULL;
4657     sa_sigaction_t sa = NULL;
4658     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4659     // retrieve the chained handler
4660     if (siginfo_flag_set) {
4661       sa = actp->sa_sigaction;
4662     } else {
4663       hand = actp->sa_handler;
4664     }
4665 
4666     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4667       actp->sa_handler = SIG_DFL;
4668     }
4669 
4670     // try to honor the signal mask
4671     sigset_t oset;
4672     sigemptyset(&oset);
4673     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4674 
4675     // call into the chained handler
4676     if (siginfo_flag_set) {
4677       (*sa)(sig, siginfo, context);
4678     } else {
4679       (*hand)(sig);
4680     }
4681 
4682     // restore the signal mask
4683     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4684   }
4685   // Tell jvm's signal handler the signal is taken care of.
4686   return true;
4687 }
4688 
4689 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4690   bool chained = false;
4691   // signal-chaining
4692   if (UseSignalChaining) {
4693     struct sigaction *actp = get_chained_signal_action(sig);
4694     if (actp != NULL) {
4695       chained = call_chained_handler(actp, sig, siginfo, context);
4696     }
4697   }
4698   return chained;
4699 }
4700 
4701 // for diagnostic
4702 int sigflags[NSIG];
4703 
4704 int os::Linux::get_our_sigflags(int sig) {
4705   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4706   return sigflags[sig];
4707 }
4708 
4709 void os::Linux::set_our_sigflags(int sig, int flags) {
4710   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4711   if (sig > 0 && sig < NSIG) {
4712     sigflags[sig] = flags;
4713   }
4714 }
4715 
4716 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4717   // Check for overwrite.
4718   struct sigaction oldAct;
4719   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4720 
4721   void* oldhand = oldAct.sa_sigaction
4722                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4723                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4724   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4725       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4726       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4727     if (AllowUserSignalHandlers || !set_installed) {
4728       // Do not overwrite; user takes responsibility to forward to us.
4729       return;
4730     } else if (UseSignalChaining) {
4731       // save the old handler in jvm
4732       os::Posix::save_preinstalled_handler(sig, oldAct);
4733       // libjsig also interposes the sigaction() call below and saves the
4734       // old sigaction on it own.
4735     } else {
4736       fatal("Encountered unexpected pre-existing sigaction handler "
4737             "%#lx for signal %d.", (long)oldhand, sig);
4738     }
4739   }
4740 
4741   struct sigaction sigAct;
4742   sigfillset(&(sigAct.sa_mask));
4743   sigAct.sa_handler = SIG_DFL;
4744   if (!set_installed) {
4745     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4746   } else {
4747     sigAct.sa_sigaction = signalHandler;
4748     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4749   }
4750   // Save flags, which are set by ours
4751   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4752   sigflags[sig] = sigAct.sa_flags;
4753 
4754   int ret = sigaction(sig, &sigAct, &oldAct);
4755   assert(ret == 0, "check");
4756 
4757   void* oldhand2  = oldAct.sa_sigaction
4758                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4759                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4760   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4761 }
4762 
4763 // install signal handlers for signals that HotSpot needs to
4764 // handle in order to support Java-level exception handling.
4765 
4766 void os::Linux::install_signal_handlers() {
4767   if (!signal_handlers_are_installed) {
4768     signal_handlers_are_installed = true;
4769 
4770     // signal-chaining
4771     typedef void (*signal_setting_t)();
4772     signal_setting_t begin_signal_setting = NULL;
4773     signal_setting_t end_signal_setting = NULL;
4774     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4775                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4776     if (begin_signal_setting != NULL) {
4777       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4778                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4779       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4780                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4781       libjsig_is_loaded = true;
4782       assert(UseSignalChaining, "should enable signal-chaining");
4783     }
4784     if (libjsig_is_loaded) {
4785       // Tell libjsig jvm is setting signal handlers
4786       (*begin_signal_setting)();
4787     }
4788 
4789     set_signal_handler(SIGSEGV, true);
4790     set_signal_handler(SIGPIPE, true);
4791     set_signal_handler(SIGBUS, true);
4792     set_signal_handler(SIGILL, true);
4793     set_signal_handler(SIGFPE, true);
4794 #if defined(PPC64)
4795     set_signal_handler(SIGTRAP, true);
4796 #endif
4797     set_signal_handler(SIGXFSZ, true);
4798 
4799     if (libjsig_is_loaded) {
4800       // Tell libjsig jvm finishes setting signal handlers
4801       (*end_signal_setting)();
4802     }
4803 
4804     // We don't activate signal checker if libjsig is in place, we trust ourselves
4805     // and if UserSignalHandler is installed all bets are off.
4806     // Log that signal checking is off only if -verbose:jni is specified.
4807     if (CheckJNICalls) {
4808       if (libjsig_is_loaded) {
4809         if (PrintJNIResolving) {
4810           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4811         }
4812         check_signals = false;
4813       }
4814       if (AllowUserSignalHandlers) {
4815         if (PrintJNIResolving) {
4816           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4817         }
4818         check_signals = false;
4819       }
4820     }
4821   }
4822 }
4823 
4824 // This is the fastest way to get thread cpu time on Linux.
4825 // Returns cpu time (user+sys) for any thread, not only for current.
4826 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4827 // It might work on 2.6.10+ with a special kernel/glibc patch.
4828 // For reference, please, see IEEE Std 1003.1-2004:
4829 //   http://www.unix.org/single_unix_specification
4830 
4831 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4832   struct timespec tp;
4833   int rc = os::Posix::clock_gettime(clockid, &tp);
4834   assert(rc == 0, "clock_gettime is expected to return 0 code");
4835 
4836   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4837 }
4838 
4839 /////
4840 // glibc on Linux platform uses non-documented flag
4841 // to indicate, that some special sort of signal
4842 // trampoline is used.
4843 // We will never set this flag, and we should
4844 // ignore this flag in our diagnostic
4845 #ifdef SIGNIFICANT_SIGNAL_MASK
4846   #undef SIGNIFICANT_SIGNAL_MASK
4847 #endif
4848 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4849 
4850 static const char* get_signal_handler_name(address handler,
4851                                            char* buf, int buflen) {
4852   int offset = 0;
4853   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4854   if (found) {
4855     // skip directory names
4856     const char *p1, *p2;
4857     p1 = buf;
4858     size_t len = strlen(os::file_separator());
4859     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4860     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4861   } else {
4862     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4863   }
4864   return buf;
4865 }
4866 
4867 static void print_signal_handler(outputStream* st, int sig,
4868                                  char* buf, size_t buflen) {
4869   struct sigaction sa;
4870 
4871   sigaction(sig, NULL, &sa);
4872 
4873   // See comment for SIGNIFICANT_SIGNAL_MASK define
4874   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4875 
4876   st->print("%s: ", os::exception_name(sig, buf, buflen));
4877 
4878   address handler = (sa.sa_flags & SA_SIGINFO)
4879     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4880     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4881 
4882   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4883     st->print("SIG_DFL");
4884   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4885     st->print("SIG_IGN");
4886   } else {
4887     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4888   }
4889 
4890   st->print(", sa_mask[0]=");
4891   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4892 
4893   address rh = VMError::get_resetted_sighandler(sig);
4894   // May be, handler was resetted by VMError?
4895   if (rh != NULL) {
4896     handler = rh;
4897     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4898   }
4899 
4900   st->print(", sa_flags=");
4901   os::Posix::print_sa_flags(st, sa.sa_flags);
4902 
4903   // Check: is it our handler?
4904   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4905       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4906     // It is our signal handler
4907     // check for flags, reset system-used one!
4908     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4909       st->print(
4910                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4911                 os::Linux::get_our_sigflags(sig));
4912     }
4913   }
4914   st->cr();
4915 }
4916 
4917 
4918 #define DO_SIGNAL_CHECK(sig)                      \
4919   do {                                            \
4920     if (!sigismember(&check_signal_done, sig)) {  \
4921       os::Linux::check_signal_handler(sig);       \
4922     }                                             \
4923   } while (0)
4924 
4925 // This method is a periodic task to check for misbehaving JNI applications
4926 // under CheckJNI, we can add any periodic checks here
4927 
4928 void os::run_periodic_checks() {
4929   if (check_signals == false) return;
4930 
4931   // SEGV and BUS if overridden could potentially prevent
4932   // generation of hs*.log in the event of a crash, debugging
4933   // such a case can be very challenging, so we absolutely
4934   // check the following for a good measure:
4935   DO_SIGNAL_CHECK(SIGSEGV);
4936   DO_SIGNAL_CHECK(SIGILL);
4937   DO_SIGNAL_CHECK(SIGFPE);
4938   DO_SIGNAL_CHECK(SIGBUS);
4939   DO_SIGNAL_CHECK(SIGPIPE);
4940   DO_SIGNAL_CHECK(SIGXFSZ);
4941 #if defined(PPC64)
4942   DO_SIGNAL_CHECK(SIGTRAP);
4943 #endif
4944 
4945   // ReduceSignalUsage allows the user to override these handlers
4946   // see comments at the very top and jvm_md.h
4947   if (!ReduceSignalUsage) {
4948     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4949     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4950     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4951     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4952   }
4953 
4954   DO_SIGNAL_CHECK(SR_signum);
4955 }
4956 
4957 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4958 
4959 static os_sigaction_t os_sigaction = NULL;
4960 
4961 void os::Linux::check_signal_handler(int sig) {
4962   char buf[O_BUFLEN];
4963   address jvmHandler = NULL;
4964 
4965 
4966   struct sigaction act;
4967   if (os_sigaction == NULL) {
4968     // only trust the default sigaction, in case it has been interposed
4969     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4970     if (os_sigaction == NULL) return;
4971   }
4972 
4973   os_sigaction(sig, (struct sigaction*)NULL, &act);
4974 
4975 
4976   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4977 
4978   address thisHandler = (act.sa_flags & SA_SIGINFO)
4979     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4980     : CAST_FROM_FN_PTR(address, act.sa_handler);
4981 
4982 
4983   switch (sig) {
4984   case SIGSEGV:
4985   case SIGBUS:
4986   case SIGFPE:
4987   case SIGPIPE:
4988   case SIGILL:
4989   case SIGXFSZ:
4990     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4991     break;
4992 
4993   case SHUTDOWN1_SIGNAL:
4994   case SHUTDOWN2_SIGNAL:
4995   case SHUTDOWN3_SIGNAL:
4996   case BREAK_SIGNAL:
4997     jvmHandler = (address)user_handler();
4998     break;
4999 
5000   default:
5001     if (sig == SR_signum) {
5002       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
5003     } else {
5004       return;
5005     }
5006     break;
5007   }
5008 
5009   if (thisHandler != jvmHandler) {
5010     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
5011     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
5012     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
5013     // No need to check this sig any longer
5014     sigaddset(&check_signal_done, sig);
5015     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
5016     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
5017       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
5018                     exception_name(sig, buf, O_BUFLEN));
5019     }
5020   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
5021     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
5022     tty->print("expected:");
5023     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
5024     tty->cr();
5025     tty->print("  found:");
5026     os::Posix::print_sa_flags(tty, act.sa_flags);
5027     tty->cr();
5028     // No need to check this sig any longer
5029     sigaddset(&check_signal_done, sig);
5030   }
5031 
5032   // Dump all the signal
5033   if (sigismember(&check_signal_done, sig)) {
5034     print_signal_handlers(tty, buf, O_BUFLEN);
5035   }
5036 }
5037 
5038 extern void report_error(char* file_name, int line_no, char* title,
5039                          char* format, ...);
5040 
5041 // this is called _before_ most of the global arguments have been parsed
5042 void os::init(void) {
5043   char dummy;   // used to get a guess on initial stack address
5044 
5045   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5046 
5047   init_random(1234567);
5048 
5049   Linux::set_page_size(sysconf(_SC_PAGESIZE));
5050   if (Linux::page_size() == -1) {
5051     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
5052           os::strerror(errno));
5053   }
5054   init_page_sizes((size_t) Linux::page_size());
5055 
5056   Linux::initialize_system_info();
5057 
5058   os::Linux::CPUPerfTicks pticks;
5059   bool res = os::Linux::get_tick_information(&pticks, -1);
5060 
5061   if (res && pticks.has_steal_ticks) {
5062     has_initial_tick_info = true;
5063     initial_total_ticks = pticks.total;
5064     initial_steal_ticks = pticks.steal;
5065   }
5066 
5067   // _main_thread points to the thread that created/loaded the JVM.
5068   Linux::_main_thread = pthread_self();
5069 
5070   // retrieve entry point for pthread_setname_np
5071   Linux::_pthread_setname_np =
5072     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5073 
5074   os::Posix::init();
5075 
5076   initial_time_count = javaTimeNanos();
5077 
5078   // Always warn if no monotonic clock available
5079   if (!os::Posix::supports_monotonic_clock()) {
5080     warning("No monotonic clock was available - timed services may "    \
5081             "be adversely affected if the time-of-day clock changes");
5082   }
5083 }
5084 
5085 // To install functions for atexit system call
5086 extern "C" {
5087   static void perfMemory_exit_helper() {
5088     perfMemory_exit();
5089   }
5090 }
5091 
5092 void os::pd_init_container_support() {
5093   OSContainer::init();
5094 }
5095 
5096 void os::Linux::numa_init() {
5097 
5098   // Java can be invoked as
5099   // 1. Without numactl and heap will be allocated/configured on all nodes as
5100   //    per the system policy.
5101   // 2. With numactl --interleave:
5102   //      Use numa_get_interleave_mask(v2) API to get nodes bitmask. The same
5103   //      API for membind case bitmask is reset.
5104   //      Interleave is only hint and Kernel can fallback to other nodes if
5105   //      no memory is available on the target nodes.
5106   // 3. With numactl --membind:
5107   //      Use numa_get_membind(v2) API to get nodes bitmask. The same API for
5108   //      interleave case returns bitmask of all nodes.
5109   // numa_all_nodes_ptr holds bitmask of all nodes.
5110   // numa_get_interleave_mask(v2) and numa_get_membind(v2) APIs returns correct
5111   // bitmask when externally configured to run on all or fewer nodes.
5112 
5113   if (!Linux::libnuma_init()) {
5114     UseNUMA = false;
5115   } else {
5116     if ((Linux::numa_max_node() < 1) || Linux::is_bound_to_single_node()) {
5117       // If there's only one node (they start from 0) or if the process
5118       // is bound explicitly to a single node using membind, disable NUMA.
5119       UseNUMA = false;
5120     } else {
5121 
5122       LogTarget(Info,os) log;
5123       LogStream ls(log);
5124 
5125       Linux::set_configured_numa_policy(Linux::identify_numa_policy());
5126 
5127       struct bitmask* bmp = Linux::_numa_membind_bitmask;
5128       const char* numa_mode = "membind";
5129 
5130       if (Linux::is_running_in_interleave_mode()) {
5131         bmp = Linux::_numa_interleave_bitmask;
5132         numa_mode = "interleave";
5133       }
5134 
5135       ls.print("UseNUMA is enabled and invoked in '%s' mode."
5136                " Heap will be configured using NUMA memory nodes:", numa_mode);
5137 
5138       for (int node = 0; node <= Linux::numa_max_node(); node++) {
5139         if (Linux::_numa_bitmask_isbitset(bmp, node)) {
5140           ls.print(" %d", node);
5141         }
5142       }
5143     }
5144   }
5145 
5146   if (UseParallelGC && UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5147     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5148     // we can make the adaptive lgrp chunk resizing work. If the user specified both
5149     // UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn
5150     // and disable adaptive resizing.
5151     if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
5152       warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, "
5153               "disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
5154       UseAdaptiveSizePolicy = false;
5155       UseAdaptiveNUMAChunkSizing = false;
5156     }
5157   }
5158 
5159   if (!UseNUMA && ForceNUMA) {
5160     UseNUMA = true;
5161   }
5162 }
5163 
5164 // this is called _after_ the global arguments have been parsed
5165 jint os::init_2(void) {
5166 
5167   // This could be set after os::Posix::init() but all platforms
5168   // have to set it the same so we have to mirror Solaris.
5169   DEBUG_ONLY(os::set_mutex_init_done();)
5170 
5171   os::Posix::init_2();
5172 
5173   Linux::fast_thread_clock_init();
5174 
5175   // initialize suspend/resume support - must do this before signal_sets_init()
5176   if (SR_initialize() != 0) {
5177     perror("SR_initialize failed");
5178     return JNI_ERR;
5179   }
5180 
5181   Linux::signal_sets_init();
5182   Linux::install_signal_handlers();
5183   // Initialize data for jdk.internal.misc.Signal
5184   if (!ReduceSignalUsage) {
5185     jdk_misc_signal_init();
5186   }
5187 
5188   if (AdjustStackSizeForTLS) {
5189     get_minstack_init();
5190   }
5191 
5192   // Check and sets minimum stack sizes against command line options
5193   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
5194     return JNI_ERR;
5195   }
5196 
5197 #if defined(IA32)
5198   // Need to ensure we've determined the process's initial stack to
5199   // perform the workaround
5200   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5201   workaround_expand_exec_shield_cs_limit();
5202 #else
5203   suppress_primordial_thread_resolution = Arguments::created_by_java_launcher();
5204   if (!suppress_primordial_thread_resolution) {
5205     Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5206   }
5207 #endif
5208 
5209   Linux::libpthread_init();
5210   Linux::sched_getcpu_init();
5211   log_info(os)("HotSpot is running with %s, %s",
5212                Linux::glibc_version(), Linux::libpthread_version());
5213 
5214   if (UseNUMA) {
5215     Linux::numa_init();
5216   }
5217 
5218   if (MaxFDLimit) {
5219     // set the number of file descriptors to max. print out error
5220     // if getrlimit/setrlimit fails but continue regardless.
5221     struct rlimit nbr_files;
5222     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5223     if (status != 0) {
5224       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
5225     } else {
5226       nbr_files.rlim_cur = nbr_files.rlim_max;
5227       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5228       if (status != 0) {
5229         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
5230       }
5231     }
5232   }
5233 
5234   // at-exit methods are called in the reverse order of their registration.
5235   // atexit functions are called on return from main or as a result of a
5236   // call to exit(3C). There can be only 32 of these functions registered
5237   // and atexit() does not set errno.
5238 
5239   if (PerfAllowAtExitRegistration) {
5240     // only register atexit functions if PerfAllowAtExitRegistration is set.
5241     // atexit functions can be delayed until process exit time, which
5242     // can be problematic for embedded VM situations. Embedded VMs should
5243     // call DestroyJavaVM() to assure that VM resources are released.
5244 
5245     // note: perfMemory_exit_helper atexit function may be removed in
5246     // the future if the appropriate cleanup code can be added to the
5247     // VM_Exit VMOperation's doit method.
5248     if (atexit(perfMemory_exit_helper) != 0) {
5249       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5250     }
5251   }
5252 
5253   // initialize thread priority policy
5254   prio_init();
5255 
5256   if (!FLAG_IS_DEFAULT(AllocateHeapAt) || !FLAG_IS_DEFAULT(AllocateOldGenAt)) {
5257     set_coredump_filter(DAX_SHARED_BIT);
5258   }
5259 
5260   if (DumpPrivateMappingsInCore) {
5261     set_coredump_filter(FILE_BACKED_PVT_BIT);
5262   }
5263 
5264   if (DumpSharedMappingsInCore) {
5265     set_coredump_filter(FILE_BACKED_SHARED_BIT);
5266   }
5267 
5268   return JNI_OK;
5269 }
5270 
5271 // Mark the polling page as unreadable
5272 void os::make_polling_page_unreadable(void) {
5273   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5274     fatal("Could not disable polling page");
5275   }
5276 }
5277 
5278 // Mark the polling page as readable
5279 void os::make_polling_page_readable(void) {
5280   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5281     fatal("Could not enable polling page");
5282   }
5283 }
5284 
5285 // older glibc versions don't have this macro (which expands to
5286 // an optimized bit-counting function) so we have to roll our own
5287 #ifndef CPU_COUNT
5288 
5289 static int _cpu_count(const cpu_set_t* cpus) {
5290   int count = 0;
5291   // only look up to the number of configured processors
5292   for (int i = 0; i < os::processor_count(); i++) {
5293     if (CPU_ISSET(i, cpus)) {
5294       count++;
5295     }
5296   }
5297   return count;
5298 }
5299 
5300 #define CPU_COUNT(cpus) _cpu_count(cpus)
5301 
5302 #endif // CPU_COUNT
5303 
5304 // Get the current number of available processors for this process.
5305 // This value can change at any time during a process's lifetime.
5306 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5307 // If it appears there may be more than 1024 processors then we do a
5308 // dynamic check - see 6515172 for details.
5309 // If anything goes wrong we fallback to returning the number of online
5310 // processors - which can be greater than the number available to the process.
5311 int os::Linux::active_processor_count() {
5312   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
5313   cpu_set_t* cpus_p = &cpus;
5314   int cpus_size = sizeof(cpu_set_t);
5315 
5316   int configured_cpus = os::processor_count();  // upper bound on available cpus
5317   int cpu_count = 0;
5318 
5319 // old build platforms may not support dynamic cpu sets
5320 #ifdef CPU_ALLOC
5321 
5322   // To enable easy testing of the dynamic path on different platforms we
5323   // introduce a diagnostic flag: UseCpuAllocPath
5324   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5325     // kernel may use a mask bigger than cpu_set_t
5326     log_trace(os)("active_processor_count: using dynamic path %s"
5327                   "- configured processors: %d",
5328                   UseCpuAllocPath ? "(forced) " : "",
5329                   configured_cpus);
5330     cpus_p = CPU_ALLOC(configured_cpus);
5331     if (cpus_p != NULL) {
5332       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5333       // zero it just to be safe
5334       CPU_ZERO_S(cpus_size, cpus_p);
5335     }
5336     else {
5337        // failed to allocate so fallback to online cpus
5338        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5339        log_trace(os)("active_processor_count: "
5340                      "CPU_ALLOC failed (%s) - using "
5341                      "online processor count: %d",
5342                      os::strerror(errno), online_cpus);
5343        return online_cpus;
5344     }
5345   }
5346   else {
5347     log_trace(os)("active_processor_count: using static path - configured processors: %d",
5348                   configured_cpus);
5349   }
5350 #else // CPU_ALLOC
5351 // these stubs won't be executed
5352 #define CPU_COUNT_S(size, cpus) -1
5353 #define CPU_FREE(cpus)
5354 
5355   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5356                 configured_cpus);
5357 #endif // CPU_ALLOC
5358 
5359   // pid 0 means the current thread - which we have to assume represents the process
5360   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5361     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5362       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5363     }
5364     else {
5365       cpu_count = CPU_COUNT(cpus_p);
5366     }
5367     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5368   }
5369   else {
5370     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5371     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5372             "which may exceed available processors", os::strerror(errno), cpu_count);
5373   }
5374 
5375   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5376     CPU_FREE(cpus_p);
5377   }
5378 
5379   assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5380   return cpu_count;
5381 }
5382 
5383 // Determine the active processor count from one of
5384 // three different sources:
5385 //
5386 // 1. User option -XX:ActiveProcessorCount
5387 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5388 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5389 //
5390 // Option 1, if specified, will always override.
5391 // If the cgroup subsystem is active and configured, we
5392 // will return the min of the cgroup and option 2 results.
5393 // This is required since tools, such as numactl, that
5394 // alter cpu affinity do not update cgroup subsystem
5395 // cpuset configuration files.
5396 int os::active_processor_count() {
5397   // User has overridden the number of active processors
5398   if (ActiveProcessorCount > 0) {
5399     log_trace(os)("active_processor_count: "
5400                   "active processor count set by user : %d",
5401                   ActiveProcessorCount);
5402     return ActiveProcessorCount;
5403   }
5404 
5405   int active_cpus;
5406   if (OSContainer::is_containerized()) {
5407     active_cpus = OSContainer::active_processor_count();
5408     log_trace(os)("active_processor_count: determined by OSContainer: %d",
5409                    active_cpus);
5410   } else {
5411     active_cpus = os::Linux::active_processor_count();
5412   }
5413 
5414   return active_cpus;
5415 }
5416 
5417 uint os::processor_id() {
5418   const int id = Linux::sched_getcpu();
5419   assert(id >= 0 && id < _processor_count, "Invalid processor id");
5420   return (uint)id;
5421 }
5422 
5423 void os::set_native_thread_name(const char *name) {
5424   if (Linux::_pthread_setname_np) {
5425     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5426     snprintf(buf, sizeof(buf), "%s", name);
5427     buf[sizeof(buf) - 1] = '\0';
5428     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5429     // ERANGE should not happen; all other errors should just be ignored.
5430     assert(rc != ERANGE, "pthread_setname_np failed");
5431   }
5432 }
5433 
5434 bool os::bind_to_processor(uint processor_id) {
5435   // Not yet implemented.
5436   return false;
5437 }
5438 
5439 ///
5440 
5441 void os::SuspendedThreadTask::internal_do_task() {
5442   if (do_suspend(_thread->osthread())) {
5443     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5444     do_task(context);
5445     do_resume(_thread->osthread());
5446   }
5447 }
5448 
5449 ////////////////////////////////////////////////////////////////////////////////
5450 // debug support
5451 
5452 bool os::find(address addr, outputStream* st) {
5453   Dl_info dlinfo;
5454   memset(&dlinfo, 0, sizeof(dlinfo));
5455   if (dladdr(addr, &dlinfo) != 0) {
5456     st->print(PTR_FORMAT ": ", p2i(addr));
5457     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5458       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5459                 p2i(addr) - p2i(dlinfo.dli_saddr));
5460     } else if (dlinfo.dli_fbase != NULL) {
5461       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5462     } else {
5463       st->print("<absolute address>");
5464     }
5465     if (dlinfo.dli_fname != NULL) {
5466       st->print(" in %s", dlinfo.dli_fname);
5467     }
5468     if (dlinfo.dli_fbase != NULL) {
5469       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5470     }
5471     st->cr();
5472 
5473     if (Verbose) {
5474       // decode some bytes around the PC
5475       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5476       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5477       address       lowest = (address) dlinfo.dli_sname;
5478       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5479       if (begin < lowest)  begin = lowest;
5480       Dl_info dlinfo2;
5481       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5482           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5483         end = (address) dlinfo2.dli_saddr;
5484       }
5485       Disassembler::decode(begin, end, st);
5486     }
5487     return true;
5488   }
5489   return false;
5490 }
5491 
5492 ////////////////////////////////////////////////////////////////////////////////
5493 // misc
5494 
5495 // This does not do anything on Linux. This is basically a hook for being
5496 // able to use structured exception handling (thread-local exception filters)
5497 // on, e.g., Win32.
5498 void
5499 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5500                          JavaCallArguments* args, Thread* thread) {
5501   f(value, method, args, thread);
5502 }
5503 
5504 void os::print_statistics() {
5505 }
5506 
5507 bool os::message_box(const char* title, const char* message) {
5508   int i;
5509   fdStream err(defaultStream::error_fd());
5510   for (i = 0; i < 78; i++) err.print_raw("=");
5511   err.cr();
5512   err.print_raw_cr(title);
5513   for (i = 0; i < 78; i++) err.print_raw("-");
5514   err.cr();
5515   err.print_raw_cr(message);
5516   for (i = 0; i < 78; i++) err.print_raw("=");
5517   err.cr();
5518 
5519   char buf[16];
5520   // Prevent process from exiting upon "read error" without consuming all CPU
5521   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5522 
5523   return buf[0] == 'y' || buf[0] == 'Y';
5524 }
5525 
5526 // Is a (classpath) directory empty?
5527 bool os::dir_is_empty(const char* path) {
5528   DIR *dir = NULL;
5529   struct dirent *ptr;
5530 
5531   dir = opendir(path);
5532   if (dir == NULL) return true;
5533 
5534   // Scan the directory
5535   bool result = true;
5536   while (result && (ptr = readdir(dir)) != NULL) {
5537     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5538       result = false;
5539     }
5540   }
5541   closedir(dir);
5542   return result;
5543 }
5544 
5545 // This code originates from JDK's sysOpen and open64_w
5546 // from src/solaris/hpi/src/system_md.c
5547 
5548 int os::open(const char *path, int oflag, int mode) {
5549   if (strlen(path) > MAX_PATH - 1) {
5550     errno = ENAMETOOLONG;
5551     return -1;
5552   }
5553 
5554   // All file descriptors that are opened in the Java process and not
5555   // specifically destined for a subprocess should have the close-on-exec
5556   // flag set.  If we don't set it, then careless 3rd party native code
5557   // might fork and exec without closing all appropriate file descriptors
5558   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5559   // turn might:
5560   //
5561   // - cause end-of-file to fail to be detected on some file
5562   //   descriptors, resulting in mysterious hangs, or
5563   //
5564   // - might cause an fopen in the subprocess to fail on a system
5565   //   suffering from bug 1085341.
5566   //
5567   // (Yes, the default setting of the close-on-exec flag is a Unix
5568   // design flaw)
5569   //
5570   // See:
5571   // 1085341: 32-bit stdio routines should support file descriptors >255
5572   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5573   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5574   //
5575   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5576   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5577   // because it saves a system call and removes a small window where the flag
5578   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5579   // and we fall back to using FD_CLOEXEC (see below).
5580 #ifdef O_CLOEXEC
5581   oflag |= O_CLOEXEC;
5582 #endif
5583 
5584   int fd = ::open64(path, oflag, mode);
5585   if (fd == -1) return -1;
5586 
5587   //If the open succeeded, the file might still be a directory
5588   {
5589     struct stat64 buf64;
5590     int ret = ::fstat64(fd, &buf64);
5591     int st_mode = buf64.st_mode;
5592 
5593     if (ret != -1) {
5594       if ((st_mode & S_IFMT) == S_IFDIR) {
5595         errno = EISDIR;
5596         ::close(fd);
5597         return -1;
5598       }
5599     } else {
5600       ::close(fd);
5601       return -1;
5602     }
5603   }
5604 
5605 #ifdef FD_CLOEXEC
5606   // Validate that the use of the O_CLOEXEC flag on open above worked.
5607   // With recent kernels, we will perform this check exactly once.
5608   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5609   if (!O_CLOEXEC_is_known_to_work) {
5610     int flags = ::fcntl(fd, F_GETFD);
5611     if (flags != -1) {
5612       if ((flags & FD_CLOEXEC) != 0)
5613         O_CLOEXEC_is_known_to_work = 1;
5614       else
5615         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5616     }
5617   }
5618 #endif
5619 
5620   return fd;
5621 }
5622 
5623 
5624 // create binary file, rewriting existing file if required
5625 int os::create_binary_file(const char* path, bool rewrite_existing) {
5626   int oflags = O_WRONLY | O_CREAT;
5627   if (!rewrite_existing) {
5628     oflags |= O_EXCL;
5629   }
5630   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5631 }
5632 
5633 // return current position of file pointer
5634 jlong os::current_file_offset(int fd) {
5635   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5636 }
5637 
5638 // move file pointer to the specified offset
5639 jlong os::seek_to_file_offset(int fd, jlong offset) {
5640   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5641 }
5642 
5643 // This code originates from JDK's sysAvailable
5644 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5645 
5646 int os::available(int fd, jlong *bytes) {
5647   jlong cur, end;
5648   int mode;
5649   struct stat64 buf64;
5650 
5651   if (::fstat64(fd, &buf64) >= 0) {
5652     mode = buf64.st_mode;
5653     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5654       int n;
5655       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5656         *bytes = n;
5657         return 1;
5658       }
5659     }
5660   }
5661   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5662     return 0;
5663   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5664     return 0;
5665   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5666     return 0;
5667   }
5668   *bytes = end - cur;
5669   return 1;
5670 }
5671 
5672 // Map a block of memory.
5673 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5674                         char *addr, size_t bytes, bool read_only,
5675                         bool allow_exec) {
5676   int prot;
5677   int flags = MAP_PRIVATE;
5678 
5679   if (read_only) {
5680     prot = PROT_READ;
5681   } else {
5682     prot = PROT_READ | PROT_WRITE;
5683   }
5684 
5685   if (allow_exec) {
5686     prot |= PROT_EXEC;
5687   }
5688 
5689   if (addr != NULL) {
5690     flags |= MAP_FIXED;
5691   }
5692 
5693   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5694                                      fd, file_offset);
5695   if (mapped_address == MAP_FAILED) {
5696     return NULL;
5697   }
5698   return mapped_address;
5699 }
5700 
5701 
5702 // Remap a block of memory.
5703 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5704                           char *addr, size_t bytes, bool read_only,
5705                           bool allow_exec) {
5706   // same as map_memory() on this OS
5707   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5708                         allow_exec);
5709 }
5710 
5711 
5712 // Unmap a block of memory.
5713 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5714   return munmap(addr, bytes) == 0;
5715 }
5716 
5717 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5718 
5719 static jlong fast_cpu_time(Thread *thread) {
5720     clockid_t clockid;
5721     int rc = os::Linux::pthread_getcpuclockid(thread->osthread()->pthread_id(),
5722                                               &clockid);
5723     if (rc == 0) {
5724       return os::Linux::fast_thread_cpu_time(clockid);
5725     } else {
5726       // It's possible to encounter a terminated native thread that failed
5727       // to detach itself from the VM - which should result in ESRCH.
5728       assert_status(rc == ESRCH, rc, "pthread_getcpuclockid failed");
5729       return -1;
5730     }
5731 }
5732 
5733 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5734 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5735 // of a thread.
5736 //
5737 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5738 // the fast estimate available on the platform.
5739 
5740 jlong os::current_thread_cpu_time() {
5741   if (os::Linux::supports_fast_thread_cpu_time()) {
5742     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5743   } else {
5744     // return user + sys since the cost is the same
5745     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5746   }
5747 }
5748 
5749 jlong os::thread_cpu_time(Thread* thread) {
5750   // consistent with what current_thread_cpu_time() returns
5751   if (os::Linux::supports_fast_thread_cpu_time()) {
5752     return fast_cpu_time(thread);
5753   } else {
5754     return slow_thread_cpu_time(thread, true /* user + sys */);
5755   }
5756 }
5757 
5758 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5759   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5760     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5761   } else {
5762     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5763   }
5764 }
5765 
5766 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5767   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5768     return fast_cpu_time(thread);
5769   } else {
5770     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5771   }
5772 }
5773 
5774 //  -1 on error.
5775 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5776   pid_t  tid = thread->osthread()->thread_id();
5777   char *s;
5778   char stat[2048];
5779   int statlen;
5780   char proc_name[64];
5781   int count;
5782   long sys_time, user_time;
5783   char cdummy;
5784   int idummy;
5785   long ldummy;
5786   FILE *fp;
5787 
5788   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5789   fp = fopen(proc_name, "r");
5790   if (fp == NULL) return -1;
5791   statlen = fread(stat, 1, 2047, fp);
5792   stat[statlen] = '\0';
5793   fclose(fp);
5794 
5795   // Skip pid and the command string. Note that we could be dealing with
5796   // weird command names, e.g. user could decide to rename java launcher
5797   // to "java 1.4.2 :)", then the stat file would look like
5798   //                1234 (java 1.4.2 :)) R ... ...
5799   // We don't really need to know the command string, just find the last
5800   // occurrence of ")" and then start parsing from there. See bug 4726580.
5801   s = strrchr(stat, ')');
5802   if (s == NULL) return -1;
5803 
5804   // Skip blank chars
5805   do { s++; } while (s && isspace(*s));
5806 
5807   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5808                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5809                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5810                  &user_time, &sys_time);
5811   if (count != 13) return -1;
5812   if (user_sys_cpu_time) {
5813     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5814   } else {
5815     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5816   }
5817 }
5818 
5819 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5820   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5821   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5822   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5823   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5824 }
5825 
5826 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5827   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5828   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5829   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5830   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5831 }
5832 
5833 bool os::is_thread_cpu_time_supported() {
5834   return true;
5835 }
5836 
5837 // System loadavg support.  Returns -1 if load average cannot be obtained.
5838 // Linux doesn't yet have a (official) notion of processor sets,
5839 // so just return the system wide load average.
5840 int os::loadavg(double loadavg[], int nelem) {
5841   return ::getloadavg(loadavg, nelem);
5842 }
5843 
5844 void os::pause() {
5845   char filename[MAX_PATH];
5846   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5847     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5848   } else {
5849     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5850   }
5851 
5852   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5853   if (fd != -1) {
5854     struct stat buf;
5855     ::close(fd);
5856     while (::stat(filename, &buf) == 0) {
5857       (void)::poll(NULL, 0, 100);
5858     }
5859   } else {
5860     jio_fprintf(stderr,
5861                 "Could not open pause file '%s', continuing immediately.\n", filename);
5862   }
5863 }
5864 
5865 extern char** environ;
5866 
5867 // Run the specified command in a separate process. Return its exit value,
5868 // or -1 on failure (e.g. can't fork a new process).
5869 // Unlike system(), this function can be called from signal handler. It
5870 // doesn't block SIGINT et al.
5871 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5872   const char * argv[4] = {"sh", "-c", cmd, NULL};
5873 
5874   pid_t pid ;
5875 
5876   if (use_vfork_if_available) {
5877     pid = vfork();
5878   } else {
5879     pid = fork();
5880   }
5881 
5882   if (pid < 0) {
5883     // fork failed
5884     return -1;
5885 
5886   } else if (pid == 0) {
5887     // child process
5888 
5889     execve("/bin/sh", (char* const*)argv, environ);
5890 
5891     // execve failed
5892     _exit(-1);
5893 
5894   } else  {
5895     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5896     // care about the actual exit code, for now.
5897 
5898     int status;
5899 
5900     // Wait for the child process to exit.  This returns immediately if
5901     // the child has already exited. */
5902     while (waitpid(pid, &status, 0) < 0) {
5903       switch (errno) {
5904       case ECHILD: return 0;
5905       case EINTR: break;
5906       default: return -1;
5907       }
5908     }
5909 
5910     if (WIFEXITED(status)) {
5911       // The child exited normally; get its exit code.
5912       return WEXITSTATUS(status);
5913     } else if (WIFSIGNALED(status)) {
5914       // The child exited because of a signal
5915       // The best value to return is 0x80 + signal number,
5916       // because that is what all Unix shells do, and because
5917       // it allows callers to distinguish between process exit and
5918       // process death by signal.
5919       return 0x80 + WTERMSIG(status);
5920     } else {
5921       // Unknown exit code; pass it through
5922       return status;
5923     }
5924   }
5925 }
5926 
5927 // Get the default path to the core file
5928 // Returns the length of the string
5929 int os::get_core_path(char* buffer, size_t bufferSize) {
5930   /*
5931    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5932    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5933    */
5934   const int core_pattern_len = 129;
5935   char core_pattern[core_pattern_len] = {0};
5936 
5937   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5938   if (core_pattern_file == -1) {
5939     return -1;
5940   }
5941 
5942   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5943   ::close(core_pattern_file);
5944   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5945     return -1;
5946   }
5947   if (core_pattern[ret-1] == '\n') {
5948     core_pattern[ret-1] = '\0';
5949   } else {
5950     core_pattern[ret] = '\0';
5951   }
5952 
5953   // Replace the %p in the core pattern with the process id. NOTE: we do this
5954   // only if the pattern doesn't start with "|", and we support only one %p in
5955   // the pattern.
5956   char *pid_pos = strstr(core_pattern, "%p");
5957   const char* tail = (pid_pos != NULL) ? (pid_pos + 2) : "";  // skip over the "%p"
5958   int written;
5959 
5960   if (core_pattern[0] == '/') {
5961     if (pid_pos != NULL) {
5962       *pid_pos = '\0';
5963       written = jio_snprintf(buffer, bufferSize, "%s%d%s", core_pattern,
5964                              current_process_id(), tail);
5965     } else {
5966       written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5967     }
5968   } else {
5969     char cwd[PATH_MAX];
5970 
5971     const char* p = get_current_directory(cwd, PATH_MAX);
5972     if (p == NULL) {
5973       return -1;
5974     }
5975 
5976     if (core_pattern[0] == '|') {
5977       written = jio_snprintf(buffer, bufferSize,
5978                              "\"%s\" (or dumping to %s/core.%d)",
5979                              &core_pattern[1], p, current_process_id());
5980     } else if (pid_pos != NULL) {
5981       *pid_pos = '\0';
5982       written = jio_snprintf(buffer, bufferSize, "%s/%s%d%s", p, core_pattern,
5983                              current_process_id(), tail);
5984     } else {
5985       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5986     }
5987   }
5988 
5989   if (written < 0) {
5990     return -1;
5991   }
5992 
5993   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5994     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5995 
5996     if (core_uses_pid_file != -1) {
5997       char core_uses_pid = 0;
5998       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5999       ::close(core_uses_pid_file);
6000 
6001       if (core_uses_pid == '1') {
6002         jio_snprintf(buffer + written, bufferSize - written,
6003                                           ".%d", current_process_id());
6004       }
6005     }
6006   }
6007 
6008   return strlen(buffer);
6009 }
6010 
6011 bool os::start_debugging(char *buf, int buflen) {
6012   int len = (int)strlen(buf);
6013   char *p = &buf[len];
6014 
6015   jio_snprintf(p, buflen-len,
6016                "\n\n"
6017                "Do you want to debug the problem?\n\n"
6018                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
6019                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
6020                "Otherwise, press RETURN to abort...",
6021                os::current_process_id(), os::current_process_id(),
6022                os::current_thread_id(), os::current_thread_id());
6023 
6024   bool yes = os::message_box("Unexpected Error", buf);
6025 
6026   if (yes) {
6027     // yes, user asked VM to launch debugger
6028     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
6029                  os::current_process_id(), os::current_process_id());
6030 
6031     os::fork_and_exec(buf);
6032     yes = false;
6033   }
6034   return yes;
6035 }
6036 
6037 
6038 // Java/Compiler thread:
6039 //
6040 //   Low memory addresses
6041 // P0 +------------------------+
6042 //    |                        |\  Java thread created by VM does not have glibc
6043 //    |    glibc guard page    | - guard page, attached Java thread usually has
6044 //    |                        |/  1 glibc guard page.
6045 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6046 //    |                        |\
6047 //    |  HotSpot Guard Pages   | - red, yellow and reserved pages
6048 //    |                        |/
6049 //    +------------------------+ JavaThread::stack_reserved_zone_base()
6050 //    |                        |\
6051 //    |      Normal Stack      | -
6052 //    |                        |/
6053 // P2 +------------------------+ Thread::stack_base()
6054 //
6055 // Non-Java thread:
6056 //
6057 //   Low memory addresses
6058 // P0 +------------------------+
6059 //    |                        |\
6060 //    |  glibc guard page      | - usually 1 page
6061 //    |                        |/
6062 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
6063 //    |                        |\
6064 //    |      Normal Stack      | -
6065 //    |                        |/
6066 // P2 +------------------------+ Thread::stack_base()
6067 //
6068 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
6069 //    returned from pthread_attr_getstack().
6070 // ** Due to NPTL implementation error, linux takes the glibc guard page out
6071 //    of the stack size given in pthread_attr. We work around this for
6072 //    threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
6073 //
6074 #ifndef ZERO
6075 static void current_stack_region(address * bottom, size_t * size) {
6076   if (os::is_primordial_thread()) {
6077     // primordial thread needs special handling because pthread_getattr_np()
6078     // may return bogus value.
6079     *bottom = os::Linux::initial_thread_stack_bottom();
6080     *size   = os::Linux::initial_thread_stack_size();
6081   } else {
6082     pthread_attr_t attr;
6083 
6084     int rslt = pthread_getattr_np(pthread_self(), &attr);
6085 
6086     // JVM needs to know exact stack location, abort if it fails
6087     if (rslt != 0) {
6088       if (rslt == ENOMEM) {
6089         vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
6090       } else {
6091         fatal("pthread_getattr_np failed with error = %d", rslt);
6092       }
6093     }
6094 
6095     if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
6096       fatal("Cannot locate current stack attributes!");
6097     }
6098 
6099     // Work around NPTL stack guard error.
6100     size_t guard_size = 0;
6101     rslt = pthread_attr_getguardsize(&attr, &guard_size);
6102     if (rslt != 0) {
6103       fatal("pthread_attr_getguardsize failed with error = %d", rslt);
6104     }
6105     *bottom += guard_size;
6106     *size   -= guard_size;
6107 
6108     pthread_attr_destroy(&attr);
6109 
6110   }
6111   assert(os::current_stack_pointer() >= *bottom &&
6112          os::current_stack_pointer() < *bottom + *size, "just checking");
6113 }
6114 
6115 address os::current_stack_base() {
6116   address bottom;
6117   size_t size;
6118   current_stack_region(&bottom, &size);
6119   return (bottom + size);
6120 }
6121 
6122 size_t os::current_stack_size() {
6123   // This stack size includes the usable stack and HotSpot guard pages
6124   // (for the threads that have Hotspot guard pages).
6125   address bottom;
6126   size_t size;
6127   current_stack_region(&bottom, &size);
6128   return size;
6129 }
6130 #endif
6131 
6132 static inline struct timespec get_mtime(const char* filename) {
6133   struct stat st;
6134   int ret = os::stat(filename, &st);
6135   assert(ret == 0, "failed to stat() file '%s': %s", filename, os::strerror(errno));
6136   return st.st_mtim;
6137 }
6138 
6139 int os::compare_file_modified_times(const char* file1, const char* file2) {
6140   struct timespec filetime1 = get_mtime(file1);
6141   struct timespec filetime2 = get_mtime(file2);
6142   int diff = filetime1.tv_sec - filetime2.tv_sec;
6143   if (diff == 0) {
6144     return filetime1.tv_nsec - filetime2.tv_nsec;
6145   }
6146   return diff;
6147 }
6148 
6149 bool os::supports_map_sync() {
6150   return true;
6151 }
6152 
6153 /////////////// Unit tests ///////////////
6154 
6155 #ifndef PRODUCT
6156 
6157 class TestReserveMemorySpecial : AllStatic {
6158  public:
6159   static void small_page_write(void* addr, size_t size) {
6160     size_t page_size = os::vm_page_size();
6161 
6162     char* end = (char*)addr + size;
6163     for (char* p = (char*)addr; p < end; p += page_size) {
6164       *p = 1;
6165     }
6166   }
6167 
6168   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6169     if (!UseHugeTLBFS) {
6170       return;
6171     }
6172 
6173     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6174 
6175     if (addr != NULL) {
6176       small_page_write(addr, size);
6177 
6178       os::Linux::release_memory_special_huge_tlbfs(addr, size);
6179     }
6180   }
6181 
6182   static void test_reserve_memory_special_huge_tlbfs_only() {
6183     if (!UseHugeTLBFS) {
6184       return;
6185     }
6186 
6187     size_t lp = os::large_page_size();
6188 
6189     for (size_t size = lp; size <= lp * 10; size += lp) {
6190       test_reserve_memory_special_huge_tlbfs_only(size);
6191     }
6192   }
6193 
6194   static void test_reserve_memory_special_huge_tlbfs_mixed() {
6195     size_t lp = os::large_page_size();
6196     size_t ag = os::vm_allocation_granularity();
6197 
6198     // sizes to test
6199     const size_t sizes[] = {
6200       lp, lp + ag, lp + lp / 2, lp * 2,
6201       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6202       lp * 10, lp * 10 + lp / 2
6203     };
6204     const int num_sizes = sizeof(sizes) / sizeof(size_t);
6205 
6206     // For each size/alignment combination, we test three scenarios:
6207     // 1) with req_addr == NULL
6208     // 2) with a non-null req_addr at which we expect to successfully allocate
6209     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6210     //    expect the allocation to either fail or to ignore req_addr
6211 
6212     // Pre-allocate two areas; they shall be as large as the largest allocation
6213     //  and aligned to the largest alignment we will be testing.
6214     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6215     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6216       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6217       -1, 0);
6218     assert(mapping1 != MAP_FAILED, "should work");
6219 
6220     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6221       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6222       -1, 0);
6223     assert(mapping2 != MAP_FAILED, "should work");
6224 
6225     // Unmap the first mapping, but leave the second mapping intact: the first
6226     // mapping will serve as a value for a "good" req_addr (case 2). The second
6227     // mapping, still intact, as "bad" req_addr (case 3).
6228     ::munmap(mapping1, mapping_size);
6229 
6230     // Case 1
6231     for (int i = 0; i < num_sizes; i++) {
6232       const size_t size = sizes[i];
6233       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6234         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6235         if (p != NULL) {
6236           assert(is_aligned(p, alignment), "must be");
6237           small_page_write(p, size);
6238           os::Linux::release_memory_special_huge_tlbfs(p, size);
6239         }
6240       }
6241     }
6242 
6243     // Case 2
6244     for (int i = 0; i < num_sizes; i++) {
6245       const size_t size = sizes[i];
6246       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6247         char* const req_addr = align_up(mapping1, alignment);
6248         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6249         if (p != NULL) {
6250           assert(p == req_addr, "must be");
6251           small_page_write(p, size);
6252           os::Linux::release_memory_special_huge_tlbfs(p, size);
6253         }
6254       }
6255     }
6256 
6257     // Case 3
6258     for (int i = 0; i < num_sizes; i++) {
6259       const size_t size = sizes[i];
6260       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6261         char* const req_addr = align_up(mapping2, alignment);
6262         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6263         // as the area around req_addr contains already existing mappings, the API should always
6264         // return NULL (as per contract, it cannot return another address)
6265         assert(p == NULL, "must be");
6266       }
6267     }
6268 
6269     ::munmap(mapping2, mapping_size);
6270 
6271   }
6272 
6273   static void test_reserve_memory_special_huge_tlbfs() {
6274     if (!UseHugeTLBFS) {
6275       return;
6276     }
6277 
6278     test_reserve_memory_special_huge_tlbfs_only();
6279     test_reserve_memory_special_huge_tlbfs_mixed();
6280   }
6281 
6282   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6283     if (!UseSHM) {
6284       return;
6285     }
6286 
6287     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6288 
6289     if (addr != NULL) {
6290       assert(is_aligned(addr, alignment), "Check");
6291       assert(is_aligned(addr, os::large_page_size()), "Check");
6292 
6293       small_page_write(addr, size);
6294 
6295       os::Linux::release_memory_special_shm(addr, size);
6296     }
6297   }
6298 
6299   static void test_reserve_memory_special_shm() {
6300     size_t lp = os::large_page_size();
6301     size_t ag = os::vm_allocation_granularity();
6302 
6303     for (size_t size = ag; size < lp * 3; size += ag) {
6304       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6305         test_reserve_memory_special_shm(size, alignment);
6306       }
6307     }
6308   }
6309 
6310   static void test() {
6311     test_reserve_memory_special_huge_tlbfs();
6312     test_reserve_memory_special_shm();
6313   }
6314 };
6315 
6316 void TestReserveMemorySpecial_test() {
6317   TestReserveMemorySpecial::test();
6318 }
6319 
6320 #endif