1 /*
   2  * Copyright (c) 1999, 2017, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "jvm.h"
  27 #include "classfile/classLoader.hpp"
  28 #include "classfile/systemDictionary.hpp"
  29 #include "classfile/vmSymbols.hpp"
  30 #include "code/icBuffer.hpp"
  31 #include "code/vtableStubs.hpp"
  32 #include "compiler/compileBroker.hpp"
  33 #include "compiler/disassembler.hpp"
  34 #include "interpreter/interpreter.hpp"
  35 #include "logging/log.hpp"
  36 #include "memory/allocation.inline.hpp"
  37 #include "memory/filemap.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_linux.inline.hpp"
  40 #include "os_share_linux.hpp"
  41 #include "osContainer_linux.hpp"
  42 #include "prims/jniFastGetField.hpp"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.hpp"
  49 #include "runtime/init.hpp"
  50 #include "runtime/java.hpp"
  51 #include "runtime/javaCalls.hpp"
  52 #include "runtime/mutexLocker.hpp"
  53 #include "runtime/objectMonitor.hpp"
  54 #include "runtime/orderAccess.inline.hpp"
  55 #include "runtime/osThread.hpp"
  56 #include "runtime/perfMemory.hpp"
  57 #include "runtime/sharedRuntime.hpp"
  58 #include "runtime/statSampler.hpp"
  59 #include "runtime/stubRoutines.hpp"
  60 #include "runtime/thread.inline.hpp"
  61 #include "runtime/threadCritical.hpp"
  62 #include "runtime/threadSMR.hpp"
  63 #include "runtime/timer.hpp"
  64 #include "semaphore_posix.hpp"
  65 #include "services/attachListener.hpp"
  66 #include "services/memTracker.hpp"
  67 #include "services/runtimeService.hpp"
  68 #include "utilities/align.hpp"
  69 #include "utilities/decoder.hpp"
  70 #include "utilities/defaultStream.hpp"
  71 #include "utilities/events.hpp"
  72 #include "utilities/elfFile.hpp"
  73 #include "utilities/growableArray.hpp"
  74 #include "utilities/macros.hpp"
  75 #include "utilities/vmError.hpp"
  76 
  77 // put OS-includes here
  78 # include <sys/types.h>
  79 # include <sys/mman.h>
  80 # include <sys/stat.h>
  81 # include <sys/select.h>
  82 # include <pthread.h>
  83 # include <signal.h>
  84 # include <errno.h>
  85 # include <dlfcn.h>
  86 # include <stdio.h>
  87 # include <unistd.h>
  88 # include <sys/resource.h>
  89 # include <pthread.h>
  90 # include <sys/stat.h>
  91 # include <sys/time.h>
  92 # include <sys/times.h>
  93 # include <sys/utsname.h>
  94 # include <sys/socket.h>
  95 # include <sys/wait.h>
  96 # include <pwd.h>
  97 # include <poll.h>
  98 # include <semaphore.h>
  99 # include <fcntl.h>
 100 # include <string.h>
 101 # include <syscall.h>
 102 # include <sys/sysinfo.h>
 103 # include <gnu/libc-version.h>
 104 # include <sys/ipc.h>
 105 # include <sys/shm.h>
 106 # include <link.h>
 107 # include <stdint.h>
 108 # include <inttypes.h>
 109 # include <sys/ioctl.h>
 110 
 111 #ifndef _GNU_SOURCE
 112   #define _GNU_SOURCE
 113   #include <sched.h>
 114   #undef _GNU_SOURCE
 115 #else
 116   #include <sched.h>
 117 #endif
 118 
 119 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 120 // getrusage() is prepared to handle the associated failure.
 121 #ifndef RUSAGE_THREAD
 122   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 123 #endif
 124 
 125 #define MAX_PATH    (2 * K)
 126 
 127 #define MAX_SECS 100000000
 128 
 129 // for timer info max values which include all bits
 130 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 131 
 132 #define LARGEPAGES_BIT (1 << 6)
 133 ////////////////////////////////////////////////////////////////////////////////
 134 // global variables
 135 julong os::Linux::_physical_memory = 0;
 136 
 137 address   os::Linux::_initial_thread_stack_bottom = NULL;
 138 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 139 
 140 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 141 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 142 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 143 Mutex* os::Linux::_createThread_lock = NULL;
 144 pthread_t os::Linux::_main_thread;
 145 int os::Linux::_page_size = -1;
 146 bool os::Linux::_supports_fast_thread_cpu_time = false;
 147 uint32_t os::Linux::_os_version = 0;
 148 const char * os::Linux::_glibc_version = NULL;
 149 const char * os::Linux::_libpthread_version = NULL;
 150 
 151 static jlong initial_time_count=0;
 152 
 153 static int clock_tics_per_sec = 100;
 154 
 155 // For diagnostics to print a message once. see run_periodic_checks
 156 static sigset_t check_signal_done;
 157 static bool check_signals = true;
 158 
 159 // Signal number used to suspend/resume a thread
 160 
 161 // do not use any signal number less than SIGSEGV, see 4355769
 162 static int SR_signum = SIGUSR2;
 163 sigset_t SR_sigset;
 164 
 165 // utility functions
 166 
 167 static int SR_initialize();
 168 
 169 julong os::available_memory() {
 170   return Linux::available_memory();
 171 }
 172 
 173 julong os::Linux::available_memory() {
 174   // values in struct sysinfo are "unsigned long"
 175   struct sysinfo si;
 176   julong avail_mem;
 177 
 178   if (OSContainer::is_containerized()) {
 179     jlong mem_limit, mem_usage;
 180     if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
 181       if ((mem_usage = OSContainer::memory_usage_in_bytes()) > 0) {
 182         if (mem_limit > mem_usage) {
 183           avail_mem = (julong)mem_limit - (julong)mem_usage;
 184         } else {
 185           avail_mem = 0;
 186         }
 187         log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
 188         return avail_mem;
 189       } else {
 190         log_debug(os,container)("container memory usage call failed: " JLONG_FORMAT, mem_usage);
 191       }
 192     } else {
 193       log_debug(os,container)("container memory unlimited or failed: " JLONG_FORMAT, mem_limit);
 194     }
 195   }
 196 
 197   sysinfo(&si);
 198   avail_mem = (julong)si.freeram * si.mem_unit;
 199   log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
 200   return avail_mem;
 201 }
 202 
 203 julong os::physical_memory() {
 204   if (OSContainer::is_containerized()) {
 205     jlong mem_limit;
 206     if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
 207       log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
 208       return (julong)mem_limit;
 209     } else {
 210       if (mem_limit == OSCONTAINER_ERROR) {
 211         log_debug(os,container)("container memory limit call failed");
 212       }
 213       if (mem_limit == -1) {
 214         log_debug(os,container)("container memory unlimited, using host value");
 215       }
 216     }
 217   }
 218 
 219   jlong phys_mem = Linux::physical_memory();
 220   log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
 221   return phys_mem;
 222 }
 223 
 224 // Return true if user is running as root.
 225 
 226 bool os::have_special_privileges() {
 227   static bool init = false;
 228   static bool privileges = false;
 229   if (!init) {
 230     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 231     init = true;
 232   }
 233   return privileges;
 234 }
 235 
 236 
 237 #ifndef SYS_gettid
 238 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 239   #ifdef __ia64__
 240     #define SYS_gettid 1105
 241   #else
 242     #ifdef __i386__
 243       #define SYS_gettid 224
 244     #else
 245       #ifdef __amd64__
 246         #define SYS_gettid 186
 247       #else
 248         #ifdef __sparc__
 249           #define SYS_gettid 143
 250         #else
 251           #error define gettid for the arch
 252         #endif
 253       #endif
 254     #endif
 255   #endif
 256 #endif
 257 
 258 
 259 // pid_t gettid()
 260 //
 261 // Returns the kernel thread id of the currently running thread. Kernel
 262 // thread id is used to access /proc.
 263 pid_t os::Linux::gettid() {
 264   int rslt = syscall(SYS_gettid);
 265   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 266   return (pid_t)rslt;
 267 }
 268 
 269 // Most versions of linux have a bug where the number of processors are
 270 // determined by looking at the /proc file system.  In a chroot environment,
 271 // the system call returns 1.  This causes the VM to act as if it is
 272 // a single processor and elide locking (see is_MP() call).
 273 static bool unsafe_chroot_detected = false;
 274 static const char *unstable_chroot_error = "/proc file system not found.\n"
 275                      "Java may be unstable running multithreaded in a chroot "
 276                      "environment on Linux when /proc filesystem is not mounted.";
 277 
 278 void os::Linux::initialize_system_info() {
 279   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 280   if (processor_count() == 1) {
 281     pid_t pid = os::Linux::gettid();
 282     char fname[32];
 283     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 284     FILE *fp = fopen(fname, "r");
 285     if (fp == NULL) {
 286       unsafe_chroot_detected = true;
 287     } else {
 288       fclose(fp);
 289     }
 290   }
 291   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 292   assert(processor_count() > 0, "linux error");
 293 }
 294 
 295 void os::init_system_properties_values() {
 296   // The next steps are taken in the product version:
 297   //
 298   // Obtain the JAVA_HOME value from the location of libjvm.so.
 299   // This library should be located at:
 300   // <JAVA_HOME>/lib/{client|server}/libjvm.so.
 301   //
 302   // If "/jre/lib/" appears at the right place in the path, then we
 303   // assume libjvm.so is installed in a JDK and we use this path.
 304   //
 305   // Otherwise exit with message: "Could not create the Java virtual machine."
 306   //
 307   // The following extra steps are taken in the debugging version:
 308   //
 309   // If "/jre/lib/" does NOT appear at the right place in the path
 310   // instead of exit check for $JAVA_HOME environment variable.
 311   //
 312   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 313   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 314   // it looks like libjvm.so is installed there
 315   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 316   //
 317   // Otherwise exit.
 318   //
 319   // Important note: if the location of libjvm.so changes this
 320   // code needs to be changed accordingly.
 321 
 322   // See ld(1):
 323   //      The linker uses the following search paths to locate required
 324   //      shared libraries:
 325   //        1: ...
 326   //        ...
 327   //        7: The default directories, normally /lib and /usr/lib.
 328 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
 329   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 330 #else
 331   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 332 #endif
 333 
 334 // Base path of extensions installed on the system.
 335 #define SYS_EXT_DIR     "/usr/java/packages"
 336 #define EXTENSIONS_DIR  "/lib/ext"
 337 
 338   // Buffer that fits several sprintfs.
 339   // Note that the space for the colon and the trailing null are provided
 340   // by the nulls included by the sizeof operator.
 341   const size_t bufsize =
 342     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 343          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 344   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 345 
 346   // sysclasspath, java_home, dll_dir
 347   {
 348     char *pslash;
 349     os::jvm_path(buf, bufsize);
 350 
 351     // Found the full path to libjvm.so.
 352     // Now cut the path to <java_home>/jre if we can.
 353     pslash = strrchr(buf, '/');
 354     if (pslash != NULL) {
 355       *pslash = '\0';            // Get rid of /libjvm.so.
 356     }
 357     pslash = strrchr(buf, '/');
 358     if (pslash != NULL) {
 359       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 360     }
 361     Arguments::set_dll_dir(buf);
 362 
 363     if (pslash != NULL) {
 364       pslash = strrchr(buf, '/');
 365       if (pslash != NULL) {
 366         *pslash = '\0';        // Get rid of /lib.
 367       }
 368     }
 369     Arguments::set_java_home(buf);
 370     set_boot_path('/', ':');
 371   }
 372 
 373   // Where to look for native libraries.
 374   //
 375   // Note: Due to a legacy implementation, most of the library path
 376   // is set in the launcher. This was to accomodate linking restrictions
 377   // on legacy Linux implementations (which are no longer supported).
 378   // Eventually, all the library path setting will be done here.
 379   //
 380   // However, to prevent the proliferation of improperly built native
 381   // libraries, the new path component /usr/java/packages is added here.
 382   // Eventually, all the library path setting will be done here.
 383   {
 384     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 385     // should always exist (until the legacy problem cited above is
 386     // addressed).
 387     const char *v = ::getenv("LD_LIBRARY_PATH");
 388     const char *v_colon = ":";
 389     if (v == NULL) { v = ""; v_colon = ""; }
 390     // That's +1 for the colon and +1 for the trailing '\0'.
 391     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 392                                                      strlen(v) + 1 +
 393                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
 394                                                      mtInternal);
 395     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
 396     Arguments::set_library_path(ld_library_path);
 397     FREE_C_HEAP_ARRAY(char, ld_library_path);
 398   }
 399 
 400   // Extensions directories.
 401   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 402   Arguments::set_ext_dirs(buf);
 403 
 404   FREE_C_HEAP_ARRAY(char, buf);
 405 
 406 #undef DEFAULT_LIBPATH
 407 #undef SYS_EXT_DIR
 408 #undef EXTENSIONS_DIR
 409 }
 410 
 411 ////////////////////////////////////////////////////////////////////////////////
 412 // breakpoint support
 413 
 414 void os::breakpoint() {
 415   BREAKPOINT;
 416 }
 417 
 418 extern "C" void breakpoint() {
 419   // use debugger to set breakpoint here
 420 }
 421 
 422 ////////////////////////////////////////////////////////////////////////////////
 423 // signal support
 424 
 425 debug_only(static bool signal_sets_initialized = false);
 426 static sigset_t unblocked_sigs, vm_sigs;
 427 
 428 bool os::Linux::is_sig_ignored(int sig) {
 429   struct sigaction oact;
 430   sigaction(sig, (struct sigaction*)NULL, &oact);
 431   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 432                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 433   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 434     return true;
 435   } else {
 436     return false;
 437   }
 438 }
 439 
 440 void os::Linux::signal_sets_init() {
 441   // Should also have an assertion stating we are still single-threaded.
 442   assert(!signal_sets_initialized, "Already initialized");
 443   // Fill in signals that are necessarily unblocked for all threads in
 444   // the VM. Currently, we unblock the following signals:
 445   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 446   //                         by -Xrs (=ReduceSignalUsage));
 447   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 448   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 449   // the dispositions or masks wrt these signals.
 450   // Programs embedding the VM that want to use the above signals for their
 451   // own purposes must, at this time, use the "-Xrs" option to prevent
 452   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 453   // (See bug 4345157, and other related bugs).
 454   // In reality, though, unblocking these signals is really a nop, since
 455   // these signals are not blocked by default.
 456   sigemptyset(&unblocked_sigs);
 457   sigaddset(&unblocked_sigs, SIGILL);
 458   sigaddset(&unblocked_sigs, SIGSEGV);
 459   sigaddset(&unblocked_sigs, SIGBUS);
 460   sigaddset(&unblocked_sigs, SIGFPE);
 461 #if defined(PPC64)
 462   sigaddset(&unblocked_sigs, SIGTRAP);
 463 #endif
 464   sigaddset(&unblocked_sigs, SR_signum);
 465 
 466   if (!ReduceSignalUsage) {
 467     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 468       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 469     }
 470     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 471       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 472     }
 473     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 474       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 475     }
 476   }
 477   // Fill in signals that are blocked by all but the VM thread.
 478   sigemptyset(&vm_sigs);
 479   if (!ReduceSignalUsage) {
 480     sigaddset(&vm_sigs, BREAK_SIGNAL);
 481   }
 482   debug_only(signal_sets_initialized = true);
 483 
 484 }
 485 
 486 // These are signals that are unblocked while a thread is running Java.
 487 // (For some reason, they get blocked by default.)
 488 sigset_t* os::Linux::unblocked_signals() {
 489   assert(signal_sets_initialized, "Not initialized");
 490   return &unblocked_sigs;
 491 }
 492 
 493 // These are the signals that are blocked while a (non-VM) thread is
 494 // running Java. Only the VM thread handles these signals.
 495 sigset_t* os::Linux::vm_signals() {
 496   assert(signal_sets_initialized, "Not initialized");
 497   return &vm_sigs;
 498 }
 499 
 500 void os::Linux::hotspot_sigmask(Thread* thread) {
 501 
 502   //Save caller's signal mask before setting VM signal mask
 503   sigset_t caller_sigmask;
 504   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 505 
 506   OSThread* osthread = thread->osthread();
 507   osthread->set_caller_sigmask(caller_sigmask);
 508 
 509   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 510 
 511   if (!ReduceSignalUsage) {
 512     if (thread->is_VM_thread()) {
 513       // Only the VM thread handles BREAK_SIGNAL ...
 514       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 515     } else {
 516       // ... all other threads block BREAK_SIGNAL
 517       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 518     }
 519   }
 520 }
 521 
 522 //////////////////////////////////////////////////////////////////////////////
 523 // detecting pthread library
 524 
 525 void os::Linux::libpthread_init() {
 526   // Save glibc and pthread version strings.
 527 #if !defined(_CS_GNU_LIBC_VERSION) || \
 528     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 529   #error "glibc too old (< 2.3.2)"
 530 #endif
 531 
 532   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 533   assert(n > 0, "cannot retrieve glibc version");
 534   char *str = (char *)malloc(n, mtInternal);
 535   confstr(_CS_GNU_LIBC_VERSION, str, n);
 536   os::Linux::set_glibc_version(str);
 537 
 538   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 539   assert(n > 0, "cannot retrieve pthread version");
 540   str = (char *)malloc(n, mtInternal);
 541   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 542   os::Linux::set_libpthread_version(str);
 543 }
 544 
 545 /////////////////////////////////////////////////////////////////////////////
 546 // thread stack expansion
 547 
 548 // os::Linux::manually_expand_stack() takes care of expanding the thread
 549 // stack. Note that this is normally not needed: pthread stacks allocate
 550 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 551 // committed. Therefore it is not necessary to expand the stack manually.
 552 //
 553 // Manually expanding the stack was historically needed on LinuxThreads
 554 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 555 // it is kept to deal with very rare corner cases:
 556 //
 557 // For one, user may run the VM on an own implementation of threads
 558 // whose stacks are - like the old LinuxThreads - implemented using
 559 // mmap(MAP_GROWSDOWN).
 560 //
 561 // Also, this coding may be needed if the VM is running on the primordial
 562 // thread. Normally we avoid running on the primordial thread; however,
 563 // user may still invoke the VM on the primordial thread.
 564 //
 565 // The following historical comment describes the details about running
 566 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 567 
 568 
 569 // Force Linux kernel to expand current thread stack. If "bottom" is close
 570 // to the stack guard, caller should block all signals.
 571 //
 572 // MAP_GROWSDOWN:
 573 //   A special mmap() flag that is used to implement thread stacks. It tells
 574 //   kernel that the memory region should extend downwards when needed. This
 575 //   allows early versions of LinuxThreads to only mmap the first few pages
 576 //   when creating a new thread. Linux kernel will automatically expand thread
 577 //   stack as needed (on page faults).
 578 //
 579 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 580 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 581 //   region, it's hard to tell if the fault is due to a legitimate stack
 582 //   access or because of reading/writing non-exist memory (e.g. buffer
 583 //   overrun). As a rule, if the fault happens below current stack pointer,
 584 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 585 //   application (see Linux kernel fault.c).
 586 //
 587 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 588 //   stack overflow detection.
 589 //
 590 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 591 //   not use MAP_GROWSDOWN.
 592 //
 593 // To get around the problem and allow stack banging on Linux, we need to
 594 // manually expand thread stack after receiving the SIGSEGV.
 595 //
 596 // There are two ways to expand thread stack to address "bottom", we used
 597 // both of them in JVM before 1.5:
 598 //   1. adjust stack pointer first so that it is below "bottom", and then
 599 //      touch "bottom"
 600 //   2. mmap() the page in question
 601 //
 602 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 603 // if current sp is already near the lower end of page 101, and we need to
 604 // call mmap() to map page 100, it is possible that part of the mmap() frame
 605 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 606 // That will destroy the mmap() frame and cause VM to crash.
 607 //
 608 // The following code works by adjusting sp first, then accessing the "bottom"
 609 // page to force a page fault. Linux kernel will then automatically expand the
 610 // stack mapping.
 611 //
 612 // _expand_stack_to() assumes its frame size is less than page size, which
 613 // should always be true if the function is not inlined.
 614 
 615 static void NOINLINE _expand_stack_to(address bottom) {
 616   address sp;
 617   size_t size;
 618   volatile char *p;
 619 
 620   // Adjust bottom to point to the largest address within the same page, it
 621   // gives us a one-page buffer if alloca() allocates slightly more memory.
 622   bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
 623   bottom += os::Linux::page_size() - 1;
 624 
 625   // sp might be slightly above current stack pointer; if that's the case, we
 626   // will alloca() a little more space than necessary, which is OK. Don't use
 627   // os::current_stack_pointer(), as its result can be slightly below current
 628   // stack pointer, causing us to not alloca enough to reach "bottom".
 629   sp = (address)&sp;
 630 
 631   if (sp > bottom) {
 632     size = sp - bottom;
 633     p = (volatile char *)alloca(size);
 634     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 635     p[0] = '\0';
 636   }
 637 }
 638 
 639 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 640   assert(t!=NULL, "just checking");
 641   assert(t->osthread()->expanding_stack(), "expand should be set");
 642   assert(t->stack_base() != NULL, "stack_base was not initialized");
 643 
 644   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 645     sigset_t mask_all, old_sigset;
 646     sigfillset(&mask_all);
 647     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 648     _expand_stack_to(addr);
 649     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 650     return true;
 651   }
 652   return false;
 653 }
 654 
 655 //////////////////////////////////////////////////////////////////////////////
 656 // create new thread
 657 
 658 // Thread start routine for all newly created threads
 659 static void *thread_native_entry(Thread *thread) {
 660   // Try to randomize the cache line index of hot stack frames.
 661   // This helps when threads of the same stack traces evict each other's
 662   // cache lines. The threads can be either from the same JVM instance, or
 663   // from different JVM instances. The benefit is especially true for
 664   // processors with hyperthreading technology.
 665   static int counter = 0;
 666   int pid = os::current_process_id();
 667   alloca(((pid ^ counter++) & 7) * 128);
 668 
 669   thread->initialize_thread_current();
 670 
 671   OSThread* osthread = thread->osthread();
 672   Monitor* sync = osthread->startThread_lock();
 673 
 674   osthread->set_thread_id(os::current_thread_id());
 675 
 676   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 677     os::current_thread_id(), (uintx) pthread_self());
 678 
 679   if (UseNUMA) {
 680     int lgrp_id = os::numa_get_group_id();
 681     if (lgrp_id != -1) {
 682       thread->set_lgrp_id(lgrp_id);
 683     }
 684   }
 685   // initialize signal mask for this thread
 686   os::Linux::hotspot_sigmask(thread);
 687 
 688   // initialize floating point control register
 689   os::Linux::init_thread_fpu_state();
 690 
 691   // handshaking with parent thread
 692   {
 693     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 694 
 695     // notify parent thread
 696     osthread->set_state(INITIALIZED);
 697     sync->notify_all();
 698 
 699     // wait until os::start_thread()
 700     while (osthread->get_state() == INITIALIZED) {
 701       sync->wait(Mutex::_no_safepoint_check_flag);
 702     }
 703   }
 704 
 705   // call one more level start routine
 706   thread->run();
 707 
 708   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 709     os::current_thread_id(), (uintx) pthread_self());
 710 
 711   // If a thread has not deleted itself ("delete this") as part of its
 712   // termination sequence, we have to ensure thread-local-storage is
 713   // cleared before we actually terminate. No threads should ever be
 714   // deleted asynchronously with respect to their termination.
 715   if (Thread::current_or_null_safe() != NULL) {
 716     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 717     thread->clear_thread_current();
 718   }
 719 
 720   return 0;
 721 }
 722 
 723 bool os::create_thread(Thread* thread, ThreadType thr_type,
 724                        size_t req_stack_size) {
 725   assert(thread->osthread() == NULL, "caller responsible");
 726 
 727   // Allocate the OSThread object
 728   OSThread* osthread = new OSThread(NULL, NULL);
 729   if (osthread == NULL) {
 730     return false;
 731   }
 732 
 733   // set the correct thread state
 734   osthread->set_thread_type(thr_type);
 735 
 736   // Initial state is ALLOCATED but not INITIALIZED
 737   osthread->set_state(ALLOCATED);
 738 
 739   thread->set_osthread(osthread);
 740 
 741   // init thread attributes
 742   pthread_attr_t attr;
 743   pthread_attr_init(&attr);
 744   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 745 
 746   // Calculate stack size if it's not specified by caller.
 747   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 748   // In the Linux NPTL pthread implementation the guard size mechanism
 749   // is not implemented properly. The posix standard requires adding
 750   // the size of the guard pages to the stack size, instead Linux
 751   // takes the space out of 'stacksize'. Thus we adapt the requested
 752   // stack_size by the size of the guard pages to mimick proper
 753   // behaviour. However, be careful not to end up with a size
 754   // of zero due to overflow. Don't add the guard page in that case.
 755   size_t guard_size = os::Linux::default_guard_size(thr_type);
 756   if (stack_size <= SIZE_MAX - guard_size) {
 757     stack_size += guard_size;
 758   }
 759   assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
 760 
 761   int status = pthread_attr_setstacksize(&attr, stack_size);
 762   assert_status(status == 0, status, "pthread_attr_setstacksize");
 763 
 764   // Configure glibc guard page.
 765   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 766 
 767   ThreadState state;
 768 
 769   {
 770     pthread_t tid;
 771     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 772 
 773     char buf[64];
 774     if (ret == 0) {
 775       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 776         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 777     } else {
 778       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 779         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 780     }
 781 
 782     pthread_attr_destroy(&attr);
 783 
 784     if (ret != 0) {
 785       // Need to clean up stuff we've allocated so far
 786       thread->set_osthread(NULL);
 787       delete osthread;
 788       return false;
 789     }
 790 
 791     // Store pthread info into the OSThread
 792     osthread->set_pthread_id(tid);
 793 
 794     // Wait until child thread is either initialized or aborted
 795     {
 796       Monitor* sync_with_child = osthread->startThread_lock();
 797       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 798       while ((state = osthread->get_state()) == ALLOCATED) {
 799         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 800       }
 801     }
 802   }
 803 
 804   // Aborted due to thread limit being reached
 805   if (state == ZOMBIE) {
 806     thread->set_osthread(NULL);
 807     delete osthread;
 808     return false;
 809   }
 810 
 811   // The thread is returned suspended (in state INITIALIZED),
 812   // and is started higher up in the call chain
 813   assert(state == INITIALIZED, "race condition");
 814   return true;
 815 }
 816 
 817 /////////////////////////////////////////////////////////////////////////////
 818 // attach existing thread
 819 
 820 // bootstrap the main thread
 821 bool os::create_main_thread(JavaThread* thread) {
 822   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 823   return create_attached_thread(thread);
 824 }
 825 
 826 bool os::create_attached_thread(JavaThread* thread) {
 827 #ifdef ASSERT
 828   thread->verify_not_published();
 829 #endif
 830 
 831   // Allocate the OSThread object
 832   OSThread* osthread = new OSThread(NULL, NULL);
 833 
 834   if (osthread == NULL) {
 835     return false;
 836   }
 837 
 838   // Store pthread info into the OSThread
 839   osthread->set_thread_id(os::Linux::gettid());
 840   osthread->set_pthread_id(::pthread_self());
 841 
 842   // initialize floating point control register
 843   os::Linux::init_thread_fpu_state();
 844 
 845   // Initial thread state is RUNNABLE
 846   osthread->set_state(RUNNABLE);
 847 
 848   thread->set_osthread(osthread);
 849 
 850   if (UseNUMA) {
 851     int lgrp_id = os::numa_get_group_id();
 852     if (lgrp_id != -1) {
 853       thread->set_lgrp_id(lgrp_id);
 854     }
 855   }
 856 
 857   if (os::Linux::is_initial_thread()) {
 858     // If current thread is initial thread, its stack is mapped on demand,
 859     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 860     // the entire stack region to avoid SEGV in stack banging.
 861     // It is also useful to get around the heap-stack-gap problem on SuSE
 862     // kernel (see 4821821 for details). We first expand stack to the top
 863     // of yellow zone, then enable stack yellow zone (order is significant,
 864     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 865     // is no gap between the last two virtual memory regions.
 866 
 867     JavaThread *jt = (JavaThread *)thread;
 868     address addr = jt->stack_reserved_zone_base();
 869     assert(addr != NULL, "initialization problem?");
 870     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 871 
 872     osthread->set_expanding_stack();
 873     os::Linux::manually_expand_stack(jt, addr);
 874     osthread->clear_expanding_stack();
 875   }
 876 
 877   // initialize signal mask for this thread
 878   // and save the caller's signal mask
 879   os::Linux::hotspot_sigmask(thread);
 880 
 881   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 882     os::current_thread_id(), (uintx) pthread_self());
 883 
 884   return true;
 885 }
 886 
 887 void os::pd_start_thread(Thread* thread) {
 888   OSThread * osthread = thread->osthread();
 889   assert(osthread->get_state() != INITIALIZED, "just checking");
 890   Monitor* sync_with_child = osthread->startThread_lock();
 891   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 892   sync_with_child->notify();
 893 }
 894 
 895 // Free Linux resources related to the OSThread
 896 void os::free_thread(OSThread* osthread) {
 897   assert(osthread != NULL, "osthread not set");
 898 
 899   // We are told to free resources of the argument thread,
 900   // but we can only really operate on the current thread.
 901   assert(Thread::current()->osthread() == osthread,
 902          "os::free_thread but not current thread");
 903 
 904 #ifdef ASSERT
 905   sigset_t current;
 906   sigemptyset(&current);
 907   pthread_sigmask(SIG_SETMASK, NULL, &current);
 908   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 909 #endif
 910 
 911   // Restore caller's signal mask
 912   sigset_t sigmask = osthread->caller_sigmask();
 913   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 914 
 915   delete osthread;
 916 }
 917 
 918 //////////////////////////////////////////////////////////////////////////////
 919 // initial thread
 920 
 921 // Check if current thread is the initial thread, similar to Solaris thr_main.
 922 bool os::Linux::is_initial_thread(void) {
 923   char dummy;
 924   // If called before init complete, thread stack bottom will be null.
 925   // Can be called if fatal error occurs before initialization.
 926   if (initial_thread_stack_bottom() == NULL) return false;
 927   assert(initial_thread_stack_bottom() != NULL &&
 928          initial_thread_stack_size()   != 0,
 929          "os::init did not locate initial thread's stack region");
 930   if ((address)&dummy >= initial_thread_stack_bottom() &&
 931       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 932     return true;
 933   } else {
 934     return false;
 935   }
 936 }
 937 
 938 // Find the virtual memory area that contains addr
 939 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 940   FILE *fp = fopen("/proc/self/maps", "r");
 941   if (fp) {
 942     address low, high;
 943     while (!feof(fp)) {
 944       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 945         if (low <= addr && addr < high) {
 946           if (vma_low)  *vma_low  = low;
 947           if (vma_high) *vma_high = high;
 948           fclose(fp);
 949           return true;
 950         }
 951       }
 952       for (;;) {
 953         int ch = fgetc(fp);
 954         if (ch == EOF || ch == (int)'\n') break;
 955       }
 956     }
 957     fclose(fp);
 958   }
 959   return false;
 960 }
 961 
 962 // Locate initial thread stack. This special handling of initial thread stack
 963 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 964 // bogus value for the primordial process thread. While the launcher has created
 965 // the VM in a new thread since JDK 6, we still have to allow for the use of the
 966 // JNI invocation API from a primordial thread.
 967 void os::Linux::capture_initial_stack(size_t max_size) {
 968 
 969   // max_size is either 0 (which means accept OS default for thread stacks) or
 970   // a user-specified value known to be at least the minimum needed. If we
 971   // are actually on the primordial thread we can make it appear that we have a
 972   // smaller max_size stack by inserting the guard pages at that location. But we
 973   // cannot do anything to emulate a larger stack than what has been provided by
 974   // the OS or threading library. In fact if we try to use a stack greater than
 975   // what is set by rlimit then we will crash the hosting process.
 976 
 977   // Maximum stack size is the easy part, get it from RLIMIT_STACK.
 978   // If this is "unlimited" then it will be a huge value.
 979   struct rlimit rlim;
 980   getrlimit(RLIMIT_STACK, &rlim);
 981   size_t stack_size = rlim.rlim_cur;
 982 
 983   // 6308388: a bug in ld.so will relocate its own .data section to the
 984   //   lower end of primordial stack; reduce ulimit -s value a little bit
 985   //   so we won't install guard page on ld.so's data section.
 986   stack_size -= 2 * page_size();
 987 
 988   // Try to figure out where the stack base (top) is. This is harder.
 989   //
 990   // When an application is started, glibc saves the initial stack pointer in
 991   // a global variable "__libc_stack_end", which is then used by system
 992   // libraries. __libc_stack_end should be pretty close to stack top. The
 993   // variable is available since the very early days. However, because it is
 994   // a private interface, it could disappear in the future.
 995   //
 996   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 997   // to __libc_stack_end, it is very close to stack top, but isn't the real
 998   // stack top. Note that /proc may not exist if VM is running as a chroot
 999   // program, so reading /proc/<pid>/stat could fail. Also the contents of
1000   // /proc/<pid>/stat could change in the future (though unlikely).
1001   //
1002   // We try __libc_stack_end first. If that doesn't work, look for
1003   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1004   // as a hint, which should work well in most cases.
1005 
1006   uintptr_t stack_start;
1007 
1008   // try __libc_stack_end first
1009   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1010   if (p && *p) {
1011     stack_start = *p;
1012   } else {
1013     // see if we can get the start_stack field from /proc/self/stat
1014     FILE *fp;
1015     int pid;
1016     char state;
1017     int ppid;
1018     int pgrp;
1019     int session;
1020     int nr;
1021     int tpgrp;
1022     unsigned long flags;
1023     unsigned long minflt;
1024     unsigned long cminflt;
1025     unsigned long majflt;
1026     unsigned long cmajflt;
1027     unsigned long utime;
1028     unsigned long stime;
1029     long cutime;
1030     long cstime;
1031     long prio;
1032     long nice;
1033     long junk;
1034     long it_real;
1035     uintptr_t start;
1036     uintptr_t vsize;
1037     intptr_t rss;
1038     uintptr_t rsslim;
1039     uintptr_t scodes;
1040     uintptr_t ecode;
1041     int i;
1042 
1043     // Figure what the primordial thread stack base is. Code is inspired
1044     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1045     // followed by command name surrounded by parentheses, state, etc.
1046     char stat[2048];
1047     int statlen;
1048 
1049     fp = fopen("/proc/self/stat", "r");
1050     if (fp) {
1051       statlen = fread(stat, 1, 2047, fp);
1052       stat[statlen] = '\0';
1053       fclose(fp);
1054 
1055       // Skip pid and the command string. Note that we could be dealing with
1056       // weird command names, e.g. user could decide to rename java launcher
1057       // to "java 1.4.2 :)", then the stat file would look like
1058       //                1234 (java 1.4.2 :)) R ... ...
1059       // We don't really need to know the command string, just find the last
1060       // occurrence of ")" and then start parsing from there. See bug 4726580.
1061       char * s = strrchr(stat, ')');
1062 
1063       i = 0;
1064       if (s) {
1065         // Skip blank chars
1066         do { s++; } while (s && isspace(*s));
1067 
1068 #define _UFM UINTX_FORMAT
1069 #define _DFM INTX_FORMAT
1070 
1071         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1072         //              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
1073         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,
1074                    &state,          // 3  %c
1075                    &ppid,           // 4  %d
1076                    &pgrp,           // 5  %d
1077                    &session,        // 6  %d
1078                    &nr,             // 7  %d
1079                    &tpgrp,          // 8  %d
1080                    &flags,          // 9  %lu
1081                    &minflt,         // 10 %lu
1082                    &cminflt,        // 11 %lu
1083                    &majflt,         // 12 %lu
1084                    &cmajflt,        // 13 %lu
1085                    &utime,          // 14 %lu
1086                    &stime,          // 15 %lu
1087                    &cutime,         // 16 %ld
1088                    &cstime,         // 17 %ld
1089                    &prio,           // 18 %ld
1090                    &nice,           // 19 %ld
1091                    &junk,           // 20 %ld
1092                    &it_real,        // 21 %ld
1093                    &start,          // 22 UINTX_FORMAT
1094                    &vsize,          // 23 UINTX_FORMAT
1095                    &rss,            // 24 INTX_FORMAT
1096                    &rsslim,         // 25 UINTX_FORMAT
1097                    &scodes,         // 26 UINTX_FORMAT
1098                    &ecode,          // 27 UINTX_FORMAT
1099                    &stack_start);   // 28 UINTX_FORMAT
1100       }
1101 
1102 #undef _UFM
1103 #undef _DFM
1104 
1105       if (i != 28 - 2) {
1106         assert(false, "Bad conversion from /proc/self/stat");
1107         // product mode - assume we are the initial thread, good luck in the
1108         // embedded case.
1109         warning("Can't detect initial thread stack location - bad conversion");
1110         stack_start = (uintptr_t) &rlim;
1111       }
1112     } else {
1113       // For some reason we can't open /proc/self/stat (for example, running on
1114       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1115       // most cases, so don't abort:
1116       warning("Can't detect initial thread stack location - no /proc/self/stat");
1117       stack_start = (uintptr_t) &rlim;
1118     }
1119   }
1120 
1121   // Now we have a pointer (stack_start) very close to the stack top, the
1122   // next thing to do is to figure out the exact location of stack top. We
1123   // can find out the virtual memory area that contains stack_start by
1124   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1125   // and its upper limit is the real stack top. (again, this would fail if
1126   // running inside chroot, because /proc may not exist.)
1127 
1128   uintptr_t stack_top;
1129   address low, high;
1130   if (find_vma((address)stack_start, &low, &high)) {
1131     // success, "high" is the true stack top. (ignore "low", because initial
1132     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1133     stack_top = (uintptr_t)high;
1134   } else {
1135     // failed, likely because /proc/self/maps does not exist
1136     warning("Can't detect initial thread stack location - find_vma failed");
1137     // best effort: stack_start is normally within a few pages below the real
1138     // stack top, use it as stack top, and reduce stack size so we won't put
1139     // guard page outside stack.
1140     stack_top = stack_start;
1141     stack_size -= 16 * page_size();
1142   }
1143 
1144   // stack_top could be partially down the page so align it
1145   stack_top = align_up(stack_top, page_size());
1146 
1147   // Allowed stack value is minimum of max_size and what we derived from rlimit
1148   if (max_size > 0) {
1149     _initial_thread_stack_size = MIN2(max_size, stack_size);
1150   } else {
1151     // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1152     // clamp it at 8MB as we do on Solaris
1153     _initial_thread_stack_size = MIN2(stack_size, 8*M);
1154   }
1155   _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1156   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1157 
1158   assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1159 
1160   if (log_is_enabled(Info, os, thread)) {
1161     // See if we seem to be on primordial process thread
1162     bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1163                       uintptr_t(&rlim) < stack_top;
1164 
1165     log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1166                          SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1167                          primordial ? "primordial" : "user", max_size / K,  _initial_thread_stack_size / K,
1168                          stack_top, intptr_t(_initial_thread_stack_bottom));
1169   }
1170 }
1171 
1172 ////////////////////////////////////////////////////////////////////////////////
1173 // time support
1174 
1175 // Time since start-up in seconds to a fine granularity.
1176 // Used by VMSelfDestructTimer and the MemProfiler.
1177 double os::elapsedTime() {
1178 
1179   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1180 }
1181 
1182 jlong os::elapsed_counter() {
1183   return javaTimeNanos() - initial_time_count;
1184 }
1185 
1186 jlong os::elapsed_frequency() {
1187   return NANOSECS_PER_SEC; // nanosecond resolution
1188 }
1189 
1190 bool os::supports_vtime() { return true; }
1191 bool os::enable_vtime()   { return false; }
1192 bool os::vtime_enabled()  { return false; }
1193 
1194 double os::elapsedVTime() {
1195   struct rusage usage;
1196   int retval = getrusage(RUSAGE_THREAD, &usage);
1197   if (retval == 0) {
1198     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);
1199   } else {
1200     // better than nothing, but not much
1201     return elapsedTime();
1202   }
1203 }
1204 
1205 jlong os::javaTimeMillis() {
1206   timeval time;
1207   int status = gettimeofday(&time, NULL);
1208   assert(status != -1, "linux error");
1209   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1210 }
1211 
1212 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1213   timeval time;
1214   int status = gettimeofday(&time, NULL);
1215   assert(status != -1, "linux error");
1216   seconds = jlong(time.tv_sec);
1217   nanos = jlong(time.tv_usec) * 1000;
1218 }
1219 
1220 
1221 #ifndef CLOCK_MONOTONIC
1222   #define CLOCK_MONOTONIC (1)
1223 #endif
1224 
1225 void os::Linux::clock_init() {
1226   // we do dlopen's in this particular order due to bug in linux
1227   // dynamical loader (see 6348968) leading to crash on exit
1228   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1229   if (handle == NULL) {
1230     handle = dlopen("librt.so", RTLD_LAZY);
1231   }
1232 
1233   if (handle) {
1234     int (*clock_getres_func)(clockid_t, struct timespec*) =
1235            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1236     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1237            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1238     if (clock_getres_func && clock_gettime_func) {
1239       // See if monotonic clock is supported by the kernel. Note that some
1240       // early implementations simply return kernel jiffies (updated every
1241       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1242       // for nano time (though the monotonic property is still nice to have).
1243       // It's fixed in newer kernels, however clock_getres() still returns
1244       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1245       // resolution for now. Hopefully as people move to new kernels, this
1246       // won't be a problem.
1247       struct timespec res;
1248       struct timespec tp;
1249       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1250           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1251         // yes, monotonic clock is supported
1252         _clock_gettime = clock_gettime_func;
1253         return;
1254       } else {
1255         // close librt if there is no monotonic clock
1256         dlclose(handle);
1257       }
1258     }
1259   }
1260   warning("No monotonic clock was available - timed services may " \
1261           "be adversely affected if the time-of-day clock changes");
1262 }
1263 
1264 #ifndef SYS_clock_getres
1265   #if defined(X86) || defined(PPC64) || defined(S390)
1266     #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1267     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1268   #else
1269     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1270     #define sys_clock_getres(x,y)  -1
1271   #endif
1272 #else
1273   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1274 #endif
1275 
1276 void os::Linux::fast_thread_clock_init() {
1277   if (!UseLinuxPosixThreadCPUClocks) {
1278     return;
1279   }
1280   clockid_t clockid;
1281   struct timespec tp;
1282   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1283       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1284 
1285   // Switch to using fast clocks for thread cpu time if
1286   // the sys_clock_getres() returns 0 error code.
1287   // Note, that some kernels may support the current thread
1288   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1289   // returned by the pthread_getcpuclockid().
1290   // If the fast Posix clocks are supported then the sys_clock_getres()
1291   // must return at least tp.tv_sec == 0 which means a resolution
1292   // better than 1 sec. This is extra check for reliability.
1293 
1294   if (pthread_getcpuclockid_func &&
1295       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1296       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1297     _supports_fast_thread_cpu_time = true;
1298     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1299   }
1300 }
1301 
1302 jlong os::javaTimeNanos() {
1303   if (os::supports_monotonic_clock()) {
1304     struct timespec tp;
1305     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1306     assert(status == 0, "gettime error");
1307     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1308     return result;
1309   } else {
1310     timeval time;
1311     int status = gettimeofday(&time, NULL);
1312     assert(status != -1, "linux error");
1313     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1314     return 1000 * usecs;
1315   }
1316 }
1317 
1318 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1319   if (os::supports_monotonic_clock()) {
1320     info_ptr->max_value = ALL_64_BITS;
1321 
1322     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1323     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1324     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1325   } else {
1326     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1327     info_ptr->max_value = ALL_64_BITS;
1328 
1329     // gettimeofday is a real time clock so it skips
1330     info_ptr->may_skip_backward = true;
1331     info_ptr->may_skip_forward = true;
1332   }
1333 
1334   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1335 }
1336 
1337 // Return the real, user, and system times in seconds from an
1338 // arbitrary fixed point in the past.
1339 bool os::getTimesSecs(double* process_real_time,
1340                       double* process_user_time,
1341                       double* process_system_time) {
1342   struct tms ticks;
1343   clock_t real_ticks = times(&ticks);
1344 
1345   if (real_ticks == (clock_t) (-1)) {
1346     return false;
1347   } else {
1348     double ticks_per_second = (double) clock_tics_per_sec;
1349     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1350     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1351     *process_real_time = ((double) real_ticks) / ticks_per_second;
1352 
1353     return true;
1354   }
1355 }
1356 
1357 
1358 char * os::local_time_string(char *buf, size_t buflen) {
1359   struct tm t;
1360   time_t long_time;
1361   time(&long_time);
1362   localtime_r(&long_time, &t);
1363   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1364                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1365                t.tm_hour, t.tm_min, t.tm_sec);
1366   return buf;
1367 }
1368 
1369 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1370   return localtime_r(clock, res);
1371 }
1372 
1373 ////////////////////////////////////////////////////////////////////////////////
1374 // runtime exit support
1375 
1376 // Note: os::shutdown() might be called very early during initialization, or
1377 // called from signal handler. Before adding something to os::shutdown(), make
1378 // sure it is async-safe and can handle partially initialized VM.
1379 void os::shutdown() {
1380 
1381   // allow PerfMemory to attempt cleanup of any persistent resources
1382   perfMemory_exit();
1383 
1384   // needs to remove object in file system
1385   AttachListener::abort();
1386 
1387   // flush buffered output, finish log files
1388   ostream_abort();
1389 
1390   // Check for abort hook
1391   abort_hook_t abort_hook = Arguments::abort_hook();
1392   if (abort_hook != NULL) {
1393     abort_hook();
1394   }
1395 
1396 }
1397 
1398 // Note: os::abort() might be called very early during initialization, or
1399 // called from signal handler. Before adding something to os::abort(), make
1400 // sure it is async-safe and can handle partially initialized VM.
1401 void os::abort(bool dump_core, void* siginfo, const void* context) {
1402   os::shutdown();
1403   if (dump_core) {
1404 #ifndef PRODUCT
1405     fdStream out(defaultStream::output_fd());
1406     out.print_raw("Current thread is ");
1407     char buf[16];
1408     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1409     out.print_raw_cr(buf);
1410     out.print_raw_cr("Dumping core ...");
1411 #endif
1412     ::abort(); // dump core
1413   }
1414 
1415   ::exit(1);
1416 }
1417 
1418 // Die immediately, no exit hook, no abort hook, no cleanup.
1419 void os::die() {
1420   ::abort();
1421 }
1422 
1423 
1424 // This method is a copy of JDK's sysGetLastErrorString
1425 // from src/solaris/hpi/src/system_md.c
1426 
1427 size_t os::lasterror(char *buf, size_t len) {
1428   if (errno == 0)  return 0;
1429 
1430   const char *s = os::strerror(errno);
1431   size_t n = ::strlen(s);
1432   if (n >= len) {
1433     n = len - 1;
1434   }
1435   ::strncpy(buf, s, n);
1436   buf[n] = '\0';
1437   return n;
1438 }
1439 
1440 // thread_id is kernel thread id (similar to Solaris LWP id)
1441 intx os::current_thread_id() { return os::Linux::gettid(); }
1442 int os::current_process_id() {
1443   return ::getpid();
1444 }
1445 
1446 // DLL functions
1447 
1448 const char* os::dll_file_extension() { return ".so"; }
1449 
1450 // This must be hard coded because it's the system's temporary
1451 // directory not the java application's temp directory, ala java.io.tmpdir.
1452 const char* os::get_temp_directory() { return "/tmp"; }
1453 
1454 static bool file_exists(const char* filename) {
1455   struct stat statbuf;
1456   if (filename == NULL || strlen(filename) == 0) {
1457     return false;
1458   }
1459   return os::stat(filename, &statbuf) == 0;
1460 }
1461 
1462 // check if addr is inside libjvm.so
1463 bool os::address_is_in_vm(address addr) {
1464   static address libjvm_base_addr;
1465   Dl_info dlinfo;
1466 
1467   if (libjvm_base_addr == NULL) {
1468     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1469       libjvm_base_addr = (address)dlinfo.dli_fbase;
1470     }
1471     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1472   }
1473 
1474   if (dladdr((void *)addr, &dlinfo) != 0) {
1475     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1476   }
1477 
1478   return false;
1479 }
1480 
1481 bool os::dll_address_to_function_name(address addr, char *buf,
1482                                       int buflen, int *offset,
1483                                       bool demangle) {
1484   // buf is not optional, but offset is optional
1485   assert(buf != NULL, "sanity check");
1486 
1487   Dl_info dlinfo;
1488 
1489   if (dladdr((void*)addr, &dlinfo) != 0) {
1490     // see if we have a matching symbol
1491     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1492       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1493         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1494       }
1495       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1496       return true;
1497     }
1498     // no matching symbol so try for just file info
1499     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1500       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1501                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1502         return true;
1503       }
1504     }
1505   }
1506 
1507   buf[0] = '\0';
1508   if (offset != NULL) *offset = -1;
1509   return false;
1510 }
1511 
1512 struct _address_to_library_name {
1513   address addr;          // input : memory address
1514   size_t  buflen;        //         size of fname
1515   char*   fname;         // output: library name
1516   address base;          //         library base addr
1517 };
1518 
1519 static int address_to_library_name_callback(struct dl_phdr_info *info,
1520                                             size_t size, void *data) {
1521   int i;
1522   bool found = false;
1523   address libbase = NULL;
1524   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1525 
1526   // iterate through all loadable segments
1527   for (i = 0; i < info->dlpi_phnum; i++) {
1528     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1529     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1530       // base address of a library is the lowest address of its loaded
1531       // segments.
1532       if (libbase == NULL || libbase > segbase) {
1533         libbase = segbase;
1534       }
1535       // see if 'addr' is within current segment
1536       if (segbase <= d->addr &&
1537           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1538         found = true;
1539       }
1540     }
1541   }
1542 
1543   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1544   // so dll_address_to_library_name() can fall through to use dladdr() which
1545   // can figure out executable name from argv[0].
1546   if (found && info->dlpi_name && info->dlpi_name[0]) {
1547     d->base = libbase;
1548     if (d->fname) {
1549       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1550     }
1551     return 1;
1552   }
1553   return 0;
1554 }
1555 
1556 bool os::dll_address_to_library_name(address addr, char* buf,
1557                                      int buflen, int* offset) {
1558   // buf is not optional, but offset is optional
1559   assert(buf != NULL, "sanity check");
1560 
1561   Dl_info dlinfo;
1562   struct _address_to_library_name data;
1563 
1564   // There is a bug in old glibc dladdr() implementation that it could resolve
1565   // to wrong library name if the .so file has a base address != NULL. Here
1566   // we iterate through the program headers of all loaded libraries to find
1567   // out which library 'addr' really belongs to. This workaround can be
1568   // removed once the minimum requirement for glibc is moved to 2.3.x.
1569   data.addr = addr;
1570   data.fname = buf;
1571   data.buflen = buflen;
1572   data.base = NULL;
1573   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1574 
1575   if (rslt) {
1576     // buf already contains library name
1577     if (offset) *offset = addr - data.base;
1578     return true;
1579   }
1580   if (dladdr((void*)addr, &dlinfo) != 0) {
1581     if (dlinfo.dli_fname != NULL) {
1582       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1583     }
1584     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1585       *offset = addr - (address)dlinfo.dli_fbase;
1586     }
1587     return true;
1588   }
1589 
1590   buf[0] = '\0';
1591   if (offset) *offset = -1;
1592   return false;
1593 }
1594 
1595 // Loads .dll/.so and
1596 // in case of error it checks if .dll/.so was built for the
1597 // same architecture as Hotspot is running on
1598 
1599 
1600 // Remember the stack's state. The Linux dynamic linker will change
1601 // the stack to 'executable' at most once, so we must safepoint only once.
1602 bool os::Linux::_stack_is_executable = false;
1603 
1604 // VM operation that loads a library.  This is necessary if stack protection
1605 // of the Java stacks can be lost during loading the library.  If we
1606 // do not stop the Java threads, they can stack overflow before the stacks
1607 // are protected again.
1608 class VM_LinuxDllLoad: public VM_Operation {
1609  private:
1610   const char *_filename;
1611   char *_ebuf;
1612   int _ebuflen;
1613   void *_lib;
1614  public:
1615   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1616     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1617   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1618   void doit() {
1619     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1620     os::Linux::_stack_is_executable = true;
1621   }
1622   void* loaded_library() { return _lib; }
1623 };
1624 
1625 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1626   void * result = NULL;
1627   bool load_attempted = false;
1628 
1629   // Check whether the library to load might change execution rights
1630   // of the stack. If they are changed, the protection of the stack
1631   // guard pages will be lost. We need a safepoint to fix this.
1632   //
1633   // See Linux man page execstack(8) for more info.
1634   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1635     if (!ElfFile::specifies_noexecstack(filename)) {
1636       if (!is_init_completed()) {
1637         os::Linux::_stack_is_executable = true;
1638         // This is OK - No Java threads have been created yet, and hence no
1639         // stack guard pages to fix.
1640         //
1641         // This should happen only when you are building JDK7 using a very
1642         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1643         //
1644         // Dynamic loader will make all stacks executable after
1645         // this function returns, and will not do that again.
1646 #ifdef ASSERT
1647         ThreadsListHandle tlh;
1648         assert(tlh.length() == 0, "no Java threads should exist yet.");
1649 #endif
1650       } else {
1651         warning("You have loaded library %s which might have disabled stack guard. "
1652                 "The VM will try to fix the stack guard now.\n"
1653                 "It's highly recommended that you fix the library with "
1654                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1655                 filename);
1656 
1657         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1658         JavaThread *jt = JavaThread::current();
1659         if (jt->thread_state() != _thread_in_native) {
1660           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1661           // that requires ExecStack. Cannot enter safe point. Let's give up.
1662           warning("Unable to fix stack guard. Giving up.");
1663         } else {
1664           if (!LoadExecStackDllInVMThread) {
1665             // This is for the case where the DLL has an static
1666             // constructor function that executes JNI code. We cannot
1667             // load such DLLs in the VMThread.
1668             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1669           }
1670 
1671           ThreadInVMfromNative tiv(jt);
1672           debug_only(VMNativeEntryWrapper vew;)
1673 
1674           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1675           VMThread::execute(&op);
1676           if (LoadExecStackDllInVMThread) {
1677             result = op.loaded_library();
1678           }
1679           load_attempted = true;
1680         }
1681       }
1682     }
1683   }
1684 
1685   if (!load_attempted) {
1686     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1687   }
1688 
1689   if (result != NULL) {
1690     // Successful loading
1691     return result;
1692   }
1693 
1694   Elf32_Ehdr elf_head;
1695   int diag_msg_max_length=ebuflen-strlen(ebuf);
1696   char* diag_msg_buf=ebuf+strlen(ebuf);
1697 
1698   if (diag_msg_max_length==0) {
1699     // No more space in ebuf for additional diagnostics message
1700     return NULL;
1701   }
1702 
1703 
1704   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1705 
1706   if (file_descriptor < 0) {
1707     // Can't open library, report dlerror() message
1708     return NULL;
1709   }
1710 
1711   bool failed_to_read_elf_head=
1712     (sizeof(elf_head)!=
1713      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1714 
1715   ::close(file_descriptor);
1716   if (failed_to_read_elf_head) {
1717     // file i/o error - report dlerror() msg
1718     return NULL;
1719   }
1720 
1721   typedef struct {
1722     Elf32_Half    code;         // Actual value as defined in elf.h
1723     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1724     unsigned char elf_class;    // 32 or 64 bit
1725     unsigned char endianess;    // MSB or LSB
1726     char*         name;         // String representation
1727   } arch_t;
1728 
1729 #ifndef EM_486
1730   #define EM_486          6               /* Intel 80486 */
1731 #endif
1732 #ifndef EM_AARCH64
1733   #define EM_AARCH64    183               /* ARM AARCH64 */
1734 #endif
1735 
1736   static const arch_t arch_array[]={
1737     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1738     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1739     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1740     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1741     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1742     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1743     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1744     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1745 #if defined(VM_LITTLE_ENDIAN)
1746     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1747     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1748 #else
1749     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1750     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1751 #endif
1752     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1753     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1754     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1755     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1756     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1757     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1758     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1759     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1760   };
1761 
1762 #if  (defined IA32)
1763   static  Elf32_Half running_arch_code=EM_386;
1764 #elif   (defined AMD64)
1765   static  Elf32_Half running_arch_code=EM_X86_64;
1766 #elif  (defined IA64)
1767   static  Elf32_Half running_arch_code=EM_IA_64;
1768 #elif  (defined __sparc) && (defined _LP64)
1769   static  Elf32_Half running_arch_code=EM_SPARCV9;
1770 #elif  (defined __sparc) && (!defined _LP64)
1771   static  Elf32_Half running_arch_code=EM_SPARC;
1772 #elif  (defined __powerpc64__)
1773   static  Elf32_Half running_arch_code=EM_PPC64;
1774 #elif  (defined __powerpc__)
1775   static  Elf32_Half running_arch_code=EM_PPC;
1776 #elif  (defined AARCH64)
1777   static  Elf32_Half running_arch_code=EM_AARCH64;
1778 #elif  (defined ARM)
1779   static  Elf32_Half running_arch_code=EM_ARM;
1780 #elif  (defined S390)
1781   static  Elf32_Half running_arch_code=EM_S390;
1782 #elif  (defined ALPHA)
1783   static  Elf32_Half running_arch_code=EM_ALPHA;
1784 #elif  (defined MIPSEL)
1785   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1786 #elif  (defined PARISC)
1787   static  Elf32_Half running_arch_code=EM_PARISC;
1788 #elif  (defined MIPS)
1789   static  Elf32_Half running_arch_code=EM_MIPS;
1790 #elif  (defined M68K)
1791   static  Elf32_Half running_arch_code=EM_68K;
1792 #elif  (defined SH)
1793   static  Elf32_Half running_arch_code=EM_SH;
1794 #else
1795     #error Method os::dll_load requires that one of following is defined:\
1796         AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1797 #endif
1798 
1799   // Identify compatability class for VM's architecture and library's architecture
1800   // Obtain string descriptions for architectures
1801 
1802   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1803   int running_arch_index=-1;
1804 
1805   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1806     if (running_arch_code == arch_array[i].code) {
1807       running_arch_index    = i;
1808     }
1809     if (lib_arch.code == arch_array[i].code) {
1810       lib_arch.compat_class = arch_array[i].compat_class;
1811       lib_arch.name         = arch_array[i].name;
1812     }
1813   }
1814 
1815   assert(running_arch_index != -1,
1816          "Didn't find running architecture code (running_arch_code) in arch_array");
1817   if (running_arch_index == -1) {
1818     // Even though running architecture detection failed
1819     // we may still continue with reporting dlerror() message
1820     return NULL;
1821   }
1822 
1823   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1824     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1825     return NULL;
1826   }
1827 
1828 #ifndef S390
1829   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1830     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1831     return NULL;
1832   }
1833 #endif // !S390
1834 
1835   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1836     if (lib_arch.name!=NULL) {
1837       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1838                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1839                  lib_arch.name, arch_array[running_arch_index].name);
1840     } else {
1841       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1842                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1843                  lib_arch.code,
1844                  arch_array[running_arch_index].name);
1845     }
1846   }
1847 
1848   return NULL;
1849 }
1850 
1851 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1852                                 int ebuflen) {
1853   void * result = ::dlopen(filename, RTLD_LAZY);
1854   if (result == NULL) {
1855     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1856     ebuf[ebuflen-1] = '\0';
1857   }
1858   return result;
1859 }
1860 
1861 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1862                                        int ebuflen) {
1863   void * result = NULL;
1864   if (LoadExecStackDllInVMThread) {
1865     result = dlopen_helper(filename, ebuf, ebuflen);
1866   }
1867 
1868   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1869   // library that requires an executable stack, or which does not have this
1870   // stack attribute set, dlopen changes the stack attribute to executable. The
1871   // read protection of the guard pages gets lost.
1872   //
1873   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1874   // may have been queued at the same time.
1875 
1876   if (!_stack_is_executable) {
1877     for (JavaThreadIteratorWithHandle jtiwh; JavaThread *jt = jtiwh.next(); ) {
1878       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1879           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1880         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1881           warning("Attempt to reguard stack yellow zone failed.");
1882         }
1883       }
1884     }
1885   }
1886 
1887   return result;
1888 }
1889 
1890 void* os::dll_lookup(void* handle, const char* name) {
1891   void* res = dlsym(handle, name);
1892   return res;
1893 }
1894 
1895 void* os::get_default_process_handle() {
1896   return (void*)::dlopen(NULL, RTLD_LAZY);
1897 }
1898 
1899 static bool _print_ascii_file(const char* filename, outputStream* st) {
1900   int fd = ::open(filename, O_RDONLY);
1901   if (fd == -1) {
1902     return false;
1903   }
1904 
1905   char buf[33];
1906   int bytes;
1907   buf[32] = '\0';
1908   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1909     st->print_raw(buf, bytes);
1910   }
1911 
1912   ::close(fd);
1913 
1914   return true;
1915 }
1916 
1917 void os::print_dll_info(outputStream *st) {
1918   st->print_cr("Dynamic libraries:");
1919 
1920   char fname[32];
1921   pid_t pid = os::Linux::gettid();
1922 
1923   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1924 
1925   if (!_print_ascii_file(fname, st)) {
1926     st->print("Can not get library information for pid = %d\n", pid);
1927   }
1928 }
1929 
1930 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1931   FILE *procmapsFile = NULL;
1932 
1933   // Open the procfs maps file for the current process
1934   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1935     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1936     char line[PATH_MAX + 100];
1937 
1938     // Read line by line from 'file'
1939     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1940       u8 base, top, offset, inode;
1941       char permissions[5];
1942       char device[6];
1943       char name[PATH_MAX + 1];
1944 
1945       // Parse fields from line
1946       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1947              &base, &top, permissions, &offset, device, &inode, name);
1948 
1949       // Filter by device id '00:00' so that we only get file system mapped files.
1950       if (strcmp(device, "00:00") != 0) {
1951 
1952         // Call callback with the fields of interest
1953         if(callback(name, (address)base, (address)top, param)) {
1954           // Oops abort, callback aborted
1955           fclose(procmapsFile);
1956           return 1;
1957         }
1958       }
1959     }
1960     fclose(procmapsFile);
1961   }
1962   return 0;
1963 }
1964 
1965 void os::print_os_info_brief(outputStream* st) {
1966   os::Linux::print_distro_info(st);
1967 
1968   os::Posix::print_uname_info(st);
1969 
1970   os::Linux::print_libversion_info(st);
1971 
1972 }
1973 
1974 void os::print_os_info(outputStream* st) {
1975   st->print("OS:");
1976 
1977   os::Linux::print_distro_info(st);
1978 
1979   os::Posix::print_uname_info(st);
1980 
1981   // Print warning if unsafe chroot environment detected
1982   if (unsafe_chroot_detected) {
1983     st->print("WARNING!! ");
1984     st->print_cr("%s", unstable_chroot_error);
1985   }
1986 
1987   os::Linux::print_libversion_info(st);
1988 
1989   os::Posix::print_rlimit_info(st);
1990 
1991   os::Posix::print_load_average(st);
1992 
1993   os::Linux::print_full_memory_info(st);
1994 
1995   os::Linux::print_container_info(st);
1996 }
1997 
1998 // Try to identify popular distros.
1999 // Most Linux distributions have a /etc/XXX-release file, which contains
2000 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2001 // file that also contains the OS version string. Some have more than one
2002 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2003 // /etc/redhat-release.), so the order is important.
2004 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2005 // their own specific XXX-release file as well as a redhat-release file.
2006 // Because of this the XXX-release file needs to be searched for before the
2007 // redhat-release file.
2008 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2009 // search for redhat-release / SuSE-release needs to be before lsb-release.
2010 // Since the lsb-release file is the new standard it needs to be searched
2011 // before the older style release files.
2012 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2013 // next to last resort.  The os-release file is a new standard that contains
2014 // distribution information and the system-release file seems to be an old
2015 // standard that has been replaced by the lsb-release and os-release files.
2016 // Searching for the debian_version file is the last resort.  It contains
2017 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2018 // "Debian " is printed before the contents of the debian_version file.
2019 
2020 const char* distro_files[] = {
2021   "/etc/oracle-release",
2022   "/etc/mandriva-release",
2023   "/etc/mandrake-release",
2024   "/etc/sun-release",
2025   "/etc/redhat-release",
2026   "/etc/SuSE-release",
2027   "/etc/lsb-release",
2028   "/etc/turbolinux-release",
2029   "/etc/gentoo-release",
2030   "/etc/ltib-release",
2031   "/etc/angstrom-version",
2032   "/etc/system-release",
2033   "/etc/os-release",
2034   NULL };
2035 
2036 void os::Linux::print_distro_info(outputStream* st) {
2037   for (int i = 0;; i++) {
2038     const char* file = distro_files[i];
2039     if (file == NULL) {
2040       break;  // done
2041     }
2042     // If file prints, we found it.
2043     if (_print_ascii_file(file, st)) {
2044       return;
2045     }
2046   }
2047 
2048   if (file_exists("/etc/debian_version")) {
2049     st->print("Debian ");
2050     _print_ascii_file("/etc/debian_version", st);
2051   } else {
2052     st->print("Linux");
2053   }
2054   st->cr();
2055 }
2056 
2057 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2058   char buf[256];
2059   while (fgets(buf, sizeof(buf), fp)) {
2060     // Edit out extra stuff in expected format
2061     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2062       char* ptr = strstr(buf, "\"");  // the name is in quotes
2063       if (ptr != NULL) {
2064         ptr++; // go beyond first quote
2065         char* nl = strchr(ptr, '\"');
2066         if (nl != NULL) *nl = '\0';
2067         strncpy(distro, ptr, length);
2068       } else {
2069         ptr = strstr(buf, "=");
2070         ptr++; // go beyond equals then
2071         char* nl = strchr(ptr, '\n');
2072         if (nl != NULL) *nl = '\0';
2073         strncpy(distro, ptr, length);
2074       }
2075       return;
2076     } else if (get_first_line) {
2077       char* nl = strchr(buf, '\n');
2078       if (nl != NULL) *nl = '\0';
2079       strncpy(distro, buf, length);
2080       return;
2081     }
2082   }
2083   // print last line and close
2084   char* nl = strchr(buf, '\n');
2085   if (nl != NULL) *nl = '\0';
2086   strncpy(distro, buf, length);
2087 }
2088 
2089 static void parse_os_info(char* distro, size_t length, const char* file) {
2090   FILE* fp = fopen(file, "r");
2091   if (fp != NULL) {
2092     // if suse format, print out first line
2093     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2094     parse_os_info_helper(fp, distro, length, get_first_line);
2095     fclose(fp);
2096   }
2097 }
2098 
2099 void os::get_summary_os_info(char* buf, size_t buflen) {
2100   for (int i = 0;; i++) {
2101     const char* file = distro_files[i];
2102     if (file == NULL) {
2103       break; // ran out of distro_files
2104     }
2105     if (file_exists(file)) {
2106       parse_os_info(buf, buflen, file);
2107       return;
2108     }
2109   }
2110   // special case for debian
2111   if (file_exists("/etc/debian_version")) {
2112     strncpy(buf, "Debian ", buflen);
2113     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2114   } else {
2115     strncpy(buf, "Linux", buflen);
2116   }
2117 }
2118 
2119 void os::Linux::print_libversion_info(outputStream* st) {
2120   // libc, pthread
2121   st->print("libc:");
2122   st->print("%s ", os::Linux::glibc_version());
2123   st->print("%s ", os::Linux::libpthread_version());
2124   st->cr();
2125 }
2126 
2127 void os::Linux::print_full_memory_info(outputStream* st) {
2128   st->print("\n/proc/meminfo:\n");
2129   _print_ascii_file("/proc/meminfo", st);
2130   st->cr();
2131 }
2132 
2133 void os::Linux::print_container_info(outputStream* st) {
2134   if (OSContainer::is_containerized()) {
2135     st->print("container (cgroup) information:\n");
2136 
2137     char *p = OSContainer::container_type();
2138     if (p == NULL)
2139       st->print("container_type() failed\n");
2140     else {
2141       st->print("container_type: %s\n", p);
2142     }
2143 
2144     p = OSContainer::cpu_cpuset_cpus();
2145     if (p == NULL)
2146       st->print("cpu_cpuset_cpus() failed\n");
2147     else {
2148       st->print("cpu_cpuset_cpus: %s\n", p);
2149       free(p);
2150     }
2151 
2152     p = OSContainer::cpu_cpuset_memory_nodes();
2153     if (p < 0)
2154       st->print("cpu_memory_nodes() failed\n");
2155     else {
2156       st->print("cpu_memory_nodes: %s\n", p);
2157       free(p);
2158     }
2159 
2160     int i = OSContainer::active_processor_count();
2161     if (i < 0)
2162       st->print("active_processor_count() failed\n");
2163     else
2164       st->print("active_processor_count: %d\n", i);
2165 
2166     i = OSContainer::cpu_quota();
2167     st->print("cpu_quota: %d\n", i);
2168 
2169     i = OSContainer::cpu_period();
2170     st->print("cpu_period: %d\n", i);
2171 
2172     i = OSContainer::cpu_shares();
2173     st->print("cpu_shares: %d\n", i);
2174 
2175     jlong j = OSContainer::memory_limit_in_bytes();
2176     st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2177 
2178     j = OSContainer::memory_and_swap_limit_in_bytes();
2179     st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2180 
2181     j = OSContainer::memory_soft_limit_in_bytes();
2182     st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2183 
2184     j = OSContainer::OSContainer::memory_usage_in_bytes();
2185     st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2186 
2187     j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2188     st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2189     st->cr();
2190   }
2191 }
2192 
2193 void os::print_memory_info(outputStream* st) {
2194 
2195   st->print("Memory:");
2196   st->print(" %dk page", os::vm_page_size()>>10);
2197 
2198   // values in struct sysinfo are "unsigned long"
2199   struct sysinfo si;
2200   sysinfo(&si);
2201 
2202   st->print(", physical " UINT64_FORMAT "k",
2203             os::physical_memory() >> 10);
2204   st->print("(" UINT64_FORMAT "k free)",
2205             os::available_memory() >> 10);
2206   st->print(", swap " UINT64_FORMAT "k",
2207             ((jlong)si.totalswap * si.mem_unit) >> 10);
2208   st->print("(" UINT64_FORMAT "k free)",
2209             ((jlong)si.freeswap * si.mem_unit) >> 10);
2210   st->cr();
2211 }
2212 
2213 // Print the first "model name" line and the first "flags" line
2214 // that we find and nothing more. We assume "model name" comes
2215 // before "flags" so if we find a second "model name", then the
2216 // "flags" field is considered missing.
2217 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2218 #if defined(IA32) || defined(AMD64)
2219   // Other platforms have less repetitive cpuinfo files
2220   FILE *fp = fopen("/proc/cpuinfo", "r");
2221   if (fp) {
2222     while (!feof(fp)) {
2223       if (fgets(buf, buflen, fp)) {
2224         // Assume model name comes before flags
2225         bool model_name_printed = false;
2226         if (strstr(buf, "model name") != NULL) {
2227           if (!model_name_printed) {
2228             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2229             st->print_raw(buf);
2230             model_name_printed = true;
2231           } else {
2232             // model name printed but not flags?  Odd, just return
2233             fclose(fp);
2234             return true;
2235           }
2236         }
2237         // print the flags line too
2238         if (strstr(buf, "flags") != NULL) {
2239           st->print_raw(buf);
2240           fclose(fp);
2241           return true;
2242         }
2243       }
2244     }
2245     fclose(fp);
2246   }
2247 #endif // x86 platforms
2248   return false;
2249 }
2250 
2251 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2252   // Only print the model name if the platform provides this as a summary
2253   if (!print_model_name_and_flags(st, buf, buflen)) {
2254     st->print("\n/proc/cpuinfo:\n");
2255     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2256       st->print_cr("  <Not Available>");
2257     }
2258   }
2259 }
2260 
2261 #if defined(AMD64) || defined(IA32) || defined(X32)
2262 const char* search_string = "model name";
2263 #elif defined(M68K)
2264 const char* search_string = "CPU";
2265 #elif defined(PPC64)
2266 const char* search_string = "cpu";
2267 #elif defined(S390)
2268 const char* search_string = "processor";
2269 #elif defined(SPARC)
2270 const char* search_string = "cpu";
2271 #else
2272 const char* search_string = "Processor";
2273 #endif
2274 
2275 // Parses the cpuinfo file for string representing the model name.
2276 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2277   FILE* fp = fopen("/proc/cpuinfo", "r");
2278   if (fp != NULL) {
2279     while (!feof(fp)) {
2280       char buf[256];
2281       if (fgets(buf, sizeof(buf), fp)) {
2282         char* start = strstr(buf, search_string);
2283         if (start != NULL) {
2284           char *ptr = start + strlen(search_string);
2285           char *end = buf + strlen(buf);
2286           while (ptr != end) {
2287              // skip whitespace and colon for the rest of the name.
2288              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2289                break;
2290              }
2291              ptr++;
2292           }
2293           if (ptr != end) {
2294             // reasonable string, get rid of newline and keep the rest
2295             char* nl = strchr(buf, '\n');
2296             if (nl != NULL) *nl = '\0';
2297             strncpy(cpuinfo, ptr, length);
2298             fclose(fp);
2299             return;
2300           }
2301         }
2302       }
2303     }
2304     fclose(fp);
2305   }
2306   // cpuinfo not found or parsing failed, just print generic string.  The entire
2307   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2308 #if   defined(AARCH64)
2309   strncpy(cpuinfo, "AArch64", length);
2310 #elif defined(AMD64)
2311   strncpy(cpuinfo, "x86_64", length);
2312 #elif defined(ARM)  // Order wrt. AARCH64 is relevant!
2313   strncpy(cpuinfo, "ARM", length);
2314 #elif defined(IA32)
2315   strncpy(cpuinfo, "x86_32", length);
2316 #elif defined(IA64)
2317   strncpy(cpuinfo, "IA64", length);
2318 #elif defined(PPC)
2319   strncpy(cpuinfo, "PPC64", length);
2320 #elif defined(S390)
2321   strncpy(cpuinfo, "S390", length);
2322 #elif defined(SPARC)
2323   strncpy(cpuinfo, "sparcv9", length);
2324 #elif defined(ZERO_LIBARCH)
2325   strncpy(cpuinfo, ZERO_LIBARCH, length);
2326 #else
2327   strncpy(cpuinfo, "unknown", length);
2328 #endif
2329 }
2330 
2331 static void print_signal_handler(outputStream* st, int sig,
2332                                  char* buf, size_t buflen);
2333 
2334 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2335   st->print_cr("Signal Handlers:");
2336   print_signal_handler(st, SIGSEGV, buf, buflen);
2337   print_signal_handler(st, SIGBUS , buf, buflen);
2338   print_signal_handler(st, SIGFPE , buf, buflen);
2339   print_signal_handler(st, SIGPIPE, buf, buflen);
2340   print_signal_handler(st, SIGXFSZ, buf, buflen);
2341   print_signal_handler(st, SIGILL , buf, buflen);
2342   print_signal_handler(st, SR_signum, buf, buflen);
2343   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2344   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2345   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2346   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2347 #if defined(PPC64)
2348   print_signal_handler(st, SIGTRAP, buf, buflen);
2349 #endif
2350 }
2351 
2352 static char saved_jvm_path[MAXPATHLEN] = {0};
2353 
2354 // Find the full path to the current module, libjvm.so
2355 void os::jvm_path(char *buf, jint buflen) {
2356   // Error checking.
2357   if (buflen < MAXPATHLEN) {
2358     assert(false, "must use a large-enough buffer");
2359     buf[0] = '\0';
2360     return;
2361   }
2362   // Lazy resolve the path to current module.
2363   if (saved_jvm_path[0] != 0) {
2364     strcpy(buf, saved_jvm_path);
2365     return;
2366   }
2367 
2368   char dli_fname[MAXPATHLEN];
2369   bool ret = dll_address_to_library_name(
2370                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2371                                          dli_fname, sizeof(dli_fname), NULL);
2372   assert(ret, "cannot locate libjvm");
2373   char *rp = NULL;
2374   if (ret && dli_fname[0] != '\0') {
2375     rp = os::Posix::realpath(dli_fname, buf, buflen);
2376   }
2377   if (rp == NULL) {
2378     return;
2379   }
2380 
2381   if (Arguments::sun_java_launcher_is_altjvm()) {
2382     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2383     // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2384     // If "/jre/lib/" appears at the right place in the string, then
2385     // assume we are installed in a JDK and we're done. Otherwise, check
2386     // for a JAVA_HOME environment variable and fix up the path so it
2387     // looks like libjvm.so is installed there (append a fake suffix
2388     // hotspot/libjvm.so).
2389     const char *p = buf + strlen(buf) - 1;
2390     for (int count = 0; p > buf && count < 5; ++count) {
2391       for (--p; p > buf && *p != '/'; --p)
2392         /* empty */ ;
2393     }
2394 
2395     if (strncmp(p, "/jre/lib/", 9) != 0) {
2396       // Look for JAVA_HOME in the environment.
2397       char* java_home_var = ::getenv("JAVA_HOME");
2398       if (java_home_var != NULL && java_home_var[0] != 0) {
2399         char* jrelib_p;
2400         int len;
2401 
2402         // Check the current module name "libjvm.so".
2403         p = strrchr(buf, '/');
2404         if (p == NULL) {
2405           return;
2406         }
2407         assert(strstr(p, "/libjvm") == p, "invalid library name");
2408 
2409         rp = os::Posix::realpath(java_home_var, buf, buflen);
2410         if (rp == NULL) {
2411           return;
2412         }
2413 
2414         // determine if this is a legacy image or modules image
2415         // modules image doesn't have "jre" subdirectory
2416         len = strlen(buf);
2417         assert(len < buflen, "Ran out of buffer room");
2418         jrelib_p = buf + len;
2419         snprintf(jrelib_p, buflen-len, "/jre/lib");
2420         if (0 != access(buf, F_OK)) {
2421           snprintf(jrelib_p, buflen-len, "/lib");
2422         }
2423 
2424         if (0 == access(buf, F_OK)) {
2425           // Use current module name "libjvm.so"
2426           len = strlen(buf);
2427           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2428         } else {
2429           // Go back to path of .so
2430           rp = os::Posix::realpath(dli_fname, buf, buflen);
2431           if (rp == NULL) {
2432             return;
2433           }
2434         }
2435       }
2436     }
2437   }
2438 
2439   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2440   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2441 }
2442 
2443 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2444   // no prefix required, not even "_"
2445 }
2446 
2447 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2448   // no suffix required
2449 }
2450 
2451 ////////////////////////////////////////////////////////////////////////////////
2452 // sun.misc.Signal support
2453 
2454 static volatile jint sigint_count = 0;
2455 
2456 static void UserHandler(int sig, void *siginfo, void *context) {
2457   // 4511530 - sem_post is serialized and handled by the manager thread. When
2458   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2459   // don't want to flood the manager thread with sem_post requests.
2460   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2461     return;
2462   }
2463 
2464   // Ctrl-C is pressed during error reporting, likely because the error
2465   // handler fails to abort. Let VM die immediately.
2466   if (sig == SIGINT && VMError::is_error_reported()) {
2467     os::die();
2468   }
2469 
2470   os::signal_notify(sig);
2471 }
2472 
2473 void* os::user_handler() {
2474   return CAST_FROM_FN_PTR(void*, UserHandler);
2475 }
2476 
2477 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2478   struct timespec ts;
2479   // Semaphore's are always associated with CLOCK_REALTIME
2480   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2481   // see unpackTime for discussion on overflow checking
2482   if (sec >= MAX_SECS) {
2483     ts.tv_sec += MAX_SECS;
2484     ts.tv_nsec = 0;
2485   } else {
2486     ts.tv_sec += sec;
2487     ts.tv_nsec += nsec;
2488     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2489       ts.tv_nsec -= NANOSECS_PER_SEC;
2490       ++ts.tv_sec; // note: this must be <= max_secs
2491     }
2492   }
2493 
2494   return ts;
2495 }
2496 
2497 extern "C" {
2498   typedef void (*sa_handler_t)(int);
2499   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2500 }
2501 
2502 void* os::signal(int signal_number, void* handler) {
2503   struct sigaction sigAct, oldSigAct;
2504 
2505   sigfillset(&(sigAct.sa_mask));
2506   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2507   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2508 
2509   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2510     // -1 means registration failed
2511     return (void *)-1;
2512   }
2513 
2514   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2515 }
2516 
2517 void os::signal_raise(int signal_number) {
2518   ::raise(signal_number);
2519 }
2520 
2521 // The following code is moved from os.cpp for making this
2522 // code platform specific, which it is by its very nature.
2523 
2524 // Will be modified when max signal is changed to be dynamic
2525 int os::sigexitnum_pd() {
2526   return NSIG;
2527 }
2528 
2529 // a counter for each possible signal value
2530 static volatile jint pending_signals[NSIG+1] = { 0 };
2531 
2532 // Linux(POSIX) specific hand shaking semaphore.
2533 static sem_t sig_sem;
2534 static PosixSemaphore sr_semaphore;
2535 
2536 void os::signal_init_pd() {
2537   // Initialize signal structures
2538   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2539 
2540   // Initialize signal semaphore
2541   ::sem_init(&sig_sem, 0, 0);
2542 }
2543 
2544 void os::signal_notify(int sig) {
2545   Atomic::inc(&pending_signals[sig]);
2546   ::sem_post(&sig_sem);
2547 }
2548 
2549 static int check_pending_signals(bool wait) {
2550   Atomic::store(0, &sigint_count);
2551   for (;;) {
2552     for (int i = 0; i < NSIG + 1; i++) {
2553       jint n = pending_signals[i];
2554       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2555         return i;
2556       }
2557     }
2558     if (!wait) {
2559       return -1;
2560     }
2561     JavaThread *thread = JavaThread::current();
2562     ThreadBlockInVM tbivm(thread);
2563 
2564     bool threadIsSuspended;
2565     do {
2566       thread->set_suspend_equivalent();
2567       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2568       ::sem_wait(&sig_sem);
2569 
2570       // were we externally suspended while we were waiting?
2571       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2572       if (threadIsSuspended) {
2573         // The semaphore has been incremented, but while we were waiting
2574         // another thread suspended us. We don't want to continue running
2575         // while suspended because that would surprise the thread that
2576         // suspended us.
2577         ::sem_post(&sig_sem);
2578 
2579         thread->java_suspend_self();
2580       }
2581     } while (threadIsSuspended);
2582   }
2583 }
2584 
2585 int os::signal_lookup() {
2586   return check_pending_signals(false);
2587 }
2588 
2589 int os::signal_wait() {
2590   return check_pending_signals(true);
2591 }
2592 
2593 ////////////////////////////////////////////////////////////////////////////////
2594 // Virtual Memory
2595 
2596 int os::vm_page_size() {
2597   // Seems redundant as all get out
2598   assert(os::Linux::page_size() != -1, "must call os::init");
2599   return os::Linux::page_size();
2600 }
2601 
2602 // Solaris allocates memory by pages.
2603 int os::vm_allocation_granularity() {
2604   assert(os::Linux::page_size() != -1, "must call os::init");
2605   return os::Linux::page_size();
2606 }
2607 
2608 // Rationale behind this function:
2609 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2610 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2611 //  samples for JITted code. Here we create private executable mapping over the code cache
2612 //  and then we can use standard (well, almost, as mapping can change) way to provide
2613 //  info for the reporting script by storing timestamp and location of symbol
2614 void linux_wrap_code(char* base, size_t size) {
2615   static volatile jint cnt = 0;
2616 
2617   if (!UseOprofile) {
2618     return;
2619   }
2620 
2621   char buf[PATH_MAX+1];
2622   int num = Atomic::add(1, &cnt);
2623 
2624   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2625            os::get_temp_directory(), os::current_process_id(), num);
2626   unlink(buf);
2627 
2628   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2629 
2630   if (fd != -1) {
2631     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2632     if (rv != (off_t)-1) {
2633       if (::write(fd, "", 1) == 1) {
2634         mmap(base, size,
2635              PROT_READ|PROT_WRITE|PROT_EXEC,
2636              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2637       }
2638     }
2639     ::close(fd);
2640     unlink(buf);
2641   }
2642 }
2643 
2644 static bool recoverable_mmap_error(int err) {
2645   // See if the error is one we can let the caller handle. This
2646   // list of errno values comes from JBS-6843484. I can't find a
2647   // Linux man page that documents this specific set of errno
2648   // values so while this list currently matches Solaris, it may
2649   // change as we gain experience with this failure mode.
2650   switch (err) {
2651   case EBADF:
2652   case EINVAL:
2653   case ENOTSUP:
2654     // let the caller deal with these errors
2655     return true;
2656 
2657   default:
2658     // Any remaining errors on this OS can cause our reserved mapping
2659     // to be lost. That can cause confusion where different data
2660     // structures think they have the same memory mapped. The worst
2661     // scenario is if both the VM and a library think they have the
2662     // same memory mapped.
2663     return false;
2664   }
2665 }
2666 
2667 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2668                                     int err) {
2669   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2670           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2671           os::strerror(err), err);
2672 }
2673 
2674 static void warn_fail_commit_memory(char* addr, size_t size,
2675                                     size_t alignment_hint, bool exec,
2676                                     int err) {
2677   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2678           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2679           alignment_hint, exec, os::strerror(err), err);
2680 }
2681 
2682 // NOTE: Linux kernel does not really reserve the pages for us.
2683 //       All it does is to check if there are enough free pages
2684 //       left at the time of mmap(). This could be a potential
2685 //       problem.
2686 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2687   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2688   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2689                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2690   if (res != (uintptr_t) MAP_FAILED) {
2691     if (UseNUMAInterleaving) {
2692       numa_make_global(addr, size);
2693     }
2694     return 0;
2695   }
2696 
2697   int err = errno;  // save errno from mmap() call above
2698 
2699   if (!recoverable_mmap_error(err)) {
2700     warn_fail_commit_memory(addr, size, exec, err);
2701     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2702   }
2703 
2704   return err;
2705 }
2706 
2707 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2708   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2709 }
2710 
2711 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2712                                   const char* mesg) {
2713   assert(mesg != NULL, "mesg must be specified");
2714   int err = os::Linux::commit_memory_impl(addr, size, exec);
2715   if (err != 0) {
2716     // the caller wants all commit errors to exit with the specified mesg:
2717     warn_fail_commit_memory(addr, size, exec, err);
2718     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2719   }
2720 }
2721 
2722 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2723 #ifndef MAP_HUGETLB
2724   #define MAP_HUGETLB 0x40000
2725 #endif
2726 
2727 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2728 #ifndef MADV_HUGEPAGE
2729   #define MADV_HUGEPAGE 14
2730 #endif
2731 
2732 int os::Linux::commit_memory_impl(char* addr, size_t size,
2733                                   size_t alignment_hint, bool exec) {
2734   int err = os::Linux::commit_memory_impl(addr, size, exec);
2735   if (err == 0) {
2736     realign_memory(addr, size, alignment_hint);
2737   }
2738   return err;
2739 }
2740 
2741 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2742                           bool exec) {
2743   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2744 }
2745 
2746 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2747                                   size_t alignment_hint, bool exec,
2748                                   const char* mesg) {
2749   assert(mesg != NULL, "mesg must be specified");
2750   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2751   if (err != 0) {
2752     // the caller wants all commit errors to exit with the specified mesg:
2753     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2754     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2755   }
2756 }
2757 
2758 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2759   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2760     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2761     // be supported or the memory may already be backed by huge pages.
2762     ::madvise(addr, bytes, MADV_HUGEPAGE);
2763   }
2764 }
2765 
2766 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2767   // This method works by doing an mmap over an existing mmaping and effectively discarding
2768   // the existing pages. However it won't work for SHM-based large pages that cannot be
2769   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2770   // small pages on top of the SHM segment. This method always works for small pages, so we
2771   // allow that in any case.
2772   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2773     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2774   }
2775 }
2776 
2777 void os::numa_make_global(char *addr, size_t bytes) {
2778   Linux::numa_interleave_memory(addr, bytes);
2779 }
2780 
2781 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2782 // bind policy to MPOL_PREFERRED for the current thread.
2783 #define USE_MPOL_PREFERRED 0
2784 
2785 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2786   // To make NUMA and large pages more robust when both enabled, we need to ease
2787   // the requirements on where the memory should be allocated. MPOL_BIND is the
2788   // default policy and it will force memory to be allocated on the specified
2789   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2790   // the specified node, but will not force it. Using this policy will prevent
2791   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2792   // free large pages.
2793   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2794   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2795 }
2796 
2797 bool os::numa_topology_changed() { return false; }
2798 
2799 size_t os::numa_get_groups_num() {
2800   // Return just the number of nodes in which it's possible to allocate memory
2801   // (in numa terminology, configured nodes).
2802   return Linux::numa_num_configured_nodes();
2803 }
2804 
2805 int os::numa_get_group_id() {
2806   int cpu_id = Linux::sched_getcpu();
2807   if (cpu_id != -1) {
2808     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2809     if (lgrp_id != -1) {
2810       return lgrp_id;
2811     }
2812   }
2813   return 0;
2814 }
2815 
2816 int os::Linux::get_existing_num_nodes() {
2817   size_t node;
2818   size_t highest_node_number = Linux::numa_max_node();
2819   int num_nodes = 0;
2820 
2821   // Get the total number of nodes in the system including nodes without memory.
2822   for (node = 0; node <= highest_node_number; node++) {
2823     if (isnode_in_existing_nodes(node)) {
2824       num_nodes++;
2825     }
2826   }
2827   return num_nodes;
2828 }
2829 
2830 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2831   size_t highest_node_number = Linux::numa_max_node();
2832   size_t i = 0;
2833 
2834   // Map all node ids in which is possible to allocate memory. Also nodes are
2835   // not always consecutively available, i.e. available from 0 to the highest
2836   // node number.
2837   for (size_t node = 0; node <= highest_node_number; node++) {
2838     if (Linux::isnode_in_configured_nodes(node)) {
2839       ids[i++] = node;
2840     }
2841   }
2842   return i;
2843 }
2844 
2845 bool os::get_page_info(char *start, page_info* info) {
2846   return false;
2847 }
2848 
2849 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2850                      page_info* page_found) {
2851   return end;
2852 }
2853 
2854 
2855 int os::Linux::sched_getcpu_syscall(void) {
2856   unsigned int cpu = 0;
2857   int retval = -1;
2858 
2859 #if defined(IA32)
2860   #ifndef SYS_getcpu
2861     #define SYS_getcpu 318
2862   #endif
2863   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2864 #elif defined(AMD64)
2865 // Unfortunately we have to bring all these macros here from vsyscall.h
2866 // to be able to compile on old linuxes.
2867   #define __NR_vgetcpu 2
2868   #define VSYSCALL_START (-10UL << 20)
2869   #define VSYSCALL_SIZE 1024
2870   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2871   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2872   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2873   retval = vgetcpu(&cpu, NULL, NULL);
2874 #endif
2875 
2876   return (retval == -1) ? retval : cpu;
2877 }
2878 
2879 void os::Linux::sched_getcpu_init() {
2880   // sched_getcpu() should be in libc.
2881   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2882                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2883 
2884   // If it's not, try a direct syscall.
2885   if (sched_getcpu() == -1) {
2886     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2887                                     (void*)&sched_getcpu_syscall));
2888   }
2889 }
2890 
2891 // Something to do with the numa-aware allocator needs these symbols
2892 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2893 extern "C" JNIEXPORT void numa_error(char *where) { }
2894 
2895 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2896 // load symbol from base version instead.
2897 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2898   void *f = dlvsym(handle, name, "libnuma_1.1");
2899   if (f == NULL) {
2900     f = dlsym(handle, name);
2901   }
2902   return f;
2903 }
2904 
2905 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2906 // Return NULL if the symbol is not defined in this particular version.
2907 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2908   return dlvsym(handle, name, "libnuma_1.2");
2909 }
2910 
2911 bool os::Linux::libnuma_init() {
2912   if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2913     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2914     if (handle != NULL) {
2915       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2916                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2917       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2918                                        libnuma_dlsym(handle, "numa_max_node")));
2919       set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2920                                                    libnuma_dlsym(handle, "numa_num_configured_nodes")));
2921       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2922                                         libnuma_dlsym(handle, "numa_available")));
2923       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2924                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2925       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2926                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2927       set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2928                                                 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2929       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2930                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2931       set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2932                                                libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2933       set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2934                                        libnuma_dlsym(handle, "numa_distance")));
2935 
2936       if (numa_available() != -1) {
2937         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2938         set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2939         set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2940         // Create an index -> node mapping, since nodes are not always consecutive
2941         _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2942         rebuild_nindex_to_node_map();
2943         // Create a cpu -> node mapping
2944         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2945         rebuild_cpu_to_node_map();
2946         return true;
2947       }
2948     }
2949   }
2950   return false;
2951 }
2952 
2953 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2954   // Creating guard page is very expensive. Java thread has HotSpot
2955   // guard pages, only enable glibc guard page for non-Java threads.
2956   // (Remember: compiler thread is a Java thread, too!)
2957   return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2958 }
2959 
2960 void os::Linux::rebuild_nindex_to_node_map() {
2961   int highest_node_number = Linux::numa_max_node();
2962 
2963   nindex_to_node()->clear();
2964   for (int node = 0; node <= highest_node_number; node++) {
2965     if (Linux::isnode_in_existing_nodes(node)) {
2966       nindex_to_node()->append(node);
2967     }
2968   }
2969 }
2970 
2971 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2972 // The table is later used in get_node_by_cpu().
2973 void os::Linux::rebuild_cpu_to_node_map() {
2974   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2975                               // in libnuma (possible values are starting from 16,
2976                               // and continuing up with every other power of 2, but less
2977                               // than the maximum number of CPUs supported by kernel), and
2978                               // is a subject to change (in libnuma version 2 the requirements
2979                               // are more reasonable) we'll just hardcode the number they use
2980                               // in the library.
2981   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2982 
2983   size_t cpu_num = processor_count();
2984   size_t cpu_map_size = NCPUS / BitsPerCLong;
2985   size_t cpu_map_valid_size =
2986     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2987 
2988   cpu_to_node()->clear();
2989   cpu_to_node()->at_grow(cpu_num - 1);
2990 
2991   size_t node_num = get_existing_num_nodes();
2992 
2993   int distance = 0;
2994   int closest_distance = INT_MAX;
2995   int closest_node = 0;
2996   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2997   for (size_t i = 0; i < node_num; i++) {
2998     // Check if node is configured (not a memory-less node). If it is not, find
2999     // the closest configured node.
3000     if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
3001       closest_distance = INT_MAX;
3002       // Check distance from all remaining nodes in the system. Ignore distance
3003       // from itself and from another non-configured node.
3004       for (size_t m = 0; m < node_num; m++) {
3005         if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
3006           distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3007           // If a closest node is found, update. There is always at least one
3008           // configured node in the system so there is always at least one node
3009           // close.
3010           if (distance != 0 && distance < closest_distance) {
3011             closest_distance = distance;
3012             closest_node = nindex_to_node()->at(m);
3013           }
3014         }
3015       }
3016      } else {
3017        // Current node is already a configured node.
3018        closest_node = nindex_to_node()->at(i);
3019      }
3020 
3021     // Get cpus from the original node and map them to the closest node. If node
3022     // is a configured node (not a memory-less node), then original node and
3023     // closest node are the same.
3024     if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3025       for (size_t j = 0; j < cpu_map_valid_size; j++) {
3026         if (cpu_map[j] != 0) {
3027           for (size_t k = 0; k < BitsPerCLong; k++) {
3028             if (cpu_map[j] & (1UL << k)) {
3029               cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3030             }
3031           }
3032         }
3033       }
3034     }
3035   }
3036   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3037 }
3038 
3039 int os::Linux::get_node_by_cpu(int cpu_id) {
3040   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3041     return cpu_to_node()->at(cpu_id);
3042   }
3043   return -1;
3044 }
3045 
3046 GrowableArray<int>* os::Linux::_cpu_to_node;
3047 GrowableArray<int>* os::Linux::_nindex_to_node;
3048 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3049 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3050 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3051 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3052 os::Linux::numa_available_func_t os::Linux::_numa_available;
3053 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3054 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3055 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3056 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3057 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3058 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3059 unsigned long* os::Linux::_numa_all_nodes;
3060 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3061 struct bitmask* os::Linux::_numa_nodes_ptr;
3062 
3063 bool os::pd_uncommit_memory(char* addr, size_t size) {
3064   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3065                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3066   return res  != (uintptr_t) MAP_FAILED;
3067 }
3068 
3069 static address get_stack_commited_bottom(address bottom, size_t size) {
3070   address nbot = bottom;
3071   address ntop = bottom + size;
3072 
3073   size_t page_sz = os::vm_page_size();
3074   unsigned pages = size / page_sz;
3075 
3076   unsigned char vec[1];
3077   unsigned imin = 1, imax = pages + 1, imid;
3078   int mincore_return_value = 0;
3079 
3080   assert(imin <= imax, "Unexpected page size");
3081 
3082   while (imin < imax) {
3083     imid = (imax + imin) / 2;
3084     nbot = ntop - (imid * page_sz);
3085 
3086     // Use a trick with mincore to check whether the page is mapped or not.
3087     // mincore sets vec to 1 if page resides in memory and to 0 if page
3088     // is swapped output but if page we are asking for is unmapped
3089     // it returns -1,ENOMEM
3090     mincore_return_value = mincore(nbot, page_sz, vec);
3091 
3092     if (mincore_return_value == -1) {
3093       // Page is not mapped go up
3094       // to find first mapped page
3095       if (errno != EAGAIN) {
3096         assert(errno == ENOMEM, "Unexpected mincore errno");
3097         imax = imid;
3098       }
3099     } else {
3100       // Page is mapped go down
3101       // to find first not mapped page
3102       imin = imid + 1;
3103     }
3104   }
3105 
3106   nbot = nbot + page_sz;
3107 
3108   // Adjust stack bottom one page up if last checked page is not mapped
3109   if (mincore_return_value == -1) {
3110     nbot = nbot + page_sz;
3111   }
3112 
3113   return nbot;
3114 }
3115 
3116 
3117 // Linux uses a growable mapping for the stack, and if the mapping for
3118 // the stack guard pages is not removed when we detach a thread the
3119 // stack cannot grow beyond the pages where the stack guard was
3120 // mapped.  If at some point later in the process the stack expands to
3121 // that point, the Linux kernel cannot expand the stack any further
3122 // because the guard pages are in the way, and a segfault occurs.
3123 //
3124 // However, it's essential not to split the stack region by unmapping
3125 // a region (leaving a hole) that's already part of the stack mapping,
3126 // so if the stack mapping has already grown beyond the guard pages at
3127 // the time we create them, we have to truncate the stack mapping.
3128 // So, we need to know the extent of the stack mapping when
3129 // create_stack_guard_pages() is called.
3130 
3131 // We only need this for stacks that are growable: at the time of
3132 // writing thread stacks don't use growable mappings (i.e. those
3133 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3134 // only applies to the main thread.
3135 
3136 // If the (growable) stack mapping already extends beyond the point
3137 // where we're going to put our guard pages, truncate the mapping at
3138 // that point by munmap()ping it.  This ensures that when we later
3139 // munmap() the guard pages we don't leave a hole in the stack
3140 // mapping. This only affects the main/initial thread
3141 
3142 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3143   if (os::Linux::is_initial_thread()) {
3144     // As we manually grow stack up to bottom inside create_attached_thread(),
3145     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3146     // we don't need to do anything special.
3147     // Check it first, before calling heavy function.
3148     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3149     unsigned char vec[1];
3150 
3151     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3152       // Fallback to slow path on all errors, including EAGAIN
3153       stack_extent = (uintptr_t) get_stack_commited_bottom(
3154                                                            os::Linux::initial_thread_stack_bottom(),
3155                                                            (size_t)addr - stack_extent);
3156     }
3157 
3158     if (stack_extent < (uintptr_t)addr) {
3159       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3160     }
3161   }
3162 
3163   return os::commit_memory(addr, size, !ExecMem);
3164 }
3165 
3166 // If this is a growable mapping, remove the guard pages entirely by
3167 // munmap()ping them.  If not, just call uncommit_memory(). This only
3168 // affects the main/initial thread, but guard against future OS changes
3169 // It's safe to always unmap guard pages for initial thread because we
3170 // always place it right after end of the mapped region
3171 
3172 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3173   uintptr_t stack_extent, stack_base;
3174 
3175   if (os::Linux::is_initial_thread()) {
3176     return ::munmap(addr, size) == 0;
3177   }
3178 
3179   return os::uncommit_memory(addr, size);
3180 }
3181 
3182 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3183 // at 'requested_addr'. If there are existing memory mappings at the same
3184 // location, however, they will be overwritten. If 'fixed' is false,
3185 // 'requested_addr' is only treated as a hint, the return value may or
3186 // may not start from the requested address. Unlike Linux mmap(), this
3187 // function returns NULL to indicate failure.
3188 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3189   char * addr;
3190   int flags;
3191 
3192   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3193   if (fixed) {
3194     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3195     flags |= MAP_FIXED;
3196   }
3197 
3198   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3199   // touch an uncommitted page. Otherwise, the read/write might
3200   // succeed if we have enough swap space to back the physical page.
3201   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3202                        flags, -1, 0);
3203 
3204   return addr == MAP_FAILED ? NULL : addr;
3205 }
3206 
3207 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3208 //   (req_addr != NULL) or with a given alignment.
3209 //  - bytes shall be a multiple of alignment.
3210 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3211 //  - alignment sets the alignment at which memory shall be allocated.
3212 //     It must be a multiple of allocation granularity.
3213 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3214 //  req_addr or NULL.
3215 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3216 
3217   size_t extra_size = bytes;
3218   if (req_addr == NULL && alignment > 0) {
3219     extra_size += alignment;
3220   }
3221 
3222   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3223     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3224     -1, 0);
3225   if (start == MAP_FAILED) {
3226     start = NULL;
3227   } else {
3228     if (req_addr != NULL) {
3229       if (start != req_addr) {
3230         ::munmap(start, extra_size);
3231         start = NULL;
3232       }
3233     } else {
3234       char* const start_aligned = align_up(start, alignment);
3235       char* const end_aligned = start_aligned + bytes;
3236       char* const end = start + extra_size;
3237       if (start_aligned > start) {
3238         ::munmap(start, start_aligned - start);
3239       }
3240       if (end_aligned < end) {
3241         ::munmap(end_aligned, end - end_aligned);
3242       }
3243       start = start_aligned;
3244     }
3245   }
3246   return start;
3247 }
3248 
3249 static int anon_munmap(char * addr, size_t size) {
3250   return ::munmap(addr, size) == 0;
3251 }
3252 
3253 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3254                             size_t alignment_hint) {
3255   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3256 }
3257 
3258 bool os::pd_release_memory(char* addr, size_t size) {
3259   return anon_munmap(addr, size);
3260 }
3261 
3262 static bool linux_mprotect(char* addr, size_t size, int prot) {
3263   // Linux wants the mprotect address argument to be page aligned.
3264   char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3265 
3266   // According to SUSv3, mprotect() should only be used with mappings
3267   // established by mmap(), and mmap() always maps whole pages. Unaligned
3268   // 'addr' likely indicates problem in the VM (e.g. trying to change
3269   // protection of malloc'ed or statically allocated memory). Check the
3270   // caller if you hit this assert.
3271   assert(addr == bottom, "sanity check");
3272 
3273   size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3274   return ::mprotect(bottom, size, prot) == 0;
3275 }
3276 
3277 // Set protections specified
3278 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3279                         bool is_committed) {
3280   unsigned int p = 0;
3281   switch (prot) {
3282   case MEM_PROT_NONE: p = PROT_NONE; break;
3283   case MEM_PROT_READ: p = PROT_READ; break;
3284   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3285   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3286   default:
3287     ShouldNotReachHere();
3288   }
3289   // is_committed is unused.
3290   return linux_mprotect(addr, bytes, p);
3291 }
3292 
3293 bool os::guard_memory(char* addr, size_t size) {
3294   return linux_mprotect(addr, size, PROT_NONE);
3295 }
3296 
3297 bool os::unguard_memory(char* addr, size_t size) {
3298   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3299 }
3300 
3301 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3302                                                     size_t page_size) {
3303   bool result = false;
3304   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3305                  MAP_ANONYMOUS|MAP_PRIVATE,
3306                  -1, 0);
3307   if (p != MAP_FAILED) {
3308     void *aligned_p = align_up(p, page_size);
3309 
3310     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3311 
3312     munmap(p, page_size * 2);
3313   }
3314 
3315   if (warn && !result) {
3316     warning("TransparentHugePages is not supported by the operating system.");
3317   }
3318 
3319   return result;
3320 }
3321 
3322 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3323   bool result = false;
3324   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3325                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3326                  -1, 0);
3327 
3328   if (p != MAP_FAILED) {
3329     // We don't know if this really is a huge page or not.
3330     FILE *fp = fopen("/proc/self/maps", "r");
3331     if (fp) {
3332       while (!feof(fp)) {
3333         char chars[257];
3334         long x = 0;
3335         if (fgets(chars, sizeof(chars), fp)) {
3336           if (sscanf(chars, "%lx-%*x", &x) == 1
3337               && x == (long)p) {
3338             if (strstr (chars, "hugepage")) {
3339               result = true;
3340               break;
3341             }
3342           }
3343         }
3344       }
3345       fclose(fp);
3346     }
3347     munmap(p, page_size);
3348   }
3349 
3350   if (warn && !result) {
3351     warning("HugeTLBFS is not supported by the operating system.");
3352   }
3353 
3354   return result;
3355 }
3356 
3357 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3358 //
3359 // From the coredump_filter documentation:
3360 //
3361 // - (bit 0) anonymous private memory
3362 // - (bit 1) anonymous shared memory
3363 // - (bit 2) file-backed private memory
3364 // - (bit 3) file-backed shared memory
3365 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3366 //           effective only if the bit 2 is cleared)
3367 // - (bit 5) hugetlb private memory
3368 // - (bit 6) hugetlb shared memory
3369 //
3370 static void set_coredump_filter(void) {
3371   FILE *f;
3372   long cdm;
3373 
3374   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3375     return;
3376   }
3377 
3378   if (fscanf(f, "%lx", &cdm) != 1) {
3379     fclose(f);
3380     return;
3381   }
3382 
3383   rewind(f);
3384 
3385   if ((cdm & LARGEPAGES_BIT) == 0) {
3386     cdm |= LARGEPAGES_BIT;
3387     fprintf(f, "%#lx", cdm);
3388   }
3389 
3390   fclose(f);
3391 }
3392 
3393 // Large page support
3394 
3395 static size_t _large_page_size = 0;
3396 
3397 size_t os::Linux::find_large_page_size() {
3398   size_t large_page_size = 0;
3399 
3400   // large_page_size on Linux is used to round up heap size. x86 uses either
3401   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3402   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3403   // page as large as 256M.
3404   //
3405   // Here we try to figure out page size by parsing /proc/meminfo and looking
3406   // for a line with the following format:
3407   //    Hugepagesize:     2048 kB
3408   //
3409   // If we can't determine the value (e.g. /proc is not mounted, or the text
3410   // format has been changed), we'll use the largest page size supported by
3411   // the processor.
3412 
3413 #ifndef ZERO
3414   large_page_size =
3415     AARCH64_ONLY(2 * M)
3416     AMD64_ONLY(2 * M)
3417     ARM32_ONLY(2 * M)
3418     IA32_ONLY(4 * M)
3419     IA64_ONLY(256 * M)
3420     PPC_ONLY(4 * M)
3421     S390_ONLY(1 * M)
3422     SPARC_ONLY(4 * M);
3423 #endif // ZERO
3424 
3425   FILE *fp = fopen("/proc/meminfo", "r");
3426   if (fp) {
3427     while (!feof(fp)) {
3428       int x = 0;
3429       char buf[16];
3430       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3431         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3432           large_page_size = x * K;
3433           break;
3434         }
3435       } else {
3436         // skip to next line
3437         for (;;) {
3438           int ch = fgetc(fp);
3439           if (ch == EOF || ch == (int)'\n') break;
3440         }
3441       }
3442     }
3443     fclose(fp);
3444   }
3445 
3446   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3447     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3448             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3449             proper_unit_for_byte_size(large_page_size));
3450   }
3451 
3452   return large_page_size;
3453 }
3454 
3455 size_t os::Linux::setup_large_page_size() {
3456   _large_page_size = Linux::find_large_page_size();
3457   const size_t default_page_size = (size_t)Linux::page_size();
3458   if (_large_page_size > default_page_size) {
3459     _page_sizes[0] = _large_page_size;
3460     _page_sizes[1] = default_page_size;
3461     _page_sizes[2] = 0;
3462   }
3463 
3464   return _large_page_size;
3465 }
3466 
3467 bool os::Linux::setup_large_page_type(size_t page_size) {
3468   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3469       FLAG_IS_DEFAULT(UseSHM) &&
3470       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3471 
3472     // The type of large pages has not been specified by the user.
3473 
3474     // Try UseHugeTLBFS and then UseSHM.
3475     UseHugeTLBFS = UseSHM = true;
3476 
3477     // Don't try UseTransparentHugePages since there are known
3478     // performance issues with it turned on. This might change in the future.
3479     UseTransparentHugePages = false;
3480   }
3481 
3482   if (UseTransparentHugePages) {
3483     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3484     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3485       UseHugeTLBFS = false;
3486       UseSHM = false;
3487       return true;
3488     }
3489     UseTransparentHugePages = false;
3490   }
3491 
3492   if (UseHugeTLBFS) {
3493     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3494     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3495       UseSHM = false;
3496       return true;
3497     }
3498     UseHugeTLBFS = false;
3499   }
3500 
3501   return UseSHM;
3502 }
3503 
3504 void os::large_page_init() {
3505   if (!UseLargePages &&
3506       !UseTransparentHugePages &&
3507       !UseHugeTLBFS &&
3508       !UseSHM) {
3509     // Not using large pages.
3510     return;
3511   }
3512 
3513   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3514     // The user explicitly turned off large pages.
3515     // Ignore the rest of the large pages flags.
3516     UseTransparentHugePages = false;
3517     UseHugeTLBFS = false;
3518     UseSHM = false;
3519     return;
3520   }
3521 
3522   size_t large_page_size = Linux::setup_large_page_size();
3523   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3524 
3525   set_coredump_filter();
3526 }
3527 
3528 #ifndef SHM_HUGETLB
3529   #define SHM_HUGETLB 04000
3530 #endif
3531 
3532 #define shm_warning_format(format, ...)              \
3533   do {                                               \
3534     if (UseLargePages &&                             \
3535         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3536          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3537          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3538       warning(format, __VA_ARGS__);                  \
3539     }                                                \
3540   } while (0)
3541 
3542 #define shm_warning(str) shm_warning_format("%s", str)
3543 
3544 #define shm_warning_with_errno(str)                \
3545   do {                                             \
3546     int err = errno;                               \
3547     shm_warning_format(str " (error = %d)", err);  \
3548   } while (0)
3549 
3550 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3551   assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3552 
3553   if (!is_aligned(alignment, SHMLBA)) {
3554     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3555     return NULL;
3556   }
3557 
3558   // To ensure that we get 'alignment' aligned memory from shmat,
3559   // we pre-reserve aligned virtual memory and then attach to that.
3560 
3561   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3562   if (pre_reserved_addr == NULL) {
3563     // Couldn't pre-reserve aligned memory.
3564     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3565     return NULL;
3566   }
3567 
3568   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3569   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3570 
3571   if ((intptr_t)addr == -1) {
3572     int err = errno;
3573     shm_warning_with_errno("Failed to attach shared memory.");
3574 
3575     assert(err != EACCES, "Unexpected error");
3576     assert(err != EIDRM,  "Unexpected error");
3577     assert(err != EINVAL, "Unexpected error");
3578 
3579     // Since we don't know if the kernel unmapped the pre-reserved memory area
3580     // we can't unmap it, since that would potentially unmap memory that was
3581     // mapped from other threads.
3582     return NULL;
3583   }
3584 
3585   return addr;
3586 }
3587 
3588 static char* shmat_at_address(int shmid, char* req_addr) {
3589   if (!is_aligned(req_addr, SHMLBA)) {
3590     assert(false, "Requested address needs to be SHMLBA aligned");
3591     return NULL;
3592   }
3593 
3594   char* addr = (char*)shmat(shmid, req_addr, 0);
3595 
3596   if ((intptr_t)addr == -1) {
3597     shm_warning_with_errno("Failed to attach shared memory.");
3598     return NULL;
3599   }
3600 
3601   return addr;
3602 }
3603 
3604 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3605   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3606   if (req_addr != NULL) {
3607     assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3608     assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3609     return shmat_at_address(shmid, req_addr);
3610   }
3611 
3612   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3613   // return large page size aligned memory addresses when req_addr == NULL.
3614   // However, if the alignment is larger than the large page size, we have
3615   // to manually ensure that the memory returned is 'alignment' aligned.
3616   if (alignment > os::large_page_size()) {
3617     assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3618     return shmat_with_alignment(shmid, bytes, alignment);
3619   } else {
3620     return shmat_at_address(shmid, NULL);
3621   }
3622 }
3623 
3624 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3625                                             char* req_addr, bool exec) {
3626   // "exec" is passed in but not used.  Creating the shared image for
3627   // the code cache doesn't have an SHM_X executable permission to check.
3628   assert(UseLargePages && UseSHM, "only for SHM large pages");
3629   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3630   assert(is_aligned(req_addr, alignment), "Unaligned address");
3631 
3632   if (!is_aligned(bytes, os::large_page_size())) {
3633     return NULL; // Fallback to small pages.
3634   }
3635 
3636   // Create a large shared memory region to attach to based on size.
3637   // Currently, size is the total size of the heap.
3638   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3639   if (shmid == -1) {
3640     // Possible reasons for shmget failure:
3641     // 1. shmmax is too small for Java heap.
3642     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3643     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3644     // 2. not enough large page memory.
3645     //    > check available large pages: cat /proc/meminfo
3646     //    > increase amount of large pages:
3647     //          echo new_value > /proc/sys/vm/nr_hugepages
3648     //      Note 1: different Linux may use different name for this property,
3649     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3650     //      Note 2: it's possible there's enough physical memory available but
3651     //            they are so fragmented after a long run that they can't
3652     //            coalesce into large pages. Try to reserve large pages when
3653     //            the system is still "fresh".
3654     shm_warning_with_errno("Failed to reserve shared memory.");
3655     return NULL;
3656   }
3657 
3658   // Attach to the region.
3659   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3660 
3661   // Remove shmid. If shmat() is successful, the actual shared memory segment
3662   // will be deleted when it's detached by shmdt() or when the process
3663   // terminates. If shmat() is not successful this will remove the shared
3664   // segment immediately.
3665   shmctl(shmid, IPC_RMID, NULL);
3666 
3667   return addr;
3668 }
3669 
3670 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3671                                         int error) {
3672   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3673 
3674   bool warn_on_failure = UseLargePages &&
3675       (!FLAG_IS_DEFAULT(UseLargePages) ||
3676        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3677        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3678 
3679   if (warn_on_failure) {
3680     char msg[128];
3681     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3682                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3683     warning("%s", msg);
3684   }
3685 }
3686 
3687 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3688                                                         char* req_addr,
3689                                                         bool exec) {
3690   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3691   assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3692   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3693 
3694   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3695   char* addr = (char*)::mmap(req_addr, bytes, prot,
3696                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3697                              -1, 0);
3698 
3699   if (addr == MAP_FAILED) {
3700     warn_on_large_pages_failure(req_addr, bytes, errno);
3701     return NULL;
3702   }
3703 
3704   assert(is_aligned(addr, os::large_page_size()), "Must be");
3705 
3706   return addr;
3707 }
3708 
3709 // Reserve memory using mmap(MAP_HUGETLB).
3710 //  - bytes shall be a multiple of alignment.
3711 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3712 //  - alignment sets the alignment at which memory shall be allocated.
3713 //     It must be a multiple of allocation granularity.
3714 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3715 //  req_addr or NULL.
3716 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3717                                                          size_t alignment,
3718                                                          char* req_addr,
3719                                                          bool exec) {
3720   size_t large_page_size = os::large_page_size();
3721   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3722 
3723   assert(is_aligned(req_addr, alignment), "Must be");
3724   assert(is_aligned(bytes, alignment), "Must be");
3725 
3726   // First reserve - but not commit - the address range in small pages.
3727   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3728 
3729   if (start == NULL) {
3730     return NULL;
3731   }
3732 
3733   assert(is_aligned(start, alignment), "Must be");
3734 
3735   char* end = start + bytes;
3736 
3737   // Find the regions of the allocated chunk that can be promoted to large pages.
3738   char* lp_start = align_up(start, large_page_size);
3739   char* lp_end   = align_down(end, large_page_size);
3740 
3741   size_t lp_bytes = lp_end - lp_start;
3742 
3743   assert(is_aligned(lp_bytes, large_page_size), "Must be");
3744 
3745   if (lp_bytes == 0) {
3746     // The mapped region doesn't even span the start and the end of a large page.
3747     // Fall back to allocate a non-special area.
3748     ::munmap(start, end - start);
3749     return NULL;
3750   }
3751 
3752   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3753 
3754   void* result;
3755 
3756   // Commit small-paged leading area.
3757   if (start != lp_start) {
3758     result = ::mmap(start, lp_start - start, prot,
3759                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3760                     -1, 0);
3761     if (result == MAP_FAILED) {
3762       ::munmap(lp_start, end - lp_start);
3763       return NULL;
3764     }
3765   }
3766 
3767   // Commit large-paged area.
3768   result = ::mmap(lp_start, lp_bytes, prot,
3769                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3770                   -1, 0);
3771   if (result == MAP_FAILED) {
3772     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3773     // If the mmap above fails, the large pages region will be unmapped and we
3774     // have regions before and after with small pages. Release these regions.
3775     //
3776     // |  mapped  |  unmapped  |  mapped  |
3777     // ^          ^            ^          ^
3778     // start      lp_start     lp_end     end
3779     //
3780     ::munmap(start, lp_start - start);
3781     ::munmap(lp_end, end - lp_end);
3782     return NULL;
3783   }
3784 
3785   // Commit small-paged trailing area.
3786   if (lp_end != end) {
3787     result = ::mmap(lp_end, end - lp_end, prot,
3788                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3789                     -1, 0);
3790     if (result == MAP_FAILED) {
3791       ::munmap(start, lp_end - start);
3792       return NULL;
3793     }
3794   }
3795 
3796   return start;
3797 }
3798 
3799 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3800                                                    size_t alignment,
3801                                                    char* req_addr,
3802                                                    bool exec) {
3803   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3804   assert(is_aligned(req_addr, alignment), "Must be");
3805   assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3806   assert(is_power_of_2(os::large_page_size()), "Must be");
3807   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3808 
3809   if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3810     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3811   } else {
3812     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3813   }
3814 }
3815 
3816 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3817                                  char* req_addr, bool exec) {
3818   assert(UseLargePages, "only for large pages");
3819 
3820   char* addr;
3821   if (UseSHM) {
3822     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3823   } else {
3824     assert(UseHugeTLBFS, "must be");
3825     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3826   }
3827 
3828   if (addr != NULL) {
3829     if (UseNUMAInterleaving) {
3830       numa_make_global(addr, bytes);
3831     }
3832 
3833     // The memory is committed
3834     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3835   }
3836 
3837   return addr;
3838 }
3839 
3840 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3841   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3842   return shmdt(base) == 0;
3843 }
3844 
3845 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3846   return pd_release_memory(base, bytes);
3847 }
3848 
3849 bool os::release_memory_special(char* base, size_t bytes) {
3850   bool res;
3851   if (MemTracker::tracking_level() > NMT_minimal) {
3852     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3853     res = os::Linux::release_memory_special_impl(base, bytes);
3854     if (res) {
3855       tkr.record((address)base, bytes);
3856     }
3857 
3858   } else {
3859     res = os::Linux::release_memory_special_impl(base, bytes);
3860   }
3861   return res;
3862 }
3863 
3864 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3865   assert(UseLargePages, "only for large pages");
3866   bool res;
3867 
3868   if (UseSHM) {
3869     res = os::Linux::release_memory_special_shm(base, bytes);
3870   } else {
3871     assert(UseHugeTLBFS, "must be");
3872     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3873   }
3874   return res;
3875 }
3876 
3877 size_t os::large_page_size() {
3878   return _large_page_size;
3879 }
3880 
3881 // With SysV SHM the entire memory region must be allocated as shared
3882 // memory.
3883 // HugeTLBFS allows application to commit large page memory on demand.
3884 // However, when committing memory with HugeTLBFS fails, the region
3885 // that was supposed to be committed will lose the old reservation
3886 // and allow other threads to steal that memory region. Because of this
3887 // behavior we can't commit HugeTLBFS memory.
3888 bool os::can_commit_large_page_memory() {
3889   return UseTransparentHugePages;
3890 }
3891 
3892 bool os::can_execute_large_page_memory() {
3893   return UseTransparentHugePages || UseHugeTLBFS;
3894 }
3895 
3896 // Reserve memory at an arbitrary address, only if that area is
3897 // available (and not reserved for something else).
3898 
3899 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3900   const int max_tries = 10;
3901   char* base[max_tries];
3902   size_t size[max_tries];
3903   const size_t gap = 0x000000;
3904 
3905   // Assert only that the size is a multiple of the page size, since
3906   // that's all that mmap requires, and since that's all we really know
3907   // about at this low abstraction level.  If we need higher alignment,
3908   // we can either pass an alignment to this method or verify alignment
3909   // in one of the methods further up the call chain.  See bug 5044738.
3910   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3911 
3912   // Repeatedly allocate blocks until the block is allocated at the
3913   // right spot.
3914 
3915   // Linux mmap allows caller to pass an address as hint; give it a try first,
3916   // if kernel honors the hint then we can return immediately.
3917   char * addr = anon_mmap(requested_addr, bytes, false);
3918   if (addr == requested_addr) {
3919     return requested_addr;
3920   }
3921 
3922   if (addr != NULL) {
3923     // mmap() is successful but it fails to reserve at the requested address
3924     anon_munmap(addr, bytes);
3925   }
3926 
3927   int i;
3928   for (i = 0; i < max_tries; ++i) {
3929     base[i] = reserve_memory(bytes);
3930 
3931     if (base[i] != NULL) {
3932       // Is this the block we wanted?
3933       if (base[i] == requested_addr) {
3934         size[i] = bytes;
3935         break;
3936       }
3937 
3938       // Does this overlap the block we wanted? Give back the overlapped
3939       // parts and try again.
3940 
3941       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3942       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3943         unmap_memory(base[i], top_overlap);
3944         base[i] += top_overlap;
3945         size[i] = bytes - top_overlap;
3946       } else {
3947         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3948         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3949           unmap_memory(requested_addr, bottom_overlap);
3950           size[i] = bytes - bottom_overlap;
3951         } else {
3952           size[i] = bytes;
3953         }
3954       }
3955     }
3956   }
3957 
3958   // Give back the unused reserved pieces.
3959 
3960   for (int j = 0; j < i; ++j) {
3961     if (base[j] != NULL) {
3962       unmap_memory(base[j], size[j]);
3963     }
3964   }
3965 
3966   if (i < max_tries) {
3967     return requested_addr;
3968   } else {
3969     return NULL;
3970   }
3971 }
3972 
3973 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3974   return ::read(fd, buf, nBytes);
3975 }
3976 
3977 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3978   return ::pread(fd, buf, nBytes, offset);
3979 }
3980 
3981 // Short sleep, direct OS call.
3982 //
3983 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3984 // sched_yield(2) will actually give up the CPU:
3985 //
3986 //   * Alone on this pariticular CPU, keeps running.
3987 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3988 //     (pre 2.6.39).
3989 //
3990 // So calling this with 0 is an alternative.
3991 //
3992 void os::naked_short_sleep(jlong ms) {
3993   struct timespec req;
3994 
3995   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3996   req.tv_sec = 0;
3997   if (ms > 0) {
3998     req.tv_nsec = (ms % 1000) * 1000000;
3999   } else {
4000     req.tv_nsec = 1;
4001   }
4002 
4003   nanosleep(&req, NULL);
4004 
4005   return;
4006 }
4007 
4008 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4009 void os::infinite_sleep() {
4010   while (true) {    // sleep forever ...
4011     ::sleep(100);   // ... 100 seconds at a time
4012   }
4013 }
4014 
4015 // Used to convert frequent JVM_Yield() to nops
4016 bool os::dont_yield() {
4017   return DontYieldALot;
4018 }
4019 
4020 void os::naked_yield() {
4021   sched_yield();
4022 }
4023 
4024 ////////////////////////////////////////////////////////////////////////////////
4025 // thread priority support
4026 
4027 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4028 // only supports dynamic priority, static priority must be zero. For real-time
4029 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4030 // However, for large multi-threaded applications, SCHED_RR is not only slower
4031 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4032 // of 5 runs - Sep 2005).
4033 //
4034 // The following code actually changes the niceness of kernel-thread/LWP. It
4035 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4036 // not the entire user process, and user level threads are 1:1 mapped to kernel
4037 // threads. It has always been the case, but could change in the future. For
4038 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4039 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4040 
4041 int os::java_to_os_priority[CriticalPriority + 1] = {
4042   19,              // 0 Entry should never be used
4043 
4044    4,              // 1 MinPriority
4045    3,              // 2
4046    2,              // 3
4047 
4048    1,              // 4
4049    0,              // 5 NormPriority
4050   -1,              // 6
4051 
4052   -2,              // 7
4053   -3,              // 8
4054   -4,              // 9 NearMaxPriority
4055 
4056   -5,              // 10 MaxPriority
4057 
4058   -5               // 11 CriticalPriority
4059 };
4060 
4061 static int prio_init() {
4062   if (ThreadPriorityPolicy == 1) {
4063     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4064     // if effective uid is not root. Perhaps, a more elegant way of doing
4065     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4066     if (geteuid() != 0) {
4067       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4068         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4069       }
4070       ThreadPriorityPolicy = 0;
4071     }
4072   }
4073   if (UseCriticalJavaThreadPriority) {
4074     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4075   }
4076   return 0;
4077 }
4078 
4079 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4080   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4081 
4082   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4083   return (ret == 0) ? OS_OK : OS_ERR;
4084 }
4085 
4086 OSReturn os::get_native_priority(const Thread* const thread,
4087                                  int *priority_ptr) {
4088   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4089     *priority_ptr = java_to_os_priority[NormPriority];
4090     return OS_OK;
4091   }
4092 
4093   errno = 0;
4094   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4095   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4096 }
4097 
4098 // Hint to the underlying OS that a task switch would not be good.
4099 // Void return because it's a hint and can fail.
4100 void os::hint_no_preempt() {}
4101 
4102 ////////////////////////////////////////////////////////////////////////////////
4103 // suspend/resume support
4104 
4105 //  The low-level signal-based suspend/resume support is a remnant from the
4106 //  old VM-suspension that used to be for java-suspension, safepoints etc,
4107 //  within hotspot. Currently used by JFR's OSThreadSampler
4108 //
4109 //  The remaining code is greatly simplified from the more general suspension
4110 //  code that used to be used.
4111 //
4112 //  The protocol is quite simple:
4113 //  - suspend:
4114 //      - sends a signal to the target thread
4115 //      - polls the suspend state of the osthread using a yield loop
4116 //      - target thread signal handler (SR_handler) sets suspend state
4117 //        and blocks in sigsuspend until continued
4118 //  - resume:
4119 //      - sets target osthread state to continue
4120 //      - sends signal to end the sigsuspend loop in the SR_handler
4121 //
4122 //  Note that the SR_lock plays no role in this suspend/resume protocol,
4123 //  but is checked for NULL in SR_handler as a thread termination indicator.
4124 //  The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4125 //
4126 //  Note that resume_clear_context() and suspend_save_context() are needed
4127 //  by SR_handler(), so that fetch_frame_from_ucontext() works,
4128 //  which in part is used by:
4129 //    - Forte Analyzer: AsyncGetCallTrace()
4130 //    - StackBanging: get_frame_at_stack_banging_point()
4131 
4132 static void resume_clear_context(OSThread *osthread) {
4133   osthread->set_ucontext(NULL);
4134   osthread->set_siginfo(NULL);
4135 }
4136 
4137 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4138                                  ucontext_t* context) {
4139   osthread->set_ucontext(context);
4140   osthread->set_siginfo(siginfo);
4141 }
4142 
4143 // Handler function invoked when a thread's execution is suspended or
4144 // resumed. We have to be careful that only async-safe functions are
4145 // called here (Note: most pthread functions are not async safe and
4146 // should be avoided.)
4147 //
4148 // Note: sigwait() is a more natural fit than sigsuspend() from an
4149 // interface point of view, but sigwait() prevents the signal hander
4150 // from being run. libpthread would get very confused by not having
4151 // its signal handlers run and prevents sigwait()'s use with the
4152 // mutex granting granting signal.
4153 //
4154 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4155 //
4156 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4157   // Save and restore errno to avoid confusing native code with EINTR
4158   // after sigsuspend.
4159   int old_errno = errno;
4160 
4161   Thread* thread = Thread::current_or_null_safe();
4162   assert(thread != NULL, "Missing current thread in SR_handler");
4163 
4164   // On some systems we have seen signal delivery get "stuck" until the signal
4165   // mask is changed as part of thread termination. Check that the current thread
4166   // has not already terminated (via SR_lock()) - else the following assertion
4167   // will fail because the thread is no longer a JavaThread as the ~JavaThread
4168   // destructor has completed.
4169 
4170   if (thread->SR_lock() == NULL) {
4171     return;
4172   }
4173 
4174   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4175 
4176   OSThread* osthread = thread->osthread();
4177 
4178   os::SuspendResume::State current = osthread->sr.state();
4179   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4180     suspend_save_context(osthread, siginfo, context);
4181 
4182     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4183     os::SuspendResume::State state = osthread->sr.suspended();
4184     if (state == os::SuspendResume::SR_SUSPENDED) {
4185       sigset_t suspend_set;  // signals for sigsuspend()
4186       sigemptyset(&suspend_set);
4187       // get current set of blocked signals and unblock resume signal
4188       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4189       sigdelset(&suspend_set, SR_signum);
4190 
4191       sr_semaphore.signal();
4192       // wait here until we are resumed
4193       while (1) {
4194         sigsuspend(&suspend_set);
4195 
4196         os::SuspendResume::State result = osthread->sr.running();
4197         if (result == os::SuspendResume::SR_RUNNING) {
4198           sr_semaphore.signal();
4199           break;
4200         }
4201       }
4202 
4203     } else if (state == os::SuspendResume::SR_RUNNING) {
4204       // request was cancelled, continue
4205     } else {
4206       ShouldNotReachHere();
4207     }
4208 
4209     resume_clear_context(osthread);
4210   } else if (current == os::SuspendResume::SR_RUNNING) {
4211     // request was cancelled, continue
4212   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4213     // ignore
4214   } else {
4215     // ignore
4216   }
4217 
4218   errno = old_errno;
4219 }
4220 
4221 static int SR_initialize() {
4222   struct sigaction act;
4223   char *s;
4224 
4225   // Get signal number to use for suspend/resume
4226   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4227     int sig = ::strtol(s, 0, 10);
4228     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4229         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4230       SR_signum = sig;
4231     } else {
4232       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4233               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4234     }
4235   }
4236 
4237   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4238          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4239 
4240   sigemptyset(&SR_sigset);
4241   sigaddset(&SR_sigset, SR_signum);
4242 
4243   // Set up signal handler for suspend/resume
4244   act.sa_flags = SA_RESTART|SA_SIGINFO;
4245   act.sa_handler = (void (*)(int)) SR_handler;
4246 
4247   // SR_signum is blocked by default.
4248   // 4528190 - We also need to block pthread restart signal (32 on all
4249   // supported Linux platforms). Note that LinuxThreads need to block
4250   // this signal for all threads to work properly. So we don't have
4251   // to use hard-coded signal number when setting up the mask.
4252   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4253 
4254   if (sigaction(SR_signum, &act, 0) == -1) {
4255     return -1;
4256   }
4257 
4258   // Save signal flag
4259   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4260   return 0;
4261 }
4262 
4263 static int sr_notify(OSThread* osthread) {
4264   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4265   assert_status(status == 0, status, "pthread_kill");
4266   return status;
4267 }
4268 
4269 // "Randomly" selected value for how long we want to spin
4270 // before bailing out on suspending a thread, also how often
4271 // we send a signal to a thread we want to resume
4272 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4273 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4274 
4275 // returns true on success and false on error - really an error is fatal
4276 // but this seems the normal response to library errors
4277 static bool do_suspend(OSThread* osthread) {
4278   assert(osthread->sr.is_running(), "thread should be running");
4279   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4280 
4281   // mark as suspended and send signal
4282   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4283     // failed to switch, state wasn't running?
4284     ShouldNotReachHere();
4285     return false;
4286   }
4287 
4288   if (sr_notify(osthread) != 0) {
4289     ShouldNotReachHere();
4290   }
4291 
4292   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4293   while (true) {
4294     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4295       break;
4296     } else {
4297       // timeout
4298       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4299       if (cancelled == os::SuspendResume::SR_RUNNING) {
4300         return false;
4301       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4302         // make sure that we consume the signal on the semaphore as well
4303         sr_semaphore.wait();
4304         break;
4305       } else {
4306         ShouldNotReachHere();
4307         return false;
4308       }
4309     }
4310   }
4311 
4312   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4313   return true;
4314 }
4315 
4316 static void do_resume(OSThread* osthread) {
4317   assert(osthread->sr.is_suspended(), "thread should be suspended");
4318   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4319 
4320   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4321     // failed to switch to WAKEUP_REQUEST
4322     ShouldNotReachHere();
4323     return;
4324   }
4325 
4326   while (true) {
4327     if (sr_notify(osthread) == 0) {
4328       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4329         if (osthread->sr.is_running()) {
4330           return;
4331         }
4332       }
4333     } else {
4334       ShouldNotReachHere();
4335     }
4336   }
4337 
4338   guarantee(osthread->sr.is_running(), "Must be running!");
4339 }
4340 
4341 ///////////////////////////////////////////////////////////////////////////////////
4342 // signal handling (except suspend/resume)
4343 
4344 // This routine may be used by user applications as a "hook" to catch signals.
4345 // The user-defined signal handler must pass unrecognized signals to this
4346 // routine, and if it returns true (non-zero), then the signal handler must
4347 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4348 // routine will never retun false (zero), but instead will execute a VM panic
4349 // routine kill the process.
4350 //
4351 // If this routine returns false, it is OK to call it again.  This allows
4352 // the user-defined signal handler to perform checks either before or after
4353 // the VM performs its own checks.  Naturally, the user code would be making
4354 // a serious error if it tried to handle an exception (such as a null check
4355 // or breakpoint) that the VM was generating for its own correct operation.
4356 //
4357 // This routine may recognize any of the following kinds of signals:
4358 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4359 // It should be consulted by handlers for any of those signals.
4360 //
4361 // The caller of this routine must pass in the three arguments supplied
4362 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4363 // field of the structure passed to sigaction().  This routine assumes that
4364 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4365 //
4366 // Note that the VM will print warnings if it detects conflicting signal
4367 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4368 //
4369 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4370                                                  siginfo_t* siginfo,
4371                                                  void* ucontext,
4372                                                  int abort_if_unrecognized);
4373 
4374 void signalHandler(int sig, siginfo_t* info, void* uc) {
4375   assert(info != NULL && uc != NULL, "it must be old kernel");
4376   int orig_errno = errno;  // Preserve errno value over signal handler.
4377   JVM_handle_linux_signal(sig, info, uc, true);
4378   errno = orig_errno;
4379 }
4380 
4381 
4382 // This boolean allows users to forward their own non-matching signals
4383 // to JVM_handle_linux_signal, harmlessly.
4384 bool os::Linux::signal_handlers_are_installed = false;
4385 
4386 // For signal-chaining
4387 struct sigaction sigact[NSIG];
4388 uint64_t sigs = 0;
4389 #if (64 < NSIG-1)
4390 #error "Not all signals can be encoded in sigs. Adapt its type!"
4391 #endif
4392 bool os::Linux::libjsig_is_loaded = false;
4393 typedef struct sigaction *(*get_signal_t)(int);
4394 get_signal_t os::Linux::get_signal_action = NULL;
4395 
4396 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4397   struct sigaction *actp = NULL;
4398 
4399   if (libjsig_is_loaded) {
4400     // Retrieve the old signal handler from libjsig
4401     actp = (*get_signal_action)(sig);
4402   }
4403   if (actp == NULL) {
4404     // Retrieve the preinstalled signal handler from jvm
4405     actp = get_preinstalled_handler(sig);
4406   }
4407 
4408   return actp;
4409 }
4410 
4411 static bool call_chained_handler(struct sigaction *actp, int sig,
4412                                  siginfo_t *siginfo, void *context) {
4413   // Call the old signal handler
4414   if (actp->sa_handler == SIG_DFL) {
4415     // It's more reasonable to let jvm treat it as an unexpected exception
4416     // instead of taking the default action.
4417     return false;
4418   } else if (actp->sa_handler != SIG_IGN) {
4419     if ((actp->sa_flags & SA_NODEFER) == 0) {
4420       // automaticlly block the signal
4421       sigaddset(&(actp->sa_mask), sig);
4422     }
4423 
4424     sa_handler_t hand = NULL;
4425     sa_sigaction_t sa = NULL;
4426     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4427     // retrieve the chained handler
4428     if (siginfo_flag_set) {
4429       sa = actp->sa_sigaction;
4430     } else {
4431       hand = actp->sa_handler;
4432     }
4433 
4434     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4435       actp->sa_handler = SIG_DFL;
4436     }
4437 
4438     // try to honor the signal mask
4439     sigset_t oset;
4440     sigemptyset(&oset);
4441     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4442 
4443     // call into the chained handler
4444     if (siginfo_flag_set) {
4445       (*sa)(sig, siginfo, context);
4446     } else {
4447       (*hand)(sig);
4448     }
4449 
4450     // restore the signal mask
4451     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4452   }
4453   // Tell jvm's signal handler the signal is taken care of.
4454   return true;
4455 }
4456 
4457 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4458   bool chained = false;
4459   // signal-chaining
4460   if (UseSignalChaining) {
4461     struct sigaction *actp = get_chained_signal_action(sig);
4462     if (actp != NULL) {
4463       chained = call_chained_handler(actp, sig, siginfo, context);
4464     }
4465   }
4466   return chained;
4467 }
4468 
4469 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4470   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4471     return &sigact[sig];
4472   }
4473   return NULL;
4474 }
4475 
4476 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4477   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4478   sigact[sig] = oldAct;
4479   sigs |= (uint64_t)1 << (sig-1);
4480 }
4481 
4482 // for diagnostic
4483 int sigflags[NSIG];
4484 
4485 int os::Linux::get_our_sigflags(int sig) {
4486   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4487   return sigflags[sig];
4488 }
4489 
4490 void os::Linux::set_our_sigflags(int sig, int flags) {
4491   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4492   if (sig > 0 && sig < NSIG) {
4493     sigflags[sig] = flags;
4494   }
4495 }
4496 
4497 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4498   // Check for overwrite.
4499   struct sigaction oldAct;
4500   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4501 
4502   void* oldhand = oldAct.sa_sigaction
4503                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4504                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4505   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4506       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4507       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4508     if (AllowUserSignalHandlers || !set_installed) {
4509       // Do not overwrite; user takes responsibility to forward to us.
4510       return;
4511     } else if (UseSignalChaining) {
4512       // save the old handler in jvm
4513       save_preinstalled_handler(sig, oldAct);
4514       // libjsig also interposes the sigaction() call below and saves the
4515       // old sigaction on it own.
4516     } else {
4517       fatal("Encountered unexpected pre-existing sigaction handler "
4518             "%#lx for signal %d.", (long)oldhand, sig);
4519     }
4520   }
4521 
4522   struct sigaction sigAct;
4523   sigfillset(&(sigAct.sa_mask));
4524   sigAct.sa_handler = SIG_DFL;
4525   if (!set_installed) {
4526     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4527   } else {
4528     sigAct.sa_sigaction = signalHandler;
4529     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4530   }
4531   // Save flags, which are set by ours
4532   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4533   sigflags[sig] = sigAct.sa_flags;
4534 
4535   int ret = sigaction(sig, &sigAct, &oldAct);
4536   assert(ret == 0, "check");
4537 
4538   void* oldhand2  = oldAct.sa_sigaction
4539                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4540                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4541   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4542 }
4543 
4544 // install signal handlers for signals that HotSpot needs to
4545 // handle in order to support Java-level exception handling.
4546 
4547 void os::Linux::install_signal_handlers() {
4548   if (!signal_handlers_are_installed) {
4549     signal_handlers_are_installed = true;
4550 
4551     // signal-chaining
4552     typedef void (*signal_setting_t)();
4553     signal_setting_t begin_signal_setting = NULL;
4554     signal_setting_t end_signal_setting = NULL;
4555     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4556                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4557     if (begin_signal_setting != NULL) {
4558       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4559                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4560       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4561                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4562       libjsig_is_loaded = true;
4563       assert(UseSignalChaining, "should enable signal-chaining");
4564     }
4565     if (libjsig_is_loaded) {
4566       // Tell libjsig jvm is setting signal handlers
4567       (*begin_signal_setting)();
4568     }
4569 
4570     set_signal_handler(SIGSEGV, true);
4571     set_signal_handler(SIGPIPE, true);
4572     set_signal_handler(SIGBUS, true);
4573     set_signal_handler(SIGILL, true);
4574     set_signal_handler(SIGFPE, true);
4575 #if defined(PPC64)
4576     set_signal_handler(SIGTRAP, true);
4577 #endif
4578     set_signal_handler(SIGXFSZ, true);
4579 
4580     if (libjsig_is_loaded) {
4581       // Tell libjsig jvm finishes setting signal handlers
4582       (*end_signal_setting)();
4583     }
4584 
4585     // We don't activate signal checker if libjsig is in place, we trust ourselves
4586     // and if UserSignalHandler is installed all bets are off.
4587     // Log that signal checking is off only if -verbose:jni is specified.
4588     if (CheckJNICalls) {
4589       if (libjsig_is_loaded) {
4590         if (PrintJNIResolving) {
4591           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4592         }
4593         check_signals = false;
4594       }
4595       if (AllowUserSignalHandlers) {
4596         if (PrintJNIResolving) {
4597           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4598         }
4599         check_signals = false;
4600       }
4601     }
4602   }
4603 }
4604 
4605 // This is the fastest way to get thread cpu time on Linux.
4606 // Returns cpu time (user+sys) for any thread, not only for current.
4607 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4608 // It might work on 2.6.10+ with a special kernel/glibc patch.
4609 // For reference, please, see IEEE Std 1003.1-2004:
4610 //   http://www.unix.org/single_unix_specification
4611 
4612 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4613   struct timespec tp;
4614   int rc = os::Linux::clock_gettime(clockid, &tp);
4615   assert(rc == 0, "clock_gettime is expected to return 0 code");
4616 
4617   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4618 }
4619 
4620 void os::Linux::initialize_os_info() {
4621   assert(_os_version == 0, "OS info already initialized");
4622 
4623   struct utsname _uname;
4624 
4625   uint32_t major;
4626   uint32_t minor;
4627   uint32_t fix;
4628 
4629   int rc;
4630 
4631   // Kernel version is unknown if
4632   // verification below fails.
4633   _os_version = 0x01000000;
4634 
4635   rc = uname(&_uname);
4636   if (rc != -1) {
4637 
4638     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4639     if (rc == 3) {
4640 
4641       if (major < 256 && minor < 256 && fix < 256) {
4642         // Kernel version format is as expected,
4643         // set it overriding unknown state.
4644         _os_version = (major << 16) |
4645                       (minor << 8 ) |
4646                       (fix   << 0 ) ;
4647       }
4648     }
4649   }
4650 }
4651 
4652 uint32_t os::Linux::os_version() {
4653   assert(_os_version != 0, "not initialized");
4654   return _os_version & 0x00FFFFFF;
4655 }
4656 
4657 bool os::Linux::os_version_is_known() {
4658   assert(_os_version != 0, "not initialized");
4659   return _os_version & 0x01000000 ? false : true;
4660 }
4661 
4662 /////
4663 // glibc on Linux platform uses non-documented flag
4664 // to indicate, that some special sort of signal
4665 // trampoline is used.
4666 // We will never set this flag, and we should
4667 // ignore this flag in our diagnostic
4668 #ifdef SIGNIFICANT_SIGNAL_MASK
4669   #undef SIGNIFICANT_SIGNAL_MASK
4670 #endif
4671 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4672 
4673 static const char* get_signal_handler_name(address handler,
4674                                            char* buf, int buflen) {
4675   int offset = 0;
4676   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4677   if (found) {
4678     // skip directory names
4679     const char *p1, *p2;
4680     p1 = buf;
4681     size_t len = strlen(os::file_separator());
4682     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4683     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4684   } else {
4685     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4686   }
4687   return buf;
4688 }
4689 
4690 static void print_signal_handler(outputStream* st, int sig,
4691                                  char* buf, size_t buflen) {
4692   struct sigaction sa;
4693 
4694   sigaction(sig, NULL, &sa);
4695 
4696   // See comment for SIGNIFICANT_SIGNAL_MASK define
4697   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4698 
4699   st->print("%s: ", os::exception_name(sig, buf, buflen));
4700 
4701   address handler = (sa.sa_flags & SA_SIGINFO)
4702     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4703     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4704 
4705   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4706     st->print("SIG_DFL");
4707   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4708     st->print("SIG_IGN");
4709   } else {
4710     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4711   }
4712 
4713   st->print(", sa_mask[0]=");
4714   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4715 
4716   address rh = VMError::get_resetted_sighandler(sig);
4717   // May be, handler was resetted by VMError?
4718   if (rh != NULL) {
4719     handler = rh;
4720     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4721   }
4722 
4723   st->print(", sa_flags=");
4724   os::Posix::print_sa_flags(st, sa.sa_flags);
4725 
4726   // Check: is it our handler?
4727   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4728       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4729     // It is our signal handler
4730     // check for flags, reset system-used one!
4731     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4732       st->print(
4733                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4734                 os::Linux::get_our_sigflags(sig));
4735     }
4736   }
4737   st->cr();
4738 }
4739 
4740 
4741 #define DO_SIGNAL_CHECK(sig)                      \
4742   do {                                            \
4743     if (!sigismember(&check_signal_done, sig)) {  \
4744       os::Linux::check_signal_handler(sig);       \
4745     }                                             \
4746   } while (0)
4747 
4748 // This method is a periodic task to check for misbehaving JNI applications
4749 // under CheckJNI, we can add any periodic checks here
4750 
4751 void os::run_periodic_checks() {
4752   if (check_signals == false) return;
4753 
4754   // SEGV and BUS if overridden could potentially prevent
4755   // generation of hs*.log in the event of a crash, debugging
4756   // such a case can be very challenging, so we absolutely
4757   // check the following for a good measure:
4758   DO_SIGNAL_CHECK(SIGSEGV);
4759   DO_SIGNAL_CHECK(SIGILL);
4760   DO_SIGNAL_CHECK(SIGFPE);
4761   DO_SIGNAL_CHECK(SIGBUS);
4762   DO_SIGNAL_CHECK(SIGPIPE);
4763   DO_SIGNAL_CHECK(SIGXFSZ);
4764 #if defined(PPC64)
4765   DO_SIGNAL_CHECK(SIGTRAP);
4766 #endif
4767 
4768   // ReduceSignalUsage allows the user to override these handlers
4769   // see comments at the very top and jvm_md.h
4770   if (!ReduceSignalUsage) {
4771     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4772     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4773     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4774     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4775   }
4776 
4777   DO_SIGNAL_CHECK(SR_signum);
4778 }
4779 
4780 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4781 
4782 static os_sigaction_t os_sigaction = NULL;
4783 
4784 void os::Linux::check_signal_handler(int sig) {
4785   char buf[O_BUFLEN];
4786   address jvmHandler = NULL;
4787 
4788 
4789   struct sigaction act;
4790   if (os_sigaction == NULL) {
4791     // only trust the default sigaction, in case it has been interposed
4792     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4793     if (os_sigaction == NULL) return;
4794   }
4795 
4796   os_sigaction(sig, (struct sigaction*)NULL, &act);
4797 
4798 
4799   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4800 
4801   address thisHandler = (act.sa_flags & SA_SIGINFO)
4802     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4803     : CAST_FROM_FN_PTR(address, act.sa_handler);
4804 
4805 
4806   switch (sig) {
4807   case SIGSEGV:
4808   case SIGBUS:
4809   case SIGFPE:
4810   case SIGPIPE:
4811   case SIGILL:
4812   case SIGXFSZ:
4813     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4814     break;
4815 
4816   case SHUTDOWN1_SIGNAL:
4817   case SHUTDOWN2_SIGNAL:
4818   case SHUTDOWN3_SIGNAL:
4819   case BREAK_SIGNAL:
4820     jvmHandler = (address)user_handler();
4821     break;
4822 
4823   default:
4824     if (sig == SR_signum) {
4825       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4826     } else {
4827       return;
4828     }
4829     break;
4830   }
4831 
4832   if (thisHandler != jvmHandler) {
4833     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4834     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4835     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4836     // No need to check this sig any longer
4837     sigaddset(&check_signal_done, sig);
4838     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4839     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4840       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4841                     exception_name(sig, buf, O_BUFLEN));
4842     }
4843   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4844     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4845     tty->print("expected:");
4846     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4847     tty->cr();
4848     tty->print("  found:");
4849     os::Posix::print_sa_flags(tty, act.sa_flags);
4850     tty->cr();
4851     // No need to check this sig any longer
4852     sigaddset(&check_signal_done, sig);
4853   }
4854 
4855   // Dump all the signal
4856   if (sigismember(&check_signal_done, sig)) {
4857     print_signal_handlers(tty, buf, O_BUFLEN);
4858   }
4859 }
4860 
4861 extern void report_error(char* file_name, int line_no, char* title,
4862                          char* format, ...);
4863 
4864 // this is called _before_ the most of global arguments have been parsed
4865 void os::init(void) {
4866   char dummy;   // used to get a guess on initial stack address
4867 //  first_hrtime = gethrtime();
4868 
4869   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4870 
4871   init_random(1234567);
4872 
4873   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4874   if (Linux::page_size() == -1) {
4875     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4876           os::strerror(errno));
4877   }
4878   init_page_sizes((size_t) Linux::page_size());
4879 
4880   Linux::initialize_system_info();
4881 
4882   Linux::initialize_os_info();
4883 
4884   // main_thread points to the aboriginal thread
4885   Linux::_main_thread = pthread_self();
4886 
4887   Linux::clock_init();
4888   initial_time_count = javaTimeNanos();
4889 
4890   // retrieve entry point for pthread_setname_np
4891   Linux::_pthread_setname_np =
4892     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4893 
4894   os::Posix::init();
4895 }
4896 
4897 // To install functions for atexit system call
4898 extern "C" {
4899   static void perfMemory_exit_helper() {
4900     perfMemory_exit();
4901   }
4902 }
4903 
4904 void os::pd_init_container_support() {
4905   OSContainer::init();
4906 }
4907 
4908 // this is called _after_ the global arguments have been parsed
4909 jint os::init_2(void) {
4910 
4911   os::Posix::init_2();
4912 
4913   Linux::fast_thread_clock_init();
4914 
4915   // initialize suspend/resume support - must do this before signal_sets_init()
4916   if (SR_initialize() != 0) {
4917     perror("SR_initialize failed");
4918     return JNI_ERR;
4919   }
4920 
4921   Linux::signal_sets_init();
4922   Linux::install_signal_handlers();
4923 
4924   // Check and sets minimum stack sizes against command line options
4925   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4926     return JNI_ERR;
4927   }
4928   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4929 
4930 #if defined(IA32)
4931   workaround_expand_exec_shield_cs_limit();
4932 #endif
4933 
4934   Linux::libpthread_init();
4935   Linux::sched_getcpu_init();
4936   log_info(os)("HotSpot is running with %s, %s",
4937                Linux::glibc_version(), Linux::libpthread_version());
4938 
4939   if (UseNUMA) {
4940     if (!Linux::libnuma_init()) {
4941       UseNUMA = false;
4942     } else {
4943       if ((Linux::numa_max_node() < 1)) {
4944         // There's only one node(they start from 0), disable NUMA.
4945         UseNUMA = false;
4946       }
4947     }
4948     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4949     // we can make the adaptive lgrp chunk resizing work. If the user specified
4950     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4951     // disable adaptive resizing.
4952     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4953       if (FLAG_IS_DEFAULT(UseNUMA)) {
4954         UseNUMA = false;
4955       } else {
4956         if (FLAG_IS_DEFAULT(UseLargePages) &&
4957             FLAG_IS_DEFAULT(UseSHM) &&
4958             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4959           UseLargePages = false;
4960         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4961           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4962           UseAdaptiveSizePolicy = false;
4963           UseAdaptiveNUMAChunkSizing = false;
4964         }
4965       }
4966     }
4967     if (!UseNUMA && ForceNUMA) {
4968       UseNUMA = true;
4969     }
4970   }
4971 
4972   if (MaxFDLimit) {
4973     // set the number of file descriptors to max. print out error
4974     // if getrlimit/setrlimit fails but continue regardless.
4975     struct rlimit nbr_files;
4976     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4977     if (status != 0) {
4978       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4979     } else {
4980       nbr_files.rlim_cur = nbr_files.rlim_max;
4981       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4982       if (status != 0) {
4983         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4984       }
4985     }
4986   }
4987 
4988   // Initialize lock used to serialize thread creation (see os::create_thread)
4989   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4990 
4991   // at-exit methods are called in the reverse order of their registration.
4992   // atexit functions are called on return from main or as a result of a
4993   // call to exit(3C). There can be only 32 of these functions registered
4994   // and atexit() does not set errno.
4995 
4996   if (PerfAllowAtExitRegistration) {
4997     // only register atexit functions if PerfAllowAtExitRegistration is set.
4998     // atexit functions can be delayed until process exit time, which
4999     // can be problematic for embedded VM situations. Embedded VMs should
5000     // call DestroyJavaVM() to assure that VM resources are released.
5001 
5002     // note: perfMemory_exit_helper atexit function may be removed in
5003     // the future if the appropriate cleanup code can be added to the
5004     // VM_Exit VMOperation's doit method.
5005     if (atexit(perfMemory_exit_helper) != 0) {
5006       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5007     }
5008   }
5009 
5010   // initialize thread priority policy
5011   prio_init();
5012 
5013   return JNI_OK;
5014 }
5015 
5016 // Mark the polling page as unreadable
5017 void os::make_polling_page_unreadable(void) {
5018   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5019     fatal("Could not disable polling page");
5020   }
5021 }
5022 
5023 // Mark the polling page as readable
5024 void os::make_polling_page_readable(void) {
5025   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5026     fatal("Could not enable polling page");
5027   }
5028 }
5029 
5030 // older glibc versions don't have this macro (which expands to
5031 // an optimized bit-counting function) so we have to roll our own
5032 #ifndef CPU_COUNT
5033 
5034 static int _cpu_count(const cpu_set_t* cpus) {
5035   int count = 0;
5036   // only look up to the number of configured processors
5037   for (int i = 0; i < os::processor_count(); i++) {
5038     if (CPU_ISSET(i, cpus)) {
5039       count++;
5040     }
5041   }
5042   return count;
5043 }
5044 
5045 #define CPU_COUNT(cpus) _cpu_count(cpus)
5046 
5047 #endif // CPU_COUNT
5048 
5049 // Get the current number of available processors for this process.
5050 // This value can change at any time during a process's lifetime.
5051 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5052 // If it appears there may be more than 1024 processors then we do a
5053 // dynamic check - see 6515172 for details.
5054 // If anything goes wrong we fallback to returning the number of online
5055 // processors - which can be greater than the number available to the process.
5056 int os::Linux::active_processor_count() {
5057   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
5058   cpu_set_t* cpus_p = &cpus;
5059   int cpus_size = sizeof(cpu_set_t);
5060 
5061   int configured_cpus = os::processor_count();  // upper bound on available cpus
5062   int cpu_count = 0;
5063 
5064 // old build platforms may not support dynamic cpu sets
5065 #ifdef CPU_ALLOC
5066 
5067   // To enable easy testing of the dynamic path on different platforms we
5068   // introduce a diagnostic flag: UseCpuAllocPath
5069   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5070     // kernel may use a mask bigger than cpu_set_t
5071     log_trace(os)("active_processor_count: using dynamic path %s"
5072                   "- configured processors: %d",
5073                   UseCpuAllocPath ? "(forced) " : "",
5074                   configured_cpus);
5075     cpus_p = CPU_ALLOC(configured_cpus);
5076     if (cpus_p != NULL) {
5077       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5078       // zero it just to be safe
5079       CPU_ZERO_S(cpus_size, cpus_p);
5080     }
5081     else {
5082        // failed to allocate so fallback to online cpus
5083        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5084        log_trace(os)("active_processor_count: "
5085                      "CPU_ALLOC failed (%s) - using "
5086                      "online processor count: %d",
5087                      os::strerror(errno), online_cpus);
5088        return online_cpus;
5089     }
5090   }
5091   else {
5092     log_trace(os)("active_processor_count: using static path - configured processors: %d",
5093                   configured_cpus);
5094   }
5095 #else // CPU_ALLOC
5096 // these stubs won't be executed
5097 #define CPU_COUNT_S(size, cpus) -1
5098 #define CPU_FREE(cpus)
5099 
5100   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5101                 configured_cpus);
5102 #endif // CPU_ALLOC
5103 
5104   // pid 0 means the current thread - which we have to assume represents the process
5105   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5106     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5107       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5108     }
5109     else {
5110       cpu_count = CPU_COUNT(cpus_p);
5111     }
5112     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5113   }
5114   else {
5115     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5116     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5117             "which may exceed available processors", os::strerror(errno), cpu_count);
5118   }
5119 
5120   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5121     CPU_FREE(cpus_p);
5122   }
5123 
5124   assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5125   return cpu_count;
5126 }
5127 
5128 // Determine the active processor count from one of
5129 // three different sources:
5130 //
5131 // 1. User option -XX:ActiveProcessorCount
5132 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5133 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5134 //
5135 // Option 1, if specified, will always override.
5136 // If the cgroup subsystem is active and configured, we
5137 // will return the min of the cgroup and option 2 results.
5138 // This is required since tools, such as numactl, that
5139 // alter cpu affinity do not update cgroup subsystem
5140 // cpuset configuration files.
5141 int os::active_processor_count() {
5142   // User has overridden the number of active processors
5143   if (ActiveProcessorCount > 0) {
5144     log_trace(os)("active_processor_count: "
5145                   "active processor count set by user : %d",
5146                   ActiveProcessorCount);
5147     return ActiveProcessorCount;
5148   }
5149 
5150   int active_cpus;
5151   if (OSContainer::is_containerized()) {
5152     active_cpus = OSContainer::active_processor_count();
5153     log_trace(os)("active_processor_count: determined by OSContainer: %d",
5154                    active_cpus);
5155   } else {
5156     active_cpus = os::Linux::active_processor_count();
5157   }
5158 
5159   return active_cpus;
5160 }
5161 
5162 void os::set_native_thread_name(const char *name) {
5163   if (Linux::_pthread_setname_np) {
5164     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5165     snprintf(buf, sizeof(buf), "%s", name);
5166     buf[sizeof(buf) - 1] = '\0';
5167     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5168     // ERANGE should not happen; all other errors should just be ignored.
5169     assert(rc != ERANGE, "pthread_setname_np failed");
5170   }
5171 }
5172 
5173 bool os::distribute_processes(uint length, uint* distribution) {
5174   // Not yet implemented.
5175   return false;
5176 }
5177 
5178 bool os::bind_to_processor(uint processor_id) {
5179   // Not yet implemented.
5180   return false;
5181 }
5182 
5183 ///
5184 
5185 void os::SuspendedThreadTask::internal_do_task() {
5186   if (do_suspend(_thread->osthread())) {
5187     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5188     do_task(context);
5189     do_resume(_thread->osthread());
5190   }
5191 }
5192 
5193 ////////////////////////////////////////////////////////////////////////////////
5194 // debug support
5195 
5196 bool os::find(address addr, outputStream* st) {
5197   Dl_info dlinfo;
5198   memset(&dlinfo, 0, sizeof(dlinfo));
5199   if (dladdr(addr, &dlinfo) != 0) {
5200     st->print(PTR_FORMAT ": ", p2i(addr));
5201     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5202       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5203                 p2i(addr) - p2i(dlinfo.dli_saddr));
5204     } else if (dlinfo.dli_fbase != NULL) {
5205       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5206     } else {
5207       st->print("<absolute address>");
5208     }
5209     if (dlinfo.dli_fname != NULL) {
5210       st->print(" in %s", dlinfo.dli_fname);
5211     }
5212     if (dlinfo.dli_fbase != NULL) {
5213       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5214     }
5215     st->cr();
5216 
5217     if (Verbose) {
5218       // decode some bytes around the PC
5219       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5220       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5221       address       lowest = (address) dlinfo.dli_sname;
5222       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5223       if (begin < lowest)  begin = lowest;
5224       Dl_info dlinfo2;
5225       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5226           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5227         end = (address) dlinfo2.dli_saddr;
5228       }
5229       Disassembler::decode(begin, end, st);
5230     }
5231     return true;
5232   }
5233   return false;
5234 }
5235 
5236 ////////////////////////////////////////////////////////////////////////////////
5237 // misc
5238 
5239 // This does not do anything on Linux. This is basically a hook for being
5240 // able to use structured exception handling (thread-local exception filters)
5241 // on, e.g., Win32.
5242 void
5243 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5244                          JavaCallArguments* args, Thread* thread) {
5245   f(value, method, args, thread);
5246 }
5247 
5248 void os::print_statistics() {
5249 }
5250 
5251 bool os::message_box(const char* title, const char* message) {
5252   int i;
5253   fdStream err(defaultStream::error_fd());
5254   for (i = 0; i < 78; i++) err.print_raw("=");
5255   err.cr();
5256   err.print_raw_cr(title);
5257   for (i = 0; i < 78; i++) err.print_raw("-");
5258   err.cr();
5259   err.print_raw_cr(message);
5260   for (i = 0; i < 78; i++) err.print_raw("=");
5261   err.cr();
5262 
5263   char buf[16];
5264   // Prevent process from exiting upon "read error" without consuming all CPU
5265   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5266 
5267   return buf[0] == 'y' || buf[0] == 'Y';
5268 }
5269 
5270 int os::stat(const char *path, struct stat *sbuf) {
5271   char pathbuf[MAX_PATH];
5272   if (strlen(path) > MAX_PATH - 1) {
5273     errno = ENAMETOOLONG;
5274     return -1;
5275   }
5276   os::native_path(strcpy(pathbuf, path));
5277   return ::stat(pathbuf, sbuf);
5278 }
5279 
5280 // Is a (classpath) directory empty?
5281 bool os::dir_is_empty(const char* path) {
5282   DIR *dir = NULL;
5283   struct dirent *ptr;
5284 
5285   dir = opendir(path);
5286   if (dir == NULL) return true;
5287 
5288   // Scan the directory
5289   bool result = true;
5290   char buf[sizeof(struct dirent) + MAX_PATH];
5291   while (result && (ptr = ::readdir(dir)) != NULL) {
5292     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5293       result = false;
5294     }
5295   }
5296   closedir(dir);
5297   return result;
5298 }
5299 
5300 // This code originates from JDK's sysOpen and open64_w
5301 // from src/solaris/hpi/src/system_md.c
5302 
5303 int os::open(const char *path, int oflag, int mode) {
5304   if (strlen(path) > MAX_PATH - 1) {
5305     errno = ENAMETOOLONG;
5306     return -1;
5307   }
5308 
5309   // All file descriptors that are opened in the Java process and not
5310   // specifically destined for a subprocess should have the close-on-exec
5311   // flag set.  If we don't set it, then careless 3rd party native code
5312   // might fork and exec without closing all appropriate file descriptors
5313   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5314   // turn might:
5315   //
5316   // - cause end-of-file to fail to be detected on some file
5317   //   descriptors, resulting in mysterious hangs, or
5318   //
5319   // - might cause an fopen in the subprocess to fail on a system
5320   //   suffering from bug 1085341.
5321   //
5322   // (Yes, the default setting of the close-on-exec flag is a Unix
5323   // design flaw)
5324   //
5325   // See:
5326   // 1085341: 32-bit stdio routines should support file descriptors >255
5327   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5328   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5329   //
5330   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5331   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5332   // because it saves a system call and removes a small window where the flag
5333   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5334   // and we fall back to using FD_CLOEXEC (see below).
5335 #ifdef O_CLOEXEC
5336   oflag |= O_CLOEXEC;
5337 #endif
5338 
5339   int fd = ::open64(path, oflag, mode);
5340   if (fd == -1) return -1;
5341 
5342   //If the open succeeded, the file might still be a directory
5343   {
5344     struct stat64 buf64;
5345     int ret = ::fstat64(fd, &buf64);
5346     int st_mode = buf64.st_mode;
5347 
5348     if (ret != -1) {
5349       if ((st_mode & S_IFMT) == S_IFDIR) {
5350         errno = EISDIR;
5351         ::close(fd);
5352         return -1;
5353       }
5354     } else {
5355       ::close(fd);
5356       return -1;
5357     }
5358   }
5359 
5360 #ifdef FD_CLOEXEC
5361   // Validate that the use of the O_CLOEXEC flag on open above worked.
5362   // With recent kernels, we will perform this check exactly once.
5363   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5364   if (!O_CLOEXEC_is_known_to_work) {
5365     int flags = ::fcntl(fd, F_GETFD);
5366     if (flags != -1) {
5367       if ((flags & FD_CLOEXEC) != 0)
5368         O_CLOEXEC_is_known_to_work = 1;
5369       else
5370         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5371     }
5372   }
5373 #endif
5374 
5375   return fd;
5376 }
5377 
5378 
5379 // create binary file, rewriting existing file if required
5380 int os::create_binary_file(const char* path, bool rewrite_existing) {
5381   int oflags = O_WRONLY | O_CREAT;
5382   if (!rewrite_existing) {
5383     oflags |= O_EXCL;
5384   }
5385   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5386 }
5387 
5388 // return current position of file pointer
5389 jlong os::current_file_offset(int fd) {
5390   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5391 }
5392 
5393 // move file pointer to the specified offset
5394 jlong os::seek_to_file_offset(int fd, jlong offset) {
5395   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5396 }
5397 
5398 // This code originates from JDK's sysAvailable
5399 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5400 
5401 int os::available(int fd, jlong *bytes) {
5402   jlong cur, end;
5403   int mode;
5404   struct stat64 buf64;
5405 
5406   if (::fstat64(fd, &buf64) >= 0) {
5407     mode = buf64.st_mode;
5408     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5409       int n;
5410       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5411         *bytes = n;
5412         return 1;
5413       }
5414     }
5415   }
5416   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5417     return 0;
5418   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5419     return 0;
5420   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5421     return 0;
5422   }
5423   *bytes = end - cur;
5424   return 1;
5425 }
5426 
5427 // Map a block of memory.
5428 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5429                         char *addr, size_t bytes, bool read_only,
5430                         bool allow_exec) {
5431   int prot;
5432   int flags = MAP_PRIVATE;
5433 
5434   if (read_only) {
5435     prot = PROT_READ;
5436   } else {
5437     prot = PROT_READ | PROT_WRITE;
5438   }
5439 
5440   if (allow_exec) {
5441     prot |= PROT_EXEC;
5442   }
5443 
5444   if (addr != NULL) {
5445     flags |= MAP_FIXED;
5446   }
5447 
5448   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5449                                      fd, file_offset);
5450   if (mapped_address == MAP_FAILED) {
5451     return NULL;
5452   }
5453   return mapped_address;
5454 }
5455 
5456 
5457 // Remap a block of memory.
5458 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5459                           char *addr, size_t bytes, bool read_only,
5460                           bool allow_exec) {
5461   // same as map_memory() on this OS
5462   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5463                         allow_exec);
5464 }
5465 
5466 
5467 // Unmap a block of memory.
5468 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5469   return munmap(addr, bytes) == 0;
5470 }
5471 
5472 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5473 
5474 static clockid_t thread_cpu_clockid(Thread* thread) {
5475   pthread_t tid = thread->osthread()->pthread_id();
5476   clockid_t clockid;
5477 
5478   // Get thread clockid
5479   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5480   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5481   return clockid;
5482 }
5483 
5484 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5485 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5486 // of a thread.
5487 //
5488 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5489 // the fast estimate available on the platform.
5490 
5491 jlong os::current_thread_cpu_time() {
5492   if (os::Linux::supports_fast_thread_cpu_time()) {
5493     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5494   } else {
5495     // return user + sys since the cost is the same
5496     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5497   }
5498 }
5499 
5500 jlong os::thread_cpu_time(Thread* thread) {
5501   // consistent with what current_thread_cpu_time() returns
5502   if (os::Linux::supports_fast_thread_cpu_time()) {
5503     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5504   } else {
5505     return slow_thread_cpu_time(thread, true /* user + sys */);
5506   }
5507 }
5508 
5509 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5510   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5511     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5512   } else {
5513     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5514   }
5515 }
5516 
5517 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5518   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5519     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5520   } else {
5521     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5522   }
5523 }
5524 
5525 //  -1 on error.
5526 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5527   pid_t  tid = thread->osthread()->thread_id();
5528   char *s;
5529   char stat[2048];
5530   int statlen;
5531   char proc_name[64];
5532   int count;
5533   long sys_time, user_time;
5534   char cdummy;
5535   int idummy;
5536   long ldummy;
5537   FILE *fp;
5538 
5539   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5540   fp = fopen(proc_name, "r");
5541   if (fp == NULL) return -1;
5542   statlen = fread(stat, 1, 2047, fp);
5543   stat[statlen] = '\0';
5544   fclose(fp);
5545 
5546   // Skip pid and the command string. Note that we could be dealing with
5547   // weird command names, e.g. user could decide to rename java launcher
5548   // to "java 1.4.2 :)", then the stat file would look like
5549   //                1234 (java 1.4.2 :)) R ... ...
5550   // We don't really need to know the command string, just find the last
5551   // occurrence of ")" and then start parsing from there. See bug 4726580.
5552   s = strrchr(stat, ')');
5553   if (s == NULL) return -1;
5554 
5555   // Skip blank chars
5556   do { s++; } while (s && isspace(*s));
5557 
5558   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5559                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5560                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5561                  &user_time, &sys_time);
5562   if (count != 13) return -1;
5563   if (user_sys_cpu_time) {
5564     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5565   } else {
5566     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5567   }
5568 }
5569 
5570 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5571   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5572   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5573   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5574   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5575 }
5576 
5577 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5578   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5579   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5580   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5581   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5582 }
5583 
5584 bool os::is_thread_cpu_time_supported() {
5585   return true;
5586 }
5587 
5588 // System loadavg support.  Returns -1 if load average cannot be obtained.
5589 // Linux doesn't yet have a (official) notion of processor sets,
5590 // so just return the system wide load average.
5591 int os::loadavg(double loadavg[], int nelem) {
5592   return ::getloadavg(loadavg, nelem);
5593 }
5594 
5595 void os::pause() {
5596   char filename[MAX_PATH];
5597   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5598     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5599   } else {
5600     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5601   }
5602 
5603   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5604   if (fd != -1) {
5605     struct stat buf;
5606     ::close(fd);
5607     while (::stat(filename, &buf) == 0) {
5608       (void)::poll(NULL, 0, 100);
5609     }
5610   } else {
5611     jio_fprintf(stderr,
5612                 "Could not open pause file '%s', continuing immediately.\n", filename);
5613   }
5614 }
5615 
5616 extern char** environ;
5617 
5618 // Run the specified command in a separate process. Return its exit value,
5619 // or -1 on failure (e.g. can't fork a new process).
5620 // Unlike system(), this function can be called from signal handler. It
5621 // doesn't block SIGINT et al.
5622 int os::fork_and_exec(char* cmd) {
5623   const char * argv[4] = {"sh", "-c", cmd, NULL};
5624 
5625   pid_t pid = fork();
5626 
5627   if (pid < 0) {
5628     // fork failed
5629     return -1;
5630 
5631   } else if (pid == 0) {
5632     // child process
5633 
5634     execve("/bin/sh", (char* const*)argv, environ);
5635 
5636     // execve failed
5637     _exit(-1);
5638 
5639   } else  {
5640     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5641     // care about the actual exit code, for now.
5642 
5643     int status;
5644 
5645     // Wait for the child process to exit.  This returns immediately if
5646     // the child has already exited. */
5647     while (waitpid(pid, &status, 0) < 0) {
5648       switch (errno) {
5649       case ECHILD: return 0;
5650       case EINTR: break;
5651       default: return -1;
5652       }
5653     }
5654 
5655     if (WIFEXITED(status)) {
5656       // The child exited normally; get its exit code.
5657       return WEXITSTATUS(status);
5658     } else if (WIFSIGNALED(status)) {
5659       // The child exited because of a signal
5660       // The best value to return is 0x80 + signal number,
5661       // because that is what all Unix shells do, and because
5662       // it allows callers to distinguish between process exit and
5663       // process death by signal.
5664       return 0x80 + WTERMSIG(status);
5665     } else {
5666       // Unknown exit code; pass it through
5667       return status;
5668     }
5669   }
5670 }
5671 
5672 // is_headless_jre()
5673 //
5674 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5675 // in order to report if we are running in a headless jre
5676 //
5677 // Since JDK8 xawt/libmawt.so was moved into the same directory
5678 // as libawt.so, and renamed libawt_xawt.so
5679 //
5680 bool os::is_headless_jre() {
5681   struct stat statbuf;
5682   char buf[MAXPATHLEN];
5683   char libmawtpath[MAXPATHLEN];
5684   const char *xawtstr  = "/xawt/libmawt.so";
5685   const char *new_xawtstr = "/libawt_xawt.so";
5686   char *p;
5687 
5688   // Get path to libjvm.so
5689   os::jvm_path(buf, sizeof(buf));
5690 
5691   // Get rid of libjvm.so
5692   p = strrchr(buf, '/');
5693   if (p == NULL) {
5694     return false;
5695   } else {
5696     *p = '\0';
5697   }
5698 
5699   // Get rid of client or server
5700   p = strrchr(buf, '/');
5701   if (p == NULL) {
5702     return false;
5703   } else {
5704     *p = '\0';
5705   }
5706 
5707   // check xawt/libmawt.so
5708   strcpy(libmawtpath, buf);
5709   strcat(libmawtpath, xawtstr);
5710   if (::stat(libmawtpath, &statbuf) == 0) return false;
5711 
5712   // check libawt_xawt.so
5713   strcpy(libmawtpath, buf);
5714   strcat(libmawtpath, new_xawtstr);
5715   if (::stat(libmawtpath, &statbuf) == 0) return false;
5716 
5717   return true;
5718 }
5719 
5720 // Get the default path to the core file
5721 // Returns the length of the string
5722 int os::get_core_path(char* buffer, size_t bufferSize) {
5723   /*
5724    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5725    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5726    */
5727   const int core_pattern_len = 129;
5728   char core_pattern[core_pattern_len] = {0};
5729 
5730   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5731   if (core_pattern_file == -1) {
5732     return -1;
5733   }
5734 
5735   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5736   ::close(core_pattern_file);
5737   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5738     return -1;
5739   }
5740   if (core_pattern[ret-1] == '\n') {
5741     core_pattern[ret-1] = '\0';
5742   } else {
5743     core_pattern[ret] = '\0';
5744   }
5745 
5746   char *pid_pos = strstr(core_pattern, "%p");
5747   int written;
5748 
5749   if (core_pattern[0] == '/') {
5750     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5751   } else {
5752     char cwd[PATH_MAX];
5753 
5754     const char* p = get_current_directory(cwd, PATH_MAX);
5755     if (p == NULL) {
5756       return -1;
5757     }
5758 
5759     if (core_pattern[0] == '|') {
5760       written = jio_snprintf(buffer, bufferSize,
5761                              "\"%s\" (or dumping to %s/core.%d)",
5762                              &core_pattern[1], p, current_process_id());
5763     } else {
5764       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5765     }
5766   }
5767 
5768   if (written < 0) {
5769     return -1;
5770   }
5771 
5772   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5773     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5774 
5775     if (core_uses_pid_file != -1) {
5776       char core_uses_pid = 0;
5777       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5778       ::close(core_uses_pid_file);
5779 
5780       if (core_uses_pid == '1') {
5781         jio_snprintf(buffer + written, bufferSize - written,
5782                                           ".%d", current_process_id());
5783       }
5784     }
5785   }
5786 
5787   return strlen(buffer);
5788 }
5789 
5790 bool os::start_debugging(char *buf, int buflen) {
5791   int len = (int)strlen(buf);
5792   char *p = &buf[len];
5793 
5794   jio_snprintf(p, buflen-len,
5795                "\n\n"
5796                "Do you want to debug the problem?\n\n"
5797                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5798                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5799                "Otherwise, press RETURN to abort...",
5800                os::current_process_id(), os::current_process_id(),
5801                os::current_thread_id(), os::current_thread_id());
5802 
5803   bool yes = os::message_box("Unexpected Error", buf);
5804 
5805   if (yes) {
5806     // yes, user asked VM to launch debugger
5807     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5808                  os::current_process_id(), os::current_process_id());
5809 
5810     os::fork_and_exec(buf);
5811     yes = false;
5812   }
5813   return yes;
5814 }
5815 
5816 
5817 // Java/Compiler thread:
5818 //
5819 //   Low memory addresses
5820 // P0 +------------------------+
5821 //    |                        |\  Java thread created by VM does not have glibc
5822 //    |    glibc guard page    | - guard page, attached Java thread usually has
5823 //    |                        |/  1 glibc guard page.
5824 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5825 //    |                        |\
5826 //    |  HotSpot Guard Pages   | - red, yellow and reserved pages
5827 //    |                        |/
5828 //    +------------------------+ JavaThread::stack_reserved_zone_base()
5829 //    |                        |\
5830 //    |      Normal Stack      | -
5831 //    |                        |/
5832 // P2 +------------------------+ Thread::stack_base()
5833 //
5834 // Non-Java thread:
5835 //
5836 //   Low memory addresses
5837 // P0 +------------------------+
5838 //    |                        |\
5839 //    |  glibc guard page      | - usually 1 page
5840 //    |                        |/
5841 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5842 //    |                        |\
5843 //    |      Normal Stack      | -
5844 //    |                        |/
5845 // P2 +------------------------+ Thread::stack_base()
5846 //
5847 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5848 //    returned from pthread_attr_getstack().
5849 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5850 //    of the stack size given in pthread_attr. We work around this for
5851 //    threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5852 //
5853 #ifndef ZERO
5854 static void current_stack_region(address * bottom, size_t * size) {
5855   if (os::Linux::is_initial_thread()) {
5856     // initial thread needs special handling because pthread_getattr_np()
5857     // may return bogus value.
5858     *bottom = os::Linux::initial_thread_stack_bottom();
5859     *size   = os::Linux::initial_thread_stack_size();
5860   } else {
5861     pthread_attr_t attr;
5862 
5863     int rslt = pthread_getattr_np(pthread_self(), &attr);
5864 
5865     // JVM needs to know exact stack location, abort if it fails
5866     if (rslt != 0) {
5867       if (rslt == ENOMEM) {
5868         vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5869       } else {
5870         fatal("pthread_getattr_np failed with error = %d", rslt);
5871       }
5872     }
5873 
5874     if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5875       fatal("Cannot locate current stack attributes!");
5876     }
5877 
5878     // Work around NPTL stack guard error.
5879     size_t guard_size = 0;
5880     rslt = pthread_attr_getguardsize(&attr, &guard_size);
5881     if (rslt != 0) {
5882       fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5883     }
5884     *bottom += guard_size;
5885     *size   -= guard_size;
5886 
5887     pthread_attr_destroy(&attr);
5888 
5889   }
5890   assert(os::current_stack_pointer() >= *bottom &&
5891          os::current_stack_pointer() < *bottom + *size, "just checking");
5892 }
5893 
5894 address os::current_stack_base() {
5895   address bottom;
5896   size_t size;
5897   current_stack_region(&bottom, &size);
5898   return (bottom + size);
5899 }
5900 
5901 size_t os::current_stack_size() {
5902   // This stack size includes the usable stack and HotSpot guard pages
5903   // (for the threads that have Hotspot guard pages).
5904   address bottom;
5905   size_t size;
5906   current_stack_region(&bottom, &size);
5907   return size;
5908 }
5909 #endif
5910 
5911 static inline struct timespec get_mtime(const char* filename) {
5912   struct stat st;
5913   int ret = os::stat(filename, &st);
5914   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5915   return st.st_mtim;
5916 }
5917 
5918 int os::compare_file_modified_times(const char* file1, const char* file2) {
5919   struct timespec filetime1 = get_mtime(file1);
5920   struct timespec filetime2 = get_mtime(file2);
5921   int diff = filetime1.tv_sec - filetime2.tv_sec;
5922   if (diff == 0) {
5923     return filetime1.tv_nsec - filetime2.tv_nsec;
5924   }
5925   return diff;
5926 }
5927 
5928 /////////////// Unit tests ///////////////
5929 
5930 #ifndef PRODUCT
5931 
5932 #define test_log(...)              \
5933   do {                             \
5934     if (VerboseInternalVMTests) {  \
5935       tty->print_cr(__VA_ARGS__);  \
5936       tty->flush();                \
5937     }                              \
5938   } while (false)
5939 
5940 class TestReserveMemorySpecial : AllStatic {
5941  public:
5942   static void small_page_write(void* addr, size_t size) {
5943     size_t page_size = os::vm_page_size();
5944 
5945     char* end = (char*)addr + size;
5946     for (char* p = (char*)addr; p < end; p += page_size) {
5947       *p = 1;
5948     }
5949   }
5950 
5951   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5952     if (!UseHugeTLBFS) {
5953       return;
5954     }
5955 
5956     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5957 
5958     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5959 
5960     if (addr != NULL) {
5961       small_page_write(addr, size);
5962 
5963       os::Linux::release_memory_special_huge_tlbfs(addr, size);
5964     }
5965   }
5966 
5967   static void test_reserve_memory_special_huge_tlbfs_only() {
5968     if (!UseHugeTLBFS) {
5969       return;
5970     }
5971 
5972     size_t lp = os::large_page_size();
5973 
5974     for (size_t size = lp; size <= lp * 10; size += lp) {
5975       test_reserve_memory_special_huge_tlbfs_only(size);
5976     }
5977   }
5978 
5979   static void test_reserve_memory_special_huge_tlbfs_mixed() {
5980     size_t lp = os::large_page_size();
5981     size_t ag = os::vm_allocation_granularity();
5982 
5983     // sizes to test
5984     const size_t sizes[] = {
5985       lp, lp + ag, lp + lp / 2, lp * 2,
5986       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
5987       lp * 10, lp * 10 + lp / 2
5988     };
5989     const int num_sizes = sizeof(sizes) / sizeof(size_t);
5990 
5991     // For each size/alignment combination, we test three scenarios:
5992     // 1) with req_addr == NULL
5993     // 2) with a non-null req_addr at which we expect to successfully allocate
5994     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
5995     //    expect the allocation to either fail or to ignore req_addr
5996 
5997     // Pre-allocate two areas; they shall be as large as the largest allocation
5998     //  and aligned to the largest alignment we will be testing.
5999     const size_t mapping_size = sizes[num_sizes - 1] * 2;
6000     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6001       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6002       -1, 0);
6003     assert(mapping1 != MAP_FAILED, "should work");
6004 
6005     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6006       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6007       -1, 0);
6008     assert(mapping2 != MAP_FAILED, "should work");
6009 
6010     // Unmap the first mapping, but leave the second mapping intact: the first
6011     // mapping will serve as a value for a "good" req_addr (case 2). The second
6012     // mapping, still intact, as "bad" req_addr (case 3).
6013     ::munmap(mapping1, mapping_size);
6014 
6015     // Case 1
6016     test_log("%s, req_addr NULL:", __FUNCTION__);
6017     test_log("size            align           result");
6018 
6019     for (int i = 0; i < num_sizes; i++) {
6020       const size_t size = sizes[i];
6021       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6022         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6023         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6024                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6025         if (p != NULL) {
6026           assert(is_aligned(p, alignment), "must be");
6027           small_page_write(p, size);
6028           os::Linux::release_memory_special_huge_tlbfs(p, size);
6029         }
6030       }
6031     }
6032 
6033     // Case 2
6034     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6035     test_log("size            align           req_addr         result");
6036 
6037     for (int i = 0; i < num_sizes; i++) {
6038       const size_t size = sizes[i];
6039       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6040         char* const req_addr = align_up(mapping1, alignment);
6041         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6042         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6043                  size, alignment, p2i(req_addr), p2i(p),
6044                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6045         if (p != NULL) {
6046           assert(p == req_addr, "must be");
6047           small_page_write(p, size);
6048           os::Linux::release_memory_special_huge_tlbfs(p, size);
6049         }
6050       }
6051     }
6052 
6053     // Case 3
6054     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6055     test_log("size            align           req_addr         result");
6056 
6057     for (int i = 0; i < num_sizes; i++) {
6058       const size_t size = sizes[i];
6059       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6060         char* const req_addr = align_up(mapping2, alignment);
6061         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6062         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6063                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6064         // as the area around req_addr contains already existing mappings, the API should always
6065         // return NULL (as per contract, it cannot return another address)
6066         assert(p == NULL, "must be");
6067       }
6068     }
6069 
6070     ::munmap(mapping2, mapping_size);
6071 
6072   }
6073 
6074   static void test_reserve_memory_special_huge_tlbfs() {
6075     if (!UseHugeTLBFS) {
6076       return;
6077     }
6078 
6079     test_reserve_memory_special_huge_tlbfs_only();
6080     test_reserve_memory_special_huge_tlbfs_mixed();
6081   }
6082 
6083   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6084     if (!UseSHM) {
6085       return;
6086     }
6087 
6088     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6089 
6090     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6091 
6092     if (addr != NULL) {
6093       assert(is_aligned(addr, alignment), "Check");
6094       assert(is_aligned(addr, os::large_page_size()), "Check");
6095 
6096       small_page_write(addr, size);
6097 
6098       os::Linux::release_memory_special_shm(addr, size);
6099     }
6100   }
6101 
6102   static void test_reserve_memory_special_shm() {
6103     size_t lp = os::large_page_size();
6104     size_t ag = os::vm_allocation_granularity();
6105 
6106     for (size_t size = ag; size < lp * 3; size += ag) {
6107       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6108         test_reserve_memory_special_shm(size, alignment);
6109       }
6110     }
6111   }
6112 
6113   static void test() {
6114     test_reserve_memory_special_huge_tlbfs();
6115     test_reserve_memory_special_shm();
6116   }
6117 };
6118 
6119 void TestReserveMemorySpecial_test() {
6120   TestReserveMemorySpecial::test();
6121 }
6122 
6123 #endif