1 /*
   2  * Copyright (c) 1997, 2013, 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 "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_solaris.h"
  35 #include "memory/allocation.inline.hpp"
  36 #include "memory/filemap.hpp"
  37 #include "mutex_solaris.inline.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_share_solaris.hpp"
  40 #include "prims/jniFastGetField.hpp"
  41 #include "prims/jvm.h"
  42 #include "prims/jvm_misc.hpp"
  43 #include "runtime/arguments.hpp"
  44 #include "runtime/extendedPC.hpp"
  45 #include "runtime/globals.hpp"
  46 #include "runtime/interfaceSupport.hpp"
  47 #include "runtime/java.hpp"
  48 #include "runtime/javaCalls.hpp"
  49 #include "runtime/mutexLocker.hpp"
  50 #include "runtime/objectMonitor.hpp"
  51 #include "runtime/osThread.hpp"
  52 #include "runtime/perfMemory.hpp"
  53 #include "runtime/sharedRuntime.hpp"
  54 #include "runtime/statSampler.hpp"
  55 #include "runtime/stubRoutines.hpp"
  56 #include "runtime/thread.inline.hpp"
  57 #include "runtime/threadCritical.hpp"
  58 #include "runtime/timer.hpp"
  59 #include "services/attachListener.hpp"
  60 #include "services/memTracker.hpp"
  61 #include "services/runtimeService.hpp"
  62 #include "utilities/decoder.hpp"
  63 #include "utilities/defaultStream.hpp"
  64 #include "utilities/events.hpp"
  65 #include "utilities/growableArray.hpp"
  66 #include "utilities/vmError.hpp"
  67 
  68 // put OS-includes here
  69 # include <dlfcn.h>
  70 # include <errno.h>
  71 # include <exception>
  72 # include <link.h>
  73 # include <poll.h>
  74 # include <pthread.h>
  75 # include <pwd.h>
  76 # include <schedctl.h>
  77 # include <setjmp.h>
  78 # include <signal.h>
  79 # include <stdio.h>
  80 # include <alloca.h>
  81 # include <sys/filio.h>
  82 # include <sys/ipc.h>
  83 # include <sys/lwp.h>
  84 # include <sys/machelf.h>     // for elf Sym structure used by dladdr1
  85 # include <sys/mman.h>
  86 # include <sys/processor.h>
  87 # include <sys/procset.h>
  88 # include <sys/pset.h>
  89 # include <sys/resource.h>
  90 # include <sys/shm.h>
  91 # include <sys/socket.h>
  92 # include <sys/stat.h>
  93 # include <sys/systeminfo.h>
  94 # include <sys/time.h>
  95 # include <sys/times.h>
  96 # include <sys/types.h>
  97 # include <sys/wait.h>
  98 # include <sys/utsname.h>
  99 # include <thread.h>
 100 # include <unistd.h>
 101 # include <sys/priocntl.h>
 102 # include <sys/rtpriocntl.h>
 103 # include <sys/tspriocntl.h>
 104 # include <sys/iapriocntl.h>
 105 # include <sys/fxpriocntl.h>
 106 # include <sys/loadavg.h>
 107 # include <string.h>
 108 # include <stdio.h>
 109 
 110 # define _STRUCTURED_PROC 1  //  this gets us the new structured proc interfaces of 5.6 & later
 111 # include <sys/procfs.h>     //  see comment in <sys/procfs.h>
 112 
 113 #define MAX_PATH (2 * K)
 114 
 115 // for timer info max values which include all bits
 116 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 117 
 118 
 119 // Here are some liblgrp types from sys/lgrp_user.h to be able to
 120 // compile on older systems without this header file.
 121 
 122 #ifndef MADV_ACCESS_LWP
 123 # define  MADV_ACCESS_LWP         7       /* next LWP to access heavily */
 124 #endif
 125 #ifndef MADV_ACCESS_MANY
 126 # define  MADV_ACCESS_MANY        8       /* many processes to access heavily */
 127 #endif
 128 
 129 #ifndef LGRP_RSRC_CPU
 130 # define LGRP_RSRC_CPU           0       /* CPU resources */
 131 #endif
 132 #ifndef LGRP_RSRC_MEM
 133 # define LGRP_RSRC_MEM           1       /* memory resources */
 134 #endif
 135 
 136 // see thr_setprio(3T) for the basis of these numbers
 137 #define MinimumPriority 0
 138 #define NormalPriority  64
 139 #define MaximumPriority 127
 140 
 141 // Values for ThreadPriorityPolicy == 1
 142 int prio_policy1[CriticalPriority+1] = {
 143   -99999,  0, 16,  32,  48,  64,
 144           80, 96, 112, 124, 127, 127 };
 145 
 146 // System parameters used internally
 147 static clock_t clock_tics_per_sec = 100;
 148 
 149 // Track if we have called enable_extended_FILE_stdio (on Solaris 10u4+)
 150 static bool enabled_extended_FILE_stdio = false;
 151 
 152 // For diagnostics to print a message once. see run_periodic_checks
 153 static bool check_addr0_done = false;
 154 static sigset_t check_signal_done;
 155 static bool check_signals = true;
 156 
 157 address os::Solaris::handler_start;  // start pc of thr_sighndlrinfo
 158 address os::Solaris::handler_end;    // end pc of thr_sighndlrinfo
 159 
 160 address os::Solaris::_main_stack_base = NULL;  // 4352906 workaround
 161 
 162 
 163 // "default" initializers for missing libc APIs
 164 extern "C" {
 165   static int lwp_mutex_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
 166   static int lwp_mutex_destroy(mutex_t *mx)                 { return 0; }
 167 
 168   static int lwp_cond_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
 169   static int lwp_cond_destroy(cond_t *cv)                   { return 0; }
 170 }
 171 
 172 // "default" initializers for pthread-based synchronization
 173 extern "C" {
 174   static int pthread_mutex_default_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
 175   static int pthread_cond_default_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
 176 }
 177 
 178 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 179 
 180 // Thread Local Storage
 181 // This is common to all Solaris platforms so it is defined here,
 182 // in this common file.
 183 // The declarations are in the os_cpu threadLS*.hpp files.
 184 //
 185 // Static member initialization for TLS
 186 Thread* ThreadLocalStorage::_get_thread_cache[ThreadLocalStorage::_pd_cache_size] = {NULL};
 187 
 188 #ifndef PRODUCT
 189 #define _PCT(n,d)       ((100.0*(double)(n))/(double)(d))
 190 
 191 int ThreadLocalStorage::_tcacheHit = 0;
 192 int ThreadLocalStorage::_tcacheMiss = 0;
 193 
 194 void ThreadLocalStorage::print_statistics() {
 195   int total = _tcacheMiss+_tcacheHit;
 196   tty->print_cr("Thread cache hits %d misses %d total %d percent %f\n",
 197                 _tcacheHit, _tcacheMiss, total, _PCT(_tcacheHit, total));
 198 }
 199 #undef _PCT
 200 #endif // PRODUCT
 201 
 202 Thread* ThreadLocalStorage::get_thread_via_cache_slowly(uintptr_t raw_id,
 203                                                         int index) {
 204   Thread *thread = get_thread_slow();
 205   if (thread != NULL) {
 206     address sp = os::current_stack_pointer();
 207     guarantee(thread->_stack_base == NULL ||
 208               (sp <= thread->_stack_base &&
 209                  sp >= thread->_stack_base - thread->_stack_size) ||
 210                is_error_reported(),
 211               "sp must be inside of selected thread stack");
 212 
 213     thread->set_self_raw_id(raw_id);  // mark for quick retrieval
 214     _get_thread_cache[ index ] = thread;
 215   }
 216   return thread;
 217 }
 218 
 219 
 220 static const double all_zero[ sizeof(Thread) / sizeof(double) + 1 ] = {0};
 221 #define NO_CACHED_THREAD ((Thread*)all_zero)
 222 
 223 void ThreadLocalStorage::pd_set_thread(Thread* thread) {
 224 
 225   // Store the new value before updating the cache to prevent a race
 226   // between get_thread_via_cache_slowly() and this store operation.
 227   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
 228 
 229   // Update thread cache with new thread if setting on thread create,
 230   // or NO_CACHED_THREAD (zeroed) thread if resetting thread on exit.
 231   uintptr_t raw = pd_raw_thread_id();
 232   int ix = pd_cache_index(raw);
 233   _get_thread_cache[ix] = thread == NULL ? NO_CACHED_THREAD : thread;
 234 }
 235 
 236 void ThreadLocalStorage::pd_init() {
 237   for (int i = 0; i < _pd_cache_size; i++) {
 238     _get_thread_cache[i] = NO_CACHED_THREAD;
 239   }
 240 }
 241 
 242 // Invalidate all the caches (happens to be the same as pd_init).
 243 void ThreadLocalStorage::pd_invalidate_all() { pd_init(); }
 244 
 245 #undef NO_CACHED_THREAD
 246 
 247 // END Thread Local Storage
 248 
 249 static inline size_t adjust_stack_size(address base, size_t size) {
 250   if ((ssize_t)size < 0) {
 251     // 4759953: Compensate for ridiculous stack size.
 252     size = max_intx;
 253   }
 254   if (size > (size_t)base) {
 255     // 4812466: Make sure size doesn't allow the stack to wrap the address space.
 256     size = (size_t)base;
 257   }
 258   return size;
 259 }
 260 
 261 static inline stack_t get_stack_info() {
 262   stack_t st;
 263   int retval = thr_stksegment(&st);
 264   st.ss_size = adjust_stack_size((address)st.ss_sp, st.ss_size);
 265   assert(retval == 0, "incorrect return value from thr_stksegment");
 266   assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
 267   assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
 268   return st;
 269 }
 270 
 271 address os::current_stack_base() {
 272   int r = thr_main() ;
 273   guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
 274   bool is_primordial_thread = r;
 275 
 276   // Workaround 4352906, avoid calls to thr_stksegment by
 277   // thr_main after the first one (it looks like we trash
 278   // some data, causing the value for ss_sp to be incorrect).
 279   if (!is_primordial_thread || os::Solaris::_main_stack_base == NULL) {
 280     stack_t st = get_stack_info();
 281     if (is_primordial_thread) {
 282       // cache initial value of stack base
 283       os::Solaris::_main_stack_base = (address)st.ss_sp;
 284     }
 285     return (address)st.ss_sp;
 286   } else {
 287     guarantee(os::Solaris::_main_stack_base != NULL, "Attempt to use null cached stack base");
 288     return os::Solaris::_main_stack_base;
 289   }
 290 }
 291 
 292 size_t os::current_stack_size() {
 293   size_t size;
 294 
 295   int r = thr_main() ;
 296   guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
 297   if(!r) {
 298     size = get_stack_info().ss_size;
 299   } else {
 300     struct rlimit limits;
 301     getrlimit(RLIMIT_STACK, &limits);
 302     size = adjust_stack_size(os::Solaris::_main_stack_base, (size_t)limits.rlim_cur);
 303   }
 304   // base may not be page aligned
 305   address base = current_stack_base();
 306   address bottom = (address)align_size_up((intptr_t)(base - size), os::vm_page_size());;
 307   return (size_t)(base - bottom);
 308 }
 309 
 310 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
 311   return localtime_r(clock, res);
 312 }
 313 
 314 // interruptible infrastructure
 315 
 316 // setup_interruptible saves the thread state before going into an
 317 // interruptible system call.
 318 // The saved state is used to restore the thread to
 319 // its former state whether or not an interrupt is received.
 320 // Used by classloader os::read
 321 // os::restartable_read calls skip this layer and stay in _thread_in_native
 322 
 323 void os::Solaris::setup_interruptible(JavaThread* thread) {
 324 
 325   JavaThreadState thread_state = thread->thread_state();
 326 
 327   assert(thread_state != _thread_blocked, "Coming from the wrong thread");
 328   assert(thread_state != _thread_in_native, "Native threads skip setup_interruptible");
 329   OSThread* osthread = thread->osthread();
 330   osthread->set_saved_interrupt_thread_state(thread_state);
 331   thread->frame_anchor()->make_walkable(thread);
 332   ThreadStateTransition::transition(thread, thread_state, _thread_blocked);
 333 }
 334 
 335 // Version of setup_interruptible() for threads that are already in
 336 // _thread_blocked. Used by os_sleep().
 337 void os::Solaris::setup_interruptible_already_blocked(JavaThread* thread) {
 338   thread->frame_anchor()->make_walkable(thread);
 339 }
 340 
 341 JavaThread* os::Solaris::setup_interruptible() {
 342   JavaThread* thread = (JavaThread*)ThreadLocalStorage::thread();
 343   setup_interruptible(thread);
 344   return thread;
 345 }
 346 
 347 void os::Solaris::try_enable_extended_io() {
 348   typedef int (*enable_extended_FILE_stdio_t)(int, int);
 349 
 350   if (!UseExtendedFileIO) {
 351     return;
 352   }
 353 
 354   enable_extended_FILE_stdio_t enabler =
 355     (enable_extended_FILE_stdio_t) dlsym(RTLD_DEFAULT,
 356                                          "enable_extended_FILE_stdio");
 357   if (enabler) {
 358     enabler(-1, -1);
 359   }
 360 }
 361 
 362 
 363 #ifdef ASSERT
 364 
 365 JavaThread* os::Solaris::setup_interruptible_native() {
 366   JavaThread* thread = (JavaThread*)ThreadLocalStorage::thread();
 367   JavaThreadState thread_state = thread->thread_state();
 368   assert(thread_state == _thread_in_native, "Assumed thread_in_native");
 369   return thread;
 370 }
 371 
 372 void os::Solaris::cleanup_interruptible_native(JavaThread* thread) {
 373   JavaThreadState thread_state = thread->thread_state();
 374   assert(thread_state == _thread_in_native, "Assumed thread_in_native");
 375 }
 376 #endif
 377 
 378 // cleanup_interruptible reverses the effects of setup_interruptible
 379 // setup_interruptible_already_blocked() does not need any cleanup.
 380 
 381 void os::Solaris::cleanup_interruptible(JavaThread* thread) {
 382   OSThread* osthread = thread->osthread();
 383 
 384   ThreadStateTransition::transition(thread, _thread_blocked, osthread->saved_interrupt_thread_state());
 385 }
 386 
 387 // I/O interruption related counters called in _INTERRUPTIBLE
 388 
 389 void os::Solaris::bump_interrupted_before_count() {
 390   RuntimeService::record_interrupted_before_count();
 391 }
 392 
 393 void os::Solaris::bump_interrupted_during_count() {
 394   RuntimeService::record_interrupted_during_count();
 395 }
 396 
 397 static int _processors_online = 0;
 398 
 399          jint os::Solaris::_os_thread_limit = 0;
 400 volatile jint os::Solaris::_os_thread_count = 0;
 401 
 402 julong os::available_memory() {
 403   return Solaris::available_memory();
 404 }
 405 
 406 julong os::Solaris::available_memory() {
 407   return (julong)sysconf(_SC_AVPHYS_PAGES) * os::vm_page_size();
 408 }
 409 
 410 julong os::Solaris::_physical_memory = 0;
 411 
 412 julong os::physical_memory() {
 413    return Solaris::physical_memory();
 414 }
 415 
 416 static hrtime_t first_hrtime = 0;
 417 static const hrtime_t hrtime_hz = 1000*1000*1000;
 418 const int LOCK_BUSY = 1;
 419 const int LOCK_FREE = 0;
 420 const int LOCK_INVALID = -1;
 421 static volatile hrtime_t max_hrtime = 0;
 422 static volatile int max_hrtime_lock = LOCK_FREE;     // Update counter with LSB as lock-in-progress
 423 
 424 
 425 void os::Solaris::initialize_system_info() {
 426   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 427   _processors_online = sysconf (_SC_NPROCESSORS_ONLN);
 428   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 429 }
 430 
 431 int os::active_processor_count() {
 432   int online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
 433   pid_t pid = getpid();
 434   psetid_t pset = PS_NONE;
 435   // Are we running in a processor set or is there any processor set around?
 436   if (pset_bind(PS_QUERY, P_PID, pid, &pset) == 0) {
 437     uint_t pset_cpus;
 438     // Query the number of cpus available to us.
 439     if (pset_info(pset, NULL, &pset_cpus, NULL) == 0) {
 440       assert(pset_cpus > 0 && pset_cpus <= online_cpus, "sanity check");
 441       _processors_online = pset_cpus;
 442       return pset_cpus;
 443     }
 444   }
 445   // Otherwise return number of online cpus
 446   return online_cpus;
 447 }
 448 
 449 static bool find_processors_in_pset(psetid_t        pset,
 450                                     processorid_t** id_array,
 451                                     uint_t*         id_length) {
 452   bool result = false;
 453   // Find the number of processors in the processor set.
 454   if (pset_info(pset, NULL, id_length, NULL) == 0) {
 455     // Make up an array to hold their ids.
 456     *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
 457     // Fill in the array with their processor ids.
 458     if (pset_info(pset, NULL, id_length, *id_array) == 0) {
 459       result = true;
 460     }
 461   }
 462   return result;
 463 }
 464 
 465 // Callers of find_processors_online() must tolerate imprecise results --
 466 // the system configuration can change asynchronously because of DR
 467 // or explicit psradm operations.
 468 //
 469 // We also need to take care that the loop (below) terminates as the
 470 // number of processors online can change between the _SC_NPROCESSORS_ONLN
 471 // request and the loop that builds the list of processor ids.   Unfortunately
 472 // there's no reliable way to determine the maximum valid processor id,
 473 // so we use a manifest constant, MAX_PROCESSOR_ID, instead.  See p_online
 474 // man pages, which claim the processor id set is "sparse, but
 475 // not too sparse".  MAX_PROCESSOR_ID is used to ensure that we eventually
 476 // exit the loop.
 477 //
 478 // In the future we'll be able to use sysconf(_SC_CPUID_MAX), but that's
 479 // not available on S8.0.
 480 
 481 static bool find_processors_online(processorid_t** id_array,
 482                                    uint*           id_length) {
 483   const processorid_t MAX_PROCESSOR_ID = 100000 ;
 484   // Find the number of processors online.
 485   *id_length = sysconf(_SC_NPROCESSORS_ONLN);
 486   // Make up an array to hold their ids.
 487   *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
 488   // Processors need not be numbered consecutively.
 489   long found = 0;
 490   processorid_t next = 0;
 491   while (found < *id_length && next < MAX_PROCESSOR_ID) {
 492     processor_info_t info;
 493     if (processor_info(next, &info) == 0) {
 494       // NB, PI_NOINTR processors are effectively online ...
 495       if (info.pi_state == P_ONLINE || info.pi_state == P_NOINTR) {
 496         (*id_array)[found] = next;
 497         found += 1;
 498       }
 499     }
 500     next += 1;
 501   }
 502   if (found < *id_length) {
 503       // The loop above didn't identify the expected number of processors.
 504       // We could always retry the operation, calling sysconf(_SC_NPROCESSORS_ONLN)
 505       // and re-running the loop, above, but there's no guarantee of progress
 506       // if the system configuration is in flux.  Instead, we just return what
 507       // we've got.  Note that in the worst case find_processors_online() could
 508       // return an empty set.  (As a fall-back in the case of the empty set we
 509       // could just return the ID of the current processor).
 510       *id_length = found ;
 511   }
 512 
 513   return true;
 514 }
 515 
 516 static bool assign_distribution(processorid_t* id_array,
 517                                 uint           id_length,
 518                                 uint*          distribution,
 519                                 uint           distribution_length) {
 520   // We assume we can assign processorid_t's to uint's.
 521   assert(sizeof(processorid_t) == sizeof(uint),
 522          "can't convert processorid_t to uint");
 523   // Quick check to see if we won't succeed.
 524   if (id_length < distribution_length) {
 525     return false;
 526   }
 527   // Assign processor ids to the distribution.
 528   // Try to shuffle processors to distribute work across boards,
 529   // assuming 4 processors per board.
 530   const uint processors_per_board = ProcessDistributionStride;
 531   // Find the maximum processor id.
 532   processorid_t max_id = 0;
 533   for (uint m = 0; m < id_length; m += 1) {
 534     max_id = MAX2(max_id, id_array[m]);
 535   }
 536   // The next id, to limit loops.
 537   const processorid_t limit_id = max_id + 1;
 538   // Make up markers for available processors.
 539   bool* available_id = NEW_C_HEAP_ARRAY(bool, limit_id, mtInternal);
 540   for (uint c = 0; c < limit_id; c += 1) {
 541     available_id[c] = false;
 542   }
 543   for (uint a = 0; a < id_length; a += 1) {
 544     available_id[id_array[a]] = true;
 545   }
 546   // Step by "boards", then by "slot", copying to "assigned".
 547   // NEEDS_CLEANUP: The assignment of processors should be stateful,
 548   //                remembering which processors have been assigned by
 549   //                previous calls, etc., so as to distribute several
 550   //                independent calls of this method.  What we'd like is
 551   //                It would be nice to have an API that let us ask
 552   //                how many processes are bound to a processor,
 553   //                but we don't have that, either.
 554   //                In the short term, "board" is static so that
 555   //                subsequent distributions don't all start at board 0.
 556   static uint board = 0;
 557   uint assigned = 0;
 558   // Until we've found enough processors ....
 559   while (assigned < distribution_length) {
 560     // ... find the next available processor in the board.
 561     for (uint slot = 0; slot < processors_per_board; slot += 1) {
 562       uint try_id = board * processors_per_board + slot;
 563       if ((try_id < limit_id) && (available_id[try_id] == true)) {
 564         distribution[assigned] = try_id;
 565         available_id[try_id] = false;
 566         assigned += 1;
 567         break;
 568       }
 569     }
 570     board += 1;
 571     if (board * processors_per_board + 0 >= limit_id) {
 572       board = 0;
 573     }
 574   }
 575   if (available_id != NULL) {
 576     FREE_C_HEAP_ARRAY(bool, available_id, mtInternal);
 577   }
 578   return true;
 579 }
 580 
 581 void os::set_native_thread_name(const char *name) {
 582   // Not yet implemented.
 583   return;
 584 }
 585 
 586 bool os::distribute_processes(uint length, uint* distribution) {
 587   bool result = false;
 588   // Find the processor id's of all the available CPUs.
 589   processorid_t* id_array  = NULL;
 590   uint           id_length = 0;
 591   // There are some races between querying information and using it,
 592   // since processor sets can change dynamically.
 593   psetid_t pset = PS_NONE;
 594   // Are we running in a processor set?
 595   if ((pset_bind(PS_QUERY, P_PID, P_MYID, &pset) == 0) && pset != PS_NONE) {
 596     result = find_processors_in_pset(pset, &id_array, &id_length);
 597   } else {
 598     result = find_processors_online(&id_array, &id_length);
 599   }
 600   if (result == true) {
 601     if (id_length >= length) {
 602       result = assign_distribution(id_array, id_length, distribution, length);
 603     } else {
 604       result = false;
 605     }
 606   }
 607   if (id_array != NULL) {
 608     FREE_C_HEAP_ARRAY(processorid_t, id_array, mtInternal);
 609   }
 610   return result;
 611 }
 612 
 613 bool os::bind_to_processor(uint processor_id) {
 614   // We assume that a processorid_t can be stored in a uint.
 615   assert(sizeof(uint) == sizeof(processorid_t),
 616          "can't convert uint to processorid_t");
 617   int bind_result =
 618     processor_bind(P_LWPID,                       // bind LWP.
 619                    P_MYID,                        // bind current LWP.
 620                    (processorid_t) processor_id,  // id.
 621                    NULL);                         // don't return old binding.
 622   return (bind_result == 0);
 623 }
 624 
 625 bool os::getenv(const char* name, char* buffer, int len) {
 626   char* val = ::getenv( name );
 627   if ( val == NULL
 628   ||   strlen(val) + 1  >  len ) {
 629     if (len > 0)  buffer[0] = 0; // return a null string
 630     return false;
 631   }
 632   strcpy( buffer, val );
 633   return true;
 634 }
 635 
 636 
 637 // Return true if user is running as root.
 638 
 639 bool os::have_special_privileges() {
 640   static bool init = false;
 641   static bool privileges = false;
 642   if (!init) {
 643     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 644     init = true;
 645   }
 646   return privileges;
 647 }
 648 
 649 
 650 void os::init_system_properties_values() {
 651   char arch[12];
 652   sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 653 
 654   // The next steps are taken in the product version:
 655   //
 656   // Obtain the JAVA_HOME value from the location of libjvm.so.
 657   // This library should be located at:
 658   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 659   //
 660   // If "/jre/lib/" appears at the right place in the path, then we
 661   // assume libjvm.so is installed in a JDK and we use this path.
 662   //
 663   // Otherwise exit with message: "Could not create the Java virtual machine."
 664   //
 665   // The following extra steps are taken in the debugging version:
 666   //
 667   // If "/jre/lib/" does NOT appear at the right place in the path
 668   // instead of exit check for $JAVA_HOME environment variable.
 669   //
 670   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 671   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 672   // it looks like libjvm.so is installed there
 673   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 674   //
 675   // Otherwise exit.
 676   //
 677   // Important note: if the location of libjvm.so changes this
 678   // code needs to be changed accordingly.
 679 
 680   // The next few definitions allow the code to be verbatim:
 681 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
 682 #define free(p) FREE_C_HEAP_ARRAY(char, p, mtInternal)
 683 #define getenv(n) ::getenv(n)
 684 
 685 #define EXTENSIONS_DIR  "/lib/ext"
 686 #define ENDORSED_DIR    "/lib/endorsed"
 687 #define COMMON_DIR      "/usr/jdk/packages"
 688 
 689   {
 690     /* sysclasspath, java_home, dll_dir */
 691     {
 692         char *home_path;
 693         char *dll_path;
 694         char *pslash;
 695         char buf[MAXPATHLEN];
 696         os::jvm_path(buf, sizeof(buf));
 697 
 698         // Found the full path to libjvm.so.
 699         // Now cut the path to <java_home>/jre if we can.
 700         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 701         pslash = strrchr(buf, '/');
 702         if (pslash != NULL)
 703             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 704         dll_path = malloc(strlen(buf) + 1);
 705         if (dll_path == NULL)
 706             return;
 707         strcpy(dll_path, buf);
 708         Arguments::set_dll_dir(dll_path);
 709 
 710         if (pslash != NULL) {
 711             pslash = strrchr(buf, '/');
 712             if (pslash != NULL) {
 713                 *pslash = '\0';       /* get rid of /<arch> */
 714                 pslash = strrchr(buf, '/');
 715                 if (pslash != NULL)
 716                     *pslash = '\0';   /* get rid of /lib */
 717             }
 718         }
 719 
 720         home_path = malloc(strlen(buf) + 1);
 721         if (home_path == NULL)
 722             return;
 723         strcpy(home_path, buf);
 724         Arguments::set_java_home(home_path);
 725 
 726         if (!set_boot_path('/', ':'))
 727             return;
 728     }
 729 
 730     /*
 731      * Where to look for native libraries
 732      */
 733     {
 734       // Use dlinfo() to determine the correct java.library.path.
 735       //
 736       // If we're launched by the Java launcher, and the user
 737       // does not set java.library.path explicitly on the commandline,
 738       // the Java launcher sets LD_LIBRARY_PATH for us and unsets
 739       // LD_LIBRARY_PATH_32 and LD_LIBRARY_PATH_64.  In this case
 740       // dlinfo returns LD_LIBRARY_PATH + crle settings (including
 741       // /usr/lib), which is exactly what we want.
 742       //
 743       // If the user does set java.library.path, it completely
 744       // overwrites this setting, and always has.
 745       //
 746       // If we're not launched by the Java launcher, we may
 747       // get here with any/all of the LD_LIBRARY_PATH[_32|64]
 748       // settings.  Again, dlinfo does exactly what we want.
 749 
 750       Dl_serinfo     _info, *info = &_info;
 751       Dl_serpath     *path;
 752       char*          library_path;
 753       char           *common_path;
 754       int            i;
 755 
 756       // determine search path count and required buffer size
 757       if (dlinfo(RTLD_SELF, RTLD_DI_SERINFOSIZE, (void *)info) == -1) {
 758         vm_exit_during_initialization("dlinfo SERINFOSIZE request", dlerror());
 759       }
 760 
 761       // allocate new buffer and initialize
 762       info = (Dl_serinfo*)malloc(_info.dls_size);
 763       if (info == NULL) {
 764         vm_exit_out_of_memory(_info.dls_size, OOM_MALLOC_ERROR,
 765                               "init_system_properties_values info");
 766       }
 767       info->dls_size = _info.dls_size;
 768       info->dls_cnt = _info.dls_cnt;
 769 
 770       // obtain search path information
 771       if (dlinfo(RTLD_SELF, RTLD_DI_SERINFO, (void *)info) == -1) {
 772         free(info);
 773         vm_exit_during_initialization("dlinfo SERINFO request", dlerror());
 774       }
 775 
 776       path = &info->dls_serpath[0];
 777 
 778       // Note: Due to a legacy implementation, most of the library path
 779       // is set in the launcher.  This was to accomodate linking restrictions
 780       // on legacy Solaris implementations (which are no longer supported).
 781       // Eventually, all the library path setting will be done here.
 782       //
 783       // However, to prevent the proliferation of improperly built native
 784       // libraries, the new path component /usr/jdk/packages is added here.
 785 
 786       // Determine the actual CPU architecture.
 787       char cpu_arch[12];
 788       sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
 789 #ifdef _LP64
 790       // If we are a 64-bit vm, perform the following translations:
 791       //   sparc   -> sparcv9
 792       //   i386    -> amd64
 793       if (strcmp(cpu_arch, "sparc") == 0)
 794         strcat(cpu_arch, "v9");
 795       else if (strcmp(cpu_arch, "i386") == 0)
 796         strcpy(cpu_arch, "amd64");
 797 #endif
 798 
 799       // Construct the invariant part of ld_library_path. Note that the
 800       // space for the colon and the trailing null are provided by the
 801       // nulls included by the sizeof operator.
 802       size_t bufsize = sizeof(COMMON_DIR) + sizeof("/lib/") + strlen(cpu_arch);
 803       common_path = malloc(bufsize);
 804       if (common_path == NULL) {
 805         free(info);
 806         vm_exit_out_of_memory(bufsize, OOM_MALLOC_ERROR,
 807                               "init_system_properties_values common_path");
 808       }
 809       sprintf(common_path, COMMON_DIR "/lib/%s", cpu_arch);
 810 
 811       // struct size is more than sufficient for the path components obtained
 812       // through the dlinfo() call, so only add additional space for the path
 813       // components explicitly added here.
 814       bufsize = info->dls_size + strlen(common_path);
 815       library_path = malloc(bufsize);
 816       if (library_path == NULL) {
 817         free(info);
 818         free(common_path);
 819         vm_exit_out_of_memory(bufsize, OOM_MALLOC_ERROR,
 820                               "init_system_properties_values library_path");
 821       }
 822       library_path[0] = '\0';
 823 
 824       // Construct the desired Java library path from the linker's library
 825       // search path.
 826       //
 827       // For compatibility, it is optimal that we insert the additional path
 828       // components specific to the Java VM after those components specified
 829       // in LD_LIBRARY_PATH (if any) but before those added by the ld.so
 830       // infrastructure.
 831       if (info->dls_cnt == 0) { // Not sure this can happen, but allow for it
 832         strcpy(library_path, common_path);
 833       } else {
 834         int inserted = 0;
 835         for (i = 0; i < info->dls_cnt; i++, path++) {
 836           uint_t flags = path->dls_flags & LA_SER_MASK;
 837           if (((flags & LA_SER_LIBPATH) == 0) && !inserted) {
 838             strcat(library_path, common_path);
 839             strcat(library_path, os::path_separator());
 840             inserted = 1;
 841           }
 842           strcat(library_path, path->dls_name);
 843           strcat(library_path, os::path_separator());
 844         }
 845         // eliminate trailing path separator
 846         library_path[strlen(library_path)-1] = '\0';
 847       }
 848 
 849       // happens before argument parsing - can't use a trace flag
 850       // tty->print_raw("init_system_properties_values: native lib path: ");
 851       // tty->print_raw_cr(library_path);
 852 
 853       // callee copies into its own buffer
 854       Arguments::set_library_path(library_path);
 855 
 856       free(common_path);
 857       free(library_path);
 858       free(info);
 859     }
 860 
 861     /*
 862      * Extensions directories.
 863      *
 864      * Note that the space for the colon and the trailing null are provided
 865      * by the nulls included by the sizeof operator (so actually one byte more
 866      * than necessary is allocated).
 867      */
 868     {
 869         char *buf = (char *) malloc(strlen(Arguments::get_java_home()) +
 870             sizeof(EXTENSIONS_DIR) + sizeof(COMMON_DIR) +
 871             sizeof(EXTENSIONS_DIR));
 872         sprintf(buf, "%s" EXTENSIONS_DIR ":" COMMON_DIR EXTENSIONS_DIR,
 873             Arguments::get_java_home());
 874         Arguments::set_ext_dirs(buf);
 875     }
 876 
 877     /* Endorsed standards default directory. */
 878     {
 879         char * buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 880         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 881         Arguments::set_endorsed_dirs(buf);
 882     }
 883   }
 884 
 885 #undef malloc
 886 #undef free
 887 #undef getenv
 888 #undef EXTENSIONS_DIR
 889 #undef ENDORSED_DIR
 890 #undef COMMON_DIR
 891 
 892 }
 893 
 894 void os::breakpoint() {
 895   BREAKPOINT;
 896 }
 897 
 898 bool os::obsolete_option(const JavaVMOption *option)
 899 {
 900   if (!strncmp(option->optionString, "-Xt", 3)) {
 901     return true;
 902   } else if (!strncmp(option->optionString, "-Xtm", 4)) {
 903     return true;
 904   } else if (!strncmp(option->optionString, "-Xverifyheap", 12)) {
 905     return true;
 906   } else if (!strncmp(option->optionString, "-Xmaxjitcodesize", 16)) {
 907     return true;
 908   }
 909   return false;
 910 }
 911 
 912 bool os::Solaris::valid_stack_address(Thread* thread, address sp) {
 913   address  stackStart  = (address)thread->stack_base();
 914   address  stackEnd    = (address)(stackStart - (address)thread->stack_size());
 915   if (sp < stackStart && sp >= stackEnd ) return true;
 916   return false;
 917 }
 918 
 919 extern "C" void breakpoint() {
 920   // use debugger to set breakpoint here
 921 }
 922 
 923 static thread_t main_thread;
 924 
 925 // Thread start routine for all new Java threads
 926 extern "C" void* java_start(void* thread_addr) {
 927   // Try to randomize the cache line index of hot stack frames.
 928   // This helps when threads of the same stack traces evict each other's
 929   // cache lines. The threads can be either from the same JVM instance, or
 930   // from different JVM instances. The benefit is especially true for
 931   // processors with hyperthreading technology.
 932   static int counter = 0;
 933   int pid = os::current_process_id();
 934   alloca(((pid ^ counter++) & 7) * 128);
 935 
 936   int prio;
 937   Thread* thread = (Thread*)thread_addr;
 938   OSThread* osthr = thread->osthread();
 939 
 940   osthr->set_lwp_id( _lwp_self() );  // Store lwp in case we are bound
 941   thread->_schedctl = (void *) schedctl_init () ;
 942 
 943   if (UseNUMA) {
 944     int lgrp_id = os::numa_get_group_id();
 945     if (lgrp_id != -1) {
 946       thread->set_lgrp_id(lgrp_id);
 947     }
 948   }
 949 
 950   // If the creator called set priority before we started,
 951   // we need to call set_native_priority now that we have an lwp.
 952   // We used to get the priority from thr_getprio (we called
 953   // thr_setprio way back in create_thread) and pass it to
 954   // set_native_priority, but Solaris scales the priority
 955   // in java_to_os_priority, so when we read it back here,
 956   // we pass trash to set_native_priority instead of what's
 957   // in java_to_os_priority. So we save the native priority
 958   // in the osThread and recall it here.
 959 
 960   if ( osthr->thread_id() != -1 ) {
 961     if ( UseThreadPriorities ) {
 962       int prio = osthr->native_priority();
 963       if (ThreadPriorityVerbose) {
 964         tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is "
 965                       INTPTR_FORMAT ", setting priority: %d\n",
 966                       osthr->thread_id(), osthr->lwp_id(), prio);
 967       }
 968       os::set_native_priority(thread, prio);
 969     }
 970   } else if (ThreadPriorityVerbose) {
 971     warning("Can't set priority in _start routine, thread id hasn't been set\n");
 972   }
 973 
 974   assert(osthr->get_state() == RUNNABLE, "invalid os thread state");
 975 
 976   // initialize signal mask for this thread
 977   os::Solaris::hotspot_sigmask(thread);
 978 
 979   thread->run();
 980 
 981   // One less thread is executing
 982   // When the VMThread gets here, the main thread may have already exited
 983   // which frees the CodeHeap containing the Atomic::dec code
 984   if (thread != VMThread::vm_thread() && VMThread::vm_thread() != NULL) {
 985     Atomic::dec(&os::Solaris::_os_thread_count);
 986   }
 987 
 988   if (UseDetachedThreads) {
 989     thr_exit(NULL);
 990     ShouldNotReachHere();
 991   }
 992   return NULL;
 993 }
 994 
 995 static OSThread* create_os_thread(Thread* thread, thread_t thread_id) {
 996   // Allocate the OSThread object
 997   OSThread* osthread = new OSThread(NULL, NULL);
 998   if (osthread == NULL) return NULL;
 999 
1000   // Store info on the Solaris thread into the OSThread
1001   osthread->set_thread_id(thread_id);
1002   osthread->set_lwp_id(_lwp_self());
1003   thread->_schedctl = (void *) schedctl_init () ;
1004 
1005   if (UseNUMA) {
1006     int lgrp_id = os::numa_get_group_id();
1007     if (lgrp_id != -1) {
1008       thread->set_lgrp_id(lgrp_id);
1009     }
1010   }
1011 
1012   if ( ThreadPriorityVerbose ) {
1013     tty->print_cr("In create_os_thread, Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT "\n",
1014                   osthread->thread_id(), osthread->lwp_id() );
1015   }
1016 
1017   // Initial thread state is INITIALIZED, not SUSPENDED
1018   osthread->set_state(INITIALIZED);
1019 
1020   return osthread;
1021 }
1022 
1023 void os::Solaris::hotspot_sigmask(Thread* thread) {
1024 
1025   //Save caller's signal mask
1026   sigset_t sigmask;
1027   thr_sigsetmask(SIG_SETMASK, NULL, &sigmask);
1028   OSThread *osthread = thread->osthread();
1029   osthread->set_caller_sigmask(sigmask);
1030 
1031   thr_sigsetmask(SIG_UNBLOCK, os::Solaris::unblocked_signals(), NULL);
1032   if (!ReduceSignalUsage) {
1033     if (thread->is_VM_thread()) {
1034       // Only the VM thread handles BREAK_SIGNAL ...
1035       thr_sigsetmask(SIG_UNBLOCK, vm_signals(), NULL);
1036     } else {
1037       // ... all other threads block BREAK_SIGNAL
1038       assert(!sigismember(vm_signals(), SIGINT), "SIGINT should not be blocked");
1039       thr_sigsetmask(SIG_BLOCK, vm_signals(), NULL);
1040     }
1041   }
1042 }
1043 
1044 bool os::create_attached_thread(JavaThread* thread) {
1045 #ifdef ASSERT
1046   thread->verify_not_published();
1047 #endif
1048   OSThread* osthread = create_os_thread(thread, thr_self());
1049   if (osthread == NULL) {
1050      return false;
1051   }
1052 
1053   // Initial thread state is RUNNABLE
1054   osthread->set_state(RUNNABLE);
1055   thread->set_osthread(osthread);
1056 
1057   // initialize signal mask for this thread
1058   // and save the caller's signal mask
1059   os::Solaris::hotspot_sigmask(thread);
1060 
1061   return true;
1062 }
1063 
1064 bool os::create_main_thread(JavaThread* thread) {
1065 #ifdef ASSERT
1066   thread->verify_not_published();
1067 #endif
1068   if (_starting_thread == NULL) {
1069     _starting_thread = create_os_thread(thread, main_thread);
1070      if (_starting_thread == NULL) {
1071         return false;
1072      }
1073   }
1074 
1075   // The primodial thread is runnable from the start
1076   _starting_thread->set_state(RUNNABLE);
1077 
1078   thread->set_osthread(_starting_thread);
1079 
1080   // initialize signal mask for this thread
1081   // and save the caller's signal mask
1082   os::Solaris::hotspot_sigmask(thread);
1083 
1084   return true;
1085 }
1086 
1087 // _T2_libthread is true if we believe we are running with the newer
1088 // SunSoft lwp/libthread.so (2.8 patch, 2.9 default)
1089 bool os::Solaris::_T2_libthread = false;
1090 
1091 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
1092   // Allocate the OSThread object
1093   OSThread* osthread = new OSThread(NULL, NULL);
1094   if (osthread == NULL) {
1095     return false;
1096   }
1097 
1098   if ( ThreadPriorityVerbose ) {
1099     char *thrtyp;
1100     switch ( thr_type ) {
1101       case vm_thread:
1102         thrtyp = (char *)"vm";
1103         break;
1104       case cgc_thread:
1105         thrtyp = (char *)"cgc";
1106         break;
1107       case pgc_thread:
1108         thrtyp = (char *)"pgc";
1109         break;
1110       case java_thread:
1111         thrtyp = (char *)"java";
1112         break;
1113       case compiler_thread:
1114         thrtyp = (char *)"compiler";
1115         break;
1116       case watcher_thread:
1117         thrtyp = (char *)"watcher";
1118         break;
1119       default:
1120         thrtyp = (char *)"unknown";
1121         break;
1122     }
1123     tty->print_cr("In create_thread, creating a %s thread\n", thrtyp);
1124   }
1125 
1126   // Calculate stack size if it's not specified by caller.
1127   if (stack_size == 0) {
1128     // The default stack size 1M (2M for LP64).
1129     stack_size = (BytesPerWord >> 2) * K * K;
1130 
1131     switch (thr_type) {
1132     case os::java_thread:
1133       // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
1134       if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
1135       break;
1136     case os::compiler_thread:
1137       if (CompilerThreadStackSize > 0) {
1138         stack_size = (size_t)(CompilerThreadStackSize * K);
1139         break;
1140       } // else fall through:
1141         // use VMThreadStackSize if CompilerThreadStackSize is not defined
1142     case os::vm_thread:
1143     case os::pgc_thread:
1144     case os::cgc_thread:
1145     case os::watcher_thread:
1146       if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
1147       break;
1148     }
1149   }
1150   stack_size = MAX2(stack_size, os::Solaris::min_stack_allowed);
1151 
1152   // Initial state is ALLOCATED but not INITIALIZED
1153   osthread->set_state(ALLOCATED);
1154 
1155   if (os::Solaris::_os_thread_count > os::Solaris::_os_thread_limit) {
1156     // We got lots of threads. Check if we still have some address space left.
1157     // Need to be at least 5Mb of unreserved address space. We do check by
1158     // trying to reserve some.
1159     const size_t VirtualMemoryBangSize = 20*K*K;
1160     char* mem = os::reserve_memory(VirtualMemoryBangSize);
1161     if (mem == NULL) {
1162       delete osthread;
1163       return false;
1164     } else {
1165       // Release the memory again
1166       os::release_memory(mem, VirtualMemoryBangSize);
1167     }
1168   }
1169 
1170   // Setup osthread because the child thread may need it.
1171   thread->set_osthread(osthread);
1172 
1173   // Create the Solaris thread
1174   // explicit THR_BOUND for T2_libthread case in case
1175   // that assumption is not accurate, but our alternate signal stack
1176   // handling is based on it which must have bound threads
1177   thread_t tid = 0;
1178   long     flags = (UseDetachedThreads ? THR_DETACHED : 0) | THR_SUSPENDED
1179                    | ((UseBoundThreads || os::Solaris::T2_libthread() ||
1180                        (thr_type == vm_thread) ||
1181                        (thr_type == cgc_thread) ||
1182                        (thr_type == pgc_thread) ||
1183                        (thr_type == compiler_thread && BackgroundCompilation)) ?
1184                       THR_BOUND : 0);
1185   int      status;
1186 
1187   // 4376845 -- libthread/kernel don't provide enough LWPs to utilize all CPUs.
1188   //
1189   // On multiprocessors systems, libthread sometimes under-provisions our
1190   // process with LWPs.  On a 30-way systems, for instance, we could have
1191   // 50 user-level threads in ready state and only 2 or 3 LWPs assigned
1192   // to our process.  This can result in under utilization of PEs.
1193   // I suspect the problem is related to libthread's LWP
1194   // pool management and to the kernel's SIGBLOCKING "last LWP parked"
1195   // upcall policy.
1196   //
1197   // The following code is palliative -- it attempts to ensure that our
1198   // process has sufficient LWPs to take advantage of multiple PEs.
1199   // Proper long-term cures include using user-level threads bound to LWPs
1200   // (THR_BOUND) or using LWP-based synchronization.  Note that there is a
1201   // slight timing window with respect to sampling _os_thread_count, but
1202   // the race is benign.  Also, we should periodically recompute
1203   // _processors_online as the min of SC_NPROCESSORS_ONLN and the
1204   // the number of PEs in our partition.  You might be tempted to use
1205   // THR_NEW_LWP here, but I'd recommend against it as that could
1206   // result in undesirable growth of the libthread's LWP pool.
1207   // The fix below isn't sufficient; for instance, it doesn't take into count
1208   // LWPs parked on IO.  It does, however, help certain CPU-bound benchmarks.
1209   //
1210   // Some pathologies this scheme doesn't handle:
1211   // *  Threads can block, releasing the LWPs.  The LWPs can age out.
1212   //    When a large number of threads become ready again there aren't
1213   //    enough LWPs available to service them.  This can occur when the
1214   //    number of ready threads oscillates.
1215   // *  LWPs/Threads park on IO, thus taking the LWP out of circulation.
1216   //
1217   // Finally, we should call thr_setconcurrency() periodically to refresh
1218   // the LWP pool and thwart the LWP age-out mechanism.
1219   // The "+3" term provides a little slop -- we want to slightly overprovision.
1220 
1221   if (AdjustConcurrency && os::Solaris::_os_thread_count < (_processors_online+3)) {
1222     if (!(flags & THR_BOUND)) {
1223       thr_setconcurrency (os::Solaris::_os_thread_count);       // avoid starvation
1224     }
1225   }
1226   // Although this doesn't hurt, we should warn of undefined behavior
1227   // when using unbound T1 threads with schedctl().  This should never
1228   // happen, as the compiler and VM threads are always created bound
1229   DEBUG_ONLY(
1230       if ((VMThreadHintNoPreempt || CompilerThreadHintNoPreempt) &&
1231           (!os::Solaris::T2_libthread() && (!(flags & THR_BOUND))) &&
1232           ((thr_type == vm_thread) || (thr_type == cgc_thread) ||
1233            (thr_type == pgc_thread) || (thr_type == compiler_thread && BackgroundCompilation))) {
1234          warning("schedctl behavior undefined when Compiler/VM/GC Threads are Unbound");
1235       }
1236   );
1237 
1238 
1239   // Mark that we don't have an lwp or thread id yet.
1240   // In case we attempt to set the priority before the thread starts.
1241   osthread->set_lwp_id(-1);
1242   osthread->set_thread_id(-1);
1243 
1244   status = thr_create(NULL, stack_size, java_start, thread, flags, &tid);
1245   if (status != 0) {
1246     if (PrintMiscellaneous && (Verbose || WizardMode)) {
1247       perror("os::create_thread");
1248     }
1249     thread->set_osthread(NULL);
1250     // Need to clean up stuff we've allocated so far
1251     delete osthread;
1252     return false;
1253   }
1254 
1255   Atomic::inc(&os::Solaris::_os_thread_count);
1256 
1257   // Store info on the Solaris thread into the OSThread
1258   osthread->set_thread_id(tid);
1259 
1260   // Remember that we created this thread so we can set priority on it
1261   osthread->set_vm_created();
1262 
1263   // Set the default thread priority.  If using bound threads, setting
1264   // lwp priority will be delayed until thread start.
1265   set_native_priority(thread,
1266                       DefaultThreadPriority == -1 ?
1267                         java_to_os_priority[NormPriority] :
1268                         DefaultThreadPriority);
1269 
1270   // Initial thread state is INITIALIZED, not SUSPENDED
1271   osthread->set_state(INITIALIZED);
1272 
1273   // The thread is returned suspended (in state INITIALIZED), and is started higher up in the call chain
1274   return true;
1275 }
1276 
1277 /* defined for >= Solaris 10. This allows builds on earlier versions
1278  *  of Solaris to take advantage of the newly reserved Solaris JVM signals
1279  *  With SIGJVM1, SIGJVM2, INTERRUPT_SIGNAL is SIGJVM1, ASYNC_SIGNAL is SIGJVM2
1280  *  and -XX:+UseAltSigs does nothing since these should have no conflict
1281  */
1282 #if !defined(SIGJVM1)
1283 #define SIGJVM1 39
1284 #define SIGJVM2 40
1285 #endif
1286 
1287 debug_only(static bool signal_sets_initialized = false);
1288 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
1289 int os::Solaris::_SIGinterrupt = INTERRUPT_SIGNAL;
1290 int os::Solaris::_SIGasync = ASYNC_SIGNAL;
1291 
1292 bool os::Solaris::is_sig_ignored(int sig) {
1293       struct sigaction oact;
1294       sigaction(sig, (struct sigaction*)NULL, &oact);
1295       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
1296                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
1297       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
1298            return true;
1299       else
1300            return false;
1301 }
1302 
1303 // Note: SIGRTMIN is a macro that calls sysconf() so it will
1304 // dynamically detect SIGRTMIN value for the system at runtime, not buildtime
1305 static bool isJVM1available() {
1306   return SIGJVM1 < SIGRTMIN;
1307 }
1308 
1309 void os::Solaris::signal_sets_init() {
1310   // Should also have an assertion stating we are still single-threaded.
1311   assert(!signal_sets_initialized, "Already initialized");
1312   // Fill in signals that are necessarily unblocked for all threads in
1313   // the VM. Currently, we unblock the following signals:
1314   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
1315   //                         by -Xrs (=ReduceSignalUsage));
1316   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
1317   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
1318   // the dispositions or masks wrt these signals.
1319   // Programs embedding the VM that want to use the above signals for their
1320   // own purposes must, at this time, use the "-Xrs" option to prevent
1321   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
1322   // (See bug 4345157, and other related bugs).
1323   // In reality, though, unblocking these signals is really a nop, since
1324   // these signals are not blocked by default.
1325   sigemptyset(&unblocked_sigs);
1326   sigemptyset(&allowdebug_blocked_sigs);
1327   sigaddset(&unblocked_sigs, SIGILL);
1328   sigaddset(&unblocked_sigs, SIGSEGV);
1329   sigaddset(&unblocked_sigs, SIGBUS);
1330   sigaddset(&unblocked_sigs, SIGFPE);
1331 
1332   if (isJVM1available) {
1333     os::Solaris::set_SIGinterrupt(SIGJVM1);
1334     os::Solaris::set_SIGasync(SIGJVM2);
1335   } else if (UseAltSigs) {
1336     os::Solaris::set_SIGinterrupt(ALT_INTERRUPT_SIGNAL);
1337     os::Solaris::set_SIGasync(ALT_ASYNC_SIGNAL);
1338   } else {
1339     os::Solaris::set_SIGinterrupt(INTERRUPT_SIGNAL);
1340     os::Solaris::set_SIGasync(ASYNC_SIGNAL);
1341   }
1342 
1343   sigaddset(&unblocked_sigs, os::Solaris::SIGinterrupt());
1344   sigaddset(&unblocked_sigs, os::Solaris::SIGasync());
1345 
1346   if (!ReduceSignalUsage) {
1347    if (!os::Solaris::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
1348       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
1349       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
1350    }
1351    if (!os::Solaris::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
1352       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
1353       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
1354    }
1355    if (!os::Solaris::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
1356       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
1357       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
1358    }
1359   }
1360   // Fill in signals that are blocked by all but the VM thread.
1361   sigemptyset(&vm_sigs);
1362   if (!ReduceSignalUsage)
1363     sigaddset(&vm_sigs, BREAK_SIGNAL);
1364   debug_only(signal_sets_initialized = true);
1365 
1366   // For diagnostics only used in run_periodic_checks
1367   sigemptyset(&check_signal_done);
1368 }
1369 
1370 // These are signals that are unblocked while a thread is running Java.
1371 // (For some reason, they get blocked by default.)
1372 sigset_t* os::Solaris::unblocked_signals() {
1373   assert(signal_sets_initialized, "Not initialized");
1374   return &unblocked_sigs;
1375 }
1376 
1377 // These are the signals that are blocked while a (non-VM) thread is
1378 // running Java. Only the VM thread handles these signals.
1379 sigset_t* os::Solaris::vm_signals() {
1380   assert(signal_sets_initialized, "Not initialized");
1381   return &vm_sigs;
1382 }
1383 
1384 // These are signals that are blocked during cond_wait to allow debugger in
1385 sigset_t* os::Solaris::allowdebug_blocked_signals() {
1386   assert(signal_sets_initialized, "Not initialized");
1387   return &allowdebug_blocked_sigs;
1388 }
1389 
1390 
1391 void _handle_uncaught_cxx_exception() {
1392   VMError err("An uncaught C++ exception");
1393   err.report_and_die();
1394 }
1395 
1396 
1397 // First crack at OS-specific initialization, from inside the new thread.
1398 void os::initialize_thread(Thread* thr) {
1399   int r = thr_main() ;
1400   guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
1401   if (r) {
1402     JavaThread* jt = (JavaThread *)thr;
1403     assert(jt != NULL,"Sanity check");
1404     size_t stack_size;
1405     address base = jt->stack_base();
1406     if (Arguments::created_by_java_launcher()) {
1407       // Use 2MB to allow for Solaris 7 64 bit mode.
1408       stack_size = JavaThread::stack_size_at_create() == 0
1409         ? 2048*K : JavaThread::stack_size_at_create();
1410 
1411       // There are rare cases when we may have already used more than
1412       // the basic stack size allotment before this method is invoked.
1413       // Attempt to allow for a normally sized java_stack.
1414       size_t current_stack_offset = (size_t)(base - (address)&stack_size);
1415       stack_size += ReservedSpace::page_align_size_down(current_stack_offset);
1416     } else {
1417       // 6269555: If we were not created by a Java launcher, i.e. if we are
1418       // running embedded in a native application, treat the primordial thread
1419       // as much like a native attached thread as possible.  This means using
1420       // the current stack size from thr_stksegment(), unless it is too large
1421       // to reliably setup guard pages.  A reasonable max size is 8MB.
1422       size_t current_size = current_stack_size();
1423       // This should never happen, but just in case....
1424       if (current_size == 0) current_size = 2 * K * K;
1425       stack_size = current_size > (8 * K * K) ? (8 * K * K) : current_size;
1426     }
1427     address bottom = (address)align_size_up((intptr_t)(base - stack_size), os::vm_page_size());;
1428     stack_size = (size_t)(base - bottom);
1429 
1430     assert(stack_size > 0, "Stack size calculation problem");
1431 
1432     if (stack_size > jt->stack_size()) {
1433       NOT_PRODUCT(
1434         struct rlimit limits;
1435         getrlimit(RLIMIT_STACK, &limits);
1436         size_t size = adjust_stack_size(base, (size_t)limits.rlim_cur);
1437         assert(size >= jt->stack_size(), "Stack size problem in main thread");
1438       )
1439       tty->print_cr(
1440         "Stack size of %d Kb exceeds current limit of %d Kb.\n"
1441         "(Stack sizes are rounded up to a multiple of the system page size.)\n"
1442         "See limit(1) to increase the stack size limit.",
1443         stack_size / K, jt->stack_size() / K);
1444       vm_exit(1);
1445     }
1446     assert(jt->stack_size() >= stack_size,
1447           "Attempt to map more stack than was allocated");
1448     jt->set_stack_size(stack_size);
1449   }
1450 
1451    // 5/22/01: Right now alternate signal stacks do not handle
1452    // throwing stack overflow exceptions, see bug 4463178
1453    // Until a fix is found for this, T2 will NOT imply alternate signal
1454    // stacks.
1455    // If using T2 libthread threads, install an alternate signal stack.
1456    // Because alternate stacks associate with LWPs on Solaris,
1457    // see sigaltstack(2), if using UNBOUND threads, or if UseBoundThreads
1458    // we prefer to explicitly stack bang.
1459    // If not using T2 libthread, but using UseBoundThreads any threads
1460    // (primordial thread, jni_attachCurrentThread) we do not create,
1461    // probably are not bound, therefore they can not have an alternate
1462    // signal stack. Since our stack banging code is generated and
1463    // is shared across threads, all threads must be bound to allow
1464    // using alternate signal stacks.  The alternative is to interpose
1465    // on _lwp_create to associate an alt sig stack with each LWP,
1466    // and this could be a problem when the JVM is embedded.
1467    // We would prefer to use alternate signal stacks with T2
1468    // Since there is currently no accurate way to detect T2
1469    // we do not. Assuming T2 when running T1 causes sig 11s or assertions
1470    // on installing alternate signal stacks
1471 
1472 
1473    // 05/09/03: removed alternate signal stack support for Solaris
1474    // The alternate signal stack mechanism is no longer needed to
1475    // handle stack overflow. This is now handled by allocating
1476    // guard pages (red zone) and stackbanging.
1477    // Initially the alternate signal stack mechanism was removed because
1478    // it did not work with T1 llibthread. Alternate
1479    // signal stacks MUST have all threads bound to lwps. Applications
1480    // can create their own threads and attach them without their being
1481    // bound under T1. This is frequently the case for the primordial thread.
1482    // If we were ever to reenable this mechanism we would need to
1483    // use the dynamic check for T2 libthread.
1484 
1485   os::Solaris::init_thread_fpu_state();
1486   std::set_terminate(_handle_uncaught_cxx_exception);
1487 }
1488 
1489 
1490 
1491 // Free Solaris resources related to the OSThread
1492 void os::free_thread(OSThread* osthread) {
1493   assert(osthread != NULL, "os::free_thread but osthread not set");
1494 
1495 
1496   // We are told to free resources of the argument thread,
1497   // but we can only really operate on the current thread.
1498   // The main thread must take the VMThread down synchronously
1499   // before the main thread exits and frees up CodeHeap
1500   guarantee((Thread::current()->osthread() == osthread
1501      || (osthread == VMThread::vm_thread()->osthread())), "os::free_thread but not current thread");
1502   if (Thread::current()->osthread() == osthread) {
1503     // Restore caller's signal mask
1504     sigset_t sigmask = osthread->caller_sigmask();
1505     thr_sigsetmask(SIG_SETMASK, &sigmask, NULL);
1506   }
1507   delete osthread;
1508 }
1509 
1510 void os::pd_start_thread(Thread* thread) {
1511   int status = thr_continue(thread->osthread()->thread_id());
1512   assert_status(status == 0, status, "thr_continue failed");
1513 }
1514 
1515 
1516 intx os::current_thread_id() {
1517   return (intx)thr_self();
1518 }
1519 
1520 static pid_t _initial_pid = 0;
1521 
1522 int os::current_process_id() {
1523   return (int)(_initial_pid ? _initial_pid : getpid());
1524 }
1525 
1526 int os::allocate_thread_local_storage() {
1527   // %%%       in Win32 this allocates a memory segment pointed to by a
1528   //           register.  Dan Stein can implement a similar feature in
1529   //           Solaris.  Alternatively, the VM can do the same thing
1530   //           explicitly: malloc some storage and keep the pointer in a
1531   //           register (which is part of the thread's context) (or keep it
1532   //           in TLS).
1533   // %%%       In current versions of Solaris, thr_self and TSD can
1534   //           be accessed via short sequences of displaced indirections.
1535   //           The value of thr_self is available as %g7(36).
1536   //           The value of thr_getspecific(k) is stored in %g7(12)(4)(k*4-4),
1537   //           assuming that the current thread already has a value bound to k.
1538   //           It may be worth experimenting with such access patterns,
1539   //           and later having the parameters formally exported from a Solaris
1540   //           interface.  I think, however, that it will be faster to
1541   //           maintain the invariant that %g2 always contains the
1542   //           JavaThread in Java code, and have stubs simply
1543   //           treat %g2 as a caller-save register, preserving it in a %lN.
1544   thread_key_t tk;
1545   if (thr_keycreate( &tk, NULL ) )
1546     fatal(err_msg("os::allocate_thread_local_storage: thr_keycreate failed "
1547                   "(%s)", strerror(errno)));
1548   return int(tk);
1549 }
1550 
1551 void os::free_thread_local_storage(int index) {
1552   // %%% don't think we need anything here
1553   // if ( pthread_key_delete((pthread_key_t) tk) )
1554   //   fatal("os::free_thread_local_storage: pthread_key_delete failed");
1555 }
1556 
1557 #define SMALLINT 32   // libthread allocate for tsd_common is a version specific
1558                       // small number - point is NO swap space available
1559 void os::thread_local_storage_at_put(int index, void* value) {
1560   // %%% this is used only in threadLocalStorage.cpp
1561   if (thr_setspecific((thread_key_t)index, value)) {
1562     if (errno == ENOMEM) {
1563        vm_exit_out_of_memory(SMALLINT, OOM_MALLOC_ERROR,
1564                              "thr_setspecific: out of swap space");
1565     } else {
1566       fatal(err_msg("os::thread_local_storage_at_put: thr_setspecific failed "
1567                     "(%s)", strerror(errno)));
1568     }
1569   } else {
1570       ThreadLocalStorage::set_thread_in_slot ((Thread *) value) ;
1571   }
1572 }
1573 
1574 // This function could be called before TLS is initialized, for example, when
1575 // VM receives an async signal or when VM causes a fatal error during
1576 // initialization. Return NULL if thr_getspecific() fails.
1577 void* os::thread_local_storage_at(int index) {
1578   // %%% this is used only in threadLocalStorage.cpp
1579   void* r = NULL;
1580   return thr_getspecific((thread_key_t)index, &r) != 0 ? NULL : r;
1581 }
1582 
1583 
1584 // gethrtime can move backwards if read from one cpu and then a different cpu
1585 // getTimeNanos is guaranteed to not move backward on Solaris
1586 // local spinloop created as faster for a CAS on an int than
1587 // a CAS on a 64bit jlong. Also Atomic::cmpxchg for jlong is not
1588 // supported on sparc v8 or pre supports_cx8 intel boxes.
1589 // oldgetTimeNanos for systems which do not support CAS on 64bit jlong
1590 // i.e. sparc v8 and pre supports_cx8 (i486) intel boxes
1591 inline hrtime_t oldgetTimeNanos() {
1592   int gotlock = LOCK_INVALID;
1593   hrtime_t newtime = gethrtime();
1594 
1595   for (;;) {
1596 // grab lock for max_hrtime
1597     int curlock = max_hrtime_lock;
1598     if (curlock & LOCK_BUSY)  continue;
1599     if (gotlock = Atomic::cmpxchg(LOCK_BUSY, &max_hrtime_lock, LOCK_FREE) != LOCK_FREE) continue;
1600     if (newtime > max_hrtime) {
1601       max_hrtime = newtime;
1602     } else {
1603       newtime = max_hrtime;
1604     }
1605     // release lock
1606     max_hrtime_lock = LOCK_FREE;
1607     return newtime;
1608   }
1609 }
1610 // gethrtime can move backwards if read from one cpu and then a different cpu
1611 // getTimeNanos is guaranteed to not move backward on Solaris
1612 inline hrtime_t getTimeNanos() {
1613   if (VM_Version::supports_cx8()) {
1614     const hrtime_t now = gethrtime();
1615     // Use atomic long load since 32-bit x86 uses 2 registers to keep long.
1616     const hrtime_t prev = Atomic::load((volatile jlong*)&max_hrtime);
1617     if (now <= prev)  return prev;   // same or retrograde time;
1618     const hrtime_t obsv = Atomic::cmpxchg(now, (volatile jlong*)&max_hrtime, prev);
1619     assert(obsv >= prev, "invariant");   // Monotonicity
1620     // If the CAS succeeded then we're done and return "now".
1621     // If the CAS failed and the observed value "obs" is >= now then
1622     // we should return "obs".  If the CAS failed and now > obs > prv then
1623     // some other thread raced this thread and installed a new value, in which case
1624     // we could either (a) retry the entire operation, (b) retry trying to install now
1625     // or (c) just return obs.  We use (c).   No loop is required although in some cases
1626     // we might discard a higher "now" value in deference to a slightly lower but freshly
1627     // installed obs value.   That's entirely benign -- it admits no new orderings compared
1628     // to (a) or (b) -- and greatly reduces coherence traffic.
1629     // We might also condition (c) on the magnitude of the delta between obs and now.
1630     // Avoiding excessive CAS operations to hot RW locations is critical.
1631     // See http://blogs.sun.com/dave/entry/cas_and_cache_trivia_invalidate
1632     return (prev == obsv) ? now : obsv ;
1633   } else {
1634     return oldgetTimeNanos();
1635   }
1636 }
1637 
1638 // Time since start-up in seconds to a fine granularity.
1639 // Used by VMSelfDestructTimer and the MemProfiler.
1640 double os::elapsedTime() {
1641   return (double)(getTimeNanos() - first_hrtime) / (double)hrtime_hz;
1642 }
1643 
1644 jlong os::elapsed_counter() {
1645   return (jlong)(getTimeNanos() - first_hrtime);
1646 }
1647 
1648 jlong os::elapsed_frequency() {
1649    return hrtime_hz;
1650 }
1651 
1652 // Return the real, user, and system times in seconds from an
1653 // arbitrary fixed point in the past.
1654 bool os::getTimesSecs(double* process_real_time,
1655                   double* process_user_time,
1656                   double* process_system_time) {
1657   struct tms ticks;
1658   clock_t real_ticks = times(&ticks);
1659 
1660   if (real_ticks == (clock_t) (-1)) {
1661     return false;
1662   } else {
1663     double ticks_per_second = (double) clock_tics_per_sec;
1664     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1665     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1666     // For consistency return the real time from getTimeNanos()
1667     // converted to seconds.
1668     *process_real_time = ((double) getTimeNanos()) / ((double) NANOUNITS);
1669 
1670     return true;
1671   }
1672 }
1673 
1674 bool os::supports_vtime() { return true; }
1675 
1676 bool os::enable_vtime() {
1677   int fd = ::open("/proc/self/ctl", O_WRONLY);
1678   if (fd == -1)
1679     return false;
1680 
1681   long cmd[] = { PCSET, PR_MSACCT };
1682   int res = ::write(fd, cmd, sizeof(long) * 2);
1683   ::close(fd);
1684   if (res != sizeof(long) * 2)
1685     return false;
1686 
1687   return true;
1688 }
1689 
1690 bool os::vtime_enabled() {
1691   int fd = ::open("/proc/self/status", O_RDONLY);
1692   if (fd == -1)
1693     return false;
1694 
1695   pstatus_t status;
1696   int res = os::read(fd, (void*) &status, sizeof(pstatus_t));
1697   ::close(fd);
1698   if (res != sizeof(pstatus_t))
1699     return false;
1700 
1701   return status.pr_flags & PR_MSACCT;
1702 }
1703 
1704 double os::elapsedVTime() {
1705   return (double)gethrvtime() / (double)hrtime_hz;
1706 }
1707 
1708 // Used internally for comparisons only
1709 // getTimeMillis guaranteed to not move backwards on Solaris
1710 jlong getTimeMillis() {
1711   jlong nanotime = getTimeNanos();
1712   return (jlong)(nanotime / NANOSECS_PER_MILLISEC);
1713 }
1714 
1715 // Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1716 jlong os::javaTimeMillis() {
1717   timeval t;
1718   if (gettimeofday( &t, NULL) == -1)
1719     fatal(err_msg("os::javaTimeMillis: gettimeofday (%s)", strerror(errno)));
1720   return jlong(t.tv_sec) * 1000  +  jlong(t.tv_usec) / 1000;
1721 }
1722 
1723 jlong os::javaTimeNanos() {
1724   return (jlong)getTimeNanos();
1725 }
1726 
1727 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1728   info_ptr->max_value = ALL_64_BITS;      // gethrtime() uses all 64 bits
1729   info_ptr->may_skip_backward = false;    // not subject to resetting or drifting
1730   info_ptr->may_skip_forward = false;     // not subject to resetting or drifting
1731   info_ptr->kind = JVMTI_TIMER_ELAPSED;   // elapsed not CPU time
1732 }
1733 
1734 char * os::local_time_string(char *buf, size_t buflen) {
1735   struct tm t;
1736   time_t long_time;
1737   time(&long_time);
1738   localtime_r(&long_time, &t);
1739   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1740                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1741                t.tm_hour, t.tm_min, t.tm_sec);
1742   return buf;
1743 }
1744 
1745 // Note: os::shutdown() might be called very early during initialization, or
1746 // called from signal handler. Before adding something to os::shutdown(), make
1747 // sure it is async-safe and can handle partially initialized VM.
1748 void os::shutdown() {
1749 
1750   // allow PerfMemory to attempt cleanup of any persistent resources
1751   perfMemory_exit();
1752 
1753   // needs to remove object in file system
1754   AttachListener::abort();
1755 
1756   // flush buffered output, finish log files
1757   ostream_abort();
1758 
1759   // Check for abort hook
1760   abort_hook_t abort_hook = Arguments::abort_hook();
1761   if (abort_hook != NULL) {
1762     abort_hook();
1763   }
1764 }
1765 
1766 // Note: os::abort() might be called very early during initialization, or
1767 // called from signal handler. Before adding something to os::abort(), make
1768 // sure it is async-safe and can handle partially initialized VM.
1769 void os::abort(bool dump_core) {
1770   os::shutdown();
1771   if (dump_core) {
1772 #ifndef PRODUCT
1773     fdStream out(defaultStream::output_fd());
1774     out.print_raw("Current thread is ");
1775     char buf[16];
1776     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1777     out.print_raw_cr(buf);
1778     out.print_raw_cr("Dumping core ...");
1779 #endif
1780     ::abort(); // dump core (for debugging)
1781   }
1782 
1783   ::exit(1);
1784 }
1785 
1786 // Die immediately, no exit hook, no abort hook, no cleanup.
1787 void os::die() {
1788   ::abort(); // dump core (for debugging)
1789 }
1790 
1791 // unused
1792 void os::set_error_file(const char *logfile) {}
1793 
1794 // DLL functions
1795 
1796 const char* os::dll_file_extension() { return ".so"; }
1797 
1798 // This must be hard coded because it's the system's temporary
1799 // directory not the java application's temp directory, ala java.io.tmpdir.
1800 const char* os::get_temp_directory() { return "/tmp"; }
1801 
1802 static bool file_exists(const char* filename) {
1803   struct stat statbuf;
1804   if (filename == NULL || strlen(filename) == 0) {
1805     return false;
1806   }
1807   return os::stat(filename, &statbuf) == 0;
1808 }
1809 
1810 bool os::dll_build_name(char* buffer, size_t buflen,
1811                         const char* pname, const char* fname) {
1812   bool retval = false;
1813   const size_t pnamelen = pname ? strlen(pname) : 0;
1814 
1815   // Return error on buffer overflow.
1816   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1817     return retval;
1818   }
1819 
1820   if (pnamelen == 0) {
1821     snprintf(buffer, buflen, "lib%s.so", fname);
1822     retval = true;
1823   } else if (strchr(pname, *os::path_separator()) != NULL) {
1824     int n;
1825     char** pelements = split_path(pname, &n);
1826     if (pelements == NULL) {
1827       return false;
1828     }
1829     for (int i = 0 ; i < n ; i++) {
1830       // really shouldn't be NULL but what the heck, check can't hurt
1831       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1832         continue; // skip the empty path values
1833       }
1834       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1835       if (file_exists(buffer)) {
1836         retval = true;
1837         break;
1838       }
1839     }
1840     // release the storage
1841     for (int i = 0 ; i < n ; i++) {
1842       if (pelements[i] != NULL) {
1843         FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1844       }
1845     }
1846     if (pelements != NULL) {
1847       FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1848     }
1849   } else {
1850     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1851     retval = true;
1852   }
1853   return retval;
1854 }
1855 
1856 // check if addr is inside libjvm.so
1857 bool os::address_is_in_vm(address addr) {
1858   static address libjvm_base_addr;
1859   Dl_info dlinfo;
1860 
1861   if (libjvm_base_addr == NULL) {
1862     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1863       libjvm_base_addr = (address)dlinfo.dli_fbase;
1864     }
1865     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1866   }
1867 
1868   if (dladdr((void *)addr, &dlinfo) != 0) {
1869     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1870   }
1871 
1872   return false;
1873 }
1874 
1875 typedef int (*dladdr1_func_type) (void *, Dl_info *, void **, int);
1876 static dladdr1_func_type dladdr1_func = NULL;
1877 
1878 bool os::dll_address_to_function_name(address addr, char *buf,
1879                                       int buflen, int * offset) {
1880   // buf is not optional, but offset is optional
1881   assert(buf != NULL, "sanity check");
1882 
1883   Dl_info dlinfo;
1884 
1885   // dladdr1_func was initialized in os::init()
1886   if (dladdr1_func != NULL) {
1887     // yes, we have dladdr1
1888 
1889     // Support for dladdr1 is checked at runtime; it may be
1890     // available even if the vm is built on a machine that does
1891     // not have dladdr1 support.  Make sure there is a value for
1892     // RTLD_DL_SYMENT.
1893     #ifndef RTLD_DL_SYMENT
1894     #define RTLD_DL_SYMENT 1
1895     #endif
1896 #ifdef _LP64
1897     Elf64_Sym * info;
1898 #else
1899     Elf32_Sym * info;
1900 #endif
1901     if (dladdr1_func((void *)addr, &dlinfo, (void **)&info,
1902                      RTLD_DL_SYMENT) != 0) {
1903       // see if we have a matching symbol that covers our address
1904       if (dlinfo.dli_saddr != NULL &&
1905           (char *)dlinfo.dli_saddr + info->st_size > (char *)addr) {
1906         if (dlinfo.dli_sname != NULL) {
1907           if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1908             jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1909           }
1910           if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1911           return true;
1912         }
1913       }
1914       // no matching symbol so try for just file info
1915       if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1916         if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1917                             buf, buflen, offset, dlinfo.dli_fname)) {
1918           return true;
1919         }
1920       }
1921     }
1922     buf[0] = '\0';
1923     if (offset != NULL) *offset  = -1;
1924     return false;
1925   }
1926 
1927   // no, only dladdr is available
1928   if (dladdr((void *)addr, &dlinfo) != 0) {
1929     // see if we have a matching symbol
1930     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1931       if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1932         jio_snprintf(buf, buflen, dlinfo.dli_sname);
1933       }
1934       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1935       return true;
1936     }
1937     // no matching symbol so try for just file info
1938     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1939       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1940                           buf, buflen, offset, dlinfo.dli_fname)) {
1941         return true;
1942       }
1943     }
1944   }
1945   buf[0] = '\0';
1946   if (offset != NULL) *offset  = -1;
1947   return false;
1948 }
1949 
1950 bool os::dll_address_to_library_name(address addr, char* buf,
1951                                      int buflen, int* offset) {
1952   // buf is not optional, but offset is optional
1953   assert(buf != NULL, "sanity check");
1954 
1955   Dl_info dlinfo;
1956 
1957   if (dladdr((void*)addr, &dlinfo) != 0) {
1958     if (dlinfo.dli_fname != NULL) {
1959       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1960     }
1961     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1962       *offset = addr - (address)dlinfo.dli_fbase;
1963     }
1964     return true;
1965   }
1966 
1967   buf[0] = '\0';
1968   if (offset) *offset = -1;
1969   return false;
1970 }
1971 
1972 // Prints the names and full paths of all opened dynamic libraries
1973 // for current process
1974 void os::print_dll_info(outputStream * st) {
1975   Dl_info dli;
1976   void *handle;
1977   Link_map *map;
1978   Link_map *p;
1979 
1980   st->print_cr("Dynamic libraries:"); st->flush();
1981 
1982   if (dladdr(CAST_FROM_FN_PTR(void *, os::print_dll_info), &dli) == 0 ||
1983       dli.dli_fname == NULL) {
1984     st->print_cr("Error: Cannot print dynamic libraries.");
1985     return;
1986   }
1987   handle = dlopen(dli.dli_fname, RTLD_LAZY);
1988   if (handle == NULL) {
1989     st->print_cr("Error: Cannot print dynamic libraries.");
1990     return;
1991   }
1992   dlinfo(handle, RTLD_DI_LINKMAP, &map);
1993   if (map == NULL) {
1994     st->print_cr("Error: Cannot print dynamic libraries.");
1995     return;
1996   }
1997 
1998   while (map->l_prev != NULL)
1999     map = map->l_prev;
2000 
2001   while (map != NULL) {
2002     st->print_cr(PTR_FORMAT " \t%s", map->l_addr, map->l_name);
2003     map = map->l_next;
2004   }
2005 
2006   dlclose(handle);
2007 }
2008 
2009   // Loads .dll/.so and
2010   // in case of error it checks if .dll/.so was built for the
2011   // same architecture as Hotspot is running on
2012 
2013 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
2014 {
2015   void * result= ::dlopen(filename, RTLD_LAZY);
2016   if (result != NULL) {
2017     // Successful loading
2018     return result;
2019   }
2020 
2021   Elf32_Ehdr elf_head;
2022 
2023   // Read system error message into ebuf
2024   // It may or may not be overwritten below
2025   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2026   ebuf[ebuflen-1]='\0';
2027   int diag_msg_max_length=ebuflen-strlen(ebuf);
2028   char* diag_msg_buf=ebuf+strlen(ebuf);
2029 
2030   if (diag_msg_max_length==0) {
2031     // No more space in ebuf for additional diagnostics message
2032     return NULL;
2033   }
2034 
2035 
2036   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
2037 
2038   if (file_descriptor < 0) {
2039     // Can't open library, report dlerror() message
2040     return NULL;
2041   }
2042 
2043   bool failed_to_read_elf_head=
2044     (sizeof(elf_head)!=
2045         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
2046 
2047   ::close(file_descriptor);
2048   if (failed_to_read_elf_head) {
2049     // file i/o error - report dlerror() msg
2050     return NULL;
2051   }
2052 
2053   typedef struct {
2054     Elf32_Half  code;         // Actual value as defined in elf.h
2055     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
2056     char        elf_class;    // 32 or 64 bit
2057     char        endianess;    // MSB or LSB
2058     char*       name;         // String representation
2059   } arch_t;
2060 
2061   static const arch_t arch_array[]={
2062     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2063     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2064     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
2065     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
2066     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2067     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2068     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
2069     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
2070     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
2071     {EM_ARM,         EM_ARM,     ELFCLASS32, ELFDATA2LSB, (char*)"ARM 32"}
2072   };
2073 
2074   #if  (defined IA32)
2075     static  Elf32_Half running_arch_code=EM_386;
2076   #elif   (defined AMD64)
2077     static  Elf32_Half running_arch_code=EM_X86_64;
2078   #elif  (defined IA64)
2079     static  Elf32_Half running_arch_code=EM_IA_64;
2080   #elif  (defined __sparc) && (defined _LP64)
2081     static  Elf32_Half running_arch_code=EM_SPARCV9;
2082   #elif  (defined __sparc) && (!defined _LP64)
2083     static  Elf32_Half running_arch_code=EM_SPARC;
2084   #elif  (defined __powerpc64__)
2085     static  Elf32_Half running_arch_code=EM_PPC64;
2086   #elif  (defined __powerpc__)
2087     static  Elf32_Half running_arch_code=EM_PPC;
2088   #elif (defined ARM)
2089     static  Elf32_Half running_arch_code=EM_ARM;
2090   #else
2091     #error Method os::dll_load requires that one of following is defined:\
2092          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, ARM
2093   #endif
2094 
2095   // Identify compatability class for VM's architecture and library's architecture
2096   // Obtain string descriptions for architectures
2097 
2098   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2099   int running_arch_index=-1;
2100 
2101   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2102     if (running_arch_code == arch_array[i].code) {
2103       running_arch_index    = i;
2104     }
2105     if (lib_arch.code == arch_array[i].code) {
2106       lib_arch.compat_class = arch_array[i].compat_class;
2107       lib_arch.name         = arch_array[i].name;
2108     }
2109   }
2110 
2111   assert(running_arch_index != -1,
2112     "Didn't find running architecture code (running_arch_code) in arch_array");
2113   if (running_arch_index == -1) {
2114     // Even though running architecture detection failed
2115     // we may still continue with reporting dlerror() message
2116     return NULL;
2117   }
2118 
2119   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2120     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2121     return NULL;
2122   }
2123 
2124   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2125     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2126     return NULL;
2127   }
2128 
2129   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2130     if ( lib_arch.name!=NULL ) {
2131       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2132         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2133         lib_arch.name, arch_array[running_arch_index].name);
2134     } else {
2135       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2136       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2137         lib_arch.code,
2138         arch_array[running_arch_index].name);
2139     }
2140   }
2141 
2142   return NULL;
2143 }
2144 
2145 void* os::dll_lookup(void* handle, const char* name) {
2146   return dlsym(handle, name);
2147 }
2148 
2149 int os::stat(const char *path, struct stat *sbuf) {
2150   char pathbuf[MAX_PATH];
2151   if (strlen(path) > MAX_PATH - 1) {
2152     errno = ENAMETOOLONG;
2153     return -1;
2154   }
2155   os::native_path(strcpy(pathbuf, path));
2156   return ::stat(pathbuf, sbuf);
2157 }
2158 
2159 static bool _print_ascii_file(const char* filename, outputStream* st) {
2160   int fd = ::open(filename, O_RDONLY);
2161   if (fd == -1) {
2162      return false;
2163   }
2164 
2165   char buf[32];
2166   int bytes;
2167   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2168     st->print_raw(buf, bytes);
2169   }
2170 
2171   ::close(fd);
2172 
2173   return true;
2174 }
2175 
2176 void os::print_os_info_brief(outputStream* st) {
2177   os::Solaris::print_distro_info(st);
2178 
2179   os::Posix::print_uname_info(st);
2180 
2181   os::Solaris::print_libversion_info(st);
2182 }
2183 
2184 void os::print_os_info(outputStream* st) {
2185   st->print("OS:");
2186 
2187   os::Solaris::print_distro_info(st);
2188 
2189   os::Posix::print_uname_info(st);
2190 
2191   os::Solaris::print_libversion_info(st);
2192 
2193   os::Posix::print_rlimit_info(st);
2194 
2195   os::Posix::print_load_average(st);
2196 }
2197 
2198 void os::Solaris::print_distro_info(outputStream* st) {
2199   if (!_print_ascii_file("/etc/release", st)) {
2200       st->print("Solaris");
2201     }
2202     st->cr();
2203 }
2204 
2205 void os::Solaris::print_libversion_info(outputStream* st) {
2206   if (os::Solaris::T2_libthread()) {
2207     st->print("  (T2 libthread)");
2208   }
2209   else {
2210     st->print("  (T1 libthread)");
2211   }
2212   st->cr();
2213 }
2214 
2215 static bool check_addr0(outputStream* st) {
2216   jboolean status = false;
2217   int fd = ::open("/proc/self/map",O_RDONLY);
2218   if (fd >= 0) {
2219     prmap_t p;
2220     while(::read(fd, &p, sizeof(p)) > 0) {
2221       if (p.pr_vaddr == 0x0) {
2222         st->print("Warning: Address: 0x%x, Size: %dK, ",p.pr_vaddr, p.pr_size/1024, p.pr_mapname);
2223         st->print("Mapped file: %s, ", p.pr_mapname[0] == '\0' ? "None" : p.pr_mapname);
2224         st->print("Access:");
2225         st->print("%s",(p.pr_mflags & MA_READ)  ? "r" : "-");
2226         st->print("%s",(p.pr_mflags & MA_WRITE) ? "w" : "-");
2227         st->print("%s",(p.pr_mflags & MA_EXEC)  ? "x" : "-");
2228         st->cr();
2229         status = true;
2230       }
2231       ::close(fd);
2232     }
2233   }
2234   return status;
2235 }
2236 
2237 void os::pd_print_cpu_info(outputStream* st) {
2238   // Nothing to do for now.
2239 }
2240 
2241 void os::print_memory_info(outputStream* st) {
2242   st->print("Memory:");
2243   st->print(" %dk page", os::vm_page_size()>>10);
2244   st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
2245   st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
2246   st->cr();
2247   (void) check_addr0(st);
2248 }
2249 
2250 // Taken from /usr/include/sys/machsig.h  Supposed to be architecture specific
2251 // but they're the same for all the solaris architectures that we support.
2252 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2253                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2254                           "ILL_COPROC", "ILL_BADSTK" };
2255 
2256 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2257                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2258                           "FPE_FLTINV", "FPE_FLTSUB" };
2259 
2260 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2261 
2262 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2263 
2264 void os::print_siginfo(outputStream* st, void* siginfo) {
2265   st->print("siginfo:");
2266 
2267   const int buflen = 100;
2268   char buf[buflen];
2269   siginfo_t *si = (siginfo_t*)siginfo;
2270   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2271   char *err = strerror(si->si_errno);
2272   if (si->si_errno != 0 && err != NULL) {
2273     st->print("si_errno=%s", err);
2274   } else {
2275     st->print("si_errno=%d", si->si_errno);
2276   }
2277   const int c = si->si_code;
2278   assert(c > 0, "unexpected si_code");
2279   switch (si->si_signo) {
2280   case SIGILL:
2281     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2282     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2283     break;
2284   case SIGFPE:
2285     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2286     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2287     break;
2288   case SIGSEGV:
2289     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2290     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2291     break;
2292   case SIGBUS:
2293     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2294     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2295     break;
2296   default:
2297     st->print(", si_code=%d", si->si_code);
2298     // no si_addr
2299   }
2300 
2301   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2302       UseSharedSpaces) {
2303     FileMapInfo* mapinfo = FileMapInfo::current_info();
2304     if (mapinfo->is_in_shared_space(si->si_addr)) {
2305       st->print("\n\nError accessing class data sharing archive."   \
2306                 " Mapped file inaccessible during execution, "      \
2307                 " possible disk/network problem.");
2308     }
2309   }
2310   st->cr();
2311 }
2312 
2313 // Moved from whole group, because we need them here for diagnostic
2314 // prints.
2315 #define OLDMAXSIGNUM 32
2316 static int Maxsignum = 0;
2317 static int *ourSigFlags = NULL;
2318 
2319 extern "C" void sigINTRHandler(int, siginfo_t*, void*);
2320 
2321 int os::Solaris::get_our_sigflags(int sig) {
2322   assert(ourSigFlags!=NULL, "signal data structure not initialized");
2323   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2324   return ourSigFlags[sig];
2325 }
2326 
2327 void os::Solaris::set_our_sigflags(int sig, int flags) {
2328   assert(ourSigFlags!=NULL, "signal data structure not initialized");
2329   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2330   ourSigFlags[sig] = flags;
2331 }
2332 
2333 
2334 static const char* get_signal_handler_name(address handler,
2335                                            char* buf, int buflen) {
2336   int offset;
2337   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
2338   if (found) {
2339     // skip directory names
2340     const char *p1, *p2;
2341     p1 = buf;
2342     size_t len = strlen(os::file_separator());
2343     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
2344     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
2345   } else {
2346     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
2347   }
2348   return buf;
2349 }
2350 
2351 static void print_signal_handler(outputStream* st, int sig,
2352                                   char* buf, size_t buflen) {
2353   struct sigaction sa;
2354 
2355   sigaction(sig, NULL, &sa);
2356 
2357   st->print("%s: ", os::exception_name(sig, buf, buflen));
2358 
2359   address handler = (sa.sa_flags & SA_SIGINFO)
2360                   ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
2361                   : CAST_FROM_FN_PTR(address, sa.sa_handler);
2362 
2363   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
2364     st->print("SIG_DFL");
2365   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
2366     st->print("SIG_IGN");
2367   } else {
2368     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
2369   }
2370 
2371   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
2372 
2373   address rh = VMError::get_resetted_sighandler(sig);
2374   // May be, handler was resetted by VMError?
2375   if(rh != NULL) {
2376     handler = rh;
2377     sa.sa_flags = VMError::get_resetted_sigflags(sig);
2378   }
2379 
2380   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
2381 
2382   // Check: is it our handler?
2383   if(handler == CAST_FROM_FN_PTR(address, signalHandler) ||
2384      handler == CAST_FROM_FN_PTR(address, sigINTRHandler)) {
2385     // It is our signal handler
2386     // check for flags
2387     if(sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
2388       st->print(
2389         ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
2390         os::Solaris::get_our_sigflags(sig));
2391     }
2392   }
2393   st->cr();
2394 }
2395 
2396 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2397   st->print_cr("Signal Handlers:");
2398   print_signal_handler(st, SIGSEGV, buf, buflen);
2399   print_signal_handler(st, SIGBUS , buf, buflen);
2400   print_signal_handler(st, SIGFPE , buf, buflen);
2401   print_signal_handler(st, SIGPIPE, buf, buflen);
2402   print_signal_handler(st, SIGXFSZ, buf, buflen);
2403   print_signal_handler(st, SIGILL , buf, buflen);
2404   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2405   print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
2406   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2407   print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
2408   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2409   print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
2410   print_signal_handler(st, os::Solaris::SIGinterrupt(), buf, buflen);
2411   print_signal_handler(st, os::Solaris::SIGasync(), buf, buflen);
2412 }
2413 
2414 static char saved_jvm_path[MAXPATHLEN] = { 0 };
2415 
2416 // Find the full path to the current module, libjvm.so
2417 void os::jvm_path(char *buf, jint buflen) {
2418   // Error checking.
2419   if (buflen < MAXPATHLEN) {
2420     assert(false, "must use a large-enough buffer");
2421     buf[0] = '\0';
2422     return;
2423   }
2424   // Lazy resolve the path to current module.
2425   if (saved_jvm_path[0] != 0) {
2426     strcpy(buf, saved_jvm_path);
2427     return;
2428   }
2429 
2430   Dl_info dlinfo;
2431   int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
2432   assert(ret != 0, "cannot locate libjvm");
2433   if (ret != 0 && dlinfo.dli_fname != NULL) {
2434     realpath((char *)dlinfo.dli_fname, buf);
2435   } else {
2436     buf[0] = '\0';
2437     return;
2438   }
2439 
2440   if (Arguments::created_by_gamma_launcher()) {
2441     // Support for the gamma launcher.  Typical value for buf is
2442     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2443     // the right place in the string, then assume we are installed in a JDK and
2444     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2445     // up the path so it looks like libjvm.so is installed there (append a
2446     // fake suffix hotspot/libjvm.so).
2447     const char *p = buf + strlen(buf) - 1;
2448     for (int count = 0; p > buf && count < 5; ++count) {
2449       for (--p; p > buf && *p != '/'; --p)
2450         /* empty */ ;
2451     }
2452 
2453     if (strncmp(p, "/jre/lib/", 9) != 0) {
2454       // Look for JAVA_HOME in the environment.
2455       char* java_home_var = ::getenv("JAVA_HOME");
2456       if (java_home_var != NULL && java_home_var[0] != 0) {
2457         char cpu_arch[12];
2458         char* jrelib_p;
2459         int   len;
2460         sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
2461 #ifdef _LP64
2462         // If we are on sparc running a 64-bit vm, look in jre/lib/sparcv9.
2463         if (strcmp(cpu_arch, "sparc") == 0) {
2464           strcat(cpu_arch, "v9");
2465         } else if (strcmp(cpu_arch, "i386") == 0) {
2466           strcpy(cpu_arch, "amd64");
2467         }
2468 #endif
2469         // Check the current module name "libjvm.so".
2470         p = strrchr(buf, '/');
2471         assert(strstr(p, "/libjvm") == p, "invalid library name");
2472 
2473         realpath(java_home_var, buf);
2474         // determine if this is a legacy image or modules image
2475         // modules image doesn't have "jre" subdirectory
2476         len = strlen(buf);
2477         jrelib_p = buf + len;
2478         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2479         if (0 != access(buf, F_OK)) {
2480           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2481         }
2482 
2483         if (0 == access(buf, F_OK)) {
2484           // Use current module name "libjvm.so"
2485           len = strlen(buf);
2486           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2487         } else {
2488           // Go back to path of .so
2489           realpath((char *)dlinfo.dli_fname, buf);
2490         }
2491       }
2492     }
2493   }
2494 
2495   strcpy(saved_jvm_path, buf);
2496 }
2497 
2498 
2499 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2500   // no prefix required, not even "_"
2501 }
2502 
2503 
2504 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2505   // no suffix required
2506 }
2507 
2508 // This method is a copy of JDK's sysGetLastErrorString
2509 // from src/solaris/hpi/src/system_md.c
2510 
2511 size_t os::lasterror(char *buf, size_t len) {
2512 
2513   if (errno == 0)  return 0;
2514 
2515   const char *s = ::strerror(errno);
2516   size_t n = ::strlen(s);
2517   if (n >= len) {
2518     n = len - 1;
2519   }
2520   ::strncpy(buf, s, n);
2521   buf[n] = '\0';
2522   return n;
2523 }
2524 
2525 
2526 // sun.misc.Signal
2527 
2528 extern "C" {
2529   static void UserHandler(int sig, void *siginfo, void *context) {
2530     // Ctrl-C is pressed during error reporting, likely because the error
2531     // handler fails to abort. Let VM die immediately.
2532     if (sig == SIGINT && is_error_reported()) {
2533        os::die();
2534     }
2535 
2536     os::signal_notify(sig);
2537     // We do not need to reinstate the signal handler each time...
2538   }
2539 }
2540 
2541 void* os::user_handler() {
2542   return CAST_FROM_FN_PTR(void*, UserHandler);
2543 }
2544 
2545 class Semaphore : public StackObj {
2546   public:
2547     Semaphore();
2548     ~Semaphore();
2549     void signal();
2550     void wait();
2551     bool trywait();
2552     bool timedwait(unsigned int sec, int nsec);
2553   private:
2554     sema_t _semaphore;
2555 };
2556 
2557 
2558 Semaphore::Semaphore() {
2559   sema_init(&_semaphore, 0, NULL, NULL);
2560 }
2561 
2562 Semaphore::~Semaphore() {
2563   sema_destroy(&_semaphore);
2564 }
2565 
2566 void Semaphore::signal() {
2567   sema_post(&_semaphore);
2568 }
2569 
2570 void Semaphore::wait() {
2571   sema_wait(&_semaphore);
2572 }
2573 
2574 bool Semaphore::trywait() {
2575   return sema_trywait(&_semaphore) == 0;
2576 }
2577 
2578 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2579   struct timespec ts;
2580   unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2581 
2582   while (1) {
2583     int result = sema_timedwait(&_semaphore, &ts);
2584     if (result == 0) {
2585       return true;
2586     } else if (errno == EINTR) {
2587       continue;
2588     } else if (errno == ETIME) {
2589       return false;
2590     } else {
2591       return false;
2592     }
2593   }
2594 }
2595 
2596 extern "C" {
2597   typedef void (*sa_handler_t)(int);
2598   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2599 }
2600 
2601 void* os::signal(int signal_number, void* handler) {
2602   struct sigaction sigAct, oldSigAct;
2603   sigfillset(&(sigAct.sa_mask));
2604   sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2605   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2606 
2607   if (sigaction(signal_number, &sigAct, &oldSigAct))
2608     // -1 means registration failed
2609     return (void *)-1;
2610 
2611   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2612 }
2613 
2614 void os::signal_raise(int signal_number) {
2615   raise(signal_number);
2616 }
2617 
2618 /*
2619  * The following code is moved from os.cpp for making this
2620  * code platform specific, which it is by its very nature.
2621  */
2622 
2623 // a counter for each possible signal value
2624 static int Sigexit = 0;
2625 static int Maxlibjsigsigs;
2626 static jint *pending_signals = NULL;
2627 static int *preinstalled_sigs = NULL;
2628 static struct sigaction *chainedsigactions = NULL;
2629 static sema_t sig_sem;
2630 typedef int (*version_getting_t)();
2631 version_getting_t os::Solaris::get_libjsig_version = NULL;
2632 static int libjsigversion = NULL;
2633 
2634 int os::sigexitnum_pd() {
2635   assert(Sigexit > 0, "signal memory not yet initialized");
2636   return Sigexit;
2637 }
2638 
2639 void os::Solaris::init_signal_mem() {
2640   // Initialize signal structures
2641   Maxsignum = SIGRTMAX;
2642   Sigexit = Maxsignum+1;
2643   assert(Maxsignum >0, "Unable to obtain max signal number");
2644 
2645   Maxlibjsigsigs = Maxsignum;
2646 
2647   // pending_signals has one int per signal
2648   // The additional signal is for SIGEXIT - exit signal to signal_thread
2649   pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1), mtInternal);
2650   memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2651 
2652   if (UseSignalChaining) {
2653      chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2654        * (Maxsignum + 1), mtInternal);
2655      memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2656      preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2657      memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2658   }
2659   ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1 ), mtInternal);
2660   memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2661 }
2662 
2663 void os::signal_init_pd() {
2664   int ret;
2665 
2666   ret = ::sema_init(&sig_sem, 0, NULL, NULL);
2667   assert(ret == 0, "sema_init() failed");
2668 }
2669 
2670 void os::signal_notify(int signal_number) {
2671   int ret;
2672 
2673   Atomic::inc(&pending_signals[signal_number]);
2674   ret = ::sema_post(&sig_sem);
2675   assert(ret == 0, "sema_post() failed");
2676 }
2677 
2678 static int check_pending_signals(bool wait_for_signal) {
2679   int ret;
2680   while (true) {
2681     for (int i = 0; i < Sigexit + 1; i++) {
2682       jint n = pending_signals[i];
2683       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2684         return i;
2685       }
2686     }
2687     if (!wait_for_signal) {
2688       return -1;
2689     }
2690     JavaThread *thread = JavaThread::current();
2691     ThreadBlockInVM tbivm(thread);
2692 
2693     bool threadIsSuspended;
2694     do {
2695       thread->set_suspend_equivalent();
2696       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2697       while((ret = ::sema_wait(&sig_sem)) == EINTR)
2698           ;
2699       assert(ret == 0, "sema_wait() failed");
2700 
2701       // were we externally suspended while we were waiting?
2702       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2703       if (threadIsSuspended) {
2704         //
2705         // The semaphore has been incremented, but while we were waiting
2706         // another thread suspended us. We don't want to continue running
2707         // while suspended because that would surprise the thread that
2708         // suspended us.
2709         //
2710         ret = ::sema_post(&sig_sem);
2711         assert(ret == 0, "sema_post() failed");
2712 
2713         thread->java_suspend_self();
2714       }
2715     } while (threadIsSuspended);
2716   }
2717 }
2718 
2719 int os::signal_lookup() {
2720   return check_pending_signals(false);
2721 }
2722 
2723 int os::signal_wait() {
2724   return check_pending_signals(true);
2725 }
2726 
2727 ////////////////////////////////////////////////////////////////////////////////
2728 // Virtual Memory
2729 
2730 static int page_size = -1;
2731 
2732 // The mmap MAP_ALIGN flag is supported on Solaris 9 and later.  init_2() will
2733 // clear this var if support is not available.
2734 static bool has_map_align = true;
2735 
2736 int os::vm_page_size() {
2737   assert(page_size != -1, "must call os::init");
2738   return page_size;
2739 }
2740 
2741 // Solaris allocates memory by pages.
2742 int os::vm_allocation_granularity() {
2743   assert(page_size != -1, "must call os::init");
2744   return page_size;
2745 }
2746 
2747 static bool recoverable_mmap_error(int err) {
2748   // See if the error is one we can let the caller handle. This
2749   // list of errno values comes from the Solaris mmap(2) man page.
2750   switch (err) {
2751   case EBADF:
2752   case EINVAL:
2753   case ENOTSUP:
2754     // let the caller deal with these errors
2755     return true;
2756 
2757   default:
2758     // Any remaining errors on this OS can cause our reserved mapping
2759     // to be lost. That can cause confusion where different data
2760     // structures think they have the same memory mapped. The worst
2761     // scenario is if both the VM and a library think they have the
2762     // same memory mapped.
2763     return false;
2764   }
2765 }
2766 
2767 static void warn_fail_commit_memory(char* addr, size_t bytes, bool exec,
2768                                     int err) {
2769   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2770           ", %d) failed; error='%s' (errno=%d)", addr, bytes, exec,
2771           strerror(err), err);
2772 }
2773 
2774 static void warn_fail_commit_memory(char* addr, size_t bytes,
2775                                     size_t alignment_hint, bool exec,
2776                                     int err) {
2777   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2778           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, bytes,
2779           alignment_hint, exec, strerror(err), err);
2780 }
2781 
2782 int os::Solaris::commit_memory_impl(char* addr, size_t bytes, bool exec) {
2783   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2784   size_t size = bytes;
2785   char *res = Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2786   if (res != NULL) {
2787     if (UseNUMAInterleaving) {
2788       numa_make_global(addr, bytes);
2789     }
2790     return 0;
2791   }
2792 
2793   int err = errno;  // save errno from mmap() call in mmap_chunk()
2794 
2795   if (!recoverable_mmap_error(err)) {
2796     warn_fail_commit_memory(addr, bytes, exec, err);
2797     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "committing reserved memory.");
2798   }
2799 
2800   return err;
2801 }
2802 
2803 bool os::pd_commit_memory(char* addr, size_t bytes, bool exec) {
2804   return Solaris::commit_memory_impl(addr, bytes, exec) == 0;
2805 }
2806 
2807 void os::pd_commit_memory_or_exit(char* addr, size_t bytes, bool exec,
2808                                   const char* mesg) {
2809   assert(mesg != NULL, "mesg must be specified");
2810   int err = os::Solaris::commit_memory_impl(addr, bytes, exec);
2811   if (err != 0) {
2812     // the caller wants all commit errors to exit with the specified mesg:
2813     warn_fail_commit_memory(addr, bytes, exec, err);
2814     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, mesg);
2815   }
2816 }
2817 
2818 int os::Solaris::commit_memory_impl(char* addr, size_t bytes,
2819                                     size_t alignment_hint, bool exec) {
2820   int err = Solaris::commit_memory_impl(addr, bytes, exec);
2821   if (err == 0) {
2822     if (UseLargePages && (alignment_hint > (size_t)vm_page_size())) {
2823       // If the large page size has been set and the VM
2824       // is using large pages, use the large page size
2825       // if it is smaller than the alignment hint. This is
2826       // a case where the VM wants to use a larger alignment size
2827       // for its own reasons but still want to use large pages
2828       // (which is what matters to setting the mpss range.
2829       size_t page_size = 0;
2830       if (large_page_size() < alignment_hint) {
2831         assert(UseLargePages, "Expected to be here for large page use only");
2832         page_size = large_page_size();
2833       } else {
2834         // If the alignment hint is less than the large page
2835         // size, the VM wants a particular alignment (thus the hint)
2836         // for internal reasons.  Try to set the mpss range using
2837         // the alignment_hint.
2838         page_size = alignment_hint;
2839       }
2840       // Since this is a hint, ignore any failures.
2841       (void)Solaris::setup_large_pages(addr, bytes, page_size);
2842     }
2843   }
2844   return err;
2845 }
2846 
2847 bool os::pd_commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2848                           bool exec) {
2849   return Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec) == 0;
2850 }
2851 
2852 void os::pd_commit_memory_or_exit(char* addr, size_t bytes,
2853                                   size_t alignment_hint, bool exec,
2854                                   const char* mesg) {
2855   assert(mesg != NULL, "mesg must be specified");
2856   int err = os::Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec);
2857   if (err != 0) {
2858     // the caller wants all commit errors to exit with the specified mesg:
2859     warn_fail_commit_memory(addr, bytes, alignment_hint, exec, err);
2860     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, mesg);
2861   }
2862 }
2863 
2864 // Uncommit the pages in a specified region.
2865 void os::pd_free_memory(char* addr, size_t bytes, size_t alignment_hint) {
2866   if (madvise(addr, bytes, MADV_FREE) < 0) {
2867     debug_only(warning("MADV_FREE failed."));
2868     return;
2869   }
2870 }
2871 
2872 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2873   return os::commit_memory(addr, size, !ExecMem);
2874 }
2875 
2876 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2877   return os::uncommit_memory(addr, size);
2878 }
2879 
2880 // Change the page size in a given range.
2881 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2882   assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2883   assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2884   if (UseLargePages) {
2885     Solaris::setup_large_pages(addr, bytes, alignment_hint);
2886   }
2887 }
2888 
2889 // Tell the OS to make the range local to the first-touching LWP
2890 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2891   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2892   if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2893     debug_only(warning("MADV_ACCESS_LWP failed."));
2894   }
2895 }
2896 
2897 // Tell the OS that this range would be accessed from different LWPs.
2898 void os::numa_make_global(char *addr, size_t bytes) {
2899   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2900   if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2901     debug_only(warning("MADV_ACCESS_MANY failed."));
2902   }
2903 }
2904 
2905 // Get the number of the locality groups.
2906 size_t os::numa_get_groups_num() {
2907   size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2908   return n != -1 ? n : 1;
2909 }
2910 
2911 // Get a list of leaf locality groups. A leaf lgroup is group that
2912 // doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2913 // board. An LWP is assigned to one of these groups upon creation.
2914 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2915    if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2916      ids[0] = 0;
2917      return 1;
2918    }
2919    int result_size = 0, top = 1, bottom = 0, cur = 0;
2920    for (int k = 0; k < size; k++) {
2921      int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2922                                     (Solaris::lgrp_id_t*)&ids[top], size - top);
2923      if (r == -1) {
2924        ids[0] = 0;
2925        return 1;
2926      }
2927      if (!r) {
2928        // That's a leaf node.
2929        assert (bottom <= cur, "Sanity check");
2930        // Check if the node has memory
2931        if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2932                                    NULL, 0, LGRP_RSRC_MEM) > 0) {
2933          ids[bottom++] = ids[cur];
2934        }
2935      }
2936      top += r;
2937      cur++;
2938    }
2939    if (bottom == 0) {
2940      // Handle a situation, when the OS reports no memory available.
2941      // Assume UMA architecture.
2942      ids[0] = 0;
2943      return 1;
2944    }
2945    return bottom;
2946 }
2947 
2948 // Detect the topology change. Typically happens during CPU plugging-unplugging.
2949 bool os::numa_topology_changed() {
2950   int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2951   if (is_stale != -1 && is_stale) {
2952     Solaris::lgrp_fini(Solaris::lgrp_cookie());
2953     Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2954     assert(c != 0, "Failure to initialize LGRP API");
2955     Solaris::set_lgrp_cookie(c);
2956     return true;
2957   }
2958   return false;
2959 }
2960 
2961 // Get the group id of the current LWP.
2962 int os::numa_get_group_id() {
2963   int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2964   if (lgrp_id == -1) {
2965     return 0;
2966   }
2967   const int size = os::numa_get_groups_num();
2968   int *ids = (int*)alloca(size * sizeof(int));
2969 
2970   // Get the ids of all lgroups with memory; r is the count.
2971   int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2972                                   (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2973   if (r <= 0) {
2974     return 0;
2975   }
2976   return ids[os::random() % r];
2977 }
2978 
2979 // Request information about the page.
2980 bool os::get_page_info(char *start, page_info* info) {
2981   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2982   uint64_t addr = (uintptr_t)start;
2983   uint64_t outdata[2];
2984   uint_t validity = 0;
2985 
2986   if (os::Solaris::meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2987     return false;
2988   }
2989 
2990   info->size = 0;
2991   info->lgrp_id = -1;
2992 
2993   if ((validity & 1) != 0) {
2994     if ((validity & 2) != 0) {
2995       info->lgrp_id = outdata[0];
2996     }
2997     if ((validity & 4) != 0) {
2998       info->size = outdata[1];
2999     }
3000     return true;
3001   }
3002   return false;
3003 }
3004 
3005 // Scan the pages from start to end until a page different than
3006 // the one described in the info parameter is encountered.
3007 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
3008   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
3009   const size_t types = sizeof(info_types) / sizeof(info_types[0]);
3010   uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT];
3011   uint_t validity[MAX_MEMINFO_CNT];
3012 
3013   size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
3014   uint64_t p = (uint64_t)start;
3015   while (p < (uint64_t)end) {
3016     addrs[0] = p;
3017     size_t addrs_count = 1;
3018     while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] + page_size < (uint64_t)end) {
3019       addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
3020       addrs_count++;
3021     }
3022 
3023     if (os::Solaris::meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
3024       return NULL;
3025     }
3026 
3027     size_t i = 0;
3028     for (; i < addrs_count; i++) {
3029       if ((validity[i] & 1) != 0) {
3030         if ((validity[i] & 4) != 0) {
3031           if (outdata[types * i + 1] != page_expected->size) {
3032             break;
3033           }
3034         } else
3035           if (page_expected->size != 0) {
3036             break;
3037           }
3038 
3039         if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
3040           if (outdata[types * i] != page_expected->lgrp_id) {
3041             break;
3042           }
3043         }
3044       } else {
3045         return NULL;
3046       }
3047     }
3048 
3049     if (i != addrs_count) {
3050       if ((validity[i] & 2) != 0) {
3051         page_found->lgrp_id = outdata[types * i];
3052       } else {
3053         page_found->lgrp_id = -1;
3054       }
3055       if ((validity[i] & 4) != 0) {
3056         page_found->size = outdata[types * i + 1];
3057       } else {
3058         page_found->size = 0;
3059       }
3060       return (char*)addrs[i];
3061     }
3062 
3063     p = addrs[addrs_count - 1] + page_size;
3064   }
3065   return end;
3066 }
3067 
3068 bool os::pd_uncommit_memory(char* addr, size_t bytes) {
3069   size_t size = bytes;
3070   // Map uncommitted pages PROT_NONE so we fail early if we touch an
3071   // uncommitted page. Otherwise, the read/write might succeed if we
3072   // have enough swap space to back the physical page.
3073   return
3074     NULL != Solaris::mmap_chunk(addr, size,
3075                                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
3076                                 PROT_NONE);
3077 }
3078 
3079 char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
3080   char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
3081 
3082   if (b == MAP_FAILED) {
3083     return NULL;
3084   }
3085   return b;
3086 }
3087 
3088 char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes, size_t alignment_hint, bool fixed) {
3089   char* addr = requested_addr;
3090   int flags = MAP_PRIVATE | MAP_NORESERVE;
3091 
3092   assert(!(fixed && (alignment_hint > 0)), "alignment hint meaningless with fixed mmap");
3093 
3094   if (fixed) {
3095     flags |= MAP_FIXED;
3096   } else if (has_map_align && (alignment_hint > (size_t) vm_page_size())) {
3097     flags |= MAP_ALIGN;
3098     addr = (char*) alignment_hint;
3099   }
3100 
3101   // Map uncommitted pages PROT_NONE so we fail early if we touch an
3102   // uncommitted page. Otherwise, the read/write might succeed if we
3103   // have enough swap space to back the physical page.
3104   return mmap_chunk(addr, bytes, flags, PROT_NONE);
3105 }
3106 
3107 char* os::pd_reserve_memory(size_t bytes, char* requested_addr, size_t alignment_hint) {
3108   char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint, (requested_addr != NULL));
3109 
3110   guarantee(requested_addr == NULL || requested_addr == addr,
3111             "OS failed to return requested mmap address.");
3112   return addr;
3113 }
3114 
3115 // Reserve memory at an arbitrary address, only if that area is
3116 // available (and not reserved for something else).
3117 
3118 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3119   const int max_tries = 10;
3120   char* base[max_tries];
3121   size_t size[max_tries];
3122 
3123   // Solaris adds a gap between mmap'ed regions.  The size of the gap
3124   // is dependent on the requested size and the MMU.  Our initial gap
3125   // value here is just a guess and will be corrected later.
3126   bool had_top_overlap = false;
3127   bool have_adjusted_gap = false;
3128   size_t gap = 0x400000;
3129 
3130   // Assert only that the size is a multiple of the page size, since
3131   // that's all that mmap requires, and since that's all we really know
3132   // about at this low abstraction level.  If we need higher alignment,
3133   // we can either pass an alignment to this method or verify alignment
3134   // in one of the methods further up the call chain.  See bug 5044738.
3135   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3136 
3137   // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
3138   // Give it a try, if the kernel honors the hint we can return immediately.
3139   char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
3140 
3141   volatile int err = errno;
3142   if (addr == requested_addr) {
3143     return addr;
3144   } else if (addr != NULL) {
3145     pd_unmap_memory(addr, bytes);
3146   }
3147 
3148   if (PrintMiscellaneous && Verbose) {
3149     char buf[256];
3150     buf[0] = '\0';
3151     if (addr == NULL) {
3152       jio_snprintf(buf, sizeof(buf), ": %s", strerror(err));
3153     }
3154     warning("attempt_reserve_memory_at: couldn't reserve " SIZE_FORMAT " bytes at "
3155             PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
3156             "%s", bytes, requested_addr, addr, buf);
3157   }
3158 
3159   // Address hint method didn't work.  Fall back to the old method.
3160   // In theory, once SNV becomes our oldest supported platform, this
3161   // code will no longer be needed.
3162   //
3163   // Repeatedly allocate blocks until the block is allocated at the
3164   // right spot. Give up after max_tries.
3165   int i;
3166   for (i = 0; i < max_tries; ++i) {
3167     base[i] = reserve_memory(bytes);
3168 
3169     if (base[i] != NULL) {
3170       // Is this the block we wanted?
3171       if (base[i] == requested_addr) {
3172         size[i] = bytes;
3173         break;
3174       }
3175 
3176       // check that the gap value is right
3177       if (had_top_overlap && !have_adjusted_gap) {
3178         size_t actual_gap = base[i-1] - base[i] - bytes;
3179         if (gap != actual_gap) {
3180           // adjust the gap value and retry the last 2 allocations
3181           assert(i > 0, "gap adjustment code problem");
3182           have_adjusted_gap = true;  // adjust the gap only once, just in case
3183           gap = actual_gap;
3184           if (PrintMiscellaneous && Verbose) {
3185             warning("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
3186           }
3187           unmap_memory(base[i], bytes);
3188           unmap_memory(base[i-1], size[i-1]);
3189           i-=2;
3190           continue;
3191         }
3192       }
3193 
3194       // Does this overlap the block we wanted? Give back the overlapped
3195       // parts and try again.
3196       //
3197       // There is still a bug in this code: if top_overlap == bytes,
3198       // the overlap is offset from requested region by the value of gap.
3199       // In this case giving back the overlapped part will not work,
3200       // because we'll give back the entire block at base[i] and
3201       // therefore the subsequent allocation will not generate a new gap.
3202       // This could be fixed with a new algorithm that used larger
3203       // or variable size chunks to find the requested region -
3204       // but such a change would introduce additional complications.
3205       // It's rare enough that the planets align for this bug,
3206       // so we'll just wait for a fix for 6204603/5003415 which
3207       // will provide a mmap flag to allow us to avoid this business.
3208 
3209       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3210       if (top_overlap >= 0 && top_overlap < bytes) {
3211         had_top_overlap = true;
3212         unmap_memory(base[i], top_overlap);
3213         base[i] += top_overlap;
3214         size[i] = bytes - top_overlap;
3215       } else {
3216         size_t bottom_overlap = base[i] + bytes - requested_addr;
3217         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3218           if (PrintMiscellaneous && Verbose && bottom_overlap == 0) {
3219             warning("attempt_reserve_memory_at: possible alignment bug");
3220           }
3221           unmap_memory(requested_addr, bottom_overlap);
3222           size[i] = bytes - bottom_overlap;
3223         } else {
3224           size[i] = bytes;
3225         }
3226       }
3227     }
3228   }
3229 
3230   // Give back the unused reserved pieces.
3231 
3232   for (int j = 0; j < i; ++j) {
3233     if (base[j] != NULL) {
3234       unmap_memory(base[j], size[j]);
3235     }
3236   }
3237 
3238   return (i < max_tries) ? requested_addr : NULL;
3239 }
3240 
3241 bool os::pd_release_memory(char* addr, size_t bytes) {
3242   size_t size = bytes;
3243   return munmap(addr, size) == 0;
3244 }
3245 
3246 static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
3247   assert(addr == (char*)align_size_down((uintptr_t)addr, os::vm_page_size()),
3248          "addr must be page aligned");
3249   int retVal = mprotect(addr, bytes, prot);
3250   return retVal == 0;
3251 }
3252 
3253 // Protect memory (Used to pass readonly pages through
3254 // JNI GetArray<type>Elements with empty arrays.)
3255 // Also, used for serialization page and for compressed oops null pointer
3256 // checking.
3257 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3258                         bool is_committed) {
3259   unsigned int p = 0;
3260   switch (prot) {
3261   case MEM_PROT_NONE: p = PROT_NONE; break;
3262   case MEM_PROT_READ: p = PROT_READ; break;
3263   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3264   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3265   default:
3266     ShouldNotReachHere();
3267   }
3268   // is_committed is unused.
3269   return solaris_mprotect(addr, bytes, p);
3270 }
3271 
3272 // guard_memory and unguard_memory only happens within stack guard pages.
3273 // Since ISM pertains only to the heap, guard and unguard memory should not
3274 /// happen with an ISM region.
3275 bool os::guard_memory(char* addr, size_t bytes) {
3276   return solaris_mprotect(addr, bytes, PROT_NONE);
3277 }
3278 
3279 bool os::unguard_memory(char* addr, size_t bytes) {
3280   return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
3281 }
3282 
3283 // Large page support
3284 static size_t _large_page_size = 0;
3285 
3286 // Insertion sort for small arrays (descending order).
3287 static void insertion_sort_descending(size_t* array, int len) {
3288   for (int i = 0; i < len; i++) {
3289     size_t val = array[i];
3290     for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
3291       size_t tmp = array[key];
3292       array[key] = array[key - 1];
3293       array[key - 1] = tmp;
3294     }
3295   }
3296 }
3297 
3298 bool os::Solaris::mpss_sanity_check(bool warn, size_t* page_size) {
3299   const unsigned int usable_count = VM_Version::page_size_count();
3300   if (usable_count == 1) {
3301     return false;
3302   }
3303 
3304   // Find the right getpagesizes interface.  When solaris 11 is the minimum
3305   // build platform, getpagesizes() (without the '2') can be called directly.
3306   typedef int (*gps_t)(size_t[], int);
3307   gps_t gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes2"));
3308   if (gps_func == NULL) {
3309     gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes"));
3310     if (gps_func == NULL) {
3311       if (warn) {
3312         warning("MPSS is not supported by the operating system.");
3313       }
3314       return false;
3315     }
3316   }
3317 
3318   // Fill the array of page sizes.
3319   int n = (*gps_func)(_page_sizes, page_sizes_max);
3320   assert(n > 0, "Solaris bug?");
3321 
3322   if (n == page_sizes_max) {
3323     // Add a sentinel value (necessary only if the array was completely filled
3324     // since it is static (zeroed at initialization)).
3325     _page_sizes[--n] = 0;
3326     DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
3327   }
3328   assert(_page_sizes[n] == 0, "missing sentinel");
3329   trace_page_sizes("available page sizes", _page_sizes, n);
3330 
3331   if (n == 1) return false;     // Only one page size available.
3332 
3333   // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
3334   // select up to usable_count elements.  First sort the array, find the first
3335   // acceptable value, then copy the usable sizes to the top of the array and
3336   // trim the rest.  Make sure to include the default page size :-).
3337   //
3338   // A better policy could get rid of the 4M limit by taking the sizes of the
3339   // important VM memory regions (java heap and possibly the code cache) into
3340   // account.
3341   insertion_sort_descending(_page_sizes, n);
3342   const size_t size_limit =
3343     FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
3344   int beg;
3345   for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */ ;
3346   const int end = MIN2((int)usable_count, n) - 1;
3347   for (int cur = 0; cur < end; ++cur, ++beg) {
3348     _page_sizes[cur] = _page_sizes[beg];
3349   }
3350   _page_sizes[end] = vm_page_size();
3351   _page_sizes[end + 1] = 0;
3352 
3353   if (_page_sizes[end] > _page_sizes[end - 1]) {
3354     // Default page size is not the smallest; sort again.
3355     insertion_sort_descending(_page_sizes, end + 1);
3356   }
3357   *page_size = _page_sizes[0];
3358 
3359   trace_page_sizes("usable page sizes", _page_sizes, end + 1);
3360   return true;
3361 }
3362 
3363 void os::large_page_init() {
3364   if (UseLargePages) {
3365     // print a warning if any large page related flag is specified on command line
3366     bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
3367                            !FLAG_IS_DEFAULT(LargePageSizeInBytes);
3368 
3369     UseLargePages = Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
3370   }
3371 }
3372 
3373 bool os::Solaris::setup_large_pages(caddr_t start, size_t bytes, size_t align) {
3374   // Signal to OS that we want large pages for addresses
3375   // from addr, addr + bytes
3376   struct memcntl_mha mpss_struct;
3377   mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
3378   mpss_struct.mha_pagesize = align;
3379   mpss_struct.mha_flags = 0;
3380   // Upon successful completion, memcntl() returns 0
3381   if (memcntl(start, bytes, MC_HAT_ADVISE, (caddr_t) &mpss_struct, 0, 0)) {
3382     debug_only(warning("Attempt to use MPSS failed."));
3383     return false;
3384   }
3385   return true;
3386 }
3387 
3388 char* os::reserve_memory_special(size_t size, size_t alignment, char* addr, bool exec) {
3389   fatal("os::reserve_memory_special should not be called on Solaris.");
3390   return NULL;
3391 }
3392 
3393 bool os::release_memory_special(char* base, size_t bytes) {
3394   fatal("os::release_memory_special should not be called on Solaris.");
3395   return false;
3396 }
3397 
3398 size_t os::large_page_size() {
3399   return _large_page_size;
3400 }
3401 
3402 // MPSS allows application to commit large page memory on demand; with ISM
3403 // the entire memory region must be allocated as shared memory.
3404 bool os::can_commit_large_page_memory() {
3405   return true;
3406 }
3407 
3408 bool os::can_execute_large_page_memory() {
3409   return true;
3410 }
3411 
3412 static int os_sleep(jlong millis, bool interruptible) {
3413   const jlong limit = INT_MAX;
3414   jlong prevtime;
3415   int res;
3416 
3417   while (millis > limit) {
3418     if ((res = os_sleep(limit, interruptible)) != OS_OK)
3419       return res;
3420     millis -= limit;
3421   }
3422 
3423   // Restart interrupted polls with new parameters until the proper delay
3424   // has been completed.
3425 
3426   prevtime = getTimeMillis();
3427 
3428   while (millis > 0) {
3429     jlong newtime;
3430 
3431     if (!interruptible) {
3432       // Following assert fails for os::yield_all:
3433       // assert(!thread->is_Java_thread(), "must not be java thread");
3434       res = poll(NULL, 0, millis);
3435     } else {
3436       JavaThread *jt = JavaThread::current();
3437 
3438       INTERRUPTIBLE_NORESTART_VM_ALWAYS(poll(NULL, 0, millis), res, jt,
3439         os::Solaris::clear_interrupted);
3440     }
3441 
3442     // INTERRUPTIBLE_NORESTART_VM_ALWAYS returns res == OS_INTRPT for
3443     // thread.Interrupt.
3444 
3445     // See c/r 6751923. Poll can return 0 before time
3446     // has elapsed if time is set via clock_settime (as NTP does).
3447     // res == 0 if poll timed out (see man poll RETURN VALUES)
3448     // using the logic below checks that we really did
3449     // sleep at least "millis" if not we'll sleep again.
3450     if( ( res == 0 ) || ((res == OS_ERR) && (errno == EINTR))) {
3451       newtime = getTimeMillis();
3452       assert(newtime >= prevtime, "time moving backwards");
3453     /* Doing prevtime and newtime in microseconds doesn't help precision,
3454        and trying to round up to avoid lost milliseconds can result in a
3455        too-short delay. */
3456       millis -= newtime - prevtime;
3457       if(millis <= 0)
3458         return OS_OK;
3459       prevtime = newtime;
3460     } else
3461       return res;
3462   }
3463 
3464   return OS_OK;
3465 }
3466 
3467 // Read calls from inside the vm need to perform state transitions
3468 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3469   INTERRUPTIBLE_RETURN_INT_VM(::read(fd, buf, nBytes), os::Solaris::clear_interrupted);
3470 }
3471 
3472 size_t os::restartable_read(int fd, void *buf, unsigned int nBytes) {
3473   INTERRUPTIBLE_RETURN_INT(::read(fd, buf, nBytes), os::Solaris::clear_interrupted);
3474 }
3475 
3476 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3477   assert(thread == Thread::current(),  "thread consistency check");
3478 
3479   // TODO-FIXME: this should be removed.
3480   // On Solaris machines (especially 2.5.1) we found that sometimes the VM gets into a live lock
3481   // situation with a JavaThread being starved out of a lwp. The kernel doesn't seem to generate
3482   // a SIGWAITING signal which would enable the threads library to create a new lwp for the starving
3483   // thread. We suspect that because the Watcher thread keeps waking up at periodic intervals the kernel
3484   // is fooled into believing that the system is making progress. In the code below we block the
3485   // the watcher thread while safepoint is in progress so that it would not appear as though the
3486   // system is making progress.
3487   if (!Solaris::T2_libthread() &&
3488       thread->is_Watcher_thread() && SafepointSynchronize::is_synchronizing() && !Arguments::has_profile()) {
3489     // We now try to acquire the threads lock. Since this lock is held by the VM thread during
3490     // the entire safepoint, the watcher thread will  line up here during the safepoint.
3491     Threads_lock->lock_without_safepoint_check();
3492     Threads_lock->unlock();
3493   }
3494 
3495   if (thread->is_Java_thread()) {
3496     // This is a JavaThread so we honor the _thread_blocked protocol
3497     // even for sleeps of 0 milliseconds. This was originally done
3498     // as a workaround for bug 4338139. However, now we also do it
3499     // to honor the suspend-equivalent protocol.
3500 
3501     JavaThread *jt = (JavaThread *) thread;
3502     ThreadBlockInVM tbivm(jt);
3503 
3504     jt->set_suspend_equivalent();
3505     // cleared by handle_special_suspend_equivalent_condition() or
3506     // java_suspend_self() via check_and_wait_while_suspended()
3507 
3508     int ret_code;
3509     if (millis <= 0) {
3510       thr_yield();
3511       ret_code = 0;
3512     } else {
3513       // The original sleep() implementation did not create an
3514       // OSThreadWaitState helper for sleeps of 0 milliseconds.
3515       // I'm preserving that decision for now.
3516       OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3517 
3518       ret_code = os_sleep(millis, interruptible);
3519     }
3520 
3521     // were we externally suspended while we were waiting?
3522     jt->check_and_wait_while_suspended();
3523 
3524     return ret_code;
3525   }
3526 
3527   // non-JavaThread from this point on:
3528 
3529   if (millis <= 0) {
3530     thr_yield();
3531     return 0;
3532   }
3533 
3534   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3535 
3536   return os_sleep(millis, interruptible);
3537 }
3538 
3539 int os::naked_sleep() {
3540   // %% make the sleep time an integer flag. for now use 1 millisec.
3541   return os_sleep(1, false);
3542 }
3543 
3544 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3545 void os::infinite_sleep() {
3546   while (true) {    // sleep forever ...
3547     ::sleep(100);   // ... 100 seconds at a time
3548   }
3549 }
3550 
3551 // Used to convert frequent JVM_Yield() to nops
3552 bool os::dont_yield() {
3553   if (DontYieldALot) {
3554     static hrtime_t last_time = 0;
3555     hrtime_t diff = getTimeNanos() - last_time;
3556 
3557     if (diff < DontYieldALotInterval * 1000000)
3558       return true;
3559 
3560     last_time += diff;
3561 
3562     return false;
3563   }
3564   else {
3565     return false;
3566   }
3567 }
3568 
3569 // Caveat: Solaris os::yield() causes a thread-state transition whereas
3570 // the linux and win32 implementations do not.  This should be checked.
3571 
3572 void os::yield() {
3573   // Yields to all threads with same or greater priority
3574   os::sleep(Thread::current(), 0, false);
3575 }
3576 
3577 // Note that yield semantics are defined by the scheduling class to which
3578 // the thread currently belongs.  Typically, yield will _not yield to
3579 // other equal or higher priority threads that reside on the dispatch queues
3580 // of other CPUs.
3581 
3582 os::YieldResult os::NakedYield() { thr_yield(); return os::YIELD_UNKNOWN; }
3583 
3584 
3585 // On Solaris we found that yield_all doesn't always yield to all other threads.
3586 // There have been cases where there is a thread ready to execute but it doesn't
3587 // get an lwp as the VM thread continues to spin with sleeps of 1 millisecond.
3588 // The 1 millisecond wait doesn't seem long enough for the kernel to issue a
3589 // SIGWAITING signal which will cause a new lwp to be created. So we count the
3590 // number of times yield_all is called in the one loop and increase the sleep
3591 // time after 8 attempts. If this fails too we increase the concurrency level
3592 // so that the starving thread would get an lwp
3593 
3594 void os::yield_all(int attempts) {
3595   // Yields to all threads, including threads with lower priorities
3596   if (attempts == 0) {
3597     os::sleep(Thread::current(), 1, false);
3598   } else {
3599     int iterations = attempts % 30;
3600     if (iterations == 0 && !os::Solaris::T2_libthread()) {
3601       // thr_setconcurrency and _getconcurrency make sense only under T1.
3602       int noofLWPS = thr_getconcurrency();
3603       if (noofLWPS < (Threads::number_of_threads() + 2)) {
3604         thr_setconcurrency(thr_getconcurrency() + 1);
3605       }
3606     } else if (iterations < 25) {
3607       os::sleep(Thread::current(), 1, false);
3608     } else {
3609       os::sleep(Thread::current(), 10, false);
3610     }
3611   }
3612 }
3613 
3614 // Called from the tight loops to possibly influence time-sharing heuristics
3615 void os::loop_breaker(int attempts) {
3616   os::yield_all(attempts);
3617 }
3618 
3619 
3620 // Interface for setting lwp priorities.  If we are using T2 libthread,
3621 // which forces the use of BoundThreads or we manually set UseBoundThreads,
3622 // all of our threads will be assigned to real lwp's.  Using the thr_setprio
3623 // function is meaningless in this mode so we must adjust the real lwp's priority
3624 // The routines below implement the getting and setting of lwp priorities.
3625 //
3626 // Note: There are three priority scales used on Solaris.  Java priotities
3627 //       which range from 1 to 10, libthread "thr_setprio" scale which range
3628 //       from 0 to 127, and the current scheduling class of the process we
3629 //       are running in.  This is typically from -60 to +60.
3630 //       The setting of the lwp priorities in done after a call to thr_setprio
3631 //       so Java priorities are mapped to libthread priorities and we map from
3632 //       the latter to lwp priorities.  We don't keep priorities stored in
3633 //       Java priorities since some of our worker threads want to set priorities
3634 //       higher than all Java threads.
3635 //
3636 // For related information:
3637 // (1)  man -s 2 priocntl
3638 // (2)  man -s 4 priocntl
3639 // (3)  man dispadmin
3640 // =    librt.so
3641 // =    libthread/common/rtsched.c - thrp_setlwpprio().
3642 // =    ps -cL <pid> ... to validate priority.
3643 // =    sched_get_priority_min and _max
3644 //              pthread_create
3645 //              sched_setparam
3646 //              pthread_setschedparam
3647 //
3648 // Assumptions:
3649 // +    We assume that all threads in the process belong to the same
3650 //              scheduling class.   IE. an homogenous process.
3651 // +    Must be root or in IA group to change change "interactive" attribute.
3652 //              Priocntl() will fail silently.  The only indication of failure is when
3653 //              we read-back the value and notice that it hasn't changed.
3654 // +    Interactive threads enter the runq at the head, non-interactive at the tail.
3655 // +    For RT, change timeslice as well.  Invariant:
3656 //              constant "priority integral"
3657 //              Konst == TimeSlice * (60-Priority)
3658 //              Given a priority, compute appropriate timeslice.
3659 // +    Higher numerical values have higher priority.
3660 
3661 // sched class attributes
3662 typedef struct {
3663         int   schedPolicy;              // classID
3664         int   maxPrio;
3665         int   minPrio;
3666 } SchedInfo;
3667 
3668 
3669 static SchedInfo tsLimits, iaLimits, rtLimits, fxLimits;
3670 
3671 #ifdef ASSERT
3672 static int  ReadBackValidate = 1;
3673 #endif
3674 static int  myClass     = 0;
3675 static int  myMin       = 0;
3676 static int  myMax       = 0;
3677 static int  myCur       = 0;
3678 static bool priocntl_enable = false;
3679 
3680 static const int criticalPrio = 60; // FX/60 is critical thread class/priority on T4
3681 static int java_MaxPriority_to_os_priority = 0; // Saved mapping
3682 
3683 
3684 // lwp_priocntl_init
3685 //
3686 // Try to determine the priority scale for our process.
3687 //
3688 // Return errno or 0 if OK.
3689 //
3690 static int lwp_priocntl_init () {
3691   int rslt;
3692   pcinfo_t ClassInfo;
3693   pcparms_t ParmInfo;
3694   int i;
3695 
3696   if (!UseThreadPriorities) return 0;
3697 
3698   // We are using Bound threads, we need to determine our priority ranges
3699   if (os::Solaris::T2_libthread() || UseBoundThreads) {
3700     // If ThreadPriorityPolicy is 1, switch tables
3701     if (ThreadPriorityPolicy == 1) {
3702       for (i = 0 ; i < CriticalPriority+1; i++)
3703         os::java_to_os_priority[i] = prio_policy1[i];
3704     }
3705     if (UseCriticalJavaThreadPriority) {
3706       // MaxPriority always maps to the FX scheduling class and criticalPrio.
3707       // See set_native_priority() and set_lwp_class_and_priority().
3708       // Save original MaxPriority mapping in case attempt to
3709       // use critical priority fails.
3710       java_MaxPriority_to_os_priority = os::java_to_os_priority[MaxPriority];
3711       // Set negative to distinguish from other priorities
3712       os::java_to_os_priority[MaxPriority] = -criticalPrio;
3713     }
3714   }
3715   // Not using Bound Threads, set to ThreadPolicy 1
3716   else {
3717     for ( i = 0 ; i < CriticalPriority+1; i++ ) {
3718       os::java_to_os_priority[i] = prio_policy1[i];
3719     }
3720     return 0;
3721   }
3722 
3723   // Get IDs for a set of well-known scheduling classes.
3724   // TODO-FIXME: GETCLINFO returns the current # of classes in the
3725   // the system.  We should have a loop that iterates over the
3726   // classID values, which are known to be "small" integers.
3727 
3728   strcpy(ClassInfo.pc_clname, "TS");
3729   ClassInfo.pc_cid = -1;
3730   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3731   if (rslt < 0) return errno;
3732   assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3733   tsLimits.schedPolicy = ClassInfo.pc_cid;
3734   tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3735   tsLimits.minPrio = -tsLimits.maxPrio;
3736 
3737   strcpy(ClassInfo.pc_clname, "IA");
3738   ClassInfo.pc_cid = -1;
3739   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3740   if (rslt < 0) return errno;
3741   assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3742   iaLimits.schedPolicy = ClassInfo.pc_cid;
3743   iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3744   iaLimits.minPrio = -iaLimits.maxPrio;
3745 
3746   strcpy(ClassInfo.pc_clname, "RT");
3747   ClassInfo.pc_cid = -1;
3748   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3749   if (rslt < 0) return errno;
3750   assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3751   rtLimits.schedPolicy = ClassInfo.pc_cid;
3752   rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3753   rtLimits.minPrio = 0;
3754 
3755   strcpy(ClassInfo.pc_clname, "FX");
3756   ClassInfo.pc_cid = -1;
3757   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3758   if (rslt < 0) return errno;
3759   assert(ClassInfo.pc_cid != -1, "cid for FX class is -1");
3760   fxLimits.schedPolicy = ClassInfo.pc_cid;
3761   fxLimits.maxPrio = ((fxinfo_t*)ClassInfo.pc_clinfo)->fx_maxupri;
3762   fxLimits.minPrio = 0;
3763 
3764   // Query our "current" scheduling class.
3765   // This will normally be IA, TS or, rarely, FX or RT.
3766   memset(&ParmInfo, 0, sizeof(ParmInfo));
3767   ParmInfo.pc_cid = PC_CLNULL;
3768   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3769   if (rslt < 0) return errno;
3770   myClass = ParmInfo.pc_cid;
3771 
3772   // We now know our scheduling classId, get specific information
3773   // about the class.
3774   ClassInfo.pc_cid = myClass;
3775   ClassInfo.pc_clname[0] = 0;
3776   rslt = priocntl((idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo);
3777   if (rslt < 0) return errno;
3778 
3779   if (ThreadPriorityVerbose) {
3780     tty->print_cr("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3781   }
3782 
3783   memset(&ParmInfo, 0, sizeof(pcparms_t));
3784   ParmInfo.pc_cid = PC_CLNULL;
3785   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3786   if (rslt < 0) return errno;
3787 
3788   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3789     myMin = rtLimits.minPrio;
3790     myMax = rtLimits.maxPrio;
3791   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3792     iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3793     myMin = iaLimits.minPrio;
3794     myMax = iaLimits.maxPrio;
3795     myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3796   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3797     tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3798     myMin = tsLimits.minPrio;
3799     myMax = tsLimits.maxPrio;
3800     myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3801   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3802     fxparms_t *fxInfo = (fxparms_t*)ParmInfo.pc_clparms;
3803     myMin = fxLimits.minPrio;
3804     myMax = fxLimits.maxPrio;
3805     myMax = MIN2(myMax, (int)fxInfo->fx_uprilim);       // clamp - restrict
3806   } else {
3807     // No clue - punt
3808     if (ThreadPriorityVerbose)
3809       tty->print_cr ("Unknown scheduling class: %s ... \n", ClassInfo.pc_clname);
3810     return EINVAL;      // no clue, punt
3811   }
3812 
3813   if (ThreadPriorityVerbose) {
3814     tty->print_cr ("Thread priority Range: [%d..%d]\n", myMin, myMax);
3815   }
3816 
3817   priocntl_enable = true;  // Enable changing priorities
3818   return 0;
3819 }
3820 
3821 #define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3822 #define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3823 #define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3824 #define FXPRI(x)        ((fxparms_t *)((x).pc_clparms))
3825 
3826 
3827 // scale_to_lwp_priority
3828 //
3829 // Convert from the libthread "thr_setprio" scale to our current
3830 // lwp scheduling class scale.
3831 //
3832 static
3833 int     scale_to_lwp_priority (int rMin, int rMax, int x)
3834 {
3835   int v;
3836 
3837   if (x == 127) return rMax;            // avoid round-down
3838     v = (((x*(rMax-rMin)))/128)+rMin;
3839   return v;
3840 }
3841 
3842 
3843 // set_lwp_class_and_priority
3844 //
3845 // Set the class and priority of the lwp.  This call should only
3846 // be made when using bound threads (T2 threads are bound by default).
3847 //
3848 int set_lwp_class_and_priority(int ThreadID, int lwpid,
3849                                int newPrio, int new_class, bool scale) {
3850   int rslt;
3851   int Actual, Expected, prv;
3852   pcparms_t ParmInfo;                   // for GET-SET
3853 #ifdef ASSERT
3854   pcparms_t ReadBack;                   // for readback
3855 #endif
3856 
3857   // Set priority via PC_GETPARMS, update, PC_SETPARMS
3858   // Query current values.
3859   // TODO: accelerate this by eliminating the PC_GETPARMS call.
3860   // Cache "pcparms_t" in global ParmCache.
3861   // TODO: elide set-to-same-value
3862 
3863   // If something went wrong on init, don't change priorities.
3864   if ( !priocntl_enable ) {
3865     if (ThreadPriorityVerbose)
3866       tty->print_cr("Trying to set priority but init failed, ignoring");
3867     return EINVAL;
3868   }
3869 
3870   // If lwp hasn't started yet, just return
3871   // the _start routine will call us again.
3872   if ( lwpid <= 0 ) {
3873     if (ThreadPriorityVerbose) {
3874       tty->print_cr ("deferring the set_lwp_class_and_priority of thread "
3875                      INTPTR_FORMAT " to %d, lwpid not set",
3876                      ThreadID, newPrio);
3877     }
3878     return 0;
3879   }
3880 
3881   if (ThreadPriorityVerbose) {
3882     tty->print_cr ("set_lwp_class_and_priority("
3883                    INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3884                    ThreadID, lwpid, newPrio);
3885   }
3886 
3887   memset(&ParmInfo, 0, sizeof(pcparms_t));
3888   ParmInfo.pc_cid = PC_CLNULL;
3889   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3890   if (rslt < 0) return errno;
3891 
3892   int cur_class = ParmInfo.pc_cid;
3893   ParmInfo.pc_cid = (id_t)new_class;
3894 
3895   if (new_class == rtLimits.schedPolicy) {
3896     rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3897     rtInfo->rt_pri     = scale ? scale_to_lwp_priority(rtLimits.minPrio,
3898                                                        rtLimits.maxPrio, newPrio)
3899                                : newPrio;
3900     rtInfo->rt_tqsecs  = RT_NOCHANGE;
3901     rtInfo->rt_tqnsecs = RT_NOCHANGE;
3902     if (ThreadPriorityVerbose) {
3903       tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3904     }
3905   } else if (new_class == iaLimits.schedPolicy) {
3906     iaparms_t* iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3907     int maxClamped     = MIN2(iaLimits.maxPrio,
3908                               cur_class == new_class
3909                                 ? (int)iaInfo->ia_uprilim : iaLimits.maxPrio);
3910     iaInfo->ia_upri    = scale ? scale_to_lwp_priority(iaLimits.minPrio,
3911                                                        maxClamped, newPrio)
3912                                : newPrio;
3913     iaInfo->ia_uprilim = cur_class == new_class
3914                            ? IA_NOCHANGE : (pri_t)iaLimits.maxPrio;
3915     iaInfo->ia_mode    = IA_NOCHANGE;
3916     if (ThreadPriorityVerbose) {
3917       tty->print_cr("IA: [%d...%d] %d->%d\n",
3918                     iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3919     }
3920   } else if (new_class == tsLimits.schedPolicy) {
3921     tsparms_t* tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3922     int maxClamped     = MIN2(tsLimits.maxPrio,
3923                               cur_class == new_class
3924                                 ? (int)tsInfo->ts_uprilim : tsLimits.maxPrio);
3925     tsInfo->ts_upri    = scale ? scale_to_lwp_priority(tsLimits.minPrio,
3926                                                        maxClamped, newPrio)
3927                                : newPrio;
3928     tsInfo->ts_uprilim = cur_class == new_class
3929                            ? TS_NOCHANGE : (pri_t)tsLimits.maxPrio;
3930     if (ThreadPriorityVerbose) {
3931       tty->print_cr("TS: [%d...%d] %d->%d\n",
3932                     tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
3933     }
3934   } else if (new_class == fxLimits.schedPolicy) {
3935     fxparms_t* fxInfo  = (fxparms_t*)ParmInfo.pc_clparms;
3936     int maxClamped     = MIN2(fxLimits.maxPrio,
3937                               cur_class == new_class
3938                                 ? (int)fxInfo->fx_uprilim : fxLimits.maxPrio);
3939     fxInfo->fx_upri    = scale ? scale_to_lwp_priority(fxLimits.minPrio,
3940                                                        maxClamped, newPrio)
3941                                : newPrio;
3942     fxInfo->fx_uprilim = cur_class == new_class
3943                            ? FX_NOCHANGE : (pri_t)fxLimits.maxPrio;
3944     fxInfo->fx_tqsecs  = FX_NOCHANGE;
3945     fxInfo->fx_tqnsecs = FX_NOCHANGE;
3946     if (ThreadPriorityVerbose) {
3947       tty->print_cr("FX: [%d...%d] %d->%d\n",
3948                     fxLimits.minPrio, maxClamped, newPrio, fxInfo->fx_upri);
3949     }
3950   } else {
3951     if (ThreadPriorityVerbose) {
3952       tty->print_cr("Unknown new scheduling class %d\n", new_class);
3953     }
3954     return EINVAL;    // no clue, punt
3955   }
3956 
3957   rslt = priocntl(P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
3958   if (ThreadPriorityVerbose && rslt) {
3959     tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
3960   }
3961   if (rslt < 0) return errno;
3962 
3963 #ifdef ASSERT
3964   // Sanity check: read back what we just attempted to set.
3965   // In theory it could have changed in the interim ...
3966   //
3967   // The priocntl system call is tricky.
3968   // Sometimes it'll validate the priority value argument and
3969   // return EINVAL if unhappy.  At other times it fails silently.
3970   // Readbacks are prudent.
3971 
3972   if (!ReadBackValidate) return 0;
3973 
3974   memset(&ReadBack, 0, sizeof(pcparms_t));
3975   ReadBack.pc_cid = PC_CLNULL;
3976   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
3977   assert(rslt >= 0, "priocntl failed");
3978   Actual = Expected = 0xBAD;
3979   assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
3980   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3981     Actual   = RTPRI(ReadBack)->rt_pri;
3982     Expected = RTPRI(ParmInfo)->rt_pri;
3983   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3984     Actual   = IAPRI(ReadBack)->ia_upri;
3985     Expected = IAPRI(ParmInfo)->ia_upri;
3986   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3987     Actual   = TSPRI(ReadBack)->ts_upri;
3988     Expected = TSPRI(ParmInfo)->ts_upri;
3989   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3990     Actual   = FXPRI(ReadBack)->fx_upri;
3991     Expected = FXPRI(ParmInfo)->fx_upri;
3992   } else {
3993     if (ThreadPriorityVerbose) {
3994       tty->print_cr("set_lwp_class_and_priority: unexpected class in readback: %d\n",
3995                     ParmInfo.pc_cid);
3996     }
3997   }
3998 
3999   if (Actual != Expected) {
4000     if (ThreadPriorityVerbose) {
4001       tty->print_cr ("set_lwp_class_and_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
4002                      lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
4003     }
4004   }
4005 #endif
4006 
4007   return 0;
4008 }
4009 
4010 // Solaris only gives access to 128 real priorities at a time,
4011 // so we expand Java's ten to fill this range.  This would be better
4012 // if we dynamically adjusted relative priorities.
4013 //
4014 // The ThreadPriorityPolicy option allows us to select 2 different
4015 // priority scales.
4016 //
4017 // ThreadPriorityPolicy=0
4018 // Since the Solaris' default priority is MaximumPriority, we do not
4019 // set a priority lower than Max unless a priority lower than
4020 // NormPriority is requested.
4021 //
4022 // ThreadPriorityPolicy=1
4023 // This mode causes the priority table to get filled with
4024 // linear values.  NormPriority get's mapped to 50% of the
4025 // Maximum priority an so on.  This will cause VM threads
4026 // to get unfair treatment against other Solaris processes
4027 // which do not explicitly alter their thread priorities.
4028 //
4029 
4030 int os::java_to_os_priority[CriticalPriority + 1] = {
4031   -99999,         // 0 Entry should never be used
4032 
4033   0,              // 1 MinPriority
4034   32,             // 2
4035   64,             // 3
4036 
4037   96,             // 4
4038   127,            // 5 NormPriority
4039   127,            // 6
4040 
4041   127,            // 7
4042   127,            // 8
4043   127,            // 9 NearMaxPriority
4044 
4045   127,            // 10 MaxPriority
4046 
4047   -criticalPrio   // 11 CriticalPriority
4048 };
4049 
4050 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4051   OSThread* osthread = thread->osthread();
4052 
4053   // Save requested priority in case the thread hasn't been started
4054   osthread->set_native_priority(newpri);
4055 
4056   // Check for critical priority request
4057   bool fxcritical = false;
4058   if (newpri == -criticalPrio) {
4059     fxcritical = true;
4060     newpri = criticalPrio;
4061   }
4062 
4063   assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
4064   if (!UseThreadPriorities) return OS_OK;
4065 
4066   int status = 0;
4067 
4068   if (!fxcritical) {
4069     // Use thr_setprio only if we have a priority that thr_setprio understands
4070     status = thr_setprio(thread->osthread()->thread_id(), newpri);
4071   }
4072 
4073   if (os::Solaris::T2_libthread() ||
4074       (UseBoundThreads && osthread->is_vm_created())) {
4075     int lwp_status =
4076       set_lwp_class_and_priority(osthread->thread_id(),
4077                                  osthread->lwp_id(),
4078                                  newpri,
4079                                  fxcritical ? fxLimits.schedPolicy : myClass,
4080                                  !fxcritical);
4081     if (lwp_status != 0 && fxcritical) {
4082       // Try again, this time without changing the scheduling class
4083       newpri = java_MaxPriority_to_os_priority;
4084       lwp_status = set_lwp_class_and_priority(osthread->thread_id(),
4085                                               osthread->lwp_id(),
4086                                               newpri, myClass, false);
4087     }
4088     status |= lwp_status;
4089   }
4090   return (status == 0) ? OS_OK : OS_ERR;
4091 }
4092 
4093 
4094 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4095   int p;
4096   if ( !UseThreadPriorities ) {
4097     *priority_ptr = NormalPriority;
4098     return OS_OK;
4099   }
4100   int status = thr_getprio(thread->osthread()->thread_id(), &p);
4101   if (status != 0) {
4102     return OS_ERR;
4103   }
4104   *priority_ptr = p;
4105   return OS_OK;
4106 }
4107 
4108 
4109 // Hint to the underlying OS that a task switch would not be good.
4110 // Void return because it's a hint and can fail.
4111 void os::hint_no_preempt() {
4112   schedctl_start(schedctl_init());
4113 }
4114 
4115 static void resume_clear_context(OSThread *osthread) {
4116   osthread->set_ucontext(NULL);
4117 }
4118 
4119 static void suspend_save_context(OSThread *osthread, ucontext_t* context) {
4120   osthread->set_ucontext(context);
4121 }
4122 
4123 static Semaphore sr_semaphore;
4124 
4125 void os::Solaris::SR_handler(Thread* thread, ucontext_t* uc) {
4126   // Save and restore errno to avoid confusing native code with EINTR
4127   // after sigsuspend.
4128   int old_errno = errno;
4129 
4130   OSThread* osthread = thread->osthread();
4131   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4132 
4133   os::SuspendResume::State current = osthread->sr.state();
4134   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4135     suspend_save_context(osthread, uc);
4136 
4137     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4138     os::SuspendResume::State state = osthread->sr.suspended();
4139     if (state == os::SuspendResume::SR_SUSPENDED) {
4140       sigset_t suspend_set;  // signals for sigsuspend()
4141 
4142       // get current set of blocked signals and unblock resume signal
4143       thr_sigsetmask(SIG_BLOCK, NULL, &suspend_set);
4144       sigdelset(&suspend_set, os::Solaris::SIGasync());
4145 
4146       sr_semaphore.signal();
4147       // wait here until we are resumed
4148       while (1) {
4149         sigsuspend(&suspend_set);
4150 
4151         os::SuspendResume::State result = osthread->sr.running();
4152         if (result == os::SuspendResume::SR_RUNNING) {
4153           sr_semaphore.signal();
4154           break;
4155         }
4156       }
4157 
4158     } else if (state == os::SuspendResume::SR_RUNNING) {
4159       // request was cancelled, continue
4160     } else {
4161       ShouldNotReachHere();
4162     }
4163 
4164     resume_clear_context(osthread);
4165   } else if (current == os::SuspendResume::SR_RUNNING) {
4166     // request was cancelled, continue
4167   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4168     // ignore
4169   } else {
4170     // ignore
4171   }
4172 
4173   errno = old_errno;
4174 }
4175 
4176 
4177 void os::interrupt(Thread* thread) {
4178   assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
4179 
4180   OSThread* osthread = thread->osthread();
4181 
4182   int isInterrupted = osthread->interrupted();
4183   if (!isInterrupted) {
4184       osthread->set_interrupted(true);
4185       OrderAccess::fence();
4186       // os::sleep() is implemented with either poll (NULL,0,timeout) or
4187       // by parking on _SleepEvent.  If the former, thr_kill will unwedge
4188       // the sleeper by SIGINTR, otherwise the unpark() will wake the sleeper.
4189       ParkEvent * const slp = thread->_SleepEvent ;
4190       if (slp != NULL) slp->unpark() ;
4191   }
4192 
4193   // For JSR166:  unpark after setting status but before thr_kill -dl
4194   if (thread->is_Java_thread()) {
4195     ((JavaThread*)thread)->parker()->unpark();
4196   }
4197 
4198   // Handle interruptible wait() ...
4199   ParkEvent * const ev = thread->_ParkEvent ;
4200   if (ev != NULL) ev->unpark() ;
4201 
4202   // When events are used everywhere for os::sleep, then this thr_kill
4203   // will only be needed if UseVMInterruptibleIO is true.
4204 
4205   if (!isInterrupted) {
4206     int status = thr_kill(osthread->thread_id(), os::Solaris::SIGinterrupt());
4207     assert_status(status == 0, status, "thr_kill");
4208 
4209     // Bump thread interruption counter
4210     RuntimeService::record_thread_interrupt_signaled_count();
4211   }
4212 }
4213 
4214 
4215 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4216   assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
4217 
4218   OSThread* osthread = thread->osthread();
4219 
4220   bool res = osthread->interrupted();
4221 
4222   // NOTE that since there is no "lock" around these two operations,
4223   // there is the possibility that the interrupted flag will be
4224   // "false" but that the interrupt event will be set. This is
4225   // intentional. The effect of this is that Object.wait() will appear
4226   // to have a spurious wakeup, which is not harmful, and the
4227   // possibility is so rare that it is not worth the added complexity
4228   // to add yet another lock. It has also been recommended not to put
4229   // the interrupted flag into the os::Solaris::Event structure,
4230   // because it hides the issue.
4231   if (res && clear_interrupted) {
4232     osthread->set_interrupted(false);
4233   }
4234   return res;
4235 }
4236 
4237 
4238 void os::print_statistics() {
4239 }
4240 
4241 int os::message_box(const char* title, const char* message) {
4242   int i;
4243   fdStream err(defaultStream::error_fd());
4244   for (i = 0; i < 78; i++) err.print_raw("=");
4245   err.cr();
4246   err.print_raw_cr(title);
4247   for (i = 0; i < 78; i++) err.print_raw("-");
4248   err.cr();
4249   err.print_raw_cr(message);
4250   for (i = 0; i < 78; i++) err.print_raw("=");
4251   err.cr();
4252 
4253   char buf[16];
4254   // Prevent process from exiting upon "read error" without consuming all CPU
4255   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4256 
4257   return buf[0] == 'y' || buf[0] == 'Y';
4258 }
4259 
4260 static int sr_notify(OSThread* osthread) {
4261   int status = thr_kill(osthread->thread_id(), os::Solaris::SIGasync());
4262   assert_status(status == 0, status, "thr_kill");
4263   return status;
4264 }
4265 
4266 // "Randomly" selected value for how long we want to spin
4267 // before bailing out on suspending a thread, also how often
4268 // we send a signal to a thread we want to resume
4269 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4270 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4271 
4272 static bool do_suspend(OSThread* osthread) {
4273   assert(osthread->sr.is_running(), "thread should be running");
4274   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4275 
4276   // mark as suspended and send signal
4277   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4278     // failed to switch, state wasn't running?
4279     ShouldNotReachHere();
4280     return false;
4281   }
4282 
4283   if (sr_notify(osthread) != 0) {
4284     ShouldNotReachHere();
4285   }
4286 
4287   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4288   while (true) {
4289     if (sr_semaphore.timedwait(0, 2000 * NANOSECS_PER_MILLISEC)) {
4290       break;
4291     } else {
4292       // timeout
4293       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4294       if (cancelled == os::SuspendResume::SR_RUNNING) {
4295         return false;
4296       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4297         // make sure that we consume the signal on the semaphore as well
4298         sr_semaphore.wait();
4299         break;
4300       } else {
4301         ShouldNotReachHere();
4302         return false;
4303       }
4304     }
4305   }
4306 
4307   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4308   return true;
4309 }
4310 
4311 static void do_resume(OSThread* osthread) {
4312   assert(osthread->sr.is_suspended(), "thread should be suspended");
4313   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4314 
4315   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4316     // failed to switch to WAKEUP_REQUEST
4317     ShouldNotReachHere();
4318     return;
4319   }
4320 
4321   while (true) {
4322     if (sr_notify(osthread) == 0) {
4323       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4324         if (osthread->sr.is_running()) {
4325           return;
4326         }
4327       }
4328     } else {
4329       ShouldNotReachHere();
4330     }
4331   }
4332 
4333   guarantee(osthread->sr.is_running(), "Must be running!");
4334 }
4335 
4336 void os::SuspendedThreadTask::internal_do_task() {
4337   if (do_suspend(_thread->osthread())) {
4338     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4339     do_task(context);
4340     do_resume(_thread->osthread());
4341   }
4342 }
4343 
4344 class PcFetcher : public os::SuspendedThreadTask {
4345 public:
4346   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4347   ExtendedPC result();
4348 protected:
4349   void do_task(const os::SuspendedThreadTaskContext& context);
4350 private:
4351   ExtendedPC _epc;
4352 };
4353 
4354 ExtendedPC PcFetcher::result() {
4355   guarantee(is_done(), "task is not done yet.");
4356   return _epc;
4357 }
4358 
4359 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4360   Thread* thread = context.thread();
4361   OSThread* osthread = thread->osthread();
4362   if (osthread->ucontext() != NULL) {
4363     _epc = os::Solaris::ucontext_get_pc((ucontext_t *) context.ucontext());
4364   } else {
4365     // NULL context is unexpected, double-check this is the VMThread
4366     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4367   }
4368 }
4369 
4370 // A lightweight implementation that does not suspend the target thread and
4371 // thus returns only a hint. Used for profiling only!
4372 ExtendedPC os::get_thread_pc(Thread* thread) {
4373   // Make sure that it is called by the watcher and the Threads lock is owned.
4374   assert(Thread::current()->is_Watcher_thread(), "Must be watcher and own Threads_lock");
4375   // For now, is only used to profile the VM Thread
4376   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4377   PcFetcher fetcher(thread);
4378   fetcher.run();
4379   return fetcher.result();
4380 }
4381 
4382 
4383 // This does not do anything on Solaris. This is basically a hook for being
4384 // able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
4385 void os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, JavaCallArguments* args, Thread* thread) {
4386   f(value, method, args, thread);
4387 }
4388 
4389 // This routine may be used by user applications as a "hook" to catch signals.
4390 // The user-defined signal handler must pass unrecognized signals to this
4391 // routine, and if it returns true (non-zero), then the signal handler must
4392 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4393 // routine will never retun false (zero), but instead will execute a VM panic
4394 // routine kill the process.
4395 //
4396 // If this routine returns false, it is OK to call it again.  This allows
4397 // the user-defined signal handler to perform checks either before or after
4398 // the VM performs its own checks.  Naturally, the user code would be making
4399 // a serious error if it tried to handle an exception (such as a null check
4400 // or breakpoint) that the VM was generating for its own correct operation.
4401 //
4402 // This routine may recognize any of the following kinds of signals:
4403 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
4404 // os::Solaris::SIGasync
4405 // It should be consulted by handlers for any of those signals.
4406 // It explicitly does not recognize os::Solaris::SIGinterrupt
4407 //
4408 // The caller of this routine must pass in the three arguments supplied
4409 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4410 // field of the structure passed to sigaction().  This routine assumes that
4411 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4412 //
4413 // Note that the VM will print warnings if it detects conflicting signal
4414 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4415 //
4416 extern "C" JNIEXPORT int
4417 JVM_handle_solaris_signal(int signo, siginfo_t* siginfo, void* ucontext,
4418                           int abort_if_unrecognized);
4419 
4420 
4421 void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
4422   int orig_errno = errno;  // Preserve errno value over signal handler.
4423   JVM_handle_solaris_signal(sig, info, ucVoid, true);
4424   errno = orig_errno;
4425 }
4426 
4427 /* Do not delete - if guarantee is ever removed,  a signal handler (even empty)
4428    is needed to provoke threads blocked on IO to return an EINTR
4429    Note: this explicitly does NOT call JVM_handle_solaris_signal and
4430    does NOT participate in signal chaining due to requirement for
4431    NOT setting SA_RESTART to make EINTR work. */
4432 extern "C" void sigINTRHandler(int sig, siginfo_t* info, void* ucVoid) {
4433    if (UseSignalChaining) {
4434       struct sigaction *actp = os::Solaris::get_chained_signal_action(sig);
4435       if (actp && actp->sa_handler) {
4436         vm_exit_during_initialization("Signal chaining detected for VM interrupt signal, try -XX:+UseAltSigs");
4437       }
4438    }
4439 }
4440 
4441 // This boolean allows users to forward their own non-matching signals
4442 // to JVM_handle_solaris_signal, harmlessly.
4443 bool os::Solaris::signal_handlers_are_installed = false;
4444 
4445 // For signal-chaining
4446 bool os::Solaris::libjsig_is_loaded = false;
4447 typedef struct sigaction *(*get_signal_t)(int);
4448 get_signal_t os::Solaris::get_signal_action = NULL;
4449 
4450 struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
4451   struct sigaction *actp = NULL;
4452 
4453   if ((libjsig_is_loaded)  && (sig <= Maxlibjsigsigs)) {
4454     // Retrieve the old signal handler from libjsig
4455     actp = (*get_signal_action)(sig);
4456   }
4457   if (actp == NULL) {
4458     // Retrieve the preinstalled signal handler from jvm
4459     actp = get_preinstalled_handler(sig);
4460   }
4461 
4462   return actp;
4463 }
4464 
4465 static bool call_chained_handler(struct sigaction *actp, int sig,
4466                                  siginfo_t *siginfo, void *context) {
4467   // Call the old signal handler
4468   if (actp->sa_handler == SIG_DFL) {
4469     // It's more reasonable to let jvm treat it as an unexpected exception
4470     // instead of taking the default action.
4471     return false;
4472   } else if (actp->sa_handler != SIG_IGN) {
4473     if ((actp->sa_flags & SA_NODEFER) == 0) {
4474       // automaticlly block the signal
4475       sigaddset(&(actp->sa_mask), sig);
4476     }
4477 
4478     sa_handler_t hand;
4479     sa_sigaction_t sa;
4480     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4481     // retrieve the chained handler
4482     if (siginfo_flag_set) {
4483       sa = actp->sa_sigaction;
4484     } else {
4485       hand = actp->sa_handler;
4486     }
4487 
4488     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4489       actp->sa_handler = SIG_DFL;
4490     }
4491 
4492     // try to honor the signal mask
4493     sigset_t oset;
4494     thr_sigsetmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4495 
4496     // call into the chained handler
4497     if (siginfo_flag_set) {
4498       (*sa)(sig, siginfo, context);
4499     } else {
4500       (*hand)(sig);
4501     }
4502 
4503     // restore the signal mask
4504     thr_sigsetmask(SIG_SETMASK, &oset, 0);
4505   }
4506   // Tell jvm's signal handler the signal is taken care of.
4507   return true;
4508 }
4509 
4510 bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4511   bool chained = false;
4512   // signal-chaining
4513   if (UseSignalChaining) {
4514     struct sigaction *actp = get_chained_signal_action(sig);
4515     if (actp != NULL) {
4516       chained = call_chained_handler(actp, sig, siginfo, context);
4517     }
4518   }
4519   return chained;
4520 }
4521 
4522 struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
4523   assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4524   if (preinstalled_sigs[sig] != 0) {
4525     return &chainedsigactions[sig];
4526   }
4527   return NULL;
4528 }
4529 
4530 void os::Solaris::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4531 
4532   assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
4533   assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4534   chainedsigactions[sig] = oldAct;
4535   preinstalled_sigs[sig] = 1;
4536 }
4537 
4538 void os::Solaris::set_signal_handler(int sig, bool set_installed, bool oktochain) {
4539   // Check for overwrite.
4540   struct sigaction oldAct;
4541   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4542   void* oldhand = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4543                                       : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4544   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4545       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4546       oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
4547     if (AllowUserSignalHandlers || !set_installed) {
4548       // Do not overwrite; user takes responsibility to forward to us.
4549       return;
4550     } else if (UseSignalChaining) {
4551       if (oktochain) {
4552         // save the old handler in jvm
4553         save_preinstalled_handler(sig, oldAct);
4554       } else {
4555         vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal, try -XX:+UseAltSigs.");
4556       }
4557       // libjsig also interposes the sigaction() call below and saves the
4558       // old sigaction on it own.
4559     } else {
4560       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4561                     "%#lx for signal %d.", (long)oldhand, sig));
4562     }
4563   }
4564 
4565   struct sigaction sigAct;
4566   sigfillset(&(sigAct.sa_mask));
4567   sigAct.sa_handler = SIG_DFL;
4568 
4569   sigAct.sa_sigaction = signalHandler;
4570   // Handle SIGSEGV on alternate signal stack if
4571   // not using stack banging
4572   if (!UseStackBanging && sig == SIGSEGV) {
4573     sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
4574   // Interruptible i/o requires SA_RESTART cleared so EINTR
4575   // is returned instead of restarting system calls
4576   } else if (sig == os::Solaris::SIGinterrupt()) {
4577     sigemptyset(&sigAct.sa_mask);
4578     sigAct.sa_handler = NULL;
4579     sigAct.sa_flags = SA_SIGINFO;
4580     sigAct.sa_sigaction = sigINTRHandler;
4581   } else {
4582     sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
4583   }
4584   os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
4585 
4586   sigaction(sig, &sigAct, &oldAct);
4587 
4588   void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4589                                        : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4590   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4591 }
4592 
4593 
4594 #define DO_SIGNAL_CHECK(sig) \
4595   if (!sigismember(&check_signal_done, sig)) \
4596     os::Solaris::check_signal_handler(sig)
4597 
4598 // This method is a periodic task to check for misbehaving JNI applications
4599 // under CheckJNI, we can add any periodic checks here
4600 
4601 void os::run_periodic_checks() {
4602   // A big source of grief is hijacking virt. addr 0x0 on Solaris,
4603   // thereby preventing a NULL checks.
4604   if(!check_addr0_done) check_addr0_done = check_addr0(tty);
4605 
4606   if (check_signals == false) return;
4607 
4608   // SEGV and BUS if overridden could potentially prevent
4609   // generation of hs*.log in the event of a crash, debugging
4610   // such a case can be very challenging, so we absolutely
4611   // check for the following for a good measure:
4612   DO_SIGNAL_CHECK(SIGSEGV);
4613   DO_SIGNAL_CHECK(SIGILL);
4614   DO_SIGNAL_CHECK(SIGFPE);
4615   DO_SIGNAL_CHECK(SIGBUS);
4616   DO_SIGNAL_CHECK(SIGPIPE);
4617   DO_SIGNAL_CHECK(SIGXFSZ);
4618 
4619   // ReduceSignalUsage allows the user to override these handlers
4620   // see comments at the very top and jvm_solaris.h
4621   if (!ReduceSignalUsage) {
4622     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4623     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4624     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4625     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4626   }
4627 
4628   // See comments above for using JVM1/JVM2 and UseAltSigs
4629   DO_SIGNAL_CHECK(os::Solaris::SIGinterrupt());
4630   DO_SIGNAL_CHECK(os::Solaris::SIGasync());
4631 
4632 }
4633 
4634 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4635 
4636 static os_sigaction_t os_sigaction = NULL;
4637 
4638 void os::Solaris::check_signal_handler(int sig) {
4639   char buf[O_BUFLEN];
4640   address jvmHandler = NULL;
4641 
4642   struct sigaction act;
4643   if (os_sigaction == NULL) {
4644     // only trust the default sigaction, in case it has been interposed
4645     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4646     if (os_sigaction == NULL) return;
4647   }
4648 
4649   os_sigaction(sig, (struct sigaction*)NULL, &act);
4650 
4651   address thisHandler = (act.sa_flags & SA_SIGINFO)
4652     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4653     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4654 
4655 
4656   switch(sig) {
4657     case SIGSEGV:
4658     case SIGBUS:
4659     case SIGFPE:
4660     case SIGPIPE:
4661     case SIGXFSZ:
4662     case SIGILL:
4663       jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4664       break;
4665 
4666     case SHUTDOWN1_SIGNAL:
4667     case SHUTDOWN2_SIGNAL:
4668     case SHUTDOWN3_SIGNAL:
4669     case BREAK_SIGNAL:
4670       jvmHandler = (address)user_handler();
4671       break;
4672 
4673     default:
4674       int intrsig = os::Solaris::SIGinterrupt();
4675       int asynsig = os::Solaris::SIGasync();
4676 
4677       if (sig == intrsig) {
4678         jvmHandler = CAST_FROM_FN_PTR(address, sigINTRHandler);
4679       } else if (sig == asynsig) {
4680         jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4681       } else {
4682         return;
4683       }
4684       break;
4685   }
4686 
4687 
4688   if (thisHandler != jvmHandler) {
4689     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4690     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4691     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4692     // No need to check this sig any longer
4693     sigaddset(&check_signal_done, sig);
4694   } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
4695     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4696     tty->print("expected:" PTR32_FORMAT, os::Solaris::get_our_sigflags(sig));
4697     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4698     // No need to check this sig any longer
4699     sigaddset(&check_signal_done, sig);
4700   }
4701 
4702   // Print all the signal handler state
4703   if (sigismember(&check_signal_done, sig)) {
4704     print_signal_handlers(tty, buf, O_BUFLEN);
4705   }
4706 
4707 }
4708 
4709 void os::Solaris::install_signal_handlers() {
4710   bool libjsigdone = false;
4711   signal_handlers_are_installed = true;
4712 
4713   // signal-chaining
4714   typedef void (*signal_setting_t)();
4715   signal_setting_t begin_signal_setting = NULL;
4716   signal_setting_t end_signal_setting = NULL;
4717   begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4718                                         dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4719   if (begin_signal_setting != NULL) {
4720     end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4721                                         dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4722     get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4723                                        dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4724     get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
4725                                          dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
4726     libjsig_is_loaded = true;
4727     if (os::Solaris::get_libjsig_version != NULL) {
4728       libjsigversion =  (*os::Solaris::get_libjsig_version)();
4729     }
4730     assert(UseSignalChaining, "should enable signal-chaining");
4731   }
4732   if (libjsig_is_loaded) {
4733     // Tell libjsig jvm is setting signal handlers
4734     (*begin_signal_setting)();
4735   }
4736 
4737   set_signal_handler(SIGSEGV, true, true);
4738   set_signal_handler(SIGPIPE, true, true);
4739   set_signal_handler(SIGXFSZ, true, true);
4740   set_signal_handler(SIGBUS, true, true);
4741   set_signal_handler(SIGILL, true, true);
4742   set_signal_handler(SIGFPE, true, true);
4743 
4744 
4745   if (os::Solaris::SIGinterrupt() > OLDMAXSIGNUM || os::Solaris::SIGasync() > OLDMAXSIGNUM) {
4746 
4747     // Pre-1.4.1 Libjsig limited to signal chaining signals <= 32 so
4748     // can not register overridable signals which might be > 32
4749     if (libjsig_is_loaded && libjsigversion <= JSIG_VERSION_1_4_1) {
4750     // Tell libjsig jvm has finished setting signal handlers
4751       (*end_signal_setting)();
4752       libjsigdone = true;
4753     }
4754   }
4755 
4756   // Never ok to chain our SIGinterrupt
4757   set_signal_handler(os::Solaris::SIGinterrupt(), true, false);
4758   set_signal_handler(os::Solaris::SIGasync(), true, true);
4759 
4760   if (libjsig_is_loaded && !libjsigdone) {
4761     // Tell libjsig jvm finishes setting signal handlers
4762     (*end_signal_setting)();
4763   }
4764 
4765   // We don't activate signal checker if libjsig is in place, we trust ourselves
4766   // and if UserSignalHandler is installed all bets are off.
4767   // Log that signal checking is off only if -verbose:jni is specified.
4768   if (CheckJNICalls) {
4769     if (libjsig_is_loaded) {
4770       if (PrintJNIResolving) {
4771         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4772       }
4773       check_signals = false;
4774     }
4775     if (AllowUserSignalHandlers) {
4776       if (PrintJNIResolving) {
4777         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4778       }
4779       check_signals = false;
4780     }
4781   }
4782 }
4783 
4784 
4785 void report_error(const char* file_name, int line_no, const char* title, const char* format, ...);
4786 
4787 const char * signames[] = {
4788   "SIG0",
4789   "SIGHUP", "SIGINT", "SIGQUIT", "SIGILL", "SIGTRAP",
4790   "SIGABRT", "SIGEMT", "SIGFPE", "SIGKILL", "SIGBUS",
4791   "SIGSEGV", "SIGSYS", "SIGPIPE", "SIGALRM", "SIGTERM",
4792   "SIGUSR1", "SIGUSR2", "SIGCLD", "SIGPWR", "SIGWINCH",
4793   "SIGURG", "SIGPOLL", "SIGSTOP", "SIGTSTP", "SIGCONT",
4794   "SIGTTIN", "SIGTTOU", "SIGVTALRM", "SIGPROF", "SIGXCPU",
4795   "SIGXFSZ", "SIGWAITING", "SIGLWP", "SIGFREEZE", "SIGTHAW",
4796   "SIGCANCEL", "SIGLOST"
4797 };
4798 
4799 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4800   if (0 < exception_code && exception_code <= SIGRTMAX) {
4801     // signal
4802     if (exception_code < sizeof(signames)/sizeof(const char*)) {
4803        jio_snprintf(buf, size, "%s", signames[exception_code]);
4804     } else {
4805        jio_snprintf(buf, size, "SIG%d", exception_code);
4806     }
4807     return buf;
4808   } else {
4809     return NULL;
4810   }
4811 }
4812 
4813 // (Static) wrappers for the new libthread API
4814 int_fnP_thread_t_iP_uP_stack_tP_gregset_t os::Solaris::_thr_getstate;
4815 int_fnP_thread_t_i_gregset_t os::Solaris::_thr_setstate;
4816 int_fnP_thread_t_i os::Solaris::_thr_setmutator;
4817 int_fnP_thread_t os::Solaris::_thr_suspend_mutator;
4818 int_fnP_thread_t os::Solaris::_thr_continue_mutator;
4819 
4820 // (Static) wrapper for getisax(2) call.
4821 os::Solaris::getisax_func_t os::Solaris::_getisax = 0;
4822 
4823 // (Static) wrappers for the liblgrp API
4824 os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4825 os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4826 os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4827 os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4828 os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4829 os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4830 os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4831 os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4832 os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4833 
4834 // (Static) wrapper for meminfo() call.
4835 os::Solaris::meminfo_func_t os::Solaris::_meminfo = 0;
4836 
4837 static address resolve_symbol_lazy(const char* name) {
4838   address addr = (address) dlsym(RTLD_DEFAULT, name);
4839   if(addr == NULL) {
4840     // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4841     addr = (address) dlsym(RTLD_NEXT, name);
4842   }
4843   return addr;
4844 }
4845 
4846 static address resolve_symbol(const char* name) {
4847   address addr = resolve_symbol_lazy(name);
4848   if(addr == NULL) {
4849     fatal(dlerror());
4850   }
4851   return addr;
4852 }
4853 
4854 
4855 
4856 // isT2_libthread()
4857 //
4858 // Routine to determine if we are currently using the new T2 libthread.
4859 //
4860 // We determine if we are using T2 by reading /proc/self/lstatus and
4861 // looking for a thread with the ASLWP bit set.  If we find this status
4862 // bit set, we must assume that we are NOT using T2.  The T2 team
4863 // has approved this algorithm.
4864 //
4865 // We need to determine if we are running with the new T2 libthread
4866 // since setting native thread priorities is handled differently
4867 // when using this library.  All threads created using T2 are bound
4868 // threads. Calling thr_setprio is meaningless in this case.
4869 //
4870 bool isT2_libthread() {
4871   static prheader_t * lwpArray = NULL;
4872   static int lwpSize = 0;
4873   static int lwpFile = -1;
4874   lwpstatus_t * that;
4875   char lwpName [128];
4876   bool isT2 = false;
4877 
4878 #define ADR(x)  ((uintptr_t)(x))
4879 #define LWPINDEX(ary,ix)   ((lwpstatus_t *)(((ary)->pr_entsize * (ix)) + (ADR((ary) + 1))))
4880 
4881   lwpFile = ::open("/proc/self/lstatus", O_RDONLY, 0);
4882   if (lwpFile < 0) {
4883       if (ThreadPriorityVerbose) warning ("Couldn't open /proc/self/lstatus\n");
4884       return false;
4885   }
4886   lwpSize = 16*1024;
4887   for (;;) {
4888     ::lseek64 (lwpFile, 0, SEEK_SET);
4889     lwpArray = (prheader_t *)NEW_C_HEAP_ARRAY(char, lwpSize, mtInternal);
4890     if (::read(lwpFile, lwpArray, lwpSize) < 0) {
4891       if (ThreadPriorityVerbose) warning("Error reading /proc/self/lstatus\n");
4892       break;
4893     }
4894     if ((lwpArray->pr_nent * lwpArray->pr_entsize) <= lwpSize) {
4895        // We got a good snapshot - now iterate over the list.
4896       int aslwpcount = 0;
4897       for (int i = 0; i < lwpArray->pr_nent; i++ ) {
4898         that = LWPINDEX(lwpArray,i);
4899         if (that->pr_flags & PR_ASLWP) {
4900           aslwpcount++;
4901         }
4902       }
4903       if (aslwpcount == 0) isT2 = true;
4904       break;
4905     }
4906     lwpSize = lwpArray->pr_nent * lwpArray->pr_entsize;
4907     FREE_C_HEAP_ARRAY(char, lwpArray, mtInternal);  // retry.
4908   }
4909 
4910   FREE_C_HEAP_ARRAY(char, lwpArray, mtInternal);
4911   ::close (lwpFile);
4912   if (ThreadPriorityVerbose) {
4913     if (isT2) tty->print_cr("We are running with a T2 libthread\n");
4914     else tty->print_cr("We are not running with a T2 libthread\n");
4915   }
4916   return isT2;
4917 }
4918 
4919 
4920 void os::Solaris::libthread_init() {
4921   address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4922 
4923   // Determine if we are running with the new T2 libthread
4924   os::Solaris::set_T2_libthread(isT2_libthread());
4925 
4926   lwp_priocntl_init();
4927 
4928   // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4929   if(func == NULL) {
4930     func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4931     // Guarantee that this VM is running on an new enough OS (5.6 or
4932     // later) that it will have a new enough libthread.so.
4933     guarantee(func != NULL, "libthread.so is too old.");
4934   }
4935 
4936   // Initialize the new libthread getstate API wrappers
4937   func = resolve_symbol("thr_getstate");
4938   os::Solaris::set_thr_getstate(CAST_TO_FN_PTR(int_fnP_thread_t_iP_uP_stack_tP_gregset_t, func));
4939 
4940   func = resolve_symbol("thr_setstate");
4941   os::Solaris::set_thr_setstate(CAST_TO_FN_PTR(int_fnP_thread_t_i_gregset_t, func));
4942 
4943   func = resolve_symbol("thr_setmutator");
4944   os::Solaris::set_thr_setmutator(CAST_TO_FN_PTR(int_fnP_thread_t_i, func));
4945 
4946   func = resolve_symbol("thr_suspend_mutator");
4947   os::Solaris::set_thr_suspend_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4948 
4949   func = resolve_symbol("thr_continue_mutator");
4950   os::Solaris::set_thr_continue_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4951 
4952   int size;
4953   void (*handler_info_func)(address *, int *);
4954   handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4955   handler_info_func(&handler_start, &size);
4956   handler_end = handler_start + size;
4957 }
4958 
4959 
4960 int_fnP_mutex_tP os::Solaris::_mutex_lock;
4961 int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4962 int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4963 int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4964 int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4965 int os::Solaris::_mutex_scope = USYNC_THREAD;
4966 
4967 int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4968 int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4969 int_fnP_cond_tP os::Solaris::_cond_signal;
4970 int_fnP_cond_tP os::Solaris::_cond_broadcast;
4971 int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4972 int_fnP_cond_tP os::Solaris::_cond_destroy;
4973 int os::Solaris::_cond_scope = USYNC_THREAD;
4974 
4975 void os::Solaris::synchronization_init() {
4976   if(UseLWPSynchronization) {
4977     os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4978     os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4979     os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4980     os::Solaris::set_mutex_init(lwp_mutex_init);
4981     os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4982     os::Solaris::set_mutex_scope(USYNC_THREAD);
4983 
4984     os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4985     os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4986     os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4987     os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4988     os::Solaris::set_cond_init(lwp_cond_init);
4989     os::Solaris::set_cond_destroy(lwp_cond_destroy);
4990     os::Solaris::set_cond_scope(USYNC_THREAD);
4991   }
4992   else {
4993     os::Solaris::set_mutex_scope(USYNC_THREAD);
4994     os::Solaris::set_cond_scope(USYNC_THREAD);
4995 
4996     if(UsePthreads) {
4997       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4998       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4999       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
5000       os::Solaris::set_mutex_init(pthread_mutex_default_init);
5001       os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
5002 
5003       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
5004       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
5005       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
5006       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
5007       os::Solaris::set_cond_init(pthread_cond_default_init);
5008       os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
5009     }
5010     else {
5011       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
5012       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
5013       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
5014       os::Solaris::set_mutex_init(::mutex_init);
5015       os::Solaris::set_mutex_destroy(::mutex_destroy);
5016 
5017       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
5018       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
5019       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
5020       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
5021       os::Solaris::set_cond_init(::cond_init);
5022       os::Solaris::set_cond_destroy(::cond_destroy);
5023     }
5024   }
5025 }
5026 
5027 bool os::Solaris::liblgrp_init() {
5028   void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
5029   if (handle != NULL) {
5030     os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
5031     os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
5032     os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
5033     os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
5034     os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
5035     os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
5036     os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
5037     os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
5038                                        dlsym(handle, "lgrp_cookie_stale")));
5039 
5040     lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
5041     set_lgrp_cookie(c);
5042     return true;
5043   }
5044   return false;
5045 }
5046 
5047 void os::Solaris::misc_sym_init() {
5048   address func;
5049 
5050   // getisax
5051   func = resolve_symbol_lazy("getisax");
5052   if (func != NULL) {
5053     os::Solaris::_getisax = CAST_TO_FN_PTR(getisax_func_t, func);
5054   }
5055 
5056   // meminfo
5057   func = resolve_symbol_lazy("meminfo");
5058   if (func != NULL) {
5059     os::Solaris::set_meminfo(CAST_TO_FN_PTR(meminfo_func_t, func));
5060   }
5061 }
5062 
5063 uint_t os::Solaris::getisax(uint32_t* array, uint_t n) {
5064   assert(_getisax != NULL, "_getisax not set");
5065   return _getisax(array, n);
5066 }
5067 
5068 // int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
5069 typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
5070 static pset_getloadavg_type pset_getloadavg_ptr = NULL;
5071 
5072 void init_pset_getloadavg_ptr(void) {
5073   pset_getloadavg_ptr =
5074     (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
5075   if (PrintMiscellaneous && Verbose && pset_getloadavg_ptr == NULL) {
5076     warning("pset_getloadavg function not found");
5077   }
5078 }
5079 
5080 int os::Solaris::_dev_zero_fd = -1;
5081 
5082 // this is called _before_ the global arguments have been parsed
5083 void os::init(void) {
5084   _initial_pid = getpid();
5085 
5086   max_hrtime = first_hrtime = gethrtime();
5087 
5088   init_random(1234567);
5089 
5090   page_size = sysconf(_SC_PAGESIZE);
5091   if (page_size == -1)
5092     fatal(err_msg("os_solaris.cpp: os::init: sysconf failed (%s)",
5093                   strerror(errno)));
5094   init_page_sizes((size_t) page_size);
5095 
5096   Solaris::initialize_system_info();
5097 
5098   // Initialize misc. symbols as soon as possible, so we can use them
5099   // if we need them.
5100   Solaris::misc_sym_init();
5101 
5102   int fd = ::open("/dev/zero", O_RDWR);
5103   if (fd < 0) {
5104     fatal(err_msg("os::init: cannot open /dev/zero (%s)", strerror(errno)));
5105   } else {
5106     Solaris::set_dev_zero_fd(fd);
5107 
5108     // Close on exec, child won't inherit.
5109     fcntl(fd, F_SETFD, FD_CLOEXEC);
5110   }
5111 
5112   clock_tics_per_sec = CLK_TCK;
5113 
5114   // check if dladdr1() exists; dladdr1 can provide more information than
5115   // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
5116   // and is available on linker patches for 5.7 and 5.8.
5117   // libdl.so must have been loaded, this call is just an entry lookup
5118   void * hdl = dlopen("libdl.so", RTLD_NOW);
5119   if (hdl)
5120     dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
5121 
5122   // (Solaris only) this switches to calls that actually do locking.
5123   ThreadCritical::initialize();
5124 
5125   main_thread = thr_self();
5126 
5127   // Constant minimum stack size allowed. It must be at least
5128   // the minimum of what the OS supports (thr_min_stack()), and
5129   // enough to allow the thread to get to user bytecode execution.
5130   Solaris::min_stack_allowed = MAX2(thr_min_stack(), Solaris::min_stack_allowed);
5131   // If the pagesize of the VM is greater than 8K determine the appropriate
5132   // number of initial guard pages.  The user can change this with the
5133   // command line arguments, if needed.
5134   if (vm_page_size() > 8*K) {
5135     StackYellowPages = 1;
5136     StackRedPages = 1;
5137     StackShadowPages = round_to((StackShadowPages*8*K), vm_page_size()) / vm_page_size();
5138   }
5139 }
5140 
5141 // To install functions for atexit system call
5142 extern "C" {
5143   static void perfMemory_exit_helper() {
5144     perfMemory_exit();
5145   }
5146 }
5147 
5148 // this is called _after_ the global arguments have been parsed
5149 jint os::init_2(void) {
5150   // try to enable extended file IO ASAP, see 6431278
5151   os::Solaris::try_enable_extended_io();
5152 
5153   // Allocate a single page and mark it as readable for safepoint polling.  Also
5154   // use this first mmap call to check support for MAP_ALIGN.
5155   address polling_page = (address)Solaris::mmap_chunk((char*)page_size,
5156                                                       page_size,
5157                                                       MAP_PRIVATE | MAP_ALIGN,
5158                                                       PROT_READ);
5159   if (polling_page == NULL) {
5160     has_map_align = false;
5161     polling_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE,
5162                                                 PROT_READ);
5163   }
5164 
5165   os::set_polling_page(polling_page);
5166 
5167 #ifndef PRODUCT
5168   if( Verbose && PrintMiscellaneous )
5169     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
5170 #endif
5171 
5172   if (!UseMembar) {
5173     address mem_serialize_page = (address)Solaris::mmap_chunk( NULL, page_size, MAP_PRIVATE, PROT_READ | PROT_WRITE );
5174     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
5175     os::set_memory_serialize_page( mem_serialize_page );
5176 
5177 #ifndef PRODUCT
5178     if(Verbose && PrintMiscellaneous)
5179       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
5180 #endif
5181   }
5182 
5183   // Check minimum allowable stack size for thread creation and to initialize
5184   // the java system classes, including StackOverflowError - depends on page
5185   // size.  Add a page for compiler2 recursion in main thread.
5186   // Add in 2*BytesPerWord times page size to account for VM stack during
5187   // class initialization depending on 32 or 64 bit VM.
5188   os::Solaris::min_stack_allowed = MAX2(os::Solaris::min_stack_allowed,
5189             (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
5190                     2*BytesPerWord COMPILER2_PRESENT(+1)) * page_size);
5191 
5192   size_t threadStackSizeInBytes = ThreadStackSize * K;
5193   if (threadStackSizeInBytes != 0 &&
5194     threadStackSizeInBytes < os::Solaris::min_stack_allowed) {
5195     tty->print_cr("\nThe stack size specified is too small, Specify at least %dk",
5196                   os::Solaris::min_stack_allowed/K);
5197     return JNI_ERR;
5198   }
5199 
5200   // For 64kbps there will be a 64kb page size, which makes
5201   // the usable default stack size quite a bit less.  Increase the
5202   // stack for 64kb (or any > than 8kb) pages, this increases
5203   // virtual memory fragmentation (since we're not creating the
5204   // stack on a power of 2 boundary.  The real fix for this
5205   // should be to fix the guard page mechanism.
5206 
5207   if (vm_page_size() > 8*K) {
5208       threadStackSizeInBytes = (threadStackSizeInBytes != 0)
5209          ? threadStackSizeInBytes +
5210            ((StackYellowPages + StackRedPages) * vm_page_size())
5211          : 0;
5212       ThreadStackSize = threadStackSizeInBytes/K;
5213   }
5214 
5215   // Make the stack size a multiple of the page size so that
5216   // the yellow/red zones can be guarded.
5217   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
5218         vm_page_size()));
5219 
5220   Solaris::libthread_init();
5221 
5222   if (UseNUMA) {
5223     if (!Solaris::liblgrp_init()) {
5224       UseNUMA = false;
5225     } else {
5226       size_t lgrp_limit = os::numa_get_groups_num();
5227       int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit, mtInternal);
5228       size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
5229       FREE_C_HEAP_ARRAY(int, lgrp_ids, mtInternal);
5230       if (lgrp_num < 2) {
5231         // There's only one locality group, disable NUMA.
5232         UseNUMA = false;
5233       }
5234     }
5235     if (!UseNUMA && ForceNUMA) {
5236       UseNUMA = true;
5237     }
5238   }
5239 
5240   Solaris::signal_sets_init();
5241   Solaris::init_signal_mem();
5242   Solaris::install_signal_handlers();
5243 
5244   if (libjsigversion < JSIG_VERSION_1_4_1) {
5245     Maxlibjsigsigs = OLDMAXSIGNUM;
5246   }
5247 
5248   // initialize synchronization primitives to use either thread or
5249   // lwp synchronization (controlled by UseLWPSynchronization)
5250   Solaris::synchronization_init();
5251 
5252   if (MaxFDLimit) {
5253     // set the number of file descriptors to max. print out error
5254     // if getrlimit/setrlimit fails but continue regardless.
5255     struct rlimit nbr_files;
5256     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5257     if (status != 0) {
5258       if (PrintMiscellaneous && (Verbose || WizardMode))
5259         perror("os::init_2 getrlimit failed");
5260     } else {
5261       nbr_files.rlim_cur = nbr_files.rlim_max;
5262       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5263       if (status != 0) {
5264         if (PrintMiscellaneous && (Verbose || WizardMode))
5265           perror("os::init_2 setrlimit failed");
5266       }
5267     }
5268   }
5269 
5270   // Calculate theoretical max. size of Threads to guard gainst
5271   // artifical out-of-memory situations, where all available address-
5272   // space has been reserved by thread stacks. Default stack size is 1Mb.
5273   size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
5274     JavaThread::stack_size_at_create() : (1*K*K);
5275   assert(pre_thread_stack_size != 0, "Must have a stack");
5276   // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
5277   // we should start doing Virtual Memory banging. Currently when the threads will
5278   // have used all but 200Mb of space.
5279   size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
5280   Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
5281 
5282   // at-exit methods are called in the reverse order of their registration.
5283   // In Solaris 7 and earlier, atexit functions are called on return from
5284   // main or as a result of a call to exit(3C). There can be only 32 of
5285   // these functions registered and atexit() does not set errno. In Solaris
5286   // 8 and later, there is no limit to the number of functions registered
5287   // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
5288   // functions are called upon dlclose(3DL) in addition to return from main
5289   // and exit(3C).
5290 
5291   if (PerfAllowAtExitRegistration) {
5292     // only register atexit functions if PerfAllowAtExitRegistration is set.
5293     // atexit functions can be delayed until process exit time, which
5294     // can be problematic for embedded VM situations. Embedded VMs should
5295     // call DestroyJavaVM() to assure that VM resources are released.
5296 
5297     // note: perfMemory_exit_helper atexit function may be removed in
5298     // the future if the appropriate cleanup code can be added to the
5299     // VM_Exit VMOperation's doit method.
5300     if (atexit(perfMemory_exit_helper) != 0) {
5301       warning("os::init2 atexit(perfMemory_exit_helper) failed");
5302     }
5303   }
5304 
5305   // Init pset_loadavg function pointer
5306   init_pset_getloadavg_ptr();
5307 
5308   return JNI_OK;
5309 }
5310 
5311 void os::init_3(void) {
5312   return;
5313 }
5314 
5315 // Mark the polling page as unreadable
5316 void os::make_polling_page_unreadable(void) {
5317   if( mprotect((char *)_polling_page, page_size, PROT_NONE) != 0 )
5318     fatal("Could not disable polling page");
5319 };
5320 
5321 // Mark the polling page as readable
5322 void os::make_polling_page_readable(void) {
5323   if( mprotect((char *)_polling_page, page_size, PROT_READ) != 0 )
5324     fatal("Could not enable polling page");
5325 };
5326 
5327 // OS interface.
5328 
5329 bool os::check_heap(bool force) { return true; }
5330 
5331 typedef int (*vsnprintf_t)(char* buf, size_t count, const char* fmt, va_list argptr);
5332 static vsnprintf_t sol_vsnprintf = NULL;
5333 
5334 int local_vsnprintf(char* buf, size_t count, const char* fmt, va_list argptr) {
5335   if (!sol_vsnprintf) {
5336     //search  for the named symbol in the objects that were loaded after libjvm
5337     void* where = RTLD_NEXT;
5338     if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5339         sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5340     if (!sol_vsnprintf){
5341       //search  for the named symbol in the objects that were loaded before libjvm
5342       where = RTLD_DEFAULT;
5343       if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5344         sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5345       assert(sol_vsnprintf != NULL, "vsnprintf not found");
5346     }
5347   }
5348   return (*sol_vsnprintf)(buf, count, fmt, argptr);
5349 }
5350 
5351 
5352 // Is a (classpath) directory empty?
5353 bool os::dir_is_empty(const char* path) {
5354   DIR *dir = NULL;
5355   struct dirent *ptr;
5356 
5357   dir = opendir(path);
5358   if (dir == NULL) return true;
5359 
5360   /* Scan the directory */
5361   bool result = true;
5362   char buf[sizeof(struct dirent) + MAX_PATH];
5363   struct dirent *dbuf = (struct dirent *) buf;
5364   while (result && (ptr = readdir(dir, dbuf)) != NULL) {
5365     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5366       result = false;
5367     }
5368   }
5369   closedir(dir);
5370   return result;
5371 }
5372 
5373 // This code originates from JDK's sysOpen and open64_w
5374 // from src/solaris/hpi/src/system_md.c
5375 
5376 #ifndef O_DELETE
5377 #define O_DELETE 0x10000
5378 #endif
5379 
5380 // Open a file. Unlink the file immediately after open returns
5381 // if the specified oflag has the O_DELETE flag set.
5382 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5383 
5384 int os::open(const char *path, int oflag, int mode) {
5385   if (strlen(path) > MAX_PATH - 1) {
5386     errno = ENAMETOOLONG;
5387     return -1;
5388   }
5389   int fd;
5390   int o_delete = (oflag & O_DELETE);
5391   oflag = oflag & ~O_DELETE;
5392 
5393   fd = ::open64(path, oflag, mode);
5394   if (fd == -1) return -1;
5395 
5396   //If the open succeeded, the file might still be a directory
5397   {
5398     struct stat64 buf64;
5399     int ret = ::fstat64(fd, &buf64);
5400     int st_mode = buf64.st_mode;
5401 
5402     if (ret != -1) {
5403       if ((st_mode & S_IFMT) == S_IFDIR) {
5404         errno = EISDIR;
5405         ::close(fd);
5406         return -1;
5407       }
5408     } else {
5409       ::close(fd);
5410       return -1;
5411     }
5412   }
5413     /*
5414      * 32-bit Solaris systems suffer from:
5415      *
5416      * - an historical default soft limit of 256 per-process file
5417      *   descriptors that is too low for many Java programs.
5418      *
5419      * - a design flaw where file descriptors created using stdio
5420      *   fopen must be less than 256, _even_ when the first limit above
5421      *   has been raised.  This can cause calls to fopen (but not calls to
5422      *   open, for example) to fail mysteriously, perhaps in 3rd party
5423      *   native code (although the JDK itself uses fopen).  One can hardly
5424      *   criticize them for using this most standard of all functions.
5425      *
5426      * We attempt to make everything work anyways by:
5427      *
5428      * - raising the soft limit on per-process file descriptors beyond
5429      *   256
5430      *
5431      * - As of Solaris 10u4, we can request that Solaris raise the 256
5432      *   stdio fopen limit by calling function enable_extended_FILE_stdio.
5433      *   This is done in init_2 and recorded in enabled_extended_FILE_stdio
5434      *
5435      * - If we are stuck on an old (pre 10u4) Solaris system, we can
5436      *   workaround the bug by remapping non-stdio file descriptors below
5437      *   256 to ones beyond 256, which is done below.
5438      *
5439      * See:
5440      * 1085341: 32-bit stdio routines should support file descriptors >255
5441      * 6533291: Work around 32-bit Solaris stdio limit of 256 open files
5442      * 6431278: Netbeans crash on 32 bit Solaris: need to call
5443      *          enable_extended_FILE_stdio() in VM initialisation
5444      * Giri Mandalika's blog
5445      * http://technopark02.blogspot.com/2005_05_01_archive.html
5446      */
5447 #ifndef  _LP64
5448      if ((!enabled_extended_FILE_stdio) && fd < 256) {
5449          int newfd = ::fcntl(fd, F_DUPFD, 256);
5450          if (newfd != -1) {
5451              ::close(fd);
5452              fd = newfd;
5453          }
5454      }
5455 #endif // 32-bit Solaris
5456     /*
5457      * All file descriptors that are opened in the JVM and not
5458      * specifically destined for a subprocess should have the
5459      * close-on-exec flag set.  If we don't set it, then careless 3rd
5460      * party native code might fork and exec without closing all
5461      * appropriate file descriptors (e.g. as we do in closeDescriptors in
5462      * UNIXProcess.c), and this in turn might:
5463      *
5464      * - cause end-of-file to fail to be detected on some file
5465      *   descriptors, resulting in mysterious hangs, or
5466      *
5467      * - might cause an fopen in the subprocess to fail on a system
5468      *   suffering from bug 1085341.
5469      *
5470      * (Yes, the default setting of the close-on-exec flag is a Unix
5471      * design flaw)
5472      *
5473      * See:
5474      * 1085341: 32-bit stdio routines should support file descriptors >255
5475      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5476      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5477      */
5478 #ifdef FD_CLOEXEC
5479     {
5480         int flags = ::fcntl(fd, F_GETFD);
5481         if (flags != -1)
5482             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5483     }
5484 #endif
5485 
5486   if (o_delete != 0) {
5487     ::unlink(path);
5488   }
5489   return fd;
5490 }
5491 
5492 // create binary file, rewriting existing file if required
5493 int os::create_binary_file(const char* path, bool rewrite_existing) {
5494   int oflags = O_WRONLY | O_CREAT;
5495   if (!rewrite_existing) {
5496     oflags |= O_EXCL;
5497   }
5498   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5499 }
5500 
5501 // return current position of file pointer
5502 jlong os::current_file_offset(int fd) {
5503   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5504 }
5505 
5506 // move file pointer to the specified offset
5507 jlong os::seek_to_file_offset(int fd, jlong offset) {
5508   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5509 }
5510 
5511 jlong os::lseek(int fd, jlong offset, int whence) {
5512   return (jlong) ::lseek64(fd, offset, whence);
5513 }
5514 
5515 char * os::native_path(char *path) {
5516   return path;
5517 }
5518 
5519 int os::ftruncate(int fd, jlong length) {
5520   return ::ftruncate64(fd, length);
5521 }
5522 
5523 int os::fsync(int fd)  {
5524   RESTARTABLE_RETURN_INT(::fsync(fd));
5525 }
5526 
5527 int os::available(int fd, jlong *bytes) {
5528   jlong cur, end;
5529   int mode;
5530   struct stat64 buf64;
5531 
5532   if (::fstat64(fd, &buf64) >= 0) {
5533     mode = buf64.st_mode;
5534     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5535       /*
5536       * XXX: is the following call interruptible? If so, this might
5537       * need to go through the INTERRUPT_IO() wrapper as for other
5538       * blocking, interruptible calls in this file.
5539       */
5540       int n,ioctl_return;
5541 
5542       INTERRUPTIBLE(::ioctl(fd, FIONREAD, &n),ioctl_return,os::Solaris::clear_interrupted);
5543       if (ioctl_return>= 0) {
5544           *bytes = n;
5545         return 1;
5546       }
5547     }
5548   }
5549   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5550     return 0;
5551   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5552     return 0;
5553   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5554     return 0;
5555   }
5556   *bytes = end - cur;
5557   return 1;
5558 }
5559 
5560 // Map a block of memory.
5561 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5562                      char *addr, size_t bytes, bool read_only,
5563                      bool allow_exec) {
5564   int prot;
5565   int flags;
5566 
5567   if (read_only) {
5568     prot = PROT_READ;
5569     flags = MAP_SHARED;
5570   } else {
5571     prot = PROT_READ | PROT_WRITE;
5572     flags = MAP_PRIVATE;
5573   }
5574 
5575   if (allow_exec) {
5576     prot |= PROT_EXEC;
5577   }
5578 
5579   if (addr != NULL) {
5580     flags |= MAP_FIXED;
5581   }
5582 
5583   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5584                                      fd, file_offset);
5585   if (mapped_address == MAP_FAILED) {
5586     return NULL;
5587   }
5588   return mapped_address;
5589 }
5590 
5591 
5592 // Remap a block of memory.
5593 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5594                        char *addr, size_t bytes, bool read_only,
5595                        bool allow_exec) {
5596   // same as map_memory() on this OS
5597   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5598                         allow_exec);
5599 }
5600 
5601 
5602 // Unmap a block of memory.
5603 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5604   return munmap(addr, bytes) == 0;
5605 }
5606 
5607 void os::pause() {
5608   char filename[MAX_PATH];
5609   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5610     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5611   } else {
5612     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5613   }
5614 
5615   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5616   if (fd != -1) {
5617     struct stat buf;
5618     ::close(fd);
5619     while (::stat(filename, &buf) == 0) {
5620       (void)::poll(NULL, 0, 100);
5621     }
5622   } else {
5623     jio_fprintf(stderr,
5624       "Could not open pause file '%s', continuing immediately.\n", filename);
5625   }
5626 }
5627 
5628 #ifndef PRODUCT
5629 #ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5630 // Turn this on if you need to trace synch operations.
5631 // Set RECORD_SYNCH_LIMIT to a large-enough value,
5632 // and call record_synch_enable and record_synch_disable
5633 // around the computation of interest.
5634 
5635 void record_synch(char* name, bool returning);  // defined below
5636 
5637 class RecordSynch {
5638   char* _name;
5639  public:
5640   RecordSynch(char* name) :_name(name)
5641                  { record_synch(_name, false); }
5642   ~RecordSynch() { record_synch(_name,   true);  }
5643 };
5644 
5645 #define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
5646 extern "C" ret name params {                                    \
5647   typedef ret name##_t params;                                  \
5648   static name##_t* implem = NULL;                               \
5649   static int callcount = 0;                                     \
5650   if (implem == NULL) {                                         \
5651     implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
5652     if (implem == NULL)  fatal(dlerror());                      \
5653   }                                                             \
5654   ++callcount;                                                  \
5655   RecordSynch _rs(#name);                                       \
5656   inner;                                                        \
5657   return implem args;                                           \
5658 }
5659 // in dbx, examine callcounts this way:
5660 // for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
5661 
5662 #define CHECK_POINTER_OK(p) \
5663   (!Universe::is_fully_initialized() || !Universe::is_reserved_heap((oop)(p)))
5664 #define CHECK_MU \
5665   if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
5666 #define CHECK_CV \
5667   if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
5668 #define CHECK_P(p) \
5669   if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
5670 
5671 #define CHECK_MUTEX(mutex_op) \
5672 CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
5673 
5674 CHECK_MUTEX(   mutex_lock)
5675 CHECK_MUTEX(  _mutex_lock)
5676 CHECK_MUTEX( mutex_unlock)
5677 CHECK_MUTEX(_mutex_unlock)
5678 CHECK_MUTEX( mutex_trylock)
5679 CHECK_MUTEX(_mutex_trylock)
5680 
5681 #define CHECK_COND(cond_op) \
5682 CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU;CHECK_CV);
5683 
5684 CHECK_COND( cond_wait);
5685 CHECK_COND(_cond_wait);
5686 CHECK_COND(_cond_wait_cancel);
5687 
5688 #define CHECK_COND2(cond_op) \
5689 CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU;CHECK_CV);
5690 
5691 CHECK_COND2( cond_timedwait);
5692 CHECK_COND2(_cond_timedwait);
5693 CHECK_COND2(_cond_timedwait_cancel);
5694 
5695 // do the _lwp_* versions too
5696 #define mutex_t lwp_mutex_t
5697 #define cond_t  lwp_cond_t
5698 CHECK_MUTEX(  _lwp_mutex_lock)
5699 CHECK_MUTEX(  _lwp_mutex_unlock)
5700 CHECK_MUTEX(  _lwp_mutex_trylock)
5701 CHECK_MUTEX( __lwp_mutex_lock)
5702 CHECK_MUTEX( __lwp_mutex_unlock)
5703 CHECK_MUTEX( __lwp_mutex_trylock)
5704 CHECK_MUTEX(___lwp_mutex_lock)
5705 CHECK_MUTEX(___lwp_mutex_unlock)
5706 
5707 CHECK_COND(  _lwp_cond_wait);
5708 CHECK_COND( __lwp_cond_wait);
5709 CHECK_COND(___lwp_cond_wait);
5710 
5711 CHECK_COND2(  _lwp_cond_timedwait);
5712 CHECK_COND2( __lwp_cond_timedwait);
5713 #undef mutex_t
5714 #undef cond_t
5715 
5716 CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5717 CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5718 CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5719 CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5720 CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5721 CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5722 CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5723 CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5724 
5725 
5726 // recording machinery:
5727 
5728 enum { RECORD_SYNCH_LIMIT = 200 };
5729 char* record_synch_name[RECORD_SYNCH_LIMIT];
5730 void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
5731 bool record_synch_returning[RECORD_SYNCH_LIMIT];
5732 thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
5733 int record_synch_count = 0;
5734 bool record_synch_enabled = false;
5735 
5736 // in dbx, examine recorded data this way:
5737 // for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
5738 
5739 void record_synch(char* name, bool returning) {
5740   if (record_synch_enabled) {
5741     if (record_synch_count < RECORD_SYNCH_LIMIT) {
5742       record_synch_name[record_synch_count] = name;
5743       record_synch_returning[record_synch_count] = returning;
5744       record_synch_thread[record_synch_count] = thr_self();
5745       record_synch_arg0ptr[record_synch_count] = &name;
5746       record_synch_count++;
5747     }
5748     // put more checking code here:
5749     // ...
5750   }
5751 }
5752 
5753 void record_synch_enable() {
5754   // start collecting trace data, if not already doing so
5755   if (!record_synch_enabled)  record_synch_count = 0;
5756   record_synch_enabled = true;
5757 }
5758 
5759 void record_synch_disable() {
5760   // stop collecting trace data
5761   record_synch_enabled = false;
5762 }
5763 
5764 #endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5765 #endif // PRODUCT
5766 
5767 const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5768 const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
5769                                (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5770 
5771 
5772 // JVMTI & JVM monitoring and management support
5773 // The thread_cpu_time() and current_thread_cpu_time() are only
5774 // supported if is_thread_cpu_time_supported() returns true.
5775 // They are not supported on Solaris T1.
5776 
5777 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5778 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5779 // of a thread.
5780 //
5781 // current_thread_cpu_time() and thread_cpu_time(Thread *)
5782 // returns the fast estimate available on the platform.
5783 
5784 // hrtime_t gethrvtime() return value includes
5785 // user time but does not include system time
5786 jlong os::current_thread_cpu_time() {
5787   return (jlong) gethrvtime();
5788 }
5789 
5790 jlong os::thread_cpu_time(Thread *thread) {
5791   // return user level CPU time only to be consistent with
5792   // what current_thread_cpu_time returns.
5793   // thread_cpu_time_info() must be changed if this changes
5794   return os::thread_cpu_time(thread, false /* user time only */);
5795 }
5796 
5797 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5798   if (user_sys_cpu_time) {
5799     return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5800   } else {
5801     return os::current_thread_cpu_time();
5802   }
5803 }
5804 
5805 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5806   char proc_name[64];
5807   int count;
5808   prusage_t prusage;
5809   jlong lwp_time;
5810   int fd;
5811 
5812   sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
5813                      getpid(),
5814                      thread->osthread()->lwp_id());
5815   fd = ::open(proc_name, O_RDONLY);
5816   if ( fd == -1 ) return -1;
5817 
5818   do {
5819     count = ::pread(fd,
5820                   (void *)&prusage.pr_utime,
5821                   thr_time_size,
5822                   thr_time_off);
5823   } while (count < 0 && errno == EINTR);
5824   ::close(fd);
5825   if ( count < 0 ) return -1;
5826 
5827   if (user_sys_cpu_time) {
5828     // user + system CPU time
5829     lwp_time = (((jlong)prusage.pr_stime.tv_sec +
5830                  (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
5831                  (jlong)prusage.pr_stime.tv_nsec +
5832                  (jlong)prusage.pr_utime.tv_nsec;
5833   } else {
5834     // user level CPU time only
5835     lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
5836                 (jlong)prusage.pr_utime.tv_nsec;
5837   }
5838 
5839   return(lwp_time);
5840 }
5841 
5842 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5843   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5844   info_ptr->may_skip_backward = false;    // elapsed time not wall time
5845   info_ptr->may_skip_forward = false;     // elapsed time not wall time
5846   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5847 }
5848 
5849 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5850   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5851   info_ptr->may_skip_backward = false;    // elapsed time not wall time
5852   info_ptr->may_skip_forward = false;     // elapsed time not wall time
5853   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5854 }
5855 
5856 bool os::is_thread_cpu_time_supported() {
5857   if ( os::Solaris::T2_libthread() || UseBoundThreads ) {
5858     return true;
5859   } else {
5860     return false;
5861   }
5862 }
5863 
5864 // System loadavg support.  Returns -1 if load average cannot be obtained.
5865 // Return the load average for our processor set if the primitive exists
5866 // (Solaris 9 and later).  Otherwise just return system wide loadavg.
5867 int os::loadavg(double loadavg[], int nelem) {
5868   if (pset_getloadavg_ptr != NULL) {
5869     return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
5870   } else {
5871     return ::getloadavg(loadavg, nelem);
5872   }
5873 }
5874 
5875 //---------------------------------------------------------------------------------
5876 
5877 bool os::find(address addr, outputStream* st) {
5878   Dl_info dlinfo;
5879   memset(&dlinfo, 0, sizeof(dlinfo));
5880   if (dladdr(addr, &dlinfo) != 0) {
5881     st->print(PTR_FORMAT ": ", addr);
5882     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5883       st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
5884     } else if (dlinfo.dli_fbase != NULL)
5885       st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
5886     else
5887       st->print("<absolute address>");
5888     if (dlinfo.dli_fname != NULL) {
5889       st->print(" in %s", dlinfo.dli_fname);
5890     }
5891     if (dlinfo.dli_fbase != NULL) {
5892       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5893     }
5894     st->cr();
5895 
5896     if (Verbose) {
5897       // decode some bytes around the PC
5898       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5899       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5900       address       lowest = (address) dlinfo.dli_sname;
5901       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5902       if (begin < lowest)  begin = lowest;
5903       Dl_info dlinfo2;
5904       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5905           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5906         end = (address) dlinfo2.dli_saddr;
5907       Disassembler::decode(begin, end, st);
5908     }
5909     return true;
5910   }
5911   return false;
5912 }
5913 
5914 // Following function has been added to support HotSparc's libjvm.so running
5915 // under Solaris production JDK 1.2.2 / 1.3.0.  These came from
5916 // src/solaris/hpi/native_threads in the EVM codebase.
5917 //
5918 // NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
5919 // libraries and should thus be removed. We will leave it behind for a while
5920 // until we no longer want to able to run on top of 1.3.0 Solaris production
5921 // JDK. See 4341971.
5922 
5923 #define STACK_SLACK 0x800
5924 
5925 extern "C" {
5926   intptr_t sysThreadAvailableStackWithSlack() {
5927     stack_t st;
5928     intptr_t retval, stack_top;
5929     retval = thr_stksegment(&st);
5930     assert(retval == 0, "incorrect return value from thr_stksegment");
5931     assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
5932     assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
5933     stack_top=(intptr_t)st.ss_sp-st.ss_size;
5934     return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
5935   }
5936 }
5937 
5938 // ObjectMonitor park-unpark infrastructure ...
5939 //
5940 // We implement Solaris and Linux PlatformEvents with the
5941 // obvious condvar-mutex-flag triple.
5942 // Another alternative that works quite well is pipes:
5943 // Each PlatformEvent consists of a pipe-pair.
5944 // The thread associated with the PlatformEvent
5945 // calls park(), which reads from the input end of the pipe.
5946 // Unpark() writes into the other end of the pipe.
5947 // The write-side of the pipe must be set NDELAY.
5948 // Unfortunately pipes consume a large # of handles.
5949 // Native solaris lwp_park() and lwp_unpark() work nicely, too.
5950 // Using pipes for the 1st few threads might be workable, however.
5951 //
5952 // park() is permitted to return spuriously.
5953 // Callers of park() should wrap the call to park() in
5954 // an appropriate loop.  A litmus test for the correct
5955 // usage of park is the following: if park() were modified
5956 // to immediately return 0 your code should still work,
5957 // albeit degenerating to a spin loop.
5958 //
5959 // An interesting optimization for park() is to use a trylock()
5960 // to attempt to acquire the mutex.  If the trylock() fails
5961 // then we know that a concurrent unpark() operation is in-progress.
5962 // in that case the park() code could simply set _count to 0
5963 // and return immediately.  The subsequent park() operation *might*
5964 // return immediately.  That's harmless as the caller of park() is
5965 // expected to loop.  By using trylock() we will have avoided a
5966 // avoided a context switch caused by contention on the per-thread mutex.
5967 //
5968 // TODO-FIXME:
5969 // 1.  Reconcile Doug's JSR166 j.u.c park-unpark with the
5970 //     objectmonitor implementation.
5971 // 2.  Collapse the JSR166 parker event, and the
5972 //     objectmonitor ParkEvent into a single "Event" construct.
5973 // 3.  In park() and unpark() add:
5974 //     assert (Thread::current() == AssociatedWith).
5975 // 4.  add spurious wakeup injection on a -XX:EarlyParkReturn=N switch.
5976 //     1-out-of-N park() operations will return immediately.
5977 //
5978 // _Event transitions in park()
5979 //   -1 => -1 : illegal
5980 //    1 =>  0 : pass - return immediately
5981 //    0 => -1 : block
5982 //
5983 // _Event serves as a restricted-range semaphore.
5984 //
5985 // Another possible encoding of _Event would be with
5986 // explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
5987 //
5988 // TODO-FIXME: add DTRACE probes for:
5989 // 1.   Tx parks
5990 // 2.   Ty unparks Tx
5991 // 3.   Tx resumes from park
5992 
5993 
5994 // value determined through experimentation
5995 #define ROUNDINGFIX 11
5996 
5997 // utility to compute the abstime argument to timedwait.
5998 // TODO-FIXME: switch from compute_abstime() to unpackTime().
5999 
6000 static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
6001   // millis is the relative timeout time
6002   // abstime will be the absolute timeout time
6003   if (millis < 0)  millis = 0;
6004   struct timeval now;
6005   int status = gettimeofday(&now, NULL);
6006   assert(status == 0, "gettimeofday");
6007   jlong seconds = millis / 1000;
6008   jlong max_wait_period;
6009 
6010   if (UseLWPSynchronization) {
6011     // forward port of fix for 4275818 (not sleeping long enough)
6012     // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
6013     // _lwp_cond_timedwait() used a round_down algorithm rather
6014     // than a round_up. For millis less than our roundfactor
6015     // it rounded down to 0 which doesn't meet the spec.
6016     // For millis > roundfactor we may return a bit sooner, but
6017     // since we can not accurately identify the patch level and
6018     // this has already been fixed in Solaris 9 and 8 we will
6019     // leave it alone rather than always rounding down.
6020 
6021     if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
6022        // It appears that when we go directly through Solaris _lwp_cond_timedwait()
6023            // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
6024            max_wait_period = 21000000;
6025   } else {
6026     max_wait_period = 50000000;
6027   }
6028   millis %= 1000;
6029   if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
6030      seconds = max_wait_period;
6031   }
6032   abstime->tv_sec = now.tv_sec  + seconds;
6033   long       usec = now.tv_usec + millis * 1000;
6034   if (usec >= 1000000) {
6035     abstime->tv_sec += 1;
6036     usec -= 1000000;
6037   }
6038   abstime->tv_nsec = usec * 1000;
6039   return abstime;
6040 }
6041 
6042 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
6043 // Conceptually TryPark() should be equivalent to park(0).
6044 
6045 int os::PlatformEvent::TryPark() {
6046   for (;;) {
6047     const int v = _Event ;
6048     guarantee ((v == 0) || (v == 1), "invariant") ;
6049     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
6050   }
6051 }
6052 
6053 void os::PlatformEvent::park() {           // AKA: down()
6054   // Invariant: Only the thread associated with the Event/PlatformEvent
6055   // may call park().
6056   int v ;
6057   for (;;) {
6058       v = _Event ;
6059       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6060   }
6061   guarantee (v >= 0, "invariant") ;
6062   if (v == 0) {
6063      // Do this the hard way by blocking ...
6064      // See http://monaco.sfbay/detail.jsf?cr=5094058.
6065      // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
6066      // Only for SPARC >= V8PlusA
6067 #if defined(__sparc) && defined(COMPILER2)
6068      if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6069 #endif
6070      int status = os::Solaris::mutex_lock(_mutex);
6071      assert_status(status == 0, status,  "mutex_lock");
6072      guarantee (_nParked == 0, "invariant") ;
6073      ++ _nParked ;
6074      while (_Event < 0) {
6075         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
6076         // Treat this the same as if the wait was interrupted
6077         // With usr/lib/lwp going to kernel, always handle ETIME
6078         status = os::Solaris::cond_wait(_cond, _mutex);
6079         if (status == ETIME) status = EINTR ;
6080         assert_status(status == 0 || status == EINTR, status, "cond_wait");
6081      }
6082      -- _nParked ;
6083      _Event = 0 ;
6084      status = os::Solaris::mutex_unlock(_mutex);
6085      assert_status(status == 0, status, "mutex_unlock");
6086     // Paranoia to ensure our locked and lock-free paths interact
6087     // correctly with each other.
6088     OrderAccess::fence();
6089   }
6090 }
6091 
6092 int os::PlatformEvent::park(jlong millis) {
6093   guarantee (_nParked == 0, "invariant") ;
6094   int v ;
6095   for (;;) {
6096       v = _Event ;
6097       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6098   }
6099   guarantee (v >= 0, "invariant") ;
6100   if (v != 0) return OS_OK ;
6101 
6102   int ret = OS_TIMEOUT;
6103   timestruc_t abst;
6104   compute_abstime (&abst, millis);
6105 
6106   // See http://monaco.sfbay/detail.jsf?cr=5094058.
6107   // For Solaris SPARC set fprs.FEF=0 prior to parking.
6108   // Only for SPARC >= V8PlusA
6109 #if defined(__sparc) && defined(COMPILER2)
6110  if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6111 #endif
6112   int status = os::Solaris::mutex_lock(_mutex);
6113   assert_status(status == 0, status, "mutex_lock");
6114   guarantee (_nParked == 0, "invariant") ;
6115   ++ _nParked ;
6116   while (_Event < 0) {
6117      int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
6118      assert_status(status == 0 || status == EINTR ||
6119                    status == ETIME || status == ETIMEDOUT,
6120                    status, "cond_timedwait");
6121      if (!FilterSpuriousWakeups) break ;                // previous semantics
6122      if (status == ETIME || status == ETIMEDOUT) break ;
6123      // We consume and ignore EINTR and spurious wakeups.
6124   }
6125   -- _nParked ;
6126   if (_Event >= 0) ret = OS_OK ;
6127   _Event = 0 ;
6128   status = os::Solaris::mutex_unlock(_mutex);
6129   assert_status(status == 0, status, "mutex_unlock");
6130   // Paranoia to ensure our locked and lock-free paths interact
6131   // correctly with each other.
6132   OrderAccess::fence();
6133   return ret;
6134 }
6135 
6136 void os::PlatformEvent::unpark() {
6137   // Transitions for _Event:
6138   //    0 :=> 1
6139   //    1 :=> 1
6140   //   -1 :=> either 0 or 1; must signal target thread
6141   //          That is, we can safely transition _Event from -1 to either
6142   //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
6143   //          unpark() calls.
6144   // See also: "Semaphores in Plan 9" by Mullender & Cox
6145   //
6146   // Note: Forcing a transition from "-1" to "1" on an unpark() means
6147   // that it will take two back-to-back park() calls for the owning
6148   // thread to block. This has the benefit of forcing a spurious return
6149   // from the first park() call after an unpark() call which will help
6150   // shake out uses of park() and unpark() without condition variables.
6151 
6152   if (Atomic::xchg(1, &_Event) >= 0) return;
6153 
6154   // If the thread associated with the event was parked, wake it.
6155   // Wait for the thread assoc with the PlatformEvent to vacate.
6156   int status = os::Solaris::mutex_lock(_mutex);
6157   assert_status(status == 0, status, "mutex_lock");
6158   int AnyWaiters = _nParked;
6159   status = os::Solaris::mutex_unlock(_mutex);
6160   assert_status(status == 0, status, "mutex_unlock");
6161   guarantee(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
6162   if (AnyWaiters != 0) {
6163     // We intentional signal *after* dropping the lock
6164     // to avoid a common class of futile wakeups.
6165     status = os::Solaris::cond_signal(_cond);
6166     assert_status(status == 0, status, "cond_signal");
6167   }
6168 }
6169 
6170 // JSR166
6171 // -------------------------------------------------------
6172 
6173 /*
6174  * The solaris and linux implementations of park/unpark are fairly
6175  * conservative for now, but can be improved. They currently use a
6176  * mutex/condvar pair, plus _counter.
6177  * Park decrements _counter if > 0, else does a condvar wait.  Unpark
6178  * sets count to 1 and signals condvar.  Only one thread ever waits
6179  * on the condvar. Contention seen when trying to park implies that someone
6180  * is unparking you, so don't wait. And spurious returns are fine, so there
6181  * is no need to track notifications.
6182  */
6183 
6184 #define MAX_SECS 100000000
6185 /*
6186  * This code is common to linux and solaris and will be moved to a
6187  * common place in dolphin.
6188  *
6189  * The passed in time value is either a relative time in nanoseconds
6190  * or an absolute time in milliseconds. Either way it has to be unpacked
6191  * into suitable seconds and nanoseconds components and stored in the
6192  * given timespec structure.
6193  * Given time is a 64-bit value and the time_t used in the timespec is only
6194  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
6195  * overflow if times way in the future are given. Further on Solaris versions
6196  * prior to 10 there is a restriction (see cond_timedwait) that the specified
6197  * number of seconds, in abstime, is less than current_time  + 100,000,000.
6198  * As it will be 28 years before "now + 100000000" will overflow we can
6199  * ignore overflow and just impose a hard-limit on seconds using the value
6200  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
6201  * years from "now".
6202  */
6203 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
6204   assert (time > 0, "convertTime");
6205 
6206   struct timeval now;
6207   int status = gettimeofday(&now, NULL);
6208   assert(status == 0, "gettimeofday");
6209 
6210   time_t max_secs = now.tv_sec + MAX_SECS;
6211 
6212   if (isAbsolute) {
6213     jlong secs = time / 1000;
6214     if (secs > max_secs) {
6215       absTime->tv_sec = max_secs;
6216     }
6217     else {
6218       absTime->tv_sec = secs;
6219     }
6220     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
6221   }
6222   else {
6223     jlong secs = time / NANOSECS_PER_SEC;
6224     if (secs >= MAX_SECS) {
6225       absTime->tv_sec = max_secs;
6226       absTime->tv_nsec = 0;
6227     }
6228     else {
6229       absTime->tv_sec = now.tv_sec + secs;
6230       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
6231       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6232         absTime->tv_nsec -= NANOSECS_PER_SEC;
6233         ++absTime->tv_sec; // note: this must be <= max_secs
6234       }
6235     }
6236   }
6237   assert(absTime->tv_sec >= 0, "tv_sec < 0");
6238   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
6239   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
6240   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
6241 }
6242 
6243 void Parker::park(bool isAbsolute, jlong time) {
6244   // Ideally we'd do something useful while spinning, such
6245   // as calling unpackTime().
6246 
6247   // Optional fast-path check:
6248   // Return immediately if a permit is available.
6249   // We depend on Atomic::xchg() having full barrier semantics
6250   // since we are doing a lock-free update to _counter.
6251   if (Atomic::xchg(0, &_counter) > 0) return;
6252 
6253   // Optional fast-exit: Check interrupt before trying to wait
6254   Thread* thread = Thread::current();
6255   assert(thread->is_Java_thread(), "Must be JavaThread");
6256   JavaThread *jt = (JavaThread *)thread;
6257   if (Thread::is_interrupted(thread, false)) {
6258     return;
6259   }
6260 
6261   // First, demultiplex/decode time arguments
6262   timespec absTime;
6263   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
6264     return;
6265   }
6266   if (time > 0) {
6267     // Warning: this code might be exposed to the old Solaris time
6268     // round-down bugs.  Grep "roundingFix" for details.
6269     unpackTime(&absTime, isAbsolute, time);
6270   }
6271 
6272   // Enter safepoint region
6273   // Beware of deadlocks such as 6317397.
6274   // The per-thread Parker:: _mutex is a classic leaf-lock.
6275   // In particular a thread must never block on the Threads_lock while
6276   // holding the Parker:: mutex.  If safepoints are pending both the
6277   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6278   ThreadBlockInVM tbivm(jt);
6279 
6280   // Don't wait if cannot get lock since interference arises from
6281   // unblocking.  Also. check interrupt before trying wait
6282   if (Thread::is_interrupted(thread, false) ||
6283       os::Solaris::mutex_trylock(_mutex) != 0) {
6284     return;
6285   }
6286 
6287   int status ;
6288 
6289   if (_counter > 0)  { // no wait needed
6290     _counter = 0;
6291     status = os::Solaris::mutex_unlock(_mutex);
6292     assert (status == 0, "invariant") ;
6293     // Paranoia to ensure our locked and lock-free paths interact
6294     // correctly with each other and Java-level accesses.
6295     OrderAccess::fence();
6296     return;
6297   }
6298 
6299 #ifdef ASSERT
6300   // Don't catch signals while blocked; let the running threads have the signals.
6301   // (This allows a debugger to break into the running thread.)
6302   sigset_t oldsigs;
6303   sigset_t* allowdebug_blocked = os::Solaris::allowdebug_blocked_signals();
6304   thr_sigsetmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6305 #endif
6306 
6307   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6308   jt->set_suspend_equivalent();
6309   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6310 
6311   // Do this the hard way by blocking ...
6312   // See http://monaco.sfbay/detail.jsf?cr=5094058.
6313   // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
6314   // Only for SPARC >= V8PlusA
6315 #if defined(__sparc) && defined(COMPILER2)
6316   if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6317 #endif
6318 
6319   if (time == 0) {
6320     status = os::Solaris::cond_wait (_cond, _mutex) ;
6321   } else {
6322     status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
6323   }
6324   // Note that an untimed cond_wait() can sometimes return ETIME on older
6325   // versions of the Solaris.
6326   assert_status(status == 0 || status == EINTR ||
6327                 status == ETIME || status == ETIMEDOUT,
6328                 status, "cond_timedwait");
6329 
6330 #ifdef ASSERT
6331   thr_sigsetmask(SIG_SETMASK, &oldsigs, NULL);
6332 #endif
6333   _counter = 0 ;
6334   status = os::Solaris::mutex_unlock(_mutex);
6335   assert_status(status == 0, status, "mutex_unlock") ;
6336   // Paranoia to ensure our locked and lock-free paths interact
6337   // correctly with each other and Java-level accesses.
6338   OrderAccess::fence();
6339 
6340   // If externally suspended while waiting, re-suspend
6341   if (jt->handle_special_suspend_equivalent_condition()) {
6342     jt->java_suspend_self();
6343   }
6344 }
6345 
6346 void Parker::unpark() {
6347   int s, status ;
6348   status = os::Solaris::mutex_lock (_mutex) ;
6349   assert (status == 0, "invariant") ;
6350   s = _counter;
6351   _counter = 1;
6352   status = os::Solaris::mutex_unlock (_mutex) ;
6353   assert (status == 0, "invariant") ;
6354 
6355   if (s < 1) {
6356     status = os::Solaris::cond_signal (_cond) ;
6357     assert (status == 0, "invariant") ;
6358   }
6359 }
6360 
6361 extern char** environ;
6362 
6363 // Run the specified command in a separate process. Return its exit value,
6364 // or -1 on failure (e.g. can't fork a new process).
6365 // Unlike system(), this function can be called from signal handler. It
6366 // doesn't block SIGINT et al.
6367 int os::fork_and_exec(char* cmd) {
6368   char * argv[4];
6369   argv[0] = (char *)"sh";
6370   argv[1] = (char *)"-c";
6371   argv[2] = cmd;
6372   argv[3] = NULL;
6373 
6374   // fork is async-safe, fork1 is not so can't use in signal handler
6375   pid_t pid;
6376   Thread* t = ThreadLocalStorage::get_thread_slow();
6377   if (t != NULL && t->is_inside_signal_handler()) {
6378     pid = fork();
6379   } else {
6380     pid = fork1();
6381   }
6382 
6383   if (pid < 0) {
6384     // fork failed
6385     warning("fork failed: %s", strerror(errno));
6386     return -1;
6387 
6388   } else if (pid == 0) {
6389     // child process
6390 
6391     // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
6392     execve("/usr/bin/sh", argv, environ);
6393 
6394     // execve failed
6395     _exit(-1);
6396 
6397   } else  {
6398     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6399     // care about the actual exit code, for now.
6400 
6401     int status;
6402 
6403     // Wait for the child process to exit.  This returns immediately if
6404     // the child has already exited. */
6405     while (waitpid(pid, &status, 0) < 0) {
6406         switch (errno) {
6407         case ECHILD: return 0;
6408         case EINTR: break;
6409         default: return -1;
6410         }
6411     }
6412 
6413     if (WIFEXITED(status)) {
6414        // The child exited normally; get its exit code.
6415        return WEXITSTATUS(status);
6416     } else if (WIFSIGNALED(status)) {
6417        // The child exited because of a signal
6418        // The best value to return is 0x80 + signal number,
6419        // because that is what all Unix shells do, and because
6420        // it allows callers to distinguish between process exit and
6421        // process death by signal.
6422        return 0x80 + WTERMSIG(status);
6423     } else {
6424        // Unknown exit code; pass it through
6425        return status;
6426     }
6427   }
6428 }
6429 
6430 // is_headless_jre()
6431 //
6432 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6433 // in order to report if we are running in a headless jre
6434 //
6435 // Since JDK8 xawt/libmawt.so was moved into the same directory
6436 // as libawt.so, and renamed libawt_xawt.so
6437 //
6438 bool os::is_headless_jre() {
6439     struct stat statbuf;
6440     char buf[MAXPATHLEN];
6441     char libmawtpath[MAXPATHLEN];
6442     const char *xawtstr  = "/xawt/libmawt.so";
6443     const char *new_xawtstr = "/libawt_xawt.so";
6444     char *p;
6445 
6446     // Get path to libjvm.so
6447     os::jvm_path(buf, sizeof(buf));
6448 
6449     // Get rid of libjvm.so
6450     p = strrchr(buf, '/');
6451     if (p == NULL) return false;
6452     else *p = '\0';
6453 
6454     // Get rid of client or server
6455     p = strrchr(buf, '/');
6456     if (p == NULL) return false;
6457     else *p = '\0';
6458 
6459     // check xawt/libmawt.so
6460     strcpy(libmawtpath, buf);
6461     strcat(libmawtpath, xawtstr);
6462     if (::stat(libmawtpath, &statbuf) == 0) return false;
6463 
6464     // check libawt_xawt.so
6465     strcpy(libmawtpath, buf);
6466     strcat(libmawtpath, new_xawtstr);
6467     if (::stat(libmawtpath, &statbuf) == 0) return false;
6468 
6469     return true;
6470 }
6471 
6472 size_t os::write(int fd, const void *buf, unsigned int nBytes) {
6473   INTERRUPTIBLE_RETURN_INT(::write(fd, buf, nBytes), os::Solaris::clear_interrupted);
6474 }
6475 
6476 int os::close(int fd) {
6477   return ::close(fd);
6478 }
6479 
6480 int os::socket_close(int fd) {
6481   return ::close(fd);
6482 }
6483 
6484 int os::recv(int fd, char* buf, size_t nBytes, uint flags) {
6485   INTERRUPTIBLE_RETURN_INT((int)::recv(fd, buf, nBytes, flags), os::Solaris::clear_interrupted);
6486 }
6487 
6488 int os::send(int fd, char* buf, size_t nBytes, uint flags) {
6489   INTERRUPTIBLE_RETURN_INT((int)::send(fd, buf, nBytes, flags), os::Solaris::clear_interrupted);
6490 }
6491 
6492 int os::raw_send(int fd, char* buf, size_t nBytes, uint flags) {
6493   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
6494 }
6495 
6496 // As both poll and select can be interrupted by signals, we have to be
6497 // prepared to restart the system call after updating the timeout, unless
6498 // a poll() is done with timeout == -1, in which case we repeat with this
6499 // "wait forever" value.
6500 
6501 int os::timeout(int fd, long timeout) {
6502   int res;
6503   struct timeval t;
6504   julong prevtime, newtime;
6505   static const char* aNull = 0;
6506   struct pollfd pfd;
6507   pfd.fd = fd;
6508   pfd.events = POLLIN;
6509 
6510   gettimeofday(&t, &aNull);
6511   prevtime = ((julong)t.tv_sec * 1000)  +  t.tv_usec / 1000;
6512 
6513   for(;;) {
6514     INTERRUPTIBLE_NORESTART(::poll(&pfd, 1, timeout), res, os::Solaris::clear_interrupted);
6515     if(res == OS_ERR && errno == EINTR) {
6516         if(timeout != -1) {
6517           gettimeofday(&t, &aNull);
6518           newtime = ((julong)t.tv_sec * 1000)  +  t.tv_usec /1000;
6519           timeout -= newtime - prevtime;
6520           if(timeout <= 0)
6521             return OS_OK;
6522           prevtime = newtime;
6523         }
6524     } else return res;
6525   }
6526 }
6527 
6528 int os::connect(int fd, struct sockaddr *him, socklen_t len) {
6529   int _result;
6530   INTERRUPTIBLE_NORESTART(::connect(fd, him, len), _result,\
6531                           os::Solaris::clear_interrupted);
6532 
6533   // Depending on when thread interruption is reset, _result could be
6534   // one of two values when errno == EINTR
6535 
6536   if (((_result == OS_INTRPT) || (_result == OS_ERR))
6537       && (errno == EINTR)) {
6538      /* restarting a connect() changes its errno semantics */
6539      INTERRUPTIBLE(::connect(fd, him, len), _result,\
6540                    os::Solaris::clear_interrupted);
6541      /* undo these changes */
6542      if (_result == OS_ERR) {
6543        if (errno == EALREADY) {
6544          errno = EINPROGRESS; /* fall through */
6545        } else if (errno == EISCONN) {
6546          errno = 0;
6547          return OS_OK;
6548        }
6549      }
6550    }
6551    return _result;
6552  }
6553 
6554 int os::accept(int fd, struct sockaddr* him, socklen_t* len) {
6555   if (fd < 0) {
6556     return OS_ERR;
6557   }
6558   INTERRUPTIBLE_RETURN_INT((int)::accept(fd, him, len),\
6559                            os::Solaris::clear_interrupted);
6560 }
6561 
6562 int os::recvfrom(int fd, char* buf, size_t nBytes, uint flags,
6563                  sockaddr* from, socklen_t* fromlen) {
6564   INTERRUPTIBLE_RETURN_INT((int)::recvfrom(fd, buf, nBytes, flags, from, fromlen),\
6565                            os::Solaris::clear_interrupted);
6566 }
6567 
6568 int os::sendto(int fd, char* buf, size_t len, uint flags,
6569                struct sockaddr* to, socklen_t tolen) {
6570   INTERRUPTIBLE_RETURN_INT((int)::sendto(fd, buf, len, flags, to, tolen),\
6571                            os::Solaris::clear_interrupted);
6572 }
6573 
6574 int os::socket_available(int fd, jint *pbytes) {
6575   if (fd < 0) {
6576     return OS_OK;
6577   }
6578   int ret;
6579   RESTARTABLE(::ioctl(fd, FIONREAD, pbytes), ret);
6580   // note: ioctl can return 0 when successful, JVM_SocketAvailable
6581   // is expected to return 0 on failure and 1 on success to the jdk.
6582   return (ret == OS_ERR) ? 0 : 1;
6583 }
6584 
6585 int os::bind(int fd, struct sockaddr* him, socklen_t len) {
6586    INTERRUPTIBLE_RETURN_INT_NORESTART(::bind(fd, him, len),\
6587                                       os::Solaris::clear_interrupted);
6588 }
6589 
6590 // Get the default path to the core file
6591 // Returns the length of the string
6592 int os::get_core_path(char* buffer, size_t bufferSize) {
6593   const char* p = get_current_directory(buffer, bufferSize);
6594 
6595   if (p == NULL) {
6596     assert(p != NULL, "failed to get current directory");
6597     return 0;
6598   }
6599 
6600   return strlen(buffer);
6601 }
6602 
6603 #ifndef PRODUCT
6604 void TestReserveMemorySpecial_test() {
6605   // No tests available for this platform
6606 }
6607 #endif