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 void* os::get_default_process_handle() {
2150   return (void*)::dlopen(NULL, RTLD_LAZY);
2151 }
2152 
2153 int os::stat(const char *path, struct stat *sbuf) {
2154   char pathbuf[MAX_PATH];
2155   if (strlen(path) > MAX_PATH - 1) {
2156     errno = ENAMETOOLONG;
2157     return -1;
2158   }
2159   os::native_path(strcpy(pathbuf, path));
2160   return ::stat(pathbuf, sbuf);
2161 }
2162 
2163 static bool _print_ascii_file(const char* filename, outputStream* st) {
2164   int fd = ::open(filename, O_RDONLY);
2165   if (fd == -1) {
2166      return false;
2167   }
2168 
2169   char buf[32];
2170   int bytes;
2171   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2172     st->print_raw(buf, bytes);
2173   }
2174 
2175   ::close(fd);
2176 
2177   return true;
2178 }
2179 
2180 void os::print_os_info_brief(outputStream* st) {
2181   os::Solaris::print_distro_info(st);
2182 
2183   os::Posix::print_uname_info(st);
2184 
2185   os::Solaris::print_libversion_info(st);
2186 }
2187 
2188 void os::print_os_info(outputStream* st) {
2189   st->print("OS:");
2190 
2191   os::Solaris::print_distro_info(st);
2192 
2193   os::Posix::print_uname_info(st);
2194 
2195   os::Solaris::print_libversion_info(st);
2196 
2197   os::Posix::print_rlimit_info(st);
2198 
2199   os::Posix::print_load_average(st);
2200 }
2201 
2202 void os::Solaris::print_distro_info(outputStream* st) {
2203   if (!_print_ascii_file("/etc/release", st)) {
2204       st->print("Solaris");
2205     }
2206     st->cr();
2207 }
2208 
2209 void os::Solaris::print_libversion_info(outputStream* st) {
2210   if (os::Solaris::T2_libthread()) {
2211     st->print("  (T2 libthread)");
2212   }
2213   else {
2214     st->print("  (T1 libthread)");
2215   }
2216   st->cr();
2217 }
2218 
2219 static bool check_addr0(outputStream* st) {
2220   jboolean status = false;
2221   int fd = ::open("/proc/self/map",O_RDONLY);
2222   if (fd >= 0) {
2223     prmap_t p;
2224     while(::read(fd, &p, sizeof(p)) > 0) {
2225       if (p.pr_vaddr == 0x0) {
2226         st->print("Warning: Address: 0x%x, Size: %dK, ",p.pr_vaddr, p.pr_size/1024, p.pr_mapname);
2227         st->print("Mapped file: %s, ", p.pr_mapname[0] == '\0' ? "None" : p.pr_mapname);
2228         st->print("Access:");
2229         st->print("%s",(p.pr_mflags & MA_READ)  ? "r" : "-");
2230         st->print("%s",(p.pr_mflags & MA_WRITE) ? "w" : "-");
2231         st->print("%s",(p.pr_mflags & MA_EXEC)  ? "x" : "-");
2232         st->cr();
2233         status = true;
2234       }
2235       ::close(fd);
2236     }
2237   }
2238   return status;
2239 }
2240 
2241 void os::pd_print_cpu_info(outputStream* st) {
2242   // Nothing to do for now.
2243 }
2244 
2245 void os::print_memory_info(outputStream* st) {
2246   st->print("Memory:");
2247   st->print(" %dk page", os::vm_page_size()>>10);
2248   st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
2249   st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
2250   st->cr();
2251   (void) check_addr0(st);
2252 }
2253 
2254 // Taken from /usr/include/sys/machsig.h  Supposed to be architecture specific
2255 // but they're the same for all the solaris architectures that we support.
2256 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2257                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2258                           "ILL_COPROC", "ILL_BADSTK" };
2259 
2260 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2261                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2262                           "FPE_FLTINV", "FPE_FLTSUB" };
2263 
2264 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2265 
2266 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2267 
2268 void os::print_siginfo(outputStream* st, void* siginfo) {
2269   st->print("siginfo:");
2270 
2271   const int buflen = 100;
2272   char buf[buflen];
2273   siginfo_t *si = (siginfo_t*)siginfo;
2274   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2275   char *err = strerror(si->si_errno);
2276   if (si->si_errno != 0 && err != NULL) {
2277     st->print("si_errno=%s", err);
2278   } else {
2279     st->print("si_errno=%d", si->si_errno);
2280   }
2281   const int c = si->si_code;
2282   assert(c > 0, "unexpected si_code");
2283   switch (si->si_signo) {
2284   case SIGILL:
2285     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2286     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2287     break;
2288   case SIGFPE:
2289     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2290     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2291     break;
2292   case SIGSEGV:
2293     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2294     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2295     break;
2296   case SIGBUS:
2297     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2298     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2299     break;
2300   default:
2301     st->print(", si_code=%d", si->si_code);
2302     // no si_addr
2303   }
2304 
2305   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2306       UseSharedSpaces) {
2307     FileMapInfo* mapinfo = FileMapInfo::current_info();
2308     if (mapinfo->is_in_shared_space(si->si_addr)) {
2309       st->print("\n\nError accessing class data sharing archive."   \
2310                 " Mapped file inaccessible during execution, "      \
2311                 " possible disk/network problem.");
2312     }
2313   }
2314   st->cr();
2315 }
2316 
2317 // Moved from whole group, because we need them here for diagnostic
2318 // prints.
2319 #define OLDMAXSIGNUM 32
2320 static int Maxsignum = 0;
2321 static int *ourSigFlags = NULL;
2322 
2323 extern "C" void sigINTRHandler(int, siginfo_t*, void*);
2324 
2325 int os::Solaris::get_our_sigflags(int sig) {
2326   assert(ourSigFlags!=NULL, "signal data structure not initialized");
2327   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2328   return ourSigFlags[sig];
2329 }
2330 
2331 void os::Solaris::set_our_sigflags(int sig, int flags) {
2332   assert(ourSigFlags!=NULL, "signal data structure not initialized");
2333   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2334   ourSigFlags[sig] = flags;
2335 }
2336 
2337 
2338 static const char* get_signal_handler_name(address handler,
2339                                            char* buf, int buflen) {
2340   int offset;
2341   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
2342   if (found) {
2343     // skip directory names
2344     const char *p1, *p2;
2345     p1 = buf;
2346     size_t len = strlen(os::file_separator());
2347     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
2348     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
2349   } else {
2350     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
2351   }
2352   return buf;
2353 }
2354 
2355 static void print_signal_handler(outputStream* st, int sig,
2356                                   char* buf, size_t buflen) {
2357   struct sigaction sa;
2358 
2359   sigaction(sig, NULL, &sa);
2360 
2361   st->print("%s: ", os::exception_name(sig, buf, buflen));
2362 
2363   address handler = (sa.sa_flags & SA_SIGINFO)
2364                   ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
2365                   : CAST_FROM_FN_PTR(address, sa.sa_handler);
2366 
2367   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
2368     st->print("SIG_DFL");
2369   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
2370     st->print("SIG_IGN");
2371   } else {
2372     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
2373   }
2374 
2375   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
2376 
2377   address rh = VMError::get_resetted_sighandler(sig);
2378   // May be, handler was resetted by VMError?
2379   if(rh != NULL) {
2380     handler = rh;
2381     sa.sa_flags = VMError::get_resetted_sigflags(sig);
2382   }
2383 
2384   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
2385 
2386   // Check: is it our handler?
2387   if(handler == CAST_FROM_FN_PTR(address, signalHandler) ||
2388      handler == CAST_FROM_FN_PTR(address, sigINTRHandler)) {
2389     // It is our signal handler
2390     // check for flags
2391     if(sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
2392       st->print(
2393         ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
2394         os::Solaris::get_our_sigflags(sig));
2395     }
2396   }
2397   st->cr();
2398 }
2399 
2400 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2401   st->print_cr("Signal Handlers:");
2402   print_signal_handler(st, SIGSEGV, buf, buflen);
2403   print_signal_handler(st, SIGBUS , buf, buflen);
2404   print_signal_handler(st, SIGFPE , buf, buflen);
2405   print_signal_handler(st, SIGPIPE, buf, buflen);
2406   print_signal_handler(st, SIGXFSZ, buf, buflen);
2407   print_signal_handler(st, SIGILL , buf, buflen);
2408   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2409   print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
2410   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2411   print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
2412   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2413   print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
2414   print_signal_handler(st, os::Solaris::SIGinterrupt(), buf, buflen);
2415   print_signal_handler(st, os::Solaris::SIGasync(), buf, buflen);
2416 }
2417 
2418 static char saved_jvm_path[MAXPATHLEN] = { 0 };
2419 
2420 // Find the full path to the current module, libjvm.so
2421 void os::jvm_path(char *buf, jint buflen) {
2422   // Error checking.
2423   if (buflen < MAXPATHLEN) {
2424     assert(false, "must use a large-enough buffer");
2425     buf[0] = '\0';
2426     return;
2427   }
2428   // Lazy resolve the path to current module.
2429   if (saved_jvm_path[0] != 0) {
2430     strcpy(buf, saved_jvm_path);
2431     return;
2432   }
2433 
2434   Dl_info dlinfo;
2435   int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
2436   assert(ret != 0, "cannot locate libjvm");
2437   if (ret != 0 && dlinfo.dli_fname != NULL) {
2438     realpath((char *)dlinfo.dli_fname, buf);
2439   } else {
2440     buf[0] = '\0';
2441     return;
2442   }
2443 
2444   if (Arguments::created_by_gamma_launcher()) {
2445     // Support for the gamma launcher.  Typical value for buf is
2446     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2447     // the right place in the string, then assume we are installed in a JDK and
2448     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2449     // up the path so it looks like libjvm.so is installed there (append a
2450     // fake suffix hotspot/libjvm.so).
2451     const char *p = buf + strlen(buf) - 1;
2452     for (int count = 0; p > buf && count < 5; ++count) {
2453       for (--p; p > buf && *p != '/'; --p)
2454         /* empty */ ;
2455     }
2456 
2457     if (strncmp(p, "/jre/lib/", 9) != 0) {
2458       // Look for JAVA_HOME in the environment.
2459       char* java_home_var = ::getenv("JAVA_HOME");
2460       if (java_home_var != NULL && java_home_var[0] != 0) {
2461         char cpu_arch[12];
2462         char* jrelib_p;
2463         int   len;
2464         sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
2465 #ifdef _LP64
2466         // If we are on sparc running a 64-bit vm, look in jre/lib/sparcv9.
2467         if (strcmp(cpu_arch, "sparc") == 0) {
2468           strcat(cpu_arch, "v9");
2469         } else if (strcmp(cpu_arch, "i386") == 0) {
2470           strcpy(cpu_arch, "amd64");
2471         }
2472 #endif
2473         // Check the current module name "libjvm.so".
2474         p = strrchr(buf, '/');
2475         assert(strstr(p, "/libjvm") == p, "invalid library name");
2476 
2477         realpath(java_home_var, buf);
2478         // determine if this is a legacy image or modules image
2479         // modules image doesn't have "jre" subdirectory
2480         len = strlen(buf);
2481         jrelib_p = buf + len;
2482         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2483         if (0 != access(buf, F_OK)) {
2484           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2485         }
2486 
2487         if (0 == access(buf, F_OK)) {
2488           // Use current module name "libjvm.so"
2489           len = strlen(buf);
2490           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2491         } else {
2492           // Go back to path of .so
2493           realpath((char *)dlinfo.dli_fname, buf);
2494         }
2495       }
2496     }
2497   }
2498 
2499   strcpy(saved_jvm_path, buf);
2500 }
2501 
2502 
2503 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2504   // no prefix required, not even "_"
2505 }
2506 
2507 
2508 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2509   // no suffix required
2510 }
2511 
2512 // This method is a copy of JDK's sysGetLastErrorString
2513 // from src/solaris/hpi/src/system_md.c
2514 
2515 size_t os::lasterror(char *buf, size_t len) {
2516 
2517   if (errno == 0)  return 0;
2518 
2519   const char *s = ::strerror(errno);
2520   size_t n = ::strlen(s);
2521   if (n >= len) {
2522     n = len - 1;
2523   }
2524   ::strncpy(buf, s, n);
2525   buf[n] = '\0';
2526   return n;
2527 }
2528 
2529 
2530 // sun.misc.Signal
2531 
2532 extern "C" {
2533   static void UserHandler(int sig, void *siginfo, void *context) {
2534     // Ctrl-C is pressed during error reporting, likely because the error
2535     // handler fails to abort. Let VM die immediately.
2536     if (sig == SIGINT && is_error_reported()) {
2537        os::die();
2538     }
2539 
2540     os::signal_notify(sig);
2541     // We do not need to reinstate the signal handler each time...
2542   }
2543 }
2544 
2545 void* os::user_handler() {
2546   return CAST_FROM_FN_PTR(void*, UserHandler);
2547 }
2548 
2549 class Semaphore : public StackObj {
2550   public:
2551     Semaphore();
2552     ~Semaphore();
2553     void signal();
2554     void wait();
2555     bool trywait();
2556     bool timedwait(unsigned int sec, int nsec);
2557   private:
2558     sema_t _semaphore;
2559 };
2560 
2561 
2562 Semaphore::Semaphore() {
2563   sema_init(&_semaphore, 0, NULL, NULL);
2564 }
2565 
2566 Semaphore::~Semaphore() {
2567   sema_destroy(&_semaphore);
2568 }
2569 
2570 void Semaphore::signal() {
2571   sema_post(&_semaphore);
2572 }
2573 
2574 void Semaphore::wait() {
2575   sema_wait(&_semaphore);
2576 }
2577 
2578 bool Semaphore::trywait() {
2579   return sema_trywait(&_semaphore) == 0;
2580 }
2581 
2582 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2583   struct timespec ts;
2584   unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2585 
2586   while (1) {
2587     int result = sema_timedwait(&_semaphore, &ts);
2588     if (result == 0) {
2589       return true;
2590     } else if (errno == EINTR) {
2591       continue;
2592     } else if (errno == ETIME) {
2593       return false;
2594     } else {
2595       return false;
2596     }
2597   }
2598 }
2599 
2600 extern "C" {
2601   typedef void (*sa_handler_t)(int);
2602   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2603 }
2604 
2605 void* os::signal(int signal_number, void* handler) {
2606   struct sigaction sigAct, oldSigAct;
2607   sigfillset(&(sigAct.sa_mask));
2608   sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2609   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2610 
2611   if (sigaction(signal_number, &sigAct, &oldSigAct))
2612     // -1 means registration failed
2613     return (void *)-1;
2614 
2615   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2616 }
2617 
2618 void os::signal_raise(int signal_number) {
2619   raise(signal_number);
2620 }
2621 
2622 /*
2623  * The following code is moved from os.cpp for making this
2624  * code platform specific, which it is by its very nature.
2625  */
2626 
2627 // a counter for each possible signal value
2628 static int Sigexit = 0;
2629 static int Maxlibjsigsigs;
2630 static jint *pending_signals = NULL;
2631 static int *preinstalled_sigs = NULL;
2632 static struct sigaction *chainedsigactions = NULL;
2633 static sema_t sig_sem;
2634 typedef int (*version_getting_t)();
2635 version_getting_t os::Solaris::get_libjsig_version = NULL;
2636 static int libjsigversion = NULL;
2637 
2638 int os::sigexitnum_pd() {
2639   assert(Sigexit > 0, "signal memory not yet initialized");
2640   return Sigexit;
2641 }
2642 
2643 void os::Solaris::init_signal_mem() {
2644   // Initialize signal structures
2645   Maxsignum = SIGRTMAX;
2646   Sigexit = Maxsignum+1;
2647   assert(Maxsignum >0, "Unable to obtain max signal number");
2648 
2649   Maxlibjsigsigs = Maxsignum;
2650 
2651   // pending_signals has one int per signal
2652   // The additional signal is for SIGEXIT - exit signal to signal_thread
2653   pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1), mtInternal);
2654   memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2655 
2656   if (UseSignalChaining) {
2657      chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2658        * (Maxsignum + 1), mtInternal);
2659      memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2660      preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2661      memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2662   }
2663   ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1 ), mtInternal);
2664   memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2665 }
2666 
2667 void os::signal_init_pd() {
2668   int ret;
2669 
2670   ret = ::sema_init(&sig_sem, 0, NULL, NULL);
2671   assert(ret == 0, "sema_init() failed");
2672 }
2673 
2674 void os::signal_notify(int signal_number) {
2675   int ret;
2676 
2677   Atomic::inc(&pending_signals[signal_number]);
2678   ret = ::sema_post(&sig_sem);
2679   assert(ret == 0, "sema_post() failed");
2680 }
2681 
2682 static int check_pending_signals(bool wait_for_signal) {
2683   int ret;
2684   while (true) {
2685     for (int i = 0; i < Sigexit + 1; i++) {
2686       jint n = pending_signals[i];
2687       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2688         return i;
2689       }
2690     }
2691     if (!wait_for_signal) {
2692       return -1;
2693     }
2694     JavaThread *thread = JavaThread::current();
2695     ThreadBlockInVM tbivm(thread);
2696 
2697     bool threadIsSuspended;
2698     do {
2699       thread->set_suspend_equivalent();
2700       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2701       while((ret = ::sema_wait(&sig_sem)) == EINTR)
2702           ;
2703       assert(ret == 0, "sema_wait() failed");
2704 
2705       // were we externally suspended while we were waiting?
2706       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2707       if (threadIsSuspended) {
2708         //
2709         // The semaphore has been incremented, but while we were waiting
2710         // another thread suspended us. We don't want to continue running
2711         // while suspended because that would surprise the thread that
2712         // suspended us.
2713         //
2714         ret = ::sema_post(&sig_sem);
2715         assert(ret == 0, "sema_post() failed");
2716 
2717         thread->java_suspend_self();
2718       }
2719     } while (threadIsSuspended);
2720   }
2721 }
2722 
2723 int os::signal_lookup() {
2724   return check_pending_signals(false);
2725 }
2726 
2727 int os::signal_wait() {
2728   return check_pending_signals(true);
2729 }
2730 
2731 ////////////////////////////////////////////////////////////////////////////////
2732 // Virtual Memory
2733 
2734 static int page_size = -1;
2735 
2736 // The mmap MAP_ALIGN flag is supported on Solaris 9 and later.  init_2() will
2737 // clear this var if support is not available.
2738 static bool has_map_align = true;
2739 
2740 int os::vm_page_size() {
2741   assert(page_size != -1, "must call os::init");
2742   return page_size;
2743 }
2744 
2745 // Solaris allocates memory by pages.
2746 int os::vm_allocation_granularity() {
2747   assert(page_size != -1, "must call os::init");
2748   return page_size;
2749 }
2750 
2751 static bool recoverable_mmap_error(int err) {
2752   // See if the error is one we can let the caller handle. This
2753   // list of errno values comes from the Solaris mmap(2) man page.
2754   switch (err) {
2755   case EBADF:
2756   case EINVAL:
2757   case ENOTSUP:
2758     // let the caller deal with these errors
2759     return true;
2760 
2761   default:
2762     // Any remaining errors on this OS can cause our reserved mapping
2763     // to be lost. That can cause confusion where different data
2764     // structures think they have the same memory mapped. The worst
2765     // scenario is if both the VM and a library think they have the
2766     // same memory mapped.
2767     return false;
2768   }
2769 }
2770 
2771 static void warn_fail_commit_memory(char* addr, size_t bytes, bool exec,
2772                                     int err) {
2773   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2774           ", %d) failed; error='%s' (errno=%d)", addr, bytes, exec,
2775           strerror(err), err);
2776 }
2777 
2778 static void warn_fail_commit_memory(char* addr, size_t bytes,
2779                                     size_t alignment_hint, bool exec,
2780                                     int err) {
2781   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2782           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, bytes,
2783           alignment_hint, exec, strerror(err), err);
2784 }
2785 
2786 int os::Solaris::commit_memory_impl(char* addr, size_t bytes, bool exec) {
2787   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2788   size_t size = bytes;
2789   char *res = Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2790   if (res != NULL) {
2791     if (UseNUMAInterleaving) {
2792       numa_make_global(addr, bytes);
2793     }
2794     return 0;
2795   }
2796 
2797   int err = errno;  // save errno from mmap() call in mmap_chunk()
2798 
2799   if (!recoverable_mmap_error(err)) {
2800     warn_fail_commit_memory(addr, bytes, exec, err);
2801     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "committing reserved memory.");
2802   }
2803 
2804   return err;
2805 }
2806 
2807 bool os::pd_commit_memory(char* addr, size_t bytes, bool exec) {
2808   return Solaris::commit_memory_impl(addr, bytes, exec) == 0;
2809 }
2810 
2811 void os::pd_commit_memory_or_exit(char* addr, size_t bytes, bool exec,
2812                                   const char* mesg) {
2813   assert(mesg != NULL, "mesg must be specified");
2814   int err = os::Solaris::commit_memory_impl(addr, bytes, exec);
2815   if (err != 0) {
2816     // the caller wants all commit errors to exit with the specified mesg:
2817     warn_fail_commit_memory(addr, bytes, exec, err);
2818     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, mesg);
2819   }
2820 }
2821 
2822 int os::Solaris::commit_memory_impl(char* addr, size_t bytes,
2823                                     size_t alignment_hint, bool exec) {
2824   int err = Solaris::commit_memory_impl(addr, bytes, exec);
2825   if (err == 0) {
2826     if (UseLargePages && (alignment_hint > (size_t)vm_page_size())) {
2827       // If the large page size has been set and the VM
2828       // is using large pages, use the large page size
2829       // if it is smaller than the alignment hint. This is
2830       // a case where the VM wants to use a larger alignment size
2831       // for its own reasons but still want to use large pages
2832       // (which is what matters to setting the mpss range.
2833       size_t page_size = 0;
2834       if (large_page_size() < alignment_hint) {
2835         assert(UseLargePages, "Expected to be here for large page use only");
2836         page_size = large_page_size();
2837       } else {
2838         // If the alignment hint is less than the large page
2839         // size, the VM wants a particular alignment (thus the hint)
2840         // for internal reasons.  Try to set the mpss range using
2841         // the alignment_hint.
2842         page_size = alignment_hint;
2843       }
2844       // Since this is a hint, ignore any failures.
2845       (void)Solaris::setup_large_pages(addr, bytes, page_size);
2846     }
2847   }
2848   return err;
2849 }
2850 
2851 bool os::pd_commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2852                           bool exec) {
2853   return Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec) == 0;
2854 }
2855 
2856 void os::pd_commit_memory_or_exit(char* addr, size_t bytes,
2857                                   size_t alignment_hint, bool exec,
2858                                   const char* mesg) {
2859   assert(mesg != NULL, "mesg must be specified");
2860   int err = os::Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec);
2861   if (err != 0) {
2862     // the caller wants all commit errors to exit with the specified mesg:
2863     warn_fail_commit_memory(addr, bytes, alignment_hint, exec, err);
2864     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, mesg);
2865   }
2866 }
2867 
2868 // Uncommit the pages in a specified region.
2869 void os::pd_free_memory(char* addr, size_t bytes, size_t alignment_hint) {
2870   if (madvise(addr, bytes, MADV_FREE) < 0) {
2871     debug_only(warning("MADV_FREE failed."));
2872     return;
2873   }
2874 }
2875 
2876 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2877   return os::commit_memory(addr, size, !ExecMem);
2878 }
2879 
2880 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2881   return os::uncommit_memory(addr, size);
2882 }
2883 
2884 // Change the page size in a given range.
2885 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2886   assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2887   assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2888   if (UseLargePages) {
2889     Solaris::setup_large_pages(addr, bytes, alignment_hint);
2890   }
2891 }
2892 
2893 // Tell the OS to make the range local to the first-touching LWP
2894 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2895   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2896   if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2897     debug_only(warning("MADV_ACCESS_LWP failed."));
2898   }
2899 }
2900 
2901 // Tell the OS that this range would be accessed from different LWPs.
2902 void os::numa_make_global(char *addr, size_t bytes) {
2903   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2904   if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2905     debug_only(warning("MADV_ACCESS_MANY failed."));
2906   }
2907 }
2908 
2909 // Get the number of the locality groups.
2910 size_t os::numa_get_groups_num() {
2911   size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2912   return n != -1 ? n : 1;
2913 }
2914 
2915 // Get a list of leaf locality groups. A leaf lgroup is group that
2916 // doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2917 // board. An LWP is assigned to one of these groups upon creation.
2918 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2919    if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2920      ids[0] = 0;
2921      return 1;
2922    }
2923    int result_size = 0, top = 1, bottom = 0, cur = 0;
2924    for (int k = 0; k < size; k++) {
2925      int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2926                                     (Solaris::lgrp_id_t*)&ids[top], size - top);
2927      if (r == -1) {
2928        ids[0] = 0;
2929        return 1;
2930      }
2931      if (!r) {
2932        // That's a leaf node.
2933        assert (bottom <= cur, "Sanity check");
2934        // Check if the node has memory
2935        if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2936                                    NULL, 0, LGRP_RSRC_MEM) > 0) {
2937          ids[bottom++] = ids[cur];
2938        }
2939      }
2940      top += r;
2941      cur++;
2942    }
2943    if (bottom == 0) {
2944      // Handle a situation, when the OS reports no memory available.
2945      // Assume UMA architecture.
2946      ids[0] = 0;
2947      return 1;
2948    }
2949    return bottom;
2950 }
2951 
2952 // Detect the topology change. Typically happens during CPU plugging-unplugging.
2953 bool os::numa_topology_changed() {
2954   int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2955   if (is_stale != -1 && is_stale) {
2956     Solaris::lgrp_fini(Solaris::lgrp_cookie());
2957     Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2958     assert(c != 0, "Failure to initialize LGRP API");
2959     Solaris::set_lgrp_cookie(c);
2960     return true;
2961   }
2962   return false;
2963 }
2964 
2965 // Get the group id of the current LWP.
2966 int os::numa_get_group_id() {
2967   int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2968   if (lgrp_id == -1) {
2969     return 0;
2970   }
2971   const int size = os::numa_get_groups_num();
2972   int *ids = (int*)alloca(size * sizeof(int));
2973 
2974   // Get the ids of all lgroups with memory; r is the count.
2975   int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2976                                   (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2977   if (r <= 0) {
2978     return 0;
2979   }
2980   return ids[os::random() % r];
2981 }
2982 
2983 // Request information about the page.
2984 bool os::get_page_info(char *start, page_info* info) {
2985   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2986   uint64_t addr = (uintptr_t)start;
2987   uint64_t outdata[2];
2988   uint_t validity = 0;
2989 
2990   if (os::Solaris::meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2991     return false;
2992   }
2993 
2994   info->size = 0;
2995   info->lgrp_id = -1;
2996 
2997   if ((validity & 1) != 0) {
2998     if ((validity & 2) != 0) {
2999       info->lgrp_id = outdata[0];
3000     }
3001     if ((validity & 4) != 0) {
3002       info->size = outdata[1];
3003     }
3004     return true;
3005   }
3006   return false;
3007 }
3008 
3009 // Scan the pages from start to end until a page different than
3010 // the one described in the info parameter is encountered.
3011 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
3012   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
3013   const size_t types = sizeof(info_types) / sizeof(info_types[0]);
3014   uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT];
3015   uint_t validity[MAX_MEMINFO_CNT];
3016 
3017   size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
3018   uint64_t p = (uint64_t)start;
3019   while (p < (uint64_t)end) {
3020     addrs[0] = p;
3021     size_t addrs_count = 1;
3022     while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] + page_size < (uint64_t)end) {
3023       addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
3024       addrs_count++;
3025     }
3026 
3027     if (os::Solaris::meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
3028       return NULL;
3029     }
3030 
3031     size_t i = 0;
3032     for (; i < addrs_count; i++) {
3033       if ((validity[i] & 1) != 0) {
3034         if ((validity[i] & 4) != 0) {
3035           if (outdata[types * i + 1] != page_expected->size) {
3036             break;
3037           }
3038         } else
3039           if (page_expected->size != 0) {
3040             break;
3041           }
3042 
3043         if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
3044           if (outdata[types * i] != page_expected->lgrp_id) {
3045             break;
3046           }
3047         }
3048       } else {
3049         return NULL;
3050       }
3051     }
3052 
3053     if (i != addrs_count) {
3054       if ((validity[i] & 2) != 0) {
3055         page_found->lgrp_id = outdata[types * i];
3056       } else {
3057         page_found->lgrp_id = -1;
3058       }
3059       if ((validity[i] & 4) != 0) {
3060         page_found->size = outdata[types * i + 1];
3061       } else {
3062         page_found->size = 0;
3063       }
3064       return (char*)addrs[i];
3065     }
3066 
3067     p = addrs[addrs_count - 1] + page_size;
3068   }
3069   return end;
3070 }
3071 
3072 bool os::pd_uncommit_memory(char* addr, size_t bytes) {
3073   size_t size = bytes;
3074   // Map uncommitted pages PROT_NONE so we fail early if we touch an
3075   // uncommitted page. Otherwise, the read/write might succeed if we
3076   // have enough swap space to back the physical page.
3077   return
3078     NULL != Solaris::mmap_chunk(addr, size,
3079                                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
3080                                 PROT_NONE);
3081 }
3082 
3083 char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
3084   char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
3085 
3086   if (b == MAP_FAILED) {
3087     return NULL;
3088   }
3089   return b;
3090 }
3091 
3092 char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes, size_t alignment_hint, bool fixed) {
3093   char* addr = requested_addr;
3094   int flags = MAP_PRIVATE | MAP_NORESERVE;
3095 
3096   assert(!(fixed && (alignment_hint > 0)), "alignment hint meaningless with fixed mmap");
3097 
3098   if (fixed) {
3099     flags |= MAP_FIXED;
3100   } else if (has_map_align && (alignment_hint > (size_t) vm_page_size())) {
3101     flags |= MAP_ALIGN;
3102     addr = (char*) alignment_hint;
3103   }
3104 
3105   // Map uncommitted pages PROT_NONE so we fail early if we touch an
3106   // uncommitted page. Otherwise, the read/write might succeed if we
3107   // have enough swap space to back the physical page.
3108   return mmap_chunk(addr, bytes, flags, PROT_NONE);
3109 }
3110 
3111 char* os::pd_reserve_memory(size_t bytes, char* requested_addr, size_t alignment_hint) {
3112   char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint, (requested_addr != NULL));
3113 
3114   guarantee(requested_addr == NULL || requested_addr == addr,
3115             "OS failed to return requested mmap address.");
3116   return addr;
3117 }
3118 
3119 // Reserve memory at an arbitrary address, only if that area is
3120 // available (and not reserved for something else).
3121 
3122 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3123   const int max_tries = 10;
3124   char* base[max_tries];
3125   size_t size[max_tries];
3126 
3127   // Solaris adds a gap between mmap'ed regions.  The size of the gap
3128   // is dependent on the requested size and the MMU.  Our initial gap
3129   // value here is just a guess and will be corrected later.
3130   bool had_top_overlap = false;
3131   bool have_adjusted_gap = false;
3132   size_t gap = 0x400000;
3133 
3134   // Assert only that the size is a multiple of the page size, since
3135   // that's all that mmap requires, and since that's all we really know
3136   // about at this low abstraction level.  If we need higher alignment,
3137   // we can either pass an alignment to this method or verify alignment
3138   // in one of the methods further up the call chain.  See bug 5044738.
3139   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3140 
3141   // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
3142   // Give it a try, if the kernel honors the hint we can return immediately.
3143   char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
3144 
3145   volatile int err = errno;
3146   if (addr == requested_addr) {
3147     return addr;
3148   } else if (addr != NULL) {
3149     pd_unmap_memory(addr, bytes);
3150   }
3151 
3152   if (PrintMiscellaneous && Verbose) {
3153     char buf[256];
3154     buf[0] = '\0';
3155     if (addr == NULL) {
3156       jio_snprintf(buf, sizeof(buf), ": %s", strerror(err));
3157     }
3158     warning("attempt_reserve_memory_at: couldn't reserve " SIZE_FORMAT " bytes at "
3159             PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
3160             "%s", bytes, requested_addr, addr, buf);
3161   }
3162 
3163   // Address hint method didn't work.  Fall back to the old method.
3164   // In theory, once SNV becomes our oldest supported platform, this
3165   // code will no longer be needed.
3166   //
3167   // Repeatedly allocate blocks until the block is allocated at the
3168   // right spot. Give up after max_tries.
3169   int i;
3170   for (i = 0; i < max_tries; ++i) {
3171     base[i] = reserve_memory(bytes);
3172 
3173     if (base[i] != NULL) {
3174       // Is this the block we wanted?
3175       if (base[i] == requested_addr) {
3176         size[i] = bytes;
3177         break;
3178       }
3179 
3180       // check that the gap value is right
3181       if (had_top_overlap && !have_adjusted_gap) {
3182         size_t actual_gap = base[i-1] - base[i] - bytes;
3183         if (gap != actual_gap) {
3184           // adjust the gap value and retry the last 2 allocations
3185           assert(i > 0, "gap adjustment code problem");
3186           have_adjusted_gap = true;  // adjust the gap only once, just in case
3187           gap = actual_gap;
3188           if (PrintMiscellaneous && Verbose) {
3189             warning("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
3190           }
3191           unmap_memory(base[i], bytes);
3192           unmap_memory(base[i-1], size[i-1]);
3193           i-=2;
3194           continue;
3195         }
3196       }
3197 
3198       // Does this overlap the block we wanted? Give back the overlapped
3199       // parts and try again.
3200       //
3201       // There is still a bug in this code: if top_overlap == bytes,
3202       // the overlap is offset from requested region by the value of gap.
3203       // In this case giving back the overlapped part will not work,
3204       // because we'll give back the entire block at base[i] and
3205       // therefore the subsequent allocation will not generate a new gap.
3206       // This could be fixed with a new algorithm that used larger
3207       // or variable size chunks to find the requested region -
3208       // but such a change would introduce additional complications.
3209       // It's rare enough that the planets align for this bug,
3210       // so we'll just wait for a fix for 6204603/5003415 which
3211       // will provide a mmap flag to allow us to avoid this business.
3212 
3213       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3214       if (top_overlap >= 0 && top_overlap < bytes) {
3215         had_top_overlap = true;
3216         unmap_memory(base[i], top_overlap);
3217         base[i] += top_overlap;
3218         size[i] = bytes - top_overlap;
3219       } else {
3220         size_t bottom_overlap = base[i] + bytes - requested_addr;
3221         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3222           if (PrintMiscellaneous && Verbose && bottom_overlap == 0) {
3223             warning("attempt_reserve_memory_at: possible alignment bug");
3224           }
3225           unmap_memory(requested_addr, bottom_overlap);
3226           size[i] = bytes - bottom_overlap;
3227         } else {
3228           size[i] = bytes;
3229         }
3230       }
3231     }
3232   }
3233 
3234   // Give back the unused reserved pieces.
3235 
3236   for (int j = 0; j < i; ++j) {
3237     if (base[j] != NULL) {
3238       unmap_memory(base[j], size[j]);
3239     }
3240   }
3241 
3242   return (i < max_tries) ? requested_addr : NULL;
3243 }
3244 
3245 bool os::pd_release_memory(char* addr, size_t bytes) {
3246   size_t size = bytes;
3247   return munmap(addr, size) == 0;
3248 }
3249 
3250 static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
3251   assert(addr == (char*)align_size_down((uintptr_t)addr, os::vm_page_size()),
3252          "addr must be page aligned");
3253   int retVal = mprotect(addr, bytes, prot);
3254   return retVal == 0;
3255 }
3256 
3257 // Protect memory (Used to pass readonly pages through
3258 // JNI GetArray<type>Elements with empty arrays.)
3259 // Also, used for serialization page and for compressed oops null pointer
3260 // checking.
3261 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3262                         bool is_committed) {
3263   unsigned int p = 0;
3264   switch (prot) {
3265   case MEM_PROT_NONE: p = PROT_NONE; break;
3266   case MEM_PROT_READ: p = PROT_READ; break;
3267   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3268   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3269   default:
3270     ShouldNotReachHere();
3271   }
3272   // is_committed is unused.
3273   return solaris_mprotect(addr, bytes, p);
3274 }
3275 
3276 // guard_memory and unguard_memory only happens within stack guard pages.
3277 // Since ISM pertains only to the heap, guard and unguard memory should not
3278 /// happen with an ISM region.
3279 bool os::guard_memory(char* addr, size_t bytes) {
3280   return solaris_mprotect(addr, bytes, PROT_NONE);
3281 }
3282 
3283 bool os::unguard_memory(char* addr, size_t bytes) {
3284   return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
3285 }
3286 
3287 // Large page support
3288 static size_t _large_page_size = 0;
3289 
3290 // Insertion sort for small arrays (descending order).
3291 static void insertion_sort_descending(size_t* array, int len) {
3292   for (int i = 0; i < len; i++) {
3293     size_t val = array[i];
3294     for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
3295       size_t tmp = array[key];
3296       array[key] = array[key - 1];
3297       array[key - 1] = tmp;
3298     }
3299   }
3300 }
3301 
3302 bool os::Solaris::mpss_sanity_check(bool warn, size_t* page_size) {
3303   const unsigned int usable_count = VM_Version::page_size_count();
3304   if (usable_count == 1) {
3305     return false;
3306   }
3307 
3308   // Find the right getpagesizes interface.  When solaris 11 is the minimum
3309   // build platform, getpagesizes() (without the '2') can be called directly.
3310   typedef int (*gps_t)(size_t[], int);
3311   gps_t gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes2"));
3312   if (gps_func == NULL) {
3313     gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes"));
3314     if (gps_func == NULL) {
3315       if (warn) {
3316         warning("MPSS is not supported by the operating system.");
3317       }
3318       return false;
3319     }
3320   }
3321 
3322   // Fill the array of page sizes.
3323   int n = (*gps_func)(_page_sizes, page_sizes_max);
3324   assert(n > 0, "Solaris bug?");
3325 
3326   if (n == page_sizes_max) {
3327     // Add a sentinel value (necessary only if the array was completely filled
3328     // since it is static (zeroed at initialization)).
3329     _page_sizes[--n] = 0;
3330     DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
3331   }
3332   assert(_page_sizes[n] == 0, "missing sentinel");
3333   trace_page_sizes("available page sizes", _page_sizes, n);
3334 
3335   if (n == 1) return false;     // Only one page size available.
3336 
3337   // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
3338   // select up to usable_count elements.  First sort the array, find the first
3339   // acceptable value, then copy the usable sizes to the top of the array and
3340   // trim the rest.  Make sure to include the default page size :-).
3341   //
3342   // A better policy could get rid of the 4M limit by taking the sizes of the
3343   // important VM memory regions (java heap and possibly the code cache) into
3344   // account.
3345   insertion_sort_descending(_page_sizes, n);
3346   const size_t size_limit =
3347     FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
3348   int beg;
3349   for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */ ;
3350   const int end = MIN2((int)usable_count, n) - 1;
3351   for (int cur = 0; cur < end; ++cur, ++beg) {
3352     _page_sizes[cur] = _page_sizes[beg];
3353   }
3354   _page_sizes[end] = vm_page_size();
3355   _page_sizes[end + 1] = 0;
3356 
3357   if (_page_sizes[end] > _page_sizes[end - 1]) {
3358     // Default page size is not the smallest; sort again.
3359     insertion_sort_descending(_page_sizes, end + 1);
3360   }
3361   *page_size = _page_sizes[0];
3362 
3363   trace_page_sizes("usable page sizes", _page_sizes, end + 1);
3364   return true;
3365 }
3366 
3367 void os::large_page_init() {
3368   if (UseLargePages) {
3369     // print a warning if any large page related flag is specified on command line
3370     bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
3371                            !FLAG_IS_DEFAULT(LargePageSizeInBytes);
3372 
3373     UseLargePages = Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
3374   }
3375 }
3376 
3377 bool os::Solaris::setup_large_pages(caddr_t start, size_t bytes, size_t align) {
3378   // Signal to OS that we want large pages for addresses
3379   // from addr, addr + bytes
3380   struct memcntl_mha mpss_struct;
3381   mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
3382   mpss_struct.mha_pagesize = align;
3383   mpss_struct.mha_flags = 0;
3384   // Upon successful completion, memcntl() returns 0
3385   if (memcntl(start, bytes, MC_HAT_ADVISE, (caddr_t) &mpss_struct, 0, 0)) {
3386     debug_only(warning("Attempt to use MPSS failed."));
3387     return false;
3388   }
3389   return true;
3390 }
3391 
3392 char* os::reserve_memory_special(size_t size, size_t alignment, char* addr, bool exec) {
3393   fatal("os::reserve_memory_special should not be called on Solaris.");
3394   return NULL;
3395 }
3396 
3397 bool os::release_memory_special(char* base, size_t bytes) {
3398   fatal("os::release_memory_special should not be called on Solaris.");
3399   return false;
3400 }
3401 
3402 size_t os::large_page_size() {
3403   return _large_page_size;
3404 }
3405 
3406 // MPSS allows application to commit large page memory on demand; with ISM
3407 // the entire memory region must be allocated as shared memory.
3408 bool os::can_commit_large_page_memory() {
3409   return true;
3410 }
3411 
3412 bool os::can_execute_large_page_memory() {
3413   return true;
3414 }
3415 
3416 static int os_sleep(jlong millis, bool interruptible) {
3417   const jlong limit = INT_MAX;
3418   jlong prevtime;
3419   int res;
3420 
3421   while (millis > limit) {
3422     if ((res = os_sleep(limit, interruptible)) != OS_OK)
3423       return res;
3424     millis -= limit;
3425   }
3426 
3427   // Restart interrupted polls with new parameters until the proper delay
3428   // has been completed.
3429 
3430   prevtime = getTimeMillis();
3431 
3432   while (millis > 0) {
3433     jlong newtime;
3434 
3435     if (!interruptible) {
3436       // Following assert fails for os::yield_all:
3437       // assert(!thread->is_Java_thread(), "must not be java thread");
3438       res = poll(NULL, 0, millis);
3439     } else {
3440       JavaThread *jt = JavaThread::current();
3441 
3442       INTERRUPTIBLE_NORESTART_VM_ALWAYS(poll(NULL, 0, millis), res, jt,
3443         os::Solaris::clear_interrupted);
3444     }
3445 
3446     // INTERRUPTIBLE_NORESTART_VM_ALWAYS returns res == OS_INTRPT for
3447     // thread.Interrupt.
3448 
3449     // See c/r 6751923. Poll can return 0 before time
3450     // has elapsed if time is set via clock_settime (as NTP does).
3451     // res == 0 if poll timed out (see man poll RETURN VALUES)
3452     // using the logic below checks that we really did
3453     // sleep at least "millis" if not we'll sleep again.
3454     if( ( res == 0 ) || ((res == OS_ERR) && (errno == EINTR))) {
3455       newtime = getTimeMillis();
3456       assert(newtime >= prevtime, "time moving backwards");
3457     /* Doing prevtime and newtime in microseconds doesn't help precision,
3458        and trying to round up to avoid lost milliseconds can result in a
3459        too-short delay. */
3460       millis -= newtime - prevtime;
3461       if(millis <= 0)
3462         return OS_OK;
3463       prevtime = newtime;
3464     } else
3465       return res;
3466   }
3467 
3468   return OS_OK;
3469 }
3470 
3471 // Read calls from inside the vm need to perform state transitions
3472 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3473   INTERRUPTIBLE_RETURN_INT_VM(::read(fd, buf, nBytes), os::Solaris::clear_interrupted);
3474 }
3475 
3476 size_t os::restartable_read(int fd, void *buf, unsigned int nBytes) {
3477   INTERRUPTIBLE_RETURN_INT(::read(fd, buf, nBytes), os::Solaris::clear_interrupted);
3478 }
3479 
3480 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3481   assert(thread == Thread::current(),  "thread consistency check");
3482 
3483   // TODO-FIXME: this should be removed.
3484   // On Solaris machines (especially 2.5.1) we found that sometimes the VM gets into a live lock
3485   // situation with a JavaThread being starved out of a lwp. The kernel doesn't seem to generate
3486   // a SIGWAITING signal which would enable the threads library to create a new lwp for the starving
3487   // thread. We suspect that because the Watcher thread keeps waking up at periodic intervals the kernel
3488   // is fooled into believing that the system is making progress. In the code below we block the
3489   // the watcher thread while safepoint is in progress so that it would not appear as though the
3490   // system is making progress.
3491   if (!Solaris::T2_libthread() &&
3492       thread->is_Watcher_thread() && SafepointSynchronize::is_synchronizing() && !Arguments::has_profile()) {
3493     // We now try to acquire the threads lock. Since this lock is held by the VM thread during
3494     // the entire safepoint, the watcher thread will  line up here during the safepoint.
3495     Threads_lock->lock_without_safepoint_check();
3496     Threads_lock->unlock();
3497   }
3498 
3499   if (thread->is_Java_thread()) {
3500     // This is a JavaThread so we honor the _thread_blocked protocol
3501     // even for sleeps of 0 milliseconds. This was originally done
3502     // as a workaround for bug 4338139. However, now we also do it
3503     // to honor the suspend-equivalent protocol.
3504 
3505     JavaThread *jt = (JavaThread *) thread;
3506     ThreadBlockInVM tbivm(jt);
3507 
3508     jt->set_suspend_equivalent();
3509     // cleared by handle_special_suspend_equivalent_condition() or
3510     // java_suspend_self() via check_and_wait_while_suspended()
3511 
3512     int ret_code;
3513     if (millis <= 0) {
3514       thr_yield();
3515       ret_code = 0;
3516     } else {
3517       // The original sleep() implementation did not create an
3518       // OSThreadWaitState helper for sleeps of 0 milliseconds.
3519       // I'm preserving that decision for now.
3520       OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3521 
3522       ret_code = os_sleep(millis, interruptible);
3523     }
3524 
3525     // were we externally suspended while we were waiting?
3526     jt->check_and_wait_while_suspended();
3527 
3528     return ret_code;
3529   }
3530 
3531   // non-JavaThread from this point on:
3532 
3533   if (millis <= 0) {
3534     thr_yield();
3535     return 0;
3536   }
3537 
3538   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3539 
3540   return os_sleep(millis, interruptible);
3541 }
3542 
3543 int os::naked_sleep() {
3544   // %% make the sleep time an integer flag. for now use 1 millisec.
3545   return os_sleep(1, false);
3546 }
3547 
3548 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3549 void os::infinite_sleep() {
3550   while (true) {    // sleep forever ...
3551     ::sleep(100);   // ... 100 seconds at a time
3552   }
3553 }
3554 
3555 // Used to convert frequent JVM_Yield() to nops
3556 bool os::dont_yield() {
3557   if (DontYieldALot) {
3558     static hrtime_t last_time = 0;
3559     hrtime_t diff = getTimeNanos() - last_time;
3560 
3561     if (diff < DontYieldALotInterval * 1000000)
3562       return true;
3563 
3564     last_time += diff;
3565 
3566     return false;
3567   }
3568   else {
3569     return false;
3570   }
3571 }
3572 
3573 // Caveat: Solaris os::yield() causes a thread-state transition whereas
3574 // the linux and win32 implementations do not.  This should be checked.
3575 
3576 void os::yield() {
3577   // Yields to all threads with same or greater priority
3578   os::sleep(Thread::current(), 0, false);
3579 }
3580 
3581 // Note that yield semantics are defined by the scheduling class to which
3582 // the thread currently belongs.  Typically, yield will _not yield to
3583 // other equal or higher priority threads that reside on the dispatch queues
3584 // of other CPUs.
3585 
3586 os::YieldResult os::NakedYield() { thr_yield(); return os::YIELD_UNKNOWN; }
3587 
3588 
3589 // On Solaris we found that yield_all doesn't always yield to all other threads.
3590 // There have been cases where there is a thread ready to execute but it doesn't
3591 // get an lwp as the VM thread continues to spin with sleeps of 1 millisecond.
3592 // The 1 millisecond wait doesn't seem long enough for the kernel to issue a
3593 // SIGWAITING signal which will cause a new lwp to be created. So we count the
3594 // number of times yield_all is called in the one loop and increase the sleep
3595 // time after 8 attempts. If this fails too we increase the concurrency level
3596 // so that the starving thread would get an lwp
3597 
3598 void os::yield_all(int attempts) {
3599   // Yields to all threads, including threads with lower priorities
3600   if (attempts == 0) {
3601     os::sleep(Thread::current(), 1, false);
3602   } else {
3603     int iterations = attempts % 30;
3604     if (iterations == 0 && !os::Solaris::T2_libthread()) {
3605       // thr_setconcurrency and _getconcurrency make sense only under T1.
3606       int noofLWPS = thr_getconcurrency();
3607       if (noofLWPS < (Threads::number_of_threads() + 2)) {
3608         thr_setconcurrency(thr_getconcurrency() + 1);
3609       }
3610     } else if (iterations < 25) {
3611       os::sleep(Thread::current(), 1, false);
3612     } else {
3613       os::sleep(Thread::current(), 10, false);
3614     }
3615   }
3616 }
3617 
3618 // Called from the tight loops to possibly influence time-sharing heuristics
3619 void os::loop_breaker(int attempts) {
3620   os::yield_all(attempts);
3621 }
3622 
3623 
3624 // Interface for setting lwp priorities.  If we are using T2 libthread,
3625 // which forces the use of BoundThreads or we manually set UseBoundThreads,
3626 // all of our threads will be assigned to real lwp's.  Using the thr_setprio
3627 // function is meaningless in this mode so we must adjust the real lwp's priority
3628 // The routines below implement the getting and setting of lwp priorities.
3629 //
3630 // Note: There are three priority scales used on Solaris.  Java priotities
3631 //       which range from 1 to 10, libthread "thr_setprio" scale which range
3632 //       from 0 to 127, and the current scheduling class of the process we
3633 //       are running in.  This is typically from -60 to +60.
3634 //       The setting of the lwp priorities in done after a call to thr_setprio
3635 //       so Java priorities are mapped to libthread priorities and we map from
3636 //       the latter to lwp priorities.  We don't keep priorities stored in
3637 //       Java priorities since some of our worker threads want to set priorities
3638 //       higher than all Java threads.
3639 //
3640 // For related information:
3641 // (1)  man -s 2 priocntl
3642 // (2)  man -s 4 priocntl
3643 // (3)  man dispadmin
3644 // =    librt.so
3645 // =    libthread/common/rtsched.c - thrp_setlwpprio().
3646 // =    ps -cL <pid> ... to validate priority.
3647 // =    sched_get_priority_min and _max
3648 //              pthread_create
3649 //              sched_setparam
3650 //              pthread_setschedparam
3651 //
3652 // Assumptions:
3653 // +    We assume that all threads in the process belong to the same
3654 //              scheduling class.   IE. an homogenous process.
3655 // +    Must be root or in IA group to change change "interactive" attribute.
3656 //              Priocntl() will fail silently.  The only indication of failure is when
3657 //              we read-back the value and notice that it hasn't changed.
3658 // +    Interactive threads enter the runq at the head, non-interactive at the tail.
3659 // +    For RT, change timeslice as well.  Invariant:
3660 //              constant "priority integral"
3661 //              Konst == TimeSlice * (60-Priority)
3662 //              Given a priority, compute appropriate timeslice.
3663 // +    Higher numerical values have higher priority.
3664 
3665 // sched class attributes
3666 typedef struct {
3667         int   schedPolicy;              // classID
3668         int   maxPrio;
3669         int   minPrio;
3670 } SchedInfo;
3671 
3672 
3673 static SchedInfo tsLimits, iaLimits, rtLimits, fxLimits;
3674 
3675 #ifdef ASSERT
3676 static int  ReadBackValidate = 1;
3677 #endif
3678 static int  myClass     = 0;
3679 static int  myMin       = 0;
3680 static int  myMax       = 0;
3681 static int  myCur       = 0;
3682 static bool priocntl_enable = false;
3683 
3684 static const int criticalPrio = 60; // FX/60 is critical thread class/priority on T4
3685 static int java_MaxPriority_to_os_priority = 0; // Saved mapping
3686 
3687 
3688 // lwp_priocntl_init
3689 //
3690 // Try to determine the priority scale for our process.
3691 //
3692 // Return errno or 0 if OK.
3693 //
3694 static int lwp_priocntl_init () {
3695   int rslt;
3696   pcinfo_t ClassInfo;
3697   pcparms_t ParmInfo;
3698   int i;
3699 
3700   if (!UseThreadPriorities) return 0;
3701 
3702   // We are using Bound threads, we need to determine our priority ranges
3703   if (os::Solaris::T2_libthread() || UseBoundThreads) {
3704     // If ThreadPriorityPolicy is 1, switch tables
3705     if (ThreadPriorityPolicy == 1) {
3706       for (i = 0 ; i < CriticalPriority+1; i++)
3707         os::java_to_os_priority[i] = prio_policy1[i];
3708     }
3709     if (UseCriticalJavaThreadPriority) {
3710       // MaxPriority always maps to the FX scheduling class and criticalPrio.
3711       // See set_native_priority() and set_lwp_class_and_priority().
3712       // Save original MaxPriority mapping in case attempt to
3713       // use critical priority fails.
3714       java_MaxPriority_to_os_priority = os::java_to_os_priority[MaxPriority];
3715       // Set negative to distinguish from other priorities
3716       os::java_to_os_priority[MaxPriority] = -criticalPrio;
3717     }
3718   }
3719   // Not using Bound Threads, set to ThreadPolicy 1
3720   else {
3721     for ( i = 0 ; i < CriticalPriority+1; i++ ) {
3722       os::java_to_os_priority[i] = prio_policy1[i];
3723     }
3724     return 0;
3725   }
3726 
3727   // Get IDs for a set of well-known scheduling classes.
3728   // TODO-FIXME: GETCLINFO returns the current # of classes in the
3729   // the system.  We should have a loop that iterates over the
3730   // classID values, which are known to be "small" integers.
3731 
3732   strcpy(ClassInfo.pc_clname, "TS");
3733   ClassInfo.pc_cid = -1;
3734   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3735   if (rslt < 0) return errno;
3736   assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3737   tsLimits.schedPolicy = ClassInfo.pc_cid;
3738   tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3739   tsLimits.minPrio = -tsLimits.maxPrio;
3740 
3741   strcpy(ClassInfo.pc_clname, "IA");
3742   ClassInfo.pc_cid = -1;
3743   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3744   if (rslt < 0) return errno;
3745   assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3746   iaLimits.schedPolicy = ClassInfo.pc_cid;
3747   iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3748   iaLimits.minPrio = -iaLimits.maxPrio;
3749 
3750   strcpy(ClassInfo.pc_clname, "RT");
3751   ClassInfo.pc_cid = -1;
3752   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3753   if (rslt < 0) return errno;
3754   assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3755   rtLimits.schedPolicy = ClassInfo.pc_cid;
3756   rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3757   rtLimits.minPrio = 0;
3758 
3759   strcpy(ClassInfo.pc_clname, "FX");
3760   ClassInfo.pc_cid = -1;
3761   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3762   if (rslt < 0) return errno;
3763   assert(ClassInfo.pc_cid != -1, "cid for FX class is -1");
3764   fxLimits.schedPolicy = ClassInfo.pc_cid;
3765   fxLimits.maxPrio = ((fxinfo_t*)ClassInfo.pc_clinfo)->fx_maxupri;
3766   fxLimits.minPrio = 0;
3767 
3768   // Query our "current" scheduling class.
3769   // This will normally be IA, TS or, rarely, FX or RT.
3770   memset(&ParmInfo, 0, sizeof(ParmInfo));
3771   ParmInfo.pc_cid = PC_CLNULL;
3772   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3773   if (rslt < 0) return errno;
3774   myClass = ParmInfo.pc_cid;
3775 
3776   // We now know our scheduling classId, get specific information
3777   // about the class.
3778   ClassInfo.pc_cid = myClass;
3779   ClassInfo.pc_clname[0] = 0;
3780   rslt = priocntl((idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo);
3781   if (rslt < 0) return errno;
3782 
3783   if (ThreadPriorityVerbose) {
3784     tty->print_cr("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3785   }
3786 
3787   memset(&ParmInfo, 0, sizeof(pcparms_t));
3788   ParmInfo.pc_cid = PC_CLNULL;
3789   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3790   if (rslt < 0) return errno;
3791 
3792   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3793     myMin = rtLimits.minPrio;
3794     myMax = rtLimits.maxPrio;
3795   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3796     iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3797     myMin = iaLimits.minPrio;
3798     myMax = iaLimits.maxPrio;
3799     myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3800   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3801     tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3802     myMin = tsLimits.minPrio;
3803     myMax = tsLimits.maxPrio;
3804     myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3805   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3806     fxparms_t *fxInfo = (fxparms_t*)ParmInfo.pc_clparms;
3807     myMin = fxLimits.minPrio;
3808     myMax = fxLimits.maxPrio;
3809     myMax = MIN2(myMax, (int)fxInfo->fx_uprilim);       // clamp - restrict
3810   } else {
3811     // No clue - punt
3812     if (ThreadPriorityVerbose)
3813       tty->print_cr ("Unknown scheduling class: %s ... \n", ClassInfo.pc_clname);
3814     return EINVAL;      // no clue, punt
3815   }
3816 
3817   if (ThreadPriorityVerbose) {
3818     tty->print_cr ("Thread priority Range: [%d..%d]\n", myMin, myMax);
3819   }
3820 
3821   priocntl_enable = true;  // Enable changing priorities
3822   return 0;
3823 }
3824 
3825 #define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3826 #define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3827 #define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3828 #define FXPRI(x)        ((fxparms_t *)((x).pc_clparms))
3829 
3830 
3831 // scale_to_lwp_priority
3832 //
3833 // Convert from the libthread "thr_setprio" scale to our current
3834 // lwp scheduling class scale.
3835 //
3836 static
3837 int     scale_to_lwp_priority (int rMin, int rMax, int x)
3838 {
3839   int v;
3840 
3841   if (x == 127) return rMax;            // avoid round-down
3842     v = (((x*(rMax-rMin)))/128)+rMin;
3843   return v;
3844 }
3845 
3846 
3847 // set_lwp_class_and_priority
3848 //
3849 // Set the class and priority of the lwp.  This call should only
3850 // be made when using bound threads (T2 threads are bound by default).
3851 //
3852 int set_lwp_class_and_priority(int ThreadID, int lwpid,
3853                                int newPrio, int new_class, bool scale) {
3854   int rslt;
3855   int Actual, Expected, prv;
3856   pcparms_t ParmInfo;                   // for GET-SET
3857 #ifdef ASSERT
3858   pcparms_t ReadBack;                   // for readback
3859 #endif
3860 
3861   // Set priority via PC_GETPARMS, update, PC_SETPARMS
3862   // Query current values.
3863   // TODO: accelerate this by eliminating the PC_GETPARMS call.
3864   // Cache "pcparms_t" in global ParmCache.
3865   // TODO: elide set-to-same-value
3866 
3867   // If something went wrong on init, don't change priorities.
3868   if ( !priocntl_enable ) {
3869     if (ThreadPriorityVerbose)
3870       tty->print_cr("Trying to set priority but init failed, ignoring");
3871     return EINVAL;
3872   }
3873 
3874   // If lwp hasn't started yet, just return
3875   // the _start routine will call us again.
3876   if ( lwpid <= 0 ) {
3877     if (ThreadPriorityVerbose) {
3878       tty->print_cr ("deferring the set_lwp_class_and_priority of thread "
3879                      INTPTR_FORMAT " to %d, lwpid not set",
3880                      ThreadID, newPrio);
3881     }
3882     return 0;
3883   }
3884 
3885   if (ThreadPriorityVerbose) {
3886     tty->print_cr ("set_lwp_class_and_priority("
3887                    INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3888                    ThreadID, lwpid, newPrio);
3889   }
3890 
3891   memset(&ParmInfo, 0, sizeof(pcparms_t));
3892   ParmInfo.pc_cid = PC_CLNULL;
3893   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3894   if (rslt < 0) return errno;
3895 
3896   int cur_class = ParmInfo.pc_cid;
3897   ParmInfo.pc_cid = (id_t)new_class;
3898 
3899   if (new_class == rtLimits.schedPolicy) {
3900     rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3901     rtInfo->rt_pri     = scale ? scale_to_lwp_priority(rtLimits.minPrio,
3902                                                        rtLimits.maxPrio, newPrio)
3903                                : newPrio;
3904     rtInfo->rt_tqsecs  = RT_NOCHANGE;
3905     rtInfo->rt_tqnsecs = RT_NOCHANGE;
3906     if (ThreadPriorityVerbose) {
3907       tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3908     }
3909   } else if (new_class == iaLimits.schedPolicy) {
3910     iaparms_t* iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3911     int maxClamped     = MIN2(iaLimits.maxPrio,
3912                               cur_class == new_class
3913                                 ? (int)iaInfo->ia_uprilim : iaLimits.maxPrio);
3914     iaInfo->ia_upri    = scale ? scale_to_lwp_priority(iaLimits.minPrio,
3915                                                        maxClamped, newPrio)
3916                                : newPrio;
3917     iaInfo->ia_uprilim = cur_class == new_class
3918                            ? IA_NOCHANGE : (pri_t)iaLimits.maxPrio;
3919     iaInfo->ia_mode    = IA_NOCHANGE;
3920     if (ThreadPriorityVerbose) {
3921       tty->print_cr("IA: [%d...%d] %d->%d\n",
3922                     iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3923     }
3924   } else if (new_class == tsLimits.schedPolicy) {
3925     tsparms_t* tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3926     int maxClamped     = MIN2(tsLimits.maxPrio,
3927                               cur_class == new_class
3928                                 ? (int)tsInfo->ts_uprilim : tsLimits.maxPrio);
3929     tsInfo->ts_upri    = scale ? scale_to_lwp_priority(tsLimits.minPrio,
3930                                                        maxClamped, newPrio)
3931                                : newPrio;
3932     tsInfo->ts_uprilim = cur_class == new_class
3933                            ? TS_NOCHANGE : (pri_t)tsLimits.maxPrio;
3934     if (ThreadPriorityVerbose) {
3935       tty->print_cr("TS: [%d...%d] %d->%d\n",
3936                     tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
3937     }
3938   } else if (new_class == fxLimits.schedPolicy) {
3939     fxparms_t* fxInfo  = (fxparms_t*)ParmInfo.pc_clparms;
3940     int maxClamped     = MIN2(fxLimits.maxPrio,
3941                               cur_class == new_class
3942                                 ? (int)fxInfo->fx_uprilim : fxLimits.maxPrio);
3943     fxInfo->fx_upri    = scale ? scale_to_lwp_priority(fxLimits.minPrio,
3944                                                        maxClamped, newPrio)
3945                                : newPrio;
3946     fxInfo->fx_uprilim = cur_class == new_class
3947                            ? FX_NOCHANGE : (pri_t)fxLimits.maxPrio;
3948     fxInfo->fx_tqsecs  = FX_NOCHANGE;
3949     fxInfo->fx_tqnsecs = FX_NOCHANGE;
3950     if (ThreadPriorityVerbose) {
3951       tty->print_cr("FX: [%d...%d] %d->%d\n",
3952                     fxLimits.minPrio, maxClamped, newPrio, fxInfo->fx_upri);
3953     }
3954   } else {
3955     if (ThreadPriorityVerbose) {
3956       tty->print_cr("Unknown new scheduling class %d\n", new_class);
3957     }
3958     return EINVAL;    // no clue, punt
3959   }
3960 
3961   rslt = priocntl(P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
3962   if (ThreadPriorityVerbose && rslt) {
3963     tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
3964   }
3965   if (rslt < 0) return errno;
3966 
3967 #ifdef ASSERT
3968   // Sanity check: read back what we just attempted to set.
3969   // In theory it could have changed in the interim ...
3970   //
3971   // The priocntl system call is tricky.
3972   // Sometimes it'll validate the priority value argument and
3973   // return EINVAL if unhappy.  At other times it fails silently.
3974   // Readbacks are prudent.
3975 
3976   if (!ReadBackValidate) return 0;
3977 
3978   memset(&ReadBack, 0, sizeof(pcparms_t));
3979   ReadBack.pc_cid = PC_CLNULL;
3980   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
3981   assert(rslt >= 0, "priocntl failed");
3982   Actual = Expected = 0xBAD;
3983   assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
3984   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3985     Actual   = RTPRI(ReadBack)->rt_pri;
3986     Expected = RTPRI(ParmInfo)->rt_pri;
3987   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3988     Actual   = IAPRI(ReadBack)->ia_upri;
3989     Expected = IAPRI(ParmInfo)->ia_upri;
3990   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3991     Actual   = TSPRI(ReadBack)->ts_upri;
3992     Expected = TSPRI(ParmInfo)->ts_upri;
3993   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3994     Actual   = FXPRI(ReadBack)->fx_upri;
3995     Expected = FXPRI(ParmInfo)->fx_upri;
3996   } else {
3997     if (ThreadPriorityVerbose) {
3998       tty->print_cr("set_lwp_class_and_priority: unexpected class in readback: %d\n",
3999                     ParmInfo.pc_cid);
4000     }
4001   }
4002 
4003   if (Actual != Expected) {
4004     if (ThreadPriorityVerbose) {
4005       tty->print_cr ("set_lwp_class_and_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
4006                      lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
4007     }
4008   }
4009 #endif
4010 
4011   return 0;
4012 }
4013 
4014 // Solaris only gives access to 128 real priorities at a time,
4015 // so we expand Java's ten to fill this range.  This would be better
4016 // if we dynamically adjusted relative priorities.
4017 //
4018 // The ThreadPriorityPolicy option allows us to select 2 different
4019 // priority scales.
4020 //
4021 // ThreadPriorityPolicy=0
4022 // Since the Solaris' default priority is MaximumPriority, we do not
4023 // set a priority lower than Max unless a priority lower than
4024 // NormPriority is requested.
4025 //
4026 // ThreadPriorityPolicy=1
4027 // This mode causes the priority table to get filled with
4028 // linear values.  NormPriority get's mapped to 50% of the
4029 // Maximum priority an so on.  This will cause VM threads
4030 // to get unfair treatment against other Solaris processes
4031 // which do not explicitly alter their thread priorities.
4032 //
4033 
4034 int os::java_to_os_priority[CriticalPriority + 1] = {
4035   -99999,         // 0 Entry should never be used
4036 
4037   0,              // 1 MinPriority
4038   32,             // 2
4039   64,             // 3
4040 
4041   96,             // 4
4042   127,            // 5 NormPriority
4043   127,            // 6
4044 
4045   127,            // 7
4046   127,            // 8
4047   127,            // 9 NearMaxPriority
4048 
4049   127,            // 10 MaxPriority
4050 
4051   -criticalPrio   // 11 CriticalPriority
4052 };
4053 
4054 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4055   OSThread* osthread = thread->osthread();
4056 
4057   // Save requested priority in case the thread hasn't been started
4058   osthread->set_native_priority(newpri);
4059 
4060   // Check for critical priority request
4061   bool fxcritical = false;
4062   if (newpri == -criticalPrio) {
4063     fxcritical = true;
4064     newpri = criticalPrio;
4065   }
4066 
4067   assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
4068   if (!UseThreadPriorities) return OS_OK;
4069 
4070   int status = 0;
4071 
4072   if (!fxcritical) {
4073     // Use thr_setprio only if we have a priority that thr_setprio understands
4074     status = thr_setprio(thread->osthread()->thread_id(), newpri);
4075   }
4076 
4077   if (os::Solaris::T2_libthread() ||
4078       (UseBoundThreads && osthread->is_vm_created())) {
4079     int lwp_status =
4080       set_lwp_class_and_priority(osthread->thread_id(),
4081                                  osthread->lwp_id(),
4082                                  newpri,
4083                                  fxcritical ? fxLimits.schedPolicy : myClass,
4084                                  !fxcritical);
4085     if (lwp_status != 0 && fxcritical) {
4086       // Try again, this time without changing the scheduling class
4087       newpri = java_MaxPriority_to_os_priority;
4088       lwp_status = set_lwp_class_and_priority(osthread->thread_id(),
4089                                               osthread->lwp_id(),
4090                                               newpri, myClass, false);
4091     }
4092     status |= lwp_status;
4093   }
4094   return (status == 0) ? OS_OK : OS_ERR;
4095 }
4096 
4097 
4098 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4099   int p;
4100   if ( !UseThreadPriorities ) {
4101     *priority_ptr = NormalPriority;
4102     return OS_OK;
4103   }
4104   int status = thr_getprio(thread->osthread()->thread_id(), &p);
4105   if (status != 0) {
4106     return OS_ERR;
4107   }
4108   *priority_ptr = p;
4109   return OS_OK;
4110 }
4111 
4112 
4113 // Hint to the underlying OS that a task switch would not be good.
4114 // Void return because it's a hint and can fail.
4115 void os::hint_no_preempt() {
4116   schedctl_start(schedctl_init());
4117 }
4118 
4119 static void resume_clear_context(OSThread *osthread) {
4120   osthread->set_ucontext(NULL);
4121 }
4122 
4123 static void suspend_save_context(OSThread *osthread, ucontext_t* context) {
4124   osthread->set_ucontext(context);
4125 }
4126 
4127 static Semaphore sr_semaphore;
4128 
4129 void os::Solaris::SR_handler(Thread* thread, ucontext_t* uc) {
4130   // Save and restore errno to avoid confusing native code with EINTR
4131   // after sigsuspend.
4132   int old_errno = errno;
4133 
4134   OSThread* osthread = thread->osthread();
4135   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4136 
4137   os::SuspendResume::State current = osthread->sr.state();
4138   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4139     suspend_save_context(osthread, uc);
4140 
4141     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4142     os::SuspendResume::State state = osthread->sr.suspended();
4143     if (state == os::SuspendResume::SR_SUSPENDED) {
4144       sigset_t suspend_set;  // signals for sigsuspend()
4145 
4146       // get current set of blocked signals and unblock resume signal
4147       thr_sigsetmask(SIG_BLOCK, NULL, &suspend_set);
4148       sigdelset(&suspend_set, os::Solaris::SIGasync());
4149 
4150       sr_semaphore.signal();
4151       // wait here until we are resumed
4152       while (1) {
4153         sigsuspend(&suspend_set);
4154 
4155         os::SuspendResume::State result = osthread->sr.running();
4156         if (result == os::SuspendResume::SR_RUNNING) {
4157           sr_semaphore.signal();
4158           break;
4159         }
4160       }
4161 
4162     } else if (state == os::SuspendResume::SR_RUNNING) {
4163       // request was cancelled, continue
4164     } else {
4165       ShouldNotReachHere();
4166     }
4167 
4168     resume_clear_context(osthread);
4169   } else if (current == os::SuspendResume::SR_RUNNING) {
4170     // request was cancelled, continue
4171   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4172     // ignore
4173   } else {
4174     // ignore
4175   }
4176 
4177   errno = old_errno;
4178 }
4179 
4180 
4181 void os::interrupt(Thread* thread) {
4182   assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
4183 
4184   OSThread* osthread = thread->osthread();
4185 
4186   int isInterrupted = osthread->interrupted();
4187   if (!isInterrupted) {
4188       osthread->set_interrupted(true);
4189       OrderAccess::fence();
4190       // os::sleep() is implemented with either poll (NULL,0,timeout) or
4191       // by parking on _SleepEvent.  If the former, thr_kill will unwedge
4192       // the sleeper by SIGINTR, otherwise the unpark() will wake the sleeper.
4193       ParkEvent * const slp = thread->_SleepEvent ;
4194       if (slp != NULL) slp->unpark() ;
4195   }
4196 
4197   // For JSR166:  unpark after setting status but before thr_kill -dl
4198   if (thread->is_Java_thread()) {
4199     ((JavaThread*)thread)->parker()->unpark();
4200   }
4201 
4202   // Handle interruptible wait() ...
4203   ParkEvent * const ev = thread->_ParkEvent ;
4204   if (ev != NULL) ev->unpark() ;
4205 
4206   // When events are used everywhere for os::sleep, then this thr_kill
4207   // will only be needed if UseVMInterruptibleIO is true.
4208 
4209   if (!isInterrupted) {
4210     int status = thr_kill(osthread->thread_id(), os::Solaris::SIGinterrupt());
4211     assert_status(status == 0, status, "thr_kill");
4212 
4213     // Bump thread interruption counter
4214     RuntimeService::record_thread_interrupt_signaled_count();
4215   }
4216 }
4217 
4218 
4219 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4220   assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
4221 
4222   OSThread* osthread = thread->osthread();
4223 
4224   bool res = osthread->interrupted();
4225 
4226   // NOTE that since there is no "lock" around these two operations,
4227   // there is the possibility that the interrupted flag will be
4228   // "false" but that the interrupt event will be set. This is
4229   // intentional. The effect of this is that Object.wait() will appear
4230   // to have a spurious wakeup, which is not harmful, and the
4231   // possibility is so rare that it is not worth the added complexity
4232   // to add yet another lock. It has also been recommended not to put
4233   // the interrupted flag into the os::Solaris::Event structure,
4234   // because it hides the issue.
4235   if (res && clear_interrupted) {
4236     osthread->set_interrupted(false);
4237   }
4238   return res;
4239 }
4240 
4241 
4242 void os::print_statistics() {
4243 }
4244 
4245 int os::message_box(const char* title, const char* message) {
4246   int i;
4247   fdStream err(defaultStream::error_fd());
4248   for (i = 0; i < 78; i++) err.print_raw("=");
4249   err.cr();
4250   err.print_raw_cr(title);
4251   for (i = 0; i < 78; i++) err.print_raw("-");
4252   err.cr();
4253   err.print_raw_cr(message);
4254   for (i = 0; i < 78; i++) err.print_raw("=");
4255   err.cr();
4256 
4257   char buf[16];
4258   // Prevent process from exiting upon "read error" without consuming all CPU
4259   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4260 
4261   return buf[0] == 'y' || buf[0] == 'Y';
4262 }
4263 
4264 static int sr_notify(OSThread* osthread) {
4265   int status = thr_kill(osthread->thread_id(), os::Solaris::SIGasync());
4266   assert_status(status == 0, status, "thr_kill");
4267   return status;
4268 }
4269 
4270 // "Randomly" selected value for how long we want to spin
4271 // before bailing out on suspending a thread, also how often
4272 // we send a signal to a thread we want to resume
4273 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4274 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4275 
4276 static bool do_suspend(OSThread* osthread) {
4277   assert(osthread->sr.is_running(), "thread should be running");
4278   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4279 
4280   // mark as suspended and send signal
4281   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4282     // failed to switch, state wasn't running?
4283     ShouldNotReachHere();
4284     return false;
4285   }
4286 
4287   if (sr_notify(osthread) != 0) {
4288     ShouldNotReachHere();
4289   }
4290 
4291   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4292   while (true) {
4293     if (sr_semaphore.timedwait(0, 2000 * NANOSECS_PER_MILLISEC)) {
4294       break;
4295     } else {
4296       // timeout
4297       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4298       if (cancelled == os::SuspendResume::SR_RUNNING) {
4299         return false;
4300       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4301         // make sure that we consume the signal on the semaphore as well
4302         sr_semaphore.wait();
4303         break;
4304       } else {
4305         ShouldNotReachHere();
4306         return false;
4307       }
4308     }
4309   }
4310 
4311   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4312   return true;
4313 }
4314 
4315 static void do_resume(OSThread* osthread) {
4316   assert(osthread->sr.is_suspended(), "thread should be suspended");
4317   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4318 
4319   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4320     // failed to switch to WAKEUP_REQUEST
4321     ShouldNotReachHere();
4322     return;
4323   }
4324 
4325   while (true) {
4326     if (sr_notify(osthread) == 0) {
4327       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4328         if (osthread->sr.is_running()) {
4329           return;
4330         }
4331       }
4332     } else {
4333       ShouldNotReachHere();
4334     }
4335   }
4336 
4337   guarantee(osthread->sr.is_running(), "Must be running!");
4338 }
4339 
4340 void os::SuspendedThreadTask::internal_do_task() {
4341   if (do_suspend(_thread->osthread())) {
4342     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4343     do_task(context);
4344     do_resume(_thread->osthread());
4345   }
4346 }
4347 
4348 class PcFetcher : public os::SuspendedThreadTask {
4349 public:
4350   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4351   ExtendedPC result();
4352 protected:
4353   void do_task(const os::SuspendedThreadTaskContext& context);
4354 private:
4355   ExtendedPC _epc;
4356 };
4357 
4358 ExtendedPC PcFetcher::result() {
4359   guarantee(is_done(), "task is not done yet.");
4360   return _epc;
4361 }
4362 
4363 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4364   Thread* thread = context.thread();
4365   OSThread* osthread = thread->osthread();
4366   if (osthread->ucontext() != NULL) {
4367     _epc = os::Solaris::ucontext_get_pc((ucontext_t *) context.ucontext());
4368   } else {
4369     // NULL context is unexpected, double-check this is the VMThread
4370     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4371   }
4372 }
4373 
4374 // A lightweight implementation that does not suspend the target thread and
4375 // thus returns only a hint. Used for profiling only!
4376 ExtendedPC os::get_thread_pc(Thread* thread) {
4377   // Make sure that it is called by the watcher and the Threads lock is owned.
4378   assert(Thread::current()->is_Watcher_thread(), "Must be watcher and own Threads_lock");
4379   // For now, is only used to profile the VM Thread
4380   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4381   PcFetcher fetcher(thread);
4382   fetcher.run();
4383   return fetcher.result();
4384 }
4385 
4386 
4387 // This does not do anything on Solaris. This is basically a hook for being
4388 // able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
4389 void os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, JavaCallArguments* args, Thread* thread) {
4390   f(value, method, args, thread);
4391 }
4392 
4393 // This routine may be used by user applications as a "hook" to catch signals.
4394 // The user-defined signal handler must pass unrecognized signals to this
4395 // routine, and if it returns true (non-zero), then the signal handler must
4396 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4397 // routine will never retun false (zero), but instead will execute a VM panic
4398 // routine kill the process.
4399 //
4400 // If this routine returns false, it is OK to call it again.  This allows
4401 // the user-defined signal handler to perform checks either before or after
4402 // the VM performs its own checks.  Naturally, the user code would be making
4403 // a serious error if it tried to handle an exception (such as a null check
4404 // or breakpoint) that the VM was generating for its own correct operation.
4405 //
4406 // This routine may recognize any of the following kinds of signals:
4407 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
4408 // os::Solaris::SIGasync
4409 // It should be consulted by handlers for any of those signals.
4410 // It explicitly does not recognize os::Solaris::SIGinterrupt
4411 //
4412 // The caller of this routine must pass in the three arguments supplied
4413 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4414 // field of the structure passed to sigaction().  This routine assumes that
4415 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4416 //
4417 // Note that the VM will print warnings if it detects conflicting signal
4418 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4419 //
4420 extern "C" JNIEXPORT int
4421 JVM_handle_solaris_signal(int signo, siginfo_t* siginfo, void* ucontext,
4422                           int abort_if_unrecognized);
4423 
4424 
4425 void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
4426   int orig_errno = errno;  // Preserve errno value over signal handler.
4427   JVM_handle_solaris_signal(sig, info, ucVoid, true);
4428   errno = orig_errno;
4429 }
4430 
4431 /* Do not delete - if guarantee is ever removed,  a signal handler (even empty)
4432    is needed to provoke threads blocked on IO to return an EINTR
4433    Note: this explicitly does NOT call JVM_handle_solaris_signal and
4434    does NOT participate in signal chaining due to requirement for
4435    NOT setting SA_RESTART to make EINTR work. */
4436 extern "C" void sigINTRHandler(int sig, siginfo_t* info, void* ucVoid) {
4437    if (UseSignalChaining) {
4438       struct sigaction *actp = os::Solaris::get_chained_signal_action(sig);
4439       if (actp && actp->sa_handler) {
4440         vm_exit_during_initialization("Signal chaining detected for VM interrupt signal, try -XX:+UseAltSigs");
4441       }
4442    }
4443 }
4444 
4445 // This boolean allows users to forward their own non-matching signals
4446 // to JVM_handle_solaris_signal, harmlessly.
4447 bool os::Solaris::signal_handlers_are_installed = false;
4448 
4449 // For signal-chaining
4450 bool os::Solaris::libjsig_is_loaded = false;
4451 typedef struct sigaction *(*get_signal_t)(int);
4452 get_signal_t os::Solaris::get_signal_action = NULL;
4453 
4454 struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
4455   struct sigaction *actp = NULL;
4456 
4457   if ((libjsig_is_loaded)  && (sig <= Maxlibjsigsigs)) {
4458     // Retrieve the old signal handler from libjsig
4459     actp = (*get_signal_action)(sig);
4460   }
4461   if (actp == NULL) {
4462     // Retrieve the preinstalled signal handler from jvm
4463     actp = get_preinstalled_handler(sig);
4464   }
4465 
4466   return actp;
4467 }
4468 
4469 static bool call_chained_handler(struct sigaction *actp, int sig,
4470                                  siginfo_t *siginfo, void *context) {
4471   // Call the old signal handler
4472   if (actp->sa_handler == SIG_DFL) {
4473     // It's more reasonable to let jvm treat it as an unexpected exception
4474     // instead of taking the default action.
4475     return false;
4476   } else if (actp->sa_handler != SIG_IGN) {
4477     if ((actp->sa_flags & SA_NODEFER) == 0) {
4478       // automaticlly block the signal
4479       sigaddset(&(actp->sa_mask), sig);
4480     }
4481 
4482     sa_handler_t hand;
4483     sa_sigaction_t sa;
4484     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4485     // retrieve the chained handler
4486     if (siginfo_flag_set) {
4487       sa = actp->sa_sigaction;
4488     } else {
4489       hand = actp->sa_handler;
4490     }
4491 
4492     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4493       actp->sa_handler = SIG_DFL;
4494     }
4495 
4496     // try to honor the signal mask
4497     sigset_t oset;
4498     thr_sigsetmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4499 
4500     // call into the chained handler
4501     if (siginfo_flag_set) {
4502       (*sa)(sig, siginfo, context);
4503     } else {
4504       (*hand)(sig);
4505     }
4506 
4507     // restore the signal mask
4508     thr_sigsetmask(SIG_SETMASK, &oset, 0);
4509   }
4510   // Tell jvm's signal handler the signal is taken care of.
4511   return true;
4512 }
4513 
4514 bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4515   bool chained = false;
4516   // signal-chaining
4517   if (UseSignalChaining) {
4518     struct sigaction *actp = get_chained_signal_action(sig);
4519     if (actp != NULL) {
4520       chained = call_chained_handler(actp, sig, siginfo, context);
4521     }
4522   }
4523   return chained;
4524 }
4525 
4526 struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
4527   assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4528   if (preinstalled_sigs[sig] != 0) {
4529     return &chainedsigactions[sig];
4530   }
4531   return NULL;
4532 }
4533 
4534 void os::Solaris::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4535 
4536   assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
4537   assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4538   chainedsigactions[sig] = oldAct;
4539   preinstalled_sigs[sig] = 1;
4540 }
4541 
4542 void os::Solaris::set_signal_handler(int sig, bool set_installed, bool oktochain) {
4543   // Check for overwrite.
4544   struct sigaction oldAct;
4545   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4546   void* oldhand = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4547                                       : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4548   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4549       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4550       oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
4551     if (AllowUserSignalHandlers || !set_installed) {
4552       // Do not overwrite; user takes responsibility to forward to us.
4553       return;
4554     } else if (UseSignalChaining) {
4555       if (oktochain) {
4556         // save the old handler in jvm
4557         save_preinstalled_handler(sig, oldAct);
4558       } else {
4559         vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal, try -XX:+UseAltSigs.");
4560       }
4561       // libjsig also interposes the sigaction() call below and saves the
4562       // old sigaction on it own.
4563     } else {
4564       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4565                     "%#lx for signal %d.", (long)oldhand, sig));
4566     }
4567   }
4568 
4569   struct sigaction sigAct;
4570   sigfillset(&(sigAct.sa_mask));
4571   sigAct.sa_handler = SIG_DFL;
4572 
4573   sigAct.sa_sigaction = signalHandler;
4574   // Handle SIGSEGV on alternate signal stack if
4575   // not using stack banging
4576   if (!UseStackBanging && sig == SIGSEGV) {
4577     sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
4578   // Interruptible i/o requires SA_RESTART cleared so EINTR
4579   // is returned instead of restarting system calls
4580   } else if (sig == os::Solaris::SIGinterrupt()) {
4581     sigemptyset(&sigAct.sa_mask);
4582     sigAct.sa_handler = NULL;
4583     sigAct.sa_flags = SA_SIGINFO;
4584     sigAct.sa_sigaction = sigINTRHandler;
4585   } else {
4586     sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
4587   }
4588   os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
4589 
4590   sigaction(sig, &sigAct, &oldAct);
4591 
4592   void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4593                                        : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4594   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4595 }
4596 
4597 
4598 #define DO_SIGNAL_CHECK(sig) \
4599   if (!sigismember(&check_signal_done, sig)) \
4600     os::Solaris::check_signal_handler(sig)
4601 
4602 // This method is a periodic task to check for misbehaving JNI applications
4603 // under CheckJNI, we can add any periodic checks here
4604 
4605 void os::run_periodic_checks() {
4606   // A big source of grief is hijacking virt. addr 0x0 on Solaris,
4607   // thereby preventing a NULL checks.
4608   if(!check_addr0_done) check_addr0_done = check_addr0(tty);
4609 
4610   if (check_signals == false) return;
4611 
4612   // SEGV and BUS if overridden could potentially prevent
4613   // generation of hs*.log in the event of a crash, debugging
4614   // such a case can be very challenging, so we absolutely
4615   // check for the following for a good measure:
4616   DO_SIGNAL_CHECK(SIGSEGV);
4617   DO_SIGNAL_CHECK(SIGILL);
4618   DO_SIGNAL_CHECK(SIGFPE);
4619   DO_SIGNAL_CHECK(SIGBUS);
4620   DO_SIGNAL_CHECK(SIGPIPE);
4621   DO_SIGNAL_CHECK(SIGXFSZ);
4622 
4623   // ReduceSignalUsage allows the user to override these handlers
4624   // see comments at the very top and jvm_solaris.h
4625   if (!ReduceSignalUsage) {
4626     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4627     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4628     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4629     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4630   }
4631 
4632   // See comments above for using JVM1/JVM2 and UseAltSigs
4633   DO_SIGNAL_CHECK(os::Solaris::SIGinterrupt());
4634   DO_SIGNAL_CHECK(os::Solaris::SIGasync());
4635 
4636 }
4637 
4638 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4639 
4640 static os_sigaction_t os_sigaction = NULL;
4641 
4642 void os::Solaris::check_signal_handler(int sig) {
4643   char buf[O_BUFLEN];
4644   address jvmHandler = NULL;
4645 
4646   struct sigaction act;
4647   if (os_sigaction == NULL) {
4648     // only trust the default sigaction, in case it has been interposed
4649     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4650     if (os_sigaction == NULL) return;
4651   }
4652 
4653   os_sigaction(sig, (struct sigaction*)NULL, &act);
4654 
4655   address thisHandler = (act.sa_flags & SA_SIGINFO)
4656     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4657     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4658 
4659 
4660   switch(sig) {
4661     case SIGSEGV:
4662     case SIGBUS:
4663     case SIGFPE:
4664     case SIGPIPE:
4665     case SIGXFSZ:
4666     case SIGILL:
4667       jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4668       break;
4669 
4670     case SHUTDOWN1_SIGNAL:
4671     case SHUTDOWN2_SIGNAL:
4672     case SHUTDOWN3_SIGNAL:
4673     case BREAK_SIGNAL:
4674       jvmHandler = (address)user_handler();
4675       break;
4676 
4677     default:
4678       int intrsig = os::Solaris::SIGinterrupt();
4679       int asynsig = os::Solaris::SIGasync();
4680 
4681       if (sig == intrsig) {
4682         jvmHandler = CAST_FROM_FN_PTR(address, sigINTRHandler);
4683       } else if (sig == asynsig) {
4684         jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4685       } else {
4686         return;
4687       }
4688       break;
4689   }
4690 
4691 
4692   if (thisHandler != jvmHandler) {
4693     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4694     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4695     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4696     // No need to check this sig any longer
4697     sigaddset(&check_signal_done, sig);
4698   } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
4699     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4700     tty->print("expected:" PTR32_FORMAT, os::Solaris::get_our_sigflags(sig));
4701     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4702     // No need to check this sig any longer
4703     sigaddset(&check_signal_done, sig);
4704   }
4705 
4706   // Print all the signal handler state
4707   if (sigismember(&check_signal_done, sig)) {
4708     print_signal_handlers(tty, buf, O_BUFLEN);
4709   }
4710 
4711 }
4712 
4713 void os::Solaris::install_signal_handlers() {
4714   bool libjsigdone = false;
4715   signal_handlers_are_installed = true;
4716 
4717   // signal-chaining
4718   typedef void (*signal_setting_t)();
4719   signal_setting_t begin_signal_setting = NULL;
4720   signal_setting_t end_signal_setting = NULL;
4721   begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4722                                         dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4723   if (begin_signal_setting != NULL) {
4724     end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4725                                         dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4726     get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4727                                        dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4728     get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
4729                                          dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
4730     libjsig_is_loaded = true;
4731     if (os::Solaris::get_libjsig_version != NULL) {
4732       libjsigversion =  (*os::Solaris::get_libjsig_version)();
4733     }
4734     assert(UseSignalChaining, "should enable signal-chaining");
4735   }
4736   if (libjsig_is_loaded) {
4737     // Tell libjsig jvm is setting signal handlers
4738     (*begin_signal_setting)();
4739   }
4740 
4741   set_signal_handler(SIGSEGV, true, true);
4742   set_signal_handler(SIGPIPE, true, true);
4743   set_signal_handler(SIGXFSZ, true, true);
4744   set_signal_handler(SIGBUS, true, true);
4745   set_signal_handler(SIGILL, true, true);
4746   set_signal_handler(SIGFPE, true, true);
4747 
4748 
4749   if (os::Solaris::SIGinterrupt() > OLDMAXSIGNUM || os::Solaris::SIGasync() > OLDMAXSIGNUM) {
4750 
4751     // Pre-1.4.1 Libjsig limited to signal chaining signals <= 32 so
4752     // can not register overridable signals which might be > 32
4753     if (libjsig_is_loaded && libjsigversion <= JSIG_VERSION_1_4_1) {
4754     // Tell libjsig jvm has finished setting signal handlers
4755       (*end_signal_setting)();
4756       libjsigdone = true;
4757     }
4758   }
4759 
4760   // Never ok to chain our SIGinterrupt
4761   set_signal_handler(os::Solaris::SIGinterrupt(), true, false);
4762   set_signal_handler(os::Solaris::SIGasync(), true, true);
4763 
4764   if (libjsig_is_loaded && !libjsigdone) {
4765     // Tell libjsig jvm finishes setting signal handlers
4766     (*end_signal_setting)();
4767   }
4768 
4769   // We don't activate signal checker if libjsig is in place, we trust ourselves
4770   // and if UserSignalHandler is installed all bets are off.
4771   // Log that signal checking is off only if -verbose:jni is specified.
4772   if (CheckJNICalls) {
4773     if (libjsig_is_loaded) {
4774       if (PrintJNIResolving) {
4775         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4776       }
4777       check_signals = false;
4778     }
4779     if (AllowUserSignalHandlers) {
4780       if (PrintJNIResolving) {
4781         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4782       }
4783       check_signals = false;
4784     }
4785   }
4786 }
4787 
4788 
4789 void report_error(const char* file_name, int line_no, const char* title, const char* format, ...);
4790 
4791 const char * signames[] = {
4792   "SIG0",
4793   "SIGHUP", "SIGINT", "SIGQUIT", "SIGILL", "SIGTRAP",
4794   "SIGABRT", "SIGEMT", "SIGFPE", "SIGKILL", "SIGBUS",
4795   "SIGSEGV", "SIGSYS", "SIGPIPE", "SIGALRM", "SIGTERM",
4796   "SIGUSR1", "SIGUSR2", "SIGCLD", "SIGPWR", "SIGWINCH",
4797   "SIGURG", "SIGPOLL", "SIGSTOP", "SIGTSTP", "SIGCONT",
4798   "SIGTTIN", "SIGTTOU", "SIGVTALRM", "SIGPROF", "SIGXCPU",
4799   "SIGXFSZ", "SIGWAITING", "SIGLWP", "SIGFREEZE", "SIGTHAW",
4800   "SIGCANCEL", "SIGLOST"
4801 };
4802 
4803 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4804   if (0 < exception_code && exception_code <= SIGRTMAX) {
4805     // signal
4806     if (exception_code < sizeof(signames)/sizeof(const char*)) {
4807        jio_snprintf(buf, size, "%s", signames[exception_code]);
4808     } else {
4809        jio_snprintf(buf, size, "SIG%d", exception_code);
4810     }
4811     return buf;
4812   } else {
4813     return NULL;
4814   }
4815 }
4816 
4817 // (Static) wrappers for the new libthread API
4818 int_fnP_thread_t_iP_uP_stack_tP_gregset_t os::Solaris::_thr_getstate;
4819 int_fnP_thread_t_i_gregset_t os::Solaris::_thr_setstate;
4820 int_fnP_thread_t_i os::Solaris::_thr_setmutator;
4821 int_fnP_thread_t os::Solaris::_thr_suspend_mutator;
4822 int_fnP_thread_t os::Solaris::_thr_continue_mutator;
4823 
4824 // (Static) wrapper for getisax(2) call.
4825 os::Solaris::getisax_func_t os::Solaris::_getisax = 0;
4826 
4827 // (Static) wrappers for the liblgrp API
4828 os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4829 os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4830 os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4831 os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4832 os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4833 os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4834 os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4835 os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4836 os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4837 
4838 // (Static) wrapper for meminfo() call.
4839 os::Solaris::meminfo_func_t os::Solaris::_meminfo = 0;
4840 
4841 static address resolve_symbol_lazy(const char* name) {
4842   address addr = (address) dlsym(RTLD_DEFAULT, name);
4843   if(addr == NULL) {
4844     // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4845     addr = (address) dlsym(RTLD_NEXT, name);
4846   }
4847   return addr;
4848 }
4849 
4850 static address resolve_symbol(const char* name) {
4851   address addr = resolve_symbol_lazy(name);
4852   if(addr == NULL) {
4853     fatal(dlerror());
4854   }
4855   return addr;
4856 }
4857 
4858 
4859 
4860 // isT2_libthread()
4861 //
4862 // Routine to determine if we are currently using the new T2 libthread.
4863 //
4864 // We determine if we are using T2 by reading /proc/self/lstatus and
4865 // looking for a thread with the ASLWP bit set.  If we find this status
4866 // bit set, we must assume that we are NOT using T2.  The T2 team
4867 // has approved this algorithm.
4868 //
4869 // We need to determine if we are running with the new T2 libthread
4870 // since setting native thread priorities is handled differently
4871 // when using this library.  All threads created using T2 are bound
4872 // threads. Calling thr_setprio is meaningless in this case.
4873 //
4874 bool isT2_libthread() {
4875   static prheader_t * lwpArray = NULL;
4876   static int lwpSize = 0;
4877   static int lwpFile = -1;
4878   lwpstatus_t * that;
4879   char lwpName [128];
4880   bool isT2 = false;
4881 
4882 #define ADR(x)  ((uintptr_t)(x))
4883 #define LWPINDEX(ary,ix)   ((lwpstatus_t *)(((ary)->pr_entsize * (ix)) + (ADR((ary) + 1))))
4884 
4885   lwpFile = ::open("/proc/self/lstatus", O_RDONLY, 0);
4886   if (lwpFile < 0) {
4887       if (ThreadPriorityVerbose) warning ("Couldn't open /proc/self/lstatus\n");
4888       return false;
4889   }
4890   lwpSize = 16*1024;
4891   for (;;) {
4892     ::lseek64 (lwpFile, 0, SEEK_SET);
4893     lwpArray = (prheader_t *)NEW_C_HEAP_ARRAY(char, lwpSize, mtInternal);
4894     if (::read(lwpFile, lwpArray, lwpSize) < 0) {
4895       if (ThreadPriorityVerbose) warning("Error reading /proc/self/lstatus\n");
4896       break;
4897     }
4898     if ((lwpArray->pr_nent * lwpArray->pr_entsize) <= lwpSize) {
4899        // We got a good snapshot - now iterate over the list.
4900       int aslwpcount = 0;
4901       for (int i = 0; i < lwpArray->pr_nent; i++ ) {
4902         that = LWPINDEX(lwpArray,i);
4903         if (that->pr_flags & PR_ASLWP) {
4904           aslwpcount++;
4905         }
4906       }
4907       if (aslwpcount == 0) isT2 = true;
4908       break;
4909     }
4910     lwpSize = lwpArray->pr_nent * lwpArray->pr_entsize;
4911     FREE_C_HEAP_ARRAY(char, lwpArray, mtInternal);  // retry.
4912   }
4913 
4914   FREE_C_HEAP_ARRAY(char, lwpArray, mtInternal);
4915   ::close (lwpFile);
4916   if (ThreadPriorityVerbose) {
4917     if (isT2) tty->print_cr("We are running with a T2 libthread\n");
4918     else tty->print_cr("We are not running with a T2 libthread\n");
4919   }
4920   return isT2;
4921 }
4922 
4923 
4924 void os::Solaris::libthread_init() {
4925   address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4926 
4927   // Determine if we are running with the new T2 libthread
4928   os::Solaris::set_T2_libthread(isT2_libthread());
4929 
4930   lwp_priocntl_init();
4931 
4932   // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4933   if(func == NULL) {
4934     func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4935     // Guarantee that this VM is running on an new enough OS (5.6 or
4936     // later) that it will have a new enough libthread.so.
4937     guarantee(func != NULL, "libthread.so is too old.");
4938   }
4939 
4940   // Initialize the new libthread getstate API wrappers
4941   func = resolve_symbol("thr_getstate");
4942   os::Solaris::set_thr_getstate(CAST_TO_FN_PTR(int_fnP_thread_t_iP_uP_stack_tP_gregset_t, func));
4943 
4944   func = resolve_symbol("thr_setstate");
4945   os::Solaris::set_thr_setstate(CAST_TO_FN_PTR(int_fnP_thread_t_i_gregset_t, func));
4946 
4947   func = resolve_symbol("thr_setmutator");
4948   os::Solaris::set_thr_setmutator(CAST_TO_FN_PTR(int_fnP_thread_t_i, func));
4949 
4950   func = resolve_symbol("thr_suspend_mutator");
4951   os::Solaris::set_thr_suspend_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4952 
4953   func = resolve_symbol("thr_continue_mutator");
4954   os::Solaris::set_thr_continue_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4955 
4956   int size;
4957   void (*handler_info_func)(address *, int *);
4958   handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4959   handler_info_func(&handler_start, &size);
4960   handler_end = handler_start + size;
4961 }
4962 
4963 
4964 int_fnP_mutex_tP os::Solaris::_mutex_lock;
4965 int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4966 int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4967 int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4968 int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4969 int os::Solaris::_mutex_scope = USYNC_THREAD;
4970 
4971 int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4972 int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4973 int_fnP_cond_tP os::Solaris::_cond_signal;
4974 int_fnP_cond_tP os::Solaris::_cond_broadcast;
4975 int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4976 int_fnP_cond_tP os::Solaris::_cond_destroy;
4977 int os::Solaris::_cond_scope = USYNC_THREAD;
4978 
4979 void os::Solaris::synchronization_init() {
4980   if(UseLWPSynchronization) {
4981     os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4982     os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4983     os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4984     os::Solaris::set_mutex_init(lwp_mutex_init);
4985     os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4986     os::Solaris::set_mutex_scope(USYNC_THREAD);
4987 
4988     os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4989     os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4990     os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4991     os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4992     os::Solaris::set_cond_init(lwp_cond_init);
4993     os::Solaris::set_cond_destroy(lwp_cond_destroy);
4994     os::Solaris::set_cond_scope(USYNC_THREAD);
4995   }
4996   else {
4997     os::Solaris::set_mutex_scope(USYNC_THREAD);
4998     os::Solaris::set_cond_scope(USYNC_THREAD);
4999 
5000     if(UsePthreads) {
5001       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
5002       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
5003       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
5004       os::Solaris::set_mutex_init(pthread_mutex_default_init);
5005       os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
5006 
5007       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
5008       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
5009       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
5010       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
5011       os::Solaris::set_cond_init(pthread_cond_default_init);
5012       os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
5013     }
5014     else {
5015       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
5016       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
5017       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
5018       os::Solaris::set_mutex_init(::mutex_init);
5019       os::Solaris::set_mutex_destroy(::mutex_destroy);
5020 
5021       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
5022       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
5023       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
5024       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
5025       os::Solaris::set_cond_init(::cond_init);
5026       os::Solaris::set_cond_destroy(::cond_destroy);
5027     }
5028   }
5029 }
5030 
5031 bool os::Solaris::liblgrp_init() {
5032   void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
5033   if (handle != NULL) {
5034     os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
5035     os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
5036     os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
5037     os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
5038     os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
5039     os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
5040     os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
5041     os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
5042                                        dlsym(handle, "lgrp_cookie_stale")));
5043 
5044     lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
5045     set_lgrp_cookie(c);
5046     return true;
5047   }
5048   return false;
5049 }
5050 
5051 void os::Solaris::misc_sym_init() {
5052   address func;
5053 
5054   // getisax
5055   func = resolve_symbol_lazy("getisax");
5056   if (func != NULL) {
5057     os::Solaris::_getisax = CAST_TO_FN_PTR(getisax_func_t, func);
5058   }
5059 
5060   // meminfo
5061   func = resolve_symbol_lazy("meminfo");
5062   if (func != NULL) {
5063     os::Solaris::set_meminfo(CAST_TO_FN_PTR(meminfo_func_t, func));
5064   }
5065 }
5066 
5067 uint_t os::Solaris::getisax(uint32_t* array, uint_t n) {
5068   assert(_getisax != NULL, "_getisax not set");
5069   return _getisax(array, n);
5070 }
5071 
5072 // int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
5073 typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
5074 static pset_getloadavg_type pset_getloadavg_ptr = NULL;
5075 
5076 void init_pset_getloadavg_ptr(void) {
5077   pset_getloadavg_ptr =
5078     (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
5079   if (PrintMiscellaneous && Verbose && pset_getloadavg_ptr == NULL) {
5080     warning("pset_getloadavg function not found");
5081   }
5082 }
5083 
5084 int os::Solaris::_dev_zero_fd = -1;
5085 
5086 // this is called _before_ the global arguments have been parsed
5087 void os::init(void) {
5088   _initial_pid = getpid();
5089 
5090   max_hrtime = first_hrtime = gethrtime();
5091 
5092   init_random(1234567);
5093 
5094   page_size = sysconf(_SC_PAGESIZE);
5095   if (page_size == -1)
5096     fatal(err_msg("os_solaris.cpp: os::init: sysconf failed (%s)",
5097                   strerror(errno)));
5098   init_page_sizes((size_t) page_size);
5099 
5100   Solaris::initialize_system_info();
5101 
5102   // Initialize misc. symbols as soon as possible, so we can use them
5103   // if we need them.
5104   Solaris::misc_sym_init();
5105 
5106   int fd = ::open("/dev/zero", O_RDWR);
5107   if (fd < 0) {
5108     fatal(err_msg("os::init: cannot open /dev/zero (%s)", strerror(errno)));
5109   } else {
5110     Solaris::set_dev_zero_fd(fd);
5111 
5112     // Close on exec, child won't inherit.
5113     fcntl(fd, F_SETFD, FD_CLOEXEC);
5114   }
5115 
5116   clock_tics_per_sec = CLK_TCK;
5117 
5118   // check if dladdr1() exists; dladdr1 can provide more information than
5119   // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
5120   // and is available on linker patches for 5.7 and 5.8.
5121   // libdl.so must have been loaded, this call is just an entry lookup
5122   void * hdl = dlopen("libdl.so", RTLD_NOW);
5123   if (hdl)
5124     dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
5125 
5126   // (Solaris only) this switches to calls that actually do locking.
5127   ThreadCritical::initialize();
5128 
5129   main_thread = thr_self();
5130 
5131   // Constant minimum stack size allowed. It must be at least
5132   // the minimum of what the OS supports (thr_min_stack()), and
5133   // enough to allow the thread to get to user bytecode execution.
5134   Solaris::min_stack_allowed = MAX2(thr_min_stack(), Solaris::min_stack_allowed);
5135   // If the pagesize of the VM is greater than 8K determine the appropriate
5136   // number of initial guard pages.  The user can change this with the
5137   // command line arguments, if needed.
5138   if (vm_page_size() > 8*K) {
5139     StackYellowPages = 1;
5140     StackRedPages = 1;
5141     StackShadowPages = round_to((StackShadowPages*8*K), vm_page_size()) / vm_page_size();
5142   }
5143 }
5144 
5145 // To install functions for atexit system call
5146 extern "C" {
5147   static void perfMemory_exit_helper() {
5148     perfMemory_exit();
5149   }
5150 }
5151 
5152 // this is called _after_ the global arguments have been parsed
5153 jint os::init_2(void) {
5154   // try to enable extended file IO ASAP, see 6431278
5155   os::Solaris::try_enable_extended_io();
5156 
5157   // Allocate a single page and mark it as readable for safepoint polling.  Also
5158   // use this first mmap call to check support for MAP_ALIGN.
5159   address polling_page = (address)Solaris::mmap_chunk((char*)page_size,
5160                                                       page_size,
5161                                                       MAP_PRIVATE | MAP_ALIGN,
5162                                                       PROT_READ);
5163   if (polling_page == NULL) {
5164     has_map_align = false;
5165     polling_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE,
5166                                                 PROT_READ);
5167   }
5168 
5169   os::set_polling_page(polling_page);
5170 
5171 #ifndef PRODUCT
5172   if( Verbose && PrintMiscellaneous )
5173     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
5174 #endif
5175 
5176   if (!UseMembar) {
5177     address mem_serialize_page = (address)Solaris::mmap_chunk( NULL, page_size, MAP_PRIVATE, PROT_READ | PROT_WRITE );
5178     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
5179     os::set_memory_serialize_page( mem_serialize_page );
5180 
5181 #ifndef PRODUCT
5182     if(Verbose && PrintMiscellaneous)
5183       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
5184 #endif
5185   }
5186 
5187   // Check minimum allowable stack size for thread creation and to initialize
5188   // the java system classes, including StackOverflowError - depends on page
5189   // size.  Add a page for compiler2 recursion in main thread.
5190   // Add in 2*BytesPerWord times page size to account for VM stack during
5191   // class initialization depending on 32 or 64 bit VM.
5192   os::Solaris::min_stack_allowed = MAX2(os::Solaris::min_stack_allowed,
5193             (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
5194                     2*BytesPerWord COMPILER2_PRESENT(+1)) * page_size);
5195 
5196   size_t threadStackSizeInBytes = ThreadStackSize * K;
5197   if (threadStackSizeInBytes != 0 &&
5198     threadStackSizeInBytes < os::Solaris::min_stack_allowed) {
5199     tty->print_cr("\nThe stack size specified is too small, Specify at least %dk",
5200                   os::Solaris::min_stack_allowed/K);
5201     return JNI_ERR;
5202   }
5203 
5204   // For 64kbps there will be a 64kb page size, which makes
5205   // the usable default stack size quite a bit less.  Increase the
5206   // stack for 64kb (or any > than 8kb) pages, this increases
5207   // virtual memory fragmentation (since we're not creating the
5208   // stack on a power of 2 boundary.  The real fix for this
5209   // should be to fix the guard page mechanism.
5210 
5211   if (vm_page_size() > 8*K) {
5212       threadStackSizeInBytes = (threadStackSizeInBytes != 0)
5213          ? threadStackSizeInBytes +
5214            ((StackYellowPages + StackRedPages) * vm_page_size())
5215          : 0;
5216       ThreadStackSize = threadStackSizeInBytes/K;
5217   }
5218 
5219   // Make the stack size a multiple of the page size so that
5220   // the yellow/red zones can be guarded.
5221   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
5222         vm_page_size()));
5223 
5224   Solaris::libthread_init();
5225 
5226   if (UseNUMA) {
5227     if (!Solaris::liblgrp_init()) {
5228       UseNUMA = false;
5229     } else {
5230       size_t lgrp_limit = os::numa_get_groups_num();
5231       int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit, mtInternal);
5232       size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
5233       FREE_C_HEAP_ARRAY(int, lgrp_ids, mtInternal);
5234       if (lgrp_num < 2) {
5235         // There's only one locality group, disable NUMA.
5236         UseNUMA = false;
5237       }
5238     }
5239     if (!UseNUMA && ForceNUMA) {
5240       UseNUMA = true;
5241     }
5242   }
5243 
5244   Solaris::signal_sets_init();
5245   Solaris::init_signal_mem();
5246   Solaris::install_signal_handlers();
5247 
5248   if (libjsigversion < JSIG_VERSION_1_4_1) {
5249     Maxlibjsigsigs = OLDMAXSIGNUM;
5250   }
5251 
5252   // initialize synchronization primitives to use either thread or
5253   // lwp synchronization (controlled by UseLWPSynchronization)
5254   Solaris::synchronization_init();
5255 
5256   if (MaxFDLimit) {
5257     // set the number of file descriptors to max. print out error
5258     // if getrlimit/setrlimit fails but continue regardless.
5259     struct rlimit nbr_files;
5260     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5261     if (status != 0) {
5262       if (PrintMiscellaneous && (Verbose || WizardMode))
5263         perror("os::init_2 getrlimit failed");
5264     } else {
5265       nbr_files.rlim_cur = nbr_files.rlim_max;
5266       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5267       if (status != 0) {
5268         if (PrintMiscellaneous && (Verbose || WizardMode))
5269           perror("os::init_2 setrlimit failed");
5270       }
5271     }
5272   }
5273 
5274   // Calculate theoretical max. size of Threads to guard gainst
5275   // artifical out-of-memory situations, where all available address-
5276   // space has been reserved by thread stacks. Default stack size is 1Mb.
5277   size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
5278     JavaThread::stack_size_at_create() : (1*K*K);
5279   assert(pre_thread_stack_size != 0, "Must have a stack");
5280   // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
5281   // we should start doing Virtual Memory banging. Currently when the threads will
5282   // have used all but 200Mb of space.
5283   size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
5284   Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
5285 
5286   // at-exit methods are called in the reverse order of their registration.
5287   // In Solaris 7 and earlier, atexit functions are called on return from
5288   // main or as a result of a call to exit(3C). There can be only 32 of
5289   // these functions registered and atexit() does not set errno. In Solaris
5290   // 8 and later, there is no limit to the number of functions registered
5291   // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
5292   // functions are called upon dlclose(3DL) in addition to return from main
5293   // and exit(3C).
5294 
5295   if (PerfAllowAtExitRegistration) {
5296     // only register atexit functions if PerfAllowAtExitRegistration is set.
5297     // atexit functions can be delayed until process exit time, which
5298     // can be problematic for embedded VM situations. Embedded VMs should
5299     // call DestroyJavaVM() to assure that VM resources are released.
5300 
5301     // note: perfMemory_exit_helper atexit function may be removed in
5302     // the future if the appropriate cleanup code can be added to the
5303     // VM_Exit VMOperation's doit method.
5304     if (atexit(perfMemory_exit_helper) != 0) {
5305       warning("os::init2 atexit(perfMemory_exit_helper) failed");
5306     }
5307   }
5308 
5309   // Init pset_loadavg function pointer
5310   init_pset_getloadavg_ptr();
5311 
5312   return JNI_OK;
5313 }
5314 
5315 void os::init_3(void) {
5316   return;
5317 }
5318 
5319 // Mark the polling page as unreadable
5320 void os::make_polling_page_unreadable(void) {
5321   if( mprotect((char *)_polling_page, page_size, PROT_NONE) != 0 )
5322     fatal("Could not disable polling page");
5323 };
5324 
5325 // Mark the polling page as readable
5326 void os::make_polling_page_readable(void) {
5327   if( mprotect((char *)_polling_page, page_size, PROT_READ) != 0 )
5328     fatal("Could not enable polling page");
5329 };
5330 
5331 // OS interface.
5332 
5333 bool os::check_heap(bool force) { return true; }
5334 
5335 typedef int (*vsnprintf_t)(char* buf, size_t count, const char* fmt, va_list argptr);
5336 static vsnprintf_t sol_vsnprintf = NULL;
5337 
5338 int local_vsnprintf(char* buf, size_t count, const char* fmt, va_list argptr) {
5339   if (!sol_vsnprintf) {
5340     //search  for the named symbol in the objects that were loaded after libjvm
5341     void* where = RTLD_NEXT;
5342     if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5343         sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5344     if (!sol_vsnprintf){
5345       //search  for the named symbol in the objects that were loaded before libjvm
5346       where = RTLD_DEFAULT;
5347       if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5348         sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5349       assert(sol_vsnprintf != NULL, "vsnprintf not found");
5350     }
5351   }
5352   return (*sol_vsnprintf)(buf, count, fmt, argptr);
5353 }
5354 
5355 
5356 // Is a (classpath) directory empty?
5357 bool os::dir_is_empty(const char* path) {
5358   DIR *dir = NULL;
5359   struct dirent *ptr;
5360 
5361   dir = opendir(path);
5362   if (dir == NULL) return true;
5363 
5364   /* Scan the directory */
5365   bool result = true;
5366   char buf[sizeof(struct dirent) + MAX_PATH];
5367   struct dirent *dbuf = (struct dirent *) buf;
5368   while (result && (ptr = readdir(dir, dbuf)) != NULL) {
5369     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5370       result = false;
5371     }
5372   }
5373   closedir(dir);
5374   return result;
5375 }
5376 
5377 // This code originates from JDK's sysOpen and open64_w
5378 // from src/solaris/hpi/src/system_md.c
5379 
5380 #ifndef O_DELETE
5381 #define O_DELETE 0x10000
5382 #endif
5383 
5384 // Open a file. Unlink the file immediately after open returns
5385 // if the specified oflag has the O_DELETE flag set.
5386 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5387 
5388 int os::open(const char *path, int oflag, int mode) {
5389   if (strlen(path) > MAX_PATH - 1) {
5390     errno = ENAMETOOLONG;
5391     return -1;
5392   }
5393   int fd;
5394   int o_delete = (oflag & O_DELETE);
5395   oflag = oflag & ~O_DELETE;
5396 
5397   fd = ::open64(path, oflag, mode);
5398   if (fd == -1) return -1;
5399 
5400   //If the open succeeded, the file might still be a directory
5401   {
5402     struct stat64 buf64;
5403     int ret = ::fstat64(fd, &buf64);
5404     int st_mode = buf64.st_mode;
5405 
5406     if (ret != -1) {
5407       if ((st_mode & S_IFMT) == S_IFDIR) {
5408         errno = EISDIR;
5409         ::close(fd);
5410         return -1;
5411       }
5412     } else {
5413       ::close(fd);
5414       return -1;
5415     }
5416   }
5417     /*
5418      * 32-bit Solaris systems suffer from:
5419      *
5420      * - an historical default soft limit of 256 per-process file
5421      *   descriptors that is too low for many Java programs.
5422      *
5423      * - a design flaw where file descriptors created using stdio
5424      *   fopen must be less than 256, _even_ when the first limit above
5425      *   has been raised.  This can cause calls to fopen (but not calls to
5426      *   open, for example) to fail mysteriously, perhaps in 3rd party
5427      *   native code (although the JDK itself uses fopen).  One can hardly
5428      *   criticize them for using this most standard of all functions.
5429      *
5430      * We attempt to make everything work anyways by:
5431      *
5432      * - raising the soft limit on per-process file descriptors beyond
5433      *   256
5434      *
5435      * - As of Solaris 10u4, we can request that Solaris raise the 256
5436      *   stdio fopen limit by calling function enable_extended_FILE_stdio.
5437      *   This is done in init_2 and recorded in enabled_extended_FILE_stdio
5438      *
5439      * - If we are stuck on an old (pre 10u4) Solaris system, we can
5440      *   workaround the bug by remapping non-stdio file descriptors below
5441      *   256 to ones beyond 256, which is done below.
5442      *
5443      * See:
5444      * 1085341: 32-bit stdio routines should support file descriptors >255
5445      * 6533291: Work around 32-bit Solaris stdio limit of 256 open files
5446      * 6431278: Netbeans crash on 32 bit Solaris: need to call
5447      *          enable_extended_FILE_stdio() in VM initialisation
5448      * Giri Mandalika's blog
5449      * http://technopark02.blogspot.com/2005_05_01_archive.html
5450      */
5451 #ifndef  _LP64
5452      if ((!enabled_extended_FILE_stdio) && fd < 256) {
5453          int newfd = ::fcntl(fd, F_DUPFD, 256);
5454          if (newfd != -1) {
5455              ::close(fd);
5456              fd = newfd;
5457          }
5458      }
5459 #endif // 32-bit Solaris
5460     /*
5461      * All file descriptors that are opened in the JVM and not
5462      * specifically destined for a subprocess should have the
5463      * close-on-exec flag set.  If we don't set it, then careless 3rd
5464      * party native code might fork and exec without closing all
5465      * appropriate file descriptors (e.g. as we do in closeDescriptors in
5466      * UNIXProcess.c), and this in turn might:
5467      *
5468      * - cause end-of-file to fail to be detected on some file
5469      *   descriptors, resulting in mysterious hangs, or
5470      *
5471      * - might cause an fopen in the subprocess to fail on a system
5472      *   suffering from bug 1085341.
5473      *
5474      * (Yes, the default setting of the close-on-exec flag is a Unix
5475      * design flaw)
5476      *
5477      * See:
5478      * 1085341: 32-bit stdio routines should support file descriptors >255
5479      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5480      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5481      */
5482 #ifdef FD_CLOEXEC
5483     {
5484         int flags = ::fcntl(fd, F_GETFD);
5485         if (flags != -1)
5486             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5487     }
5488 #endif
5489 
5490   if (o_delete != 0) {
5491     ::unlink(path);
5492   }
5493   return fd;
5494 }
5495 
5496 // create binary file, rewriting existing file if required
5497 int os::create_binary_file(const char* path, bool rewrite_existing) {
5498   int oflags = O_WRONLY | O_CREAT;
5499   if (!rewrite_existing) {
5500     oflags |= O_EXCL;
5501   }
5502   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5503 }
5504 
5505 // return current position of file pointer
5506 jlong os::current_file_offset(int fd) {
5507   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5508 }
5509 
5510 // move file pointer to the specified offset
5511 jlong os::seek_to_file_offset(int fd, jlong offset) {
5512   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5513 }
5514 
5515 jlong os::lseek(int fd, jlong offset, int whence) {
5516   return (jlong) ::lseek64(fd, offset, whence);
5517 }
5518 
5519 char * os::native_path(char *path) {
5520   return path;
5521 }
5522 
5523 int os::ftruncate(int fd, jlong length) {
5524   return ::ftruncate64(fd, length);
5525 }
5526 
5527 int os::fsync(int fd)  {
5528   RESTARTABLE_RETURN_INT(::fsync(fd));
5529 }
5530 
5531 int os::available(int fd, jlong *bytes) {
5532   jlong cur, end;
5533   int mode;
5534   struct stat64 buf64;
5535 
5536   if (::fstat64(fd, &buf64) >= 0) {
5537     mode = buf64.st_mode;
5538     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5539       /*
5540       * XXX: is the following call interruptible? If so, this might
5541       * need to go through the INTERRUPT_IO() wrapper as for other
5542       * blocking, interruptible calls in this file.
5543       */
5544       int n,ioctl_return;
5545 
5546       INTERRUPTIBLE(::ioctl(fd, FIONREAD, &n),ioctl_return,os::Solaris::clear_interrupted);
5547       if (ioctl_return>= 0) {
5548           *bytes = n;
5549         return 1;
5550       }
5551     }
5552   }
5553   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5554     return 0;
5555   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5556     return 0;
5557   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5558     return 0;
5559   }
5560   *bytes = end - cur;
5561   return 1;
5562 }
5563 
5564 // Map a block of memory.
5565 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5566                      char *addr, size_t bytes, bool read_only,
5567                      bool allow_exec) {
5568   int prot;
5569   int flags;
5570 
5571   if (read_only) {
5572     prot = PROT_READ;
5573     flags = MAP_SHARED;
5574   } else {
5575     prot = PROT_READ | PROT_WRITE;
5576     flags = MAP_PRIVATE;
5577   }
5578 
5579   if (allow_exec) {
5580     prot |= PROT_EXEC;
5581   }
5582 
5583   if (addr != NULL) {
5584     flags |= MAP_FIXED;
5585   }
5586 
5587   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5588                                      fd, file_offset);
5589   if (mapped_address == MAP_FAILED) {
5590     return NULL;
5591   }
5592   return mapped_address;
5593 }
5594 
5595 
5596 // Remap a block of memory.
5597 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5598                        char *addr, size_t bytes, bool read_only,
5599                        bool allow_exec) {
5600   // same as map_memory() on this OS
5601   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5602                         allow_exec);
5603 }
5604 
5605 
5606 // Unmap a block of memory.
5607 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5608   return munmap(addr, bytes) == 0;
5609 }
5610 
5611 void os::pause() {
5612   char filename[MAX_PATH];
5613   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5614     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5615   } else {
5616     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5617   }
5618 
5619   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5620   if (fd != -1) {
5621     struct stat buf;
5622     ::close(fd);
5623     while (::stat(filename, &buf) == 0) {
5624       (void)::poll(NULL, 0, 100);
5625     }
5626   } else {
5627     jio_fprintf(stderr,
5628       "Could not open pause file '%s', continuing immediately.\n", filename);
5629   }
5630 }
5631 
5632 #ifndef PRODUCT
5633 #ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5634 // Turn this on if you need to trace synch operations.
5635 // Set RECORD_SYNCH_LIMIT to a large-enough value,
5636 // and call record_synch_enable and record_synch_disable
5637 // around the computation of interest.
5638 
5639 void record_synch(char* name, bool returning);  // defined below
5640 
5641 class RecordSynch {
5642   char* _name;
5643  public:
5644   RecordSynch(char* name) :_name(name)
5645                  { record_synch(_name, false); }
5646   ~RecordSynch() { record_synch(_name,   true);  }
5647 };
5648 
5649 #define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
5650 extern "C" ret name params {                                    \
5651   typedef ret name##_t params;                                  \
5652   static name##_t* implem = NULL;                               \
5653   static int callcount = 0;                                     \
5654   if (implem == NULL) {                                         \
5655     implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
5656     if (implem == NULL)  fatal(dlerror());                      \
5657   }                                                             \
5658   ++callcount;                                                  \
5659   RecordSynch _rs(#name);                                       \
5660   inner;                                                        \
5661   return implem args;                                           \
5662 }
5663 // in dbx, examine callcounts this way:
5664 // for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
5665 
5666 #define CHECK_POINTER_OK(p) \
5667   (!Universe::is_fully_initialized() || !Universe::is_reserved_heap((oop)(p)))
5668 #define CHECK_MU \
5669   if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
5670 #define CHECK_CV \
5671   if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
5672 #define CHECK_P(p) \
5673   if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
5674 
5675 #define CHECK_MUTEX(mutex_op) \
5676 CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
5677 
5678 CHECK_MUTEX(   mutex_lock)
5679 CHECK_MUTEX(  _mutex_lock)
5680 CHECK_MUTEX( mutex_unlock)
5681 CHECK_MUTEX(_mutex_unlock)
5682 CHECK_MUTEX( mutex_trylock)
5683 CHECK_MUTEX(_mutex_trylock)
5684 
5685 #define CHECK_COND(cond_op) \
5686 CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU;CHECK_CV);
5687 
5688 CHECK_COND( cond_wait);
5689 CHECK_COND(_cond_wait);
5690 CHECK_COND(_cond_wait_cancel);
5691 
5692 #define CHECK_COND2(cond_op) \
5693 CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU;CHECK_CV);
5694 
5695 CHECK_COND2( cond_timedwait);
5696 CHECK_COND2(_cond_timedwait);
5697 CHECK_COND2(_cond_timedwait_cancel);
5698 
5699 // do the _lwp_* versions too
5700 #define mutex_t lwp_mutex_t
5701 #define cond_t  lwp_cond_t
5702 CHECK_MUTEX(  _lwp_mutex_lock)
5703 CHECK_MUTEX(  _lwp_mutex_unlock)
5704 CHECK_MUTEX(  _lwp_mutex_trylock)
5705 CHECK_MUTEX( __lwp_mutex_lock)
5706 CHECK_MUTEX( __lwp_mutex_unlock)
5707 CHECK_MUTEX( __lwp_mutex_trylock)
5708 CHECK_MUTEX(___lwp_mutex_lock)
5709 CHECK_MUTEX(___lwp_mutex_unlock)
5710 
5711 CHECK_COND(  _lwp_cond_wait);
5712 CHECK_COND( __lwp_cond_wait);
5713 CHECK_COND(___lwp_cond_wait);
5714 
5715 CHECK_COND2(  _lwp_cond_timedwait);
5716 CHECK_COND2( __lwp_cond_timedwait);
5717 #undef mutex_t
5718 #undef cond_t
5719 
5720 CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5721 CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5722 CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5723 CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5724 CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5725 CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5726 CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5727 CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5728 
5729 
5730 // recording machinery:
5731 
5732 enum { RECORD_SYNCH_LIMIT = 200 };
5733 char* record_synch_name[RECORD_SYNCH_LIMIT];
5734 void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
5735 bool record_synch_returning[RECORD_SYNCH_LIMIT];
5736 thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
5737 int record_synch_count = 0;
5738 bool record_synch_enabled = false;
5739 
5740 // in dbx, examine recorded data this way:
5741 // for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
5742 
5743 void record_synch(char* name, bool returning) {
5744   if (record_synch_enabled) {
5745     if (record_synch_count < RECORD_SYNCH_LIMIT) {
5746       record_synch_name[record_synch_count] = name;
5747       record_synch_returning[record_synch_count] = returning;
5748       record_synch_thread[record_synch_count] = thr_self();
5749       record_synch_arg0ptr[record_synch_count] = &name;
5750       record_synch_count++;
5751     }
5752     // put more checking code here:
5753     // ...
5754   }
5755 }
5756 
5757 void record_synch_enable() {
5758   // start collecting trace data, if not already doing so
5759   if (!record_synch_enabled)  record_synch_count = 0;
5760   record_synch_enabled = true;
5761 }
5762 
5763 void record_synch_disable() {
5764   // stop collecting trace data
5765   record_synch_enabled = false;
5766 }
5767 
5768 #endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5769 #endif // PRODUCT
5770 
5771 const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5772 const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
5773                                (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5774 
5775 
5776 // JVMTI & JVM monitoring and management support
5777 // The thread_cpu_time() and current_thread_cpu_time() are only
5778 // supported if is_thread_cpu_time_supported() returns true.
5779 // They are not supported on Solaris T1.
5780 
5781 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5782 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5783 // of a thread.
5784 //
5785 // current_thread_cpu_time() and thread_cpu_time(Thread *)
5786 // returns the fast estimate available on the platform.
5787 
5788 // hrtime_t gethrvtime() return value includes
5789 // user time but does not include system time
5790 jlong os::current_thread_cpu_time() {
5791   return (jlong) gethrvtime();
5792 }
5793 
5794 jlong os::thread_cpu_time(Thread *thread) {
5795   // return user level CPU time only to be consistent with
5796   // what current_thread_cpu_time returns.
5797   // thread_cpu_time_info() must be changed if this changes
5798   return os::thread_cpu_time(thread, false /* user time only */);
5799 }
5800 
5801 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5802   if (user_sys_cpu_time) {
5803     return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5804   } else {
5805     return os::current_thread_cpu_time();
5806   }
5807 }
5808 
5809 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5810   char proc_name[64];
5811   int count;
5812   prusage_t prusage;
5813   jlong lwp_time;
5814   int fd;
5815 
5816   sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
5817                      getpid(),
5818                      thread->osthread()->lwp_id());
5819   fd = ::open(proc_name, O_RDONLY);
5820   if ( fd == -1 ) return -1;
5821 
5822   do {
5823     count = ::pread(fd,
5824                   (void *)&prusage.pr_utime,
5825                   thr_time_size,
5826                   thr_time_off);
5827   } while (count < 0 && errno == EINTR);
5828   ::close(fd);
5829   if ( count < 0 ) return -1;
5830 
5831   if (user_sys_cpu_time) {
5832     // user + system CPU time
5833     lwp_time = (((jlong)prusage.pr_stime.tv_sec +
5834                  (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
5835                  (jlong)prusage.pr_stime.tv_nsec +
5836                  (jlong)prusage.pr_utime.tv_nsec;
5837   } else {
5838     // user level CPU time only
5839     lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
5840                 (jlong)prusage.pr_utime.tv_nsec;
5841   }
5842 
5843   return(lwp_time);
5844 }
5845 
5846 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5847   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5848   info_ptr->may_skip_backward = false;    // elapsed time not wall time
5849   info_ptr->may_skip_forward = false;     // elapsed time not wall time
5850   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5851 }
5852 
5853 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5854   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5855   info_ptr->may_skip_backward = false;    // elapsed time not wall time
5856   info_ptr->may_skip_forward = false;     // elapsed time not wall time
5857   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5858 }
5859 
5860 bool os::is_thread_cpu_time_supported() {
5861   if ( os::Solaris::T2_libthread() || UseBoundThreads ) {
5862     return true;
5863   } else {
5864     return false;
5865   }
5866 }
5867 
5868 // System loadavg support.  Returns -1 if load average cannot be obtained.
5869 // Return the load average for our processor set if the primitive exists
5870 // (Solaris 9 and later).  Otherwise just return system wide loadavg.
5871 int os::loadavg(double loadavg[], int nelem) {
5872   if (pset_getloadavg_ptr != NULL) {
5873     return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
5874   } else {
5875     return ::getloadavg(loadavg, nelem);
5876   }
5877 }
5878 
5879 //---------------------------------------------------------------------------------
5880 
5881 bool os::find(address addr, outputStream* st) {
5882   Dl_info dlinfo;
5883   memset(&dlinfo, 0, sizeof(dlinfo));
5884   if (dladdr(addr, &dlinfo) != 0) {
5885     st->print(PTR_FORMAT ": ", addr);
5886     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5887       st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
5888     } else if (dlinfo.dli_fbase != NULL)
5889       st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
5890     else
5891       st->print("<absolute address>");
5892     if (dlinfo.dli_fname != NULL) {
5893       st->print(" in %s", dlinfo.dli_fname);
5894     }
5895     if (dlinfo.dli_fbase != NULL) {
5896       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5897     }
5898     st->cr();
5899 
5900     if (Verbose) {
5901       // decode some bytes around the PC
5902       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5903       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5904       address       lowest = (address) dlinfo.dli_sname;
5905       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5906       if (begin < lowest)  begin = lowest;
5907       Dl_info dlinfo2;
5908       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5909           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5910         end = (address) dlinfo2.dli_saddr;
5911       Disassembler::decode(begin, end, st);
5912     }
5913     return true;
5914   }
5915   return false;
5916 }
5917 
5918 // Following function has been added to support HotSparc's libjvm.so running
5919 // under Solaris production JDK 1.2.2 / 1.3.0.  These came from
5920 // src/solaris/hpi/native_threads in the EVM codebase.
5921 //
5922 // NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
5923 // libraries and should thus be removed. We will leave it behind for a while
5924 // until we no longer want to able to run on top of 1.3.0 Solaris production
5925 // JDK. See 4341971.
5926 
5927 #define STACK_SLACK 0x800
5928 
5929 extern "C" {
5930   intptr_t sysThreadAvailableStackWithSlack() {
5931     stack_t st;
5932     intptr_t retval, stack_top;
5933     retval = thr_stksegment(&st);
5934     assert(retval == 0, "incorrect return value from thr_stksegment");
5935     assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
5936     assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
5937     stack_top=(intptr_t)st.ss_sp-st.ss_size;
5938     return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
5939   }
5940 }
5941 
5942 // ObjectMonitor park-unpark infrastructure ...
5943 //
5944 // We implement Solaris and Linux PlatformEvents with the
5945 // obvious condvar-mutex-flag triple.
5946 // Another alternative that works quite well is pipes:
5947 // Each PlatformEvent consists of a pipe-pair.
5948 // The thread associated with the PlatformEvent
5949 // calls park(), which reads from the input end of the pipe.
5950 // Unpark() writes into the other end of the pipe.
5951 // The write-side of the pipe must be set NDELAY.
5952 // Unfortunately pipes consume a large # of handles.
5953 // Native solaris lwp_park() and lwp_unpark() work nicely, too.
5954 // Using pipes for the 1st few threads might be workable, however.
5955 //
5956 // park() is permitted to return spuriously.
5957 // Callers of park() should wrap the call to park() in
5958 // an appropriate loop.  A litmus test for the correct
5959 // usage of park is the following: if park() were modified
5960 // to immediately return 0 your code should still work,
5961 // albeit degenerating to a spin loop.
5962 //
5963 // An interesting optimization for park() is to use a trylock()
5964 // to attempt to acquire the mutex.  If the trylock() fails
5965 // then we know that a concurrent unpark() operation is in-progress.
5966 // in that case the park() code could simply set _count to 0
5967 // and return immediately.  The subsequent park() operation *might*
5968 // return immediately.  That's harmless as the caller of park() is
5969 // expected to loop.  By using trylock() we will have avoided a
5970 // avoided a context switch caused by contention on the per-thread mutex.
5971 //
5972 // TODO-FIXME:
5973 // 1.  Reconcile Doug's JSR166 j.u.c park-unpark with the
5974 //     objectmonitor implementation.
5975 // 2.  Collapse the JSR166 parker event, and the
5976 //     objectmonitor ParkEvent into a single "Event" construct.
5977 // 3.  In park() and unpark() add:
5978 //     assert (Thread::current() == AssociatedWith).
5979 // 4.  add spurious wakeup injection on a -XX:EarlyParkReturn=N switch.
5980 //     1-out-of-N park() operations will return immediately.
5981 //
5982 // _Event transitions in park()
5983 //   -1 => -1 : illegal
5984 //    1 =>  0 : pass - return immediately
5985 //    0 => -1 : block
5986 //
5987 // _Event serves as a restricted-range semaphore.
5988 //
5989 // Another possible encoding of _Event would be with
5990 // explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
5991 //
5992 // TODO-FIXME: add DTRACE probes for:
5993 // 1.   Tx parks
5994 // 2.   Ty unparks Tx
5995 // 3.   Tx resumes from park
5996 
5997 
5998 // value determined through experimentation
5999 #define ROUNDINGFIX 11
6000 
6001 // utility to compute the abstime argument to timedwait.
6002 // TODO-FIXME: switch from compute_abstime() to unpackTime().
6003 
6004 static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
6005   // millis is the relative timeout time
6006   // abstime will be the absolute timeout time
6007   if (millis < 0)  millis = 0;
6008   struct timeval now;
6009   int status = gettimeofday(&now, NULL);
6010   assert(status == 0, "gettimeofday");
6011   jlong seconds = millis / 1000;
6012   jlong max_wait_period;
6013 
6014   if (UseLWPSynchronization) {
6015     // forward port of fix for 4275818 (not sleeping long enough)
6016     // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
6017     // _lwp_cond_timedwait() used a round_down algorithm rather
6018     // than a round_up. For millis less than our roundfactor
6019     // it rounded down to 0 which doesn't meet the spec.
6020     // For millis > roundfactor we may return a bit sooner, but
6021     // since we can not accurately identify the patch level and
6022     // this has already been fixed in Solaris 9 and 8 we will
6023     // leave it alone rather than always rounding down.
6024 
6025     if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
6026        // It appears that when we go directly through Solaris _lwp_cond_timedwait()
6027            // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
6028            max_wait_period = 21000000;
6029   } else {
6030     max_wait_period = 50000000;
6031   }
6032   millis %= 1000;
6033   if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
6034      seconds = max_wait_period;
6035   }
6036   abstime->tv_sec = now.tv_sec  + seconds;
6037   long       usec = now.tv_usec + millis * 1000;
6038   if (usec >= 1000000) {
6039     abstime->tv_sec += 1;
6040     usec -= 1000000;
6041   }
6042   abstime->tv_nsec = usec * 1000;
6043   return abstime;
6044 }
6045 
6046 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
6047 // Conceptually TryPark() should be equivalent to park(0).
6048 
6049 int os::PlatformEvent::TryPark() {
6050   for (;;) {
6051     const int v = _Event ;
6052     guarantee ((v == 0) || (v == 1), "invariant") ;
6053     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
6054   }
6055 }
6056 
6057 void os::PlatformEvent::park() {           // AKA: down()
6058   // Invariant: Only the thread associated with the Event/PlatformEvent
6059   // may call park().
6060   int v ;
6061   for (;;) {
6062       v = _Event ;
6063       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6064   }
6065   guarantee (v >= 0, "invariant") ;
6066   if (v == 0) {
6067      // Do this the hard way by blocking ...
6068      // See http://monaco.sfbay/detail.jsf?cr=5094058.
6069      // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
6070      // Only for SPARC >= V8PlusA
6071 #if defined(__sparc) && defined(COMPILER2)
6072      if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6073 #endif
6074      int status = os::Solaris::mutex_lock(_mutex);
6075      assert_status(status == 0, status,  "mutex_lock");
6076      guarantee (_nParked == 0, "invariant") ;
6077      ++ _nParked ;
6078      while (_Event < 0) {
6079         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
6080         // Treat this the same as if the wait was interrupted
6081         // With usr/lib/lwp going to kernel, always handle ETIME
6082         status = os::Solaris::cond_wait(_cond, _mutex);
6083         if (status == ETIME) status = EINTR ;
6084         assert_status(status == 0 || status == EINTR, status, "cond_wait");
6085      }
6086      -- _nParked ;
6087      _Event = 0 ;
6088      status = os::Solaris::mutex_unlock(_mutex);
6089      assert_status(status == 0, status, "mutex_unlock");
6090     // Paranoia to ensure our locked and lock-free paths interact
6091     // correctly with each other.
6092     OrderAccess::fence();
6093   }
6094 }
6095 
6096 int os::PlatformEvent::park(jlong millis) {
6097   guarantee (_nParked == 0, "invariant") ;
6098   int v ;
6099   for (;;) {
6100       v = _Event ;
6101       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6102   }
6103   guarantee (v >= 0, "invariant") ;
6104   if (v != 0) return OS_OK ;
6105 
6106   int ret = OS_TIMEOUT;
6107   timestruc_t abst;
6108   compute_abstime (&abst, millis);
6109 
6110   // See http://monaco.sfbay/detail.jsf?cr=5094058.
6111   // For Solaris SPARC set fprs.FEF=0 prior to parking.
6112   // Only for SPARC >= V8PlusA
6113 #if defined(__sparc) && defined(COMPILER2)
6114  if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6115 #endif
6116   int status = os::Solaris::mutex_lock(_mutex);
6117   assert_status(status == 0, status, "mutex_lock");
6118   guarantee (_nParked == 0, "invariant") ;
6119   ++ _nParked ;
6120   while (_Event < 0) {
6121      int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
6122      assert_status(status == 0 || status == EINTR ||
6123                    status == ETIME || status == ETIMEDOUT,
6124                    status, "cond_timedwait");
6125      if (!FilterSpuriousWakeups) break ;                // previous semantics
6126      if (status == ETIME || status == ETIMEDOUT) break ;
6127      // We consume and ignore EINTR and spurious wakeups.
6128   }
6129   -- _nParked ;
6130   if (_Event >= 0) ret = OS_OK ;
6131   _Event = 0 ;
6132   status = os::Solaris::mutex_unlock(_mutex);
6133   assert_status(status == 0, status, "mutex_unlock");
6134   // Paranoia to ensure our locked and lock-free paths interact
6135   // correctly with each other.
6136   OrderAccess::fence();
6137   return ret;
6138 }
6139 
6140 void os::PlatformEvent::unpark() {
6141   // Transitions for _Event:
6142   //    0 :=> 1
6143   //    1 :=> 1
6144   //   -1 :=> either 0 or 1; must signal target thread
6145   //          That is, we can safely transition _Event from -1 to either
6146   //          0 or 1. Forcing 1 is slightly more efficient for back-to-back
6147   //          unpark() calls.
6148   // See also: "Semaphores in Plan 9" by Mullender & Cox
6149   //
6150   // Note: Forcing a transition from "-1" to "1" on an unpark() means
6151   // that it will take two back-to-back park() calls for the owning
6152   // thread to block. This has the benefit of forcing a spurious return
6153   // from the first park() call after an unpark() call which will help
6154   // shake out uses of park() and unpark() without condition variables.
6155 
6156   if (Atomic::xchg(1, &_Event) >= 0) return;
6157 
6158   // If the thread associated with the event was parked, wake it.
6159   // Wait for the thread assoc with the PlatformEvent to vacate.
6160   int status = os::Solaris::mutex_lock(_mutex);
6161   assert_status(status == 0, status, "mutex_lock");
6162   int AnyWaiters = _nParked;
6163   status = os::Solaris::mutex_unlock(_mutex);
6164   assert_status(status == 0, status, "mutex_unlock");
6165   guarantee(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
6166   if (AnyWaiters != 0) {
6167     // We intentional signal *after* dropping the lock
6168     // to avoid a common class of futile wakeups.
6169     status = os::Solaris::cond_signal(_cond);
6170     assert_status(status == 0, status, "cond_signal");
6171   }
6172 }
6173 
6174 // JSR166
6175 // -------------------------------------------------------
6176 
6177 /*
6178  * The solaris and linux implementations of park/unpark are fairly
6179  * conservative for now, but can be improved. They currently use a
6180  * mutex/condvar pair, plus _counter.
6181  * Park decrements _counter if > 0, else does a condvar wait.  Unpark
6182  * sets count to 1 and signals condvar.  Only one thread ever waits
6183  * on the condvar. Contention seen when trying to park implies that someone
6184  * is unparking you, so don't wait. And spurious returns are fine, so there
6185  * is no need to track notifications.
6186  */
6187 
6188 #define MAX_SECS 100000000
6189 /*
6190  * This code is common to linux and solaris and will be moved to a
6191  * common place in dolphin.
6192  *
6193  * The passed in time value is either a relative time in nanoseconds
6194  * or an absolute time in milliseconds. Either way it has to be unpacked
6195  * into suitable seconds and nanoseconds components and stored in the
6196  * given timespec structure.
6197  * Given time is a 64-bit value and the time_t used in the timespec is only
6198  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
6199  * overflow if times way in the future are given. Further on Solaris versions
6200  * prior to 10 there is a restriction (see cond_timedwait) that the specified
6201  * number of seconds, in abstime, is less than current_time  + 100,000,000.
6202  * As it will be 28 years before "now + 100000000" will overflow we can
6203  * ignore overflow and just impose a hard-limit on seconds using the value
6204  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
6205  * years from "now".
6206  */
6207 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
6208   assert (time > 0, "convertTime");
6209 
6210   struct timeval now;
6211   int status = gettimeofday(&now, NULL);
6212   assert(status == 0, "gettimeofday");
6213 
6214   time_t max_secs = now.tv_sec + MAX_SECS;
6215 
6216   if (isAbsolute) {
6217     jlong secs = time / 1000;
6218     if (secs > max_secs) {
6219       absTime->tv_sec = max_secs;
6220     }
6221     else {
6222       absTime->tv_sec = secs;
6223     }
6224     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
6225   }
6226   else {
6227     jlong secs = time / NANOSECS_PER_SEC;
6228     if (secs >= MAX_SECS) {
6229       absTime->tv_sec = max_secs;
6230       absTime->tv_nsec = 0;
6231     }
6232     else {
6233       absTime->tv_sec = now.tv_sec + secs;
6234       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
6235       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6236         absTime->tv_nsec -= NANOSECS_PER_SEC;
6237         ++absTime->tv_sec; // note: this must be <= max_secs
6238       }
6239     }
6240   }
6241   assert(absTime->tv_sec >= 0, "tv_sec < 0");
6242   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
6243   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
6244   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
6245 }
6246 
6247 void Parker::park(bool isAbsolute, jlong time) {
6248   // Ideally we'd do something useful while spinning, such
6249   // as calling unpackTime().
6250 
6251   // Optional fast-path check:
6252   // Return immediately if a permit is available.
6253   // We depend on Atomic::xchg() having full barrier semantics
6254   // since we are doing a lock-free update to _counter.
6255   if (Atomic::xchg(0, &_counter) > 0) return;
6256 
6257   // Optional fast-exit: Check interrupt before trying to wait
6258   Thread* thread = Thread::current();
6259   assert(thread->is_Java_thread(), "Must be JavaThread");
6260   JavaThread *jt = (JavaThread *)thread;
6261   if (Thread::is_interrupted(thread, false)) {
6262     return;
6263   }
6264 
6265   // First, demultiplex/decode time arguments
6266   timespec absTime;
6267   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
6268     return;
6269   }
6270   if (time > 0) {
6271     // Warning: this code might be exposed to the old Solaris time
6272     // round-down bugs.  Grep "roundingFix" for details.
6273     unpackTime(&absTime, isAbsolute, time);
6274   }
6275 
6276   // Enter safepoint region
6277   // Beware of deadlocks such as 6317397.
6278   // The per-thread Parker:: _mutex is a classic leaf-lock.
6279   // In particular a thread must never block on the Threads_lock while
6280   // holding the Parker:: mutex.  If safepoints are pending both the
6281   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6282   ThreadBlockInVM tbivm(jt);
6283 
6284   // Don't wait if cannot get lock since interference arises from
6285   // unblocking.  Also. check interrupt before trying wait
6286   if (Thread::is_interrupted(thread, false) ||
6287       os::Solaris::mutex_trylock(_mutex) != 0) {
6288     return;
6289   }
6290 
6291   int status ;
6292 
6293   if (_counter > 0)  { // no wait needed
6294     _counter = 0;
6295     status = os::Solaris::mutex_unlock(_mutex);
6296     assert (status == 0, "invariant") ;
6297     // Paranoia to ensure our locked and lock-free paths interact
6298     // correctly with each other and Java-level accesses.
6299     OrderAccess::fence();
6300     return;
6301   }
6302 
6303 #ifdef ASSERT
6304   // Don't catch signals while blocked; let the running threads have the signals.
6305   // (This allows a debugger to break into the running thread.)
6306   sigset_t oldsigs;
6307   sigset_t* allowdebug_blocked = os::Solaris::allowdebug_blocked_signals();
6308   thr_sigsetmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6309 #endif
6310 
6311   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6312   jt->set_suspend_equivalent();
6313   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6314 
6315   // Do this the hard way by blocking ...
6316   // See http://monaco.sfbay/detail.jsf?cr=5094058.
6317   // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
6318   // Only for SPARC >= V8PlusA
6319 #if defined(__sparc) && defined(COMPILER2)
6320   if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6321 #endif
6322 
6323   if (time == 0) {
6324     status = os::Solaris::cond_wait (_cond, _mutex) ;
6325   } else {
6326     status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
6327   }
6328   // Note that an untimed cond_wait() can sometimes return ETIME on older
6329   // versions of the Solaris.
6330   assert_status(status == 0 || status == EINTR ||
6331                 status == ETIME || status == ETIMEDOUT,
6332                 status, "cond_timedwait");
6333 
6334 #ifdef ASSERT
6335   thr_sigsetmask(SIG_SETMASK, &oldsigs, NULL);
6336 #endif
6337   _counter = 0 ;
6338   status = os::Solaris::mutex_unlock(_mutex);
6339   assert_status(status == 0, status, "mutex_unlock") ;
6340   // Paranoia to ensure our locked and lock-free paths interact
6341   // correctly with each other and Java-level accesses.
6342   OrderAccess::fence();
6343 
6344   // If externally suspended while waiting, re-suspend
6345   if (jt->handle_special_suspend_equivalent_condition()) {
6346     jt->java_suspend_self();
6347   }
6348 }
6349 
6350 void Parker::unpark() {
6351   int s, status ;
6352   status = os::Solaris::mutex_lock (_mutex) ;
6353   assert (status == 0, "invariant") ;
6354   s = _counter;
6355   _counter = 1;
6356   status = os::Solaris::mutex_unlock (_mutex) ;
6357   assert (status == 0, "invariant") ;
6358 
6359   if (s < 1) {
6360     status = os::Solaris::cond_signal (_cond) ;
6361     assert (status == 0, "invariant") ;
6362   }
6363 }
6364 
6365 extern char** environ;
6366 
6367 // Run the specified command in a separate process. Return its exit value,
6368 // or -1 on failure (e.g. can't fork a new process).
6369 // Unlike system(), this function can be called from signal handler. It
6370 // doesn't block SIGINT et al.
6371 int os::fork_and_exec(char* cmd) {
6372   char * argv[4];
6373   argv[0] = (char *)"sh";
6374   argv[1] = (char *)"-c";
6375   argv[2] = cmd;
6376   argv[3] = NULL;
6377 
6378   // fork is async-safe, fork1 is not so can't use in signal handler
6379   pid_t pid;
6380   Thread* t = ThreadLocalStorage::get_thread_slow();
6381   if (t != NULL && t->is_inside_signal_handler()) {
6382     pid = fork();
6383   } else {
6384     pid = fork1();
6385   }
6386 
6387   if (pid < 0) {
6388     // fork failed
6389     warning("fork failed: %s", strerror(errno));
6390     return -1;
6391 
6392   } else if (pid == 0) {
6393     // child process
6394 
6395     // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
6396     execve("/usr/bin/sh", argv, environ);
6397 
6398     // execve failed
6399     _exit(-1);
6400 
6401   } else  {
6402     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6403     // care about the actual exit code, for now.
6404 
6405     int status;
6406 
6407     // Wait for the child process to exit.  This returns immediately if
6408     // the child has already exited. */
6409     while (waitpid(pid, &status, 0) < 0) {
6410         switch (errno) {
6411         case ECHILD: return 0;
6412         case EINTR: break;
6413         default: return -1;
6414         }
6415     }
6416 
6417     if (WIFEXITED(status)) {
6418        // The child exited normally; get its exit code.
6419        return WEXITSTATUS(status);
6420     } else if (WIFSIGNALED(status)) {
6421        // The child exited because of a signal
6422        // The best value to return is 0x80 + signal number,
6423        // because that is what all Unix shells do, and because
6424        // it allows callers to distinguish between process exit and
6425        // process death by signal.
6426        return 0x80 + WTERMSIG(status);
6427     } else {
6428        // Unknown exit code; pass it through
6429        return status;
6430     }
6431   }
6432 }
6433 
6434 // is_headless_jre()
6435 //
6436 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6437 // in order to report if we are running in a headless jre
6438 //
6439 // Since JDK8 xawt/libmawt.so was moved into the same directory
6440 // as libawt.so, and renamed libawt_xawt.so
6441 //
6442 bool os::is_headless_jre() {
6443     struct stat statbuf;
6444     char buf[MAXPATHLEN];
6445     char libmawtpath[MAXPATHLEN];
6446     const char *xawtstr  = "/xawt/libmawt.so";
6447     const char *new_xawtstr = "/libawt_xawt.so";
6448     char *p;
6449 
6450     // Get path to libjvm.so
6451     os::jvm_path(buf, sizeof(buf));
6452 
6453     // Get rid of libjvm.so
6454     p = strrchr(buf, '/');
6455     if (p == NULL) return false;
6456     else *p = '\0';
6457 
6458     // Get rid of client or server
6459     p = strrchr(buf, '/');
6460     if (p == NULL) return false;
6461     else *p = '\0';
6462 
6463     // check xawt/libmawt.so
6464     strcpy(libmawtpath, buf);
6465     strcat(libmawtpath, xawtstr);
6466     if (::stat(libmawtpath, &statbuf) == 0) return false;
6467 
6468     // check libawt_xawt.so
6469     strcpy(libmawtpath, buf);
6470     strcat(libmawtpath, new_xawtstr);
6471     if (::stat(libmawtpath, &statbuf) == 0) return false;
6472 
6473     return true;
6474 }
6475 
6476 size_t os::write(int fd, const void *buf, unsigned int nBytes) {
6477   INTERRUPTIBLE_RETURN_INT(::write(fd, buf, nBytes), os::Solaris::clear_interrupted);
6478 }
6479 
6480 int os::close(int fd) {
6481   return ::close(fd);
6482 }
6483 
6484 int os::socket_close(int fd) {
6485   return ::close(fd);
6486 }
6487 
6488 int os::recv(int fd, char* buf, size_t nBytes, uint flags) {
6489   INTERRUPTIBLE_RETURN_INT((int)::recv(fd, buf, nBytes, flags), os::Solaris::clear_interrupted);
6490 }
6491 
6492 int os::send(int fd, char* buf, size_t nBytes, uint flags) {
6493   INTERRUPTIBLE_RETURN_INT((int)::send(fd, buf, nBytes, flags), os::Solaris::clear_interrupted);
6494 }
6495 
6496 int os::raw_send(int fd, char* buf, size_t nBytes, uint flags) {
6497   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
6498 }
6499 
6500 // As both poll and select can be interrupted by signals, we have to be
6501 // prepared to restart the system call after updating the timeout, unless
6502 // a poll() is done with timeout == -1, in which case we repeat with this
6503 // "wait forever" value.
6504 
6505 int os::timeout(int fd, long timeout) {
6506   int res;
6507   struct timeval t;
6508   julong prevtime, newtime;
6509   static const char* aNull = 0;
6510   struct pollfd pfd;
6511   pfd.fd = fd;
6512   pfd.events = POLLIN;
6513 
6514   gettimeofday(&t, &aNull);
6515   prevtime = ((julong)t.tv_sec * 1000)  +  t.tv_usec / 1000;
6516 
6517   for(;;) {
6518     INTERRUPTIBLE_NORESTART(::poll(&pfd, 1, timeout), res, os::Solaris::clear_interrupted);
6519     if(res == OS_ERR && errno == EINTR) {
6520         if(timeout != -1) {
6521           gettimeofday(&t, &aNull);
6522           newtime = ((julong)t.tv_sec * 1000)  +  t.tv_usec /1000;
6523           timeout -= newtime - prevtime;
6524           if(timeout <= 0)
6525             return OS_OK;
6526           prevtime = newtime;
6527         }
6528     } else return res;
6529   }
6530 }
6531 
6532 int os::connect(int fd, struct sockaddr *him, socklen_t len) {
6533   int _result;
6534   INTERRUPTIBLE_NORESTART(::connect(fd, him, len), _result,\
6535                           os::Solaris::clear_interrupted);
6536 
6537   // Depending on when thread interruption is reset, _result could be
6538   // one of two values when errno == EINTR
6539 
6540   if (((_result == OS_INTRPT) || (_result == OS_ERR))
6541       && (errno == EINTR)) {
6542      /* restarting a connect() changes its errno semantics */
6543      INTERRUPTIBLE(::connect(fd, him, len), _result,\
6544                    os::Solaris::clear_interrupted);
6545      /* undo these changes */
6546      if (_result == OS_ERR) {
6547        if (errno == EALREADY) {
6548          errno = EINPROGRESS; /* fall through */
6549        } else if (errno == EISCONN) {
6550          errno = 0;
6551          return OS_OK;
6552        }
6553      }
6554    }
6555    return _result;
6556  }
6557 
6558 int os::accept(int fd, struct sockaddr* him, socklen_t* len) {
6559   if (fd < 0) {
6560     return OS_ERR;
6561   }
6562   INTERRUPTIBLE_RETURN_INT((int)::accept(fd, him, len),\
6563                            os::Solaris::clear_interrupted);
6564 }
6565 
6566 int os::recvfrom(int fd, char* buf, size_t nBytes, uint flags,
6567                  sockaddr* from, socklen_t* fromlen) {
6568   INTERRUPTIBLE_RETURN_INT((int)::recvfrom(fd, buf, nBytes, flags, from, fromlen),\
6569                            os::Solaris::clear_interrupted);
6570 }
6571 
6572 int os::sendto(int fd, char* buf, size_t len, uint flags,
6573                struct sockaddr* to, socklen_t tolen) {
6574   INTERRUPTIBLE_RETURN_INT((int)::sendto(fd, buf, len, flags, to, tolen),\
6575                            os::Solaris::clear_interrupted);
6576 }
6577 
6578 int os::socket_available(int fd, jint *pbytes) {
6579   if (fd < 0) {
6580     return OS_OK;
6581   }
6582   int ret;
6583   RESTARTABLE(::ioctl(fd, FIONREAD, pbytes), ret);
6584   // note: ioctl can return 0 when successful, JVM_SocketAvailable
6585   // is expected to return 0 on failure and 1 on success to the jdk.
6586   return (ret == OS_ERR) ? 0 : 1;
6587 }
6588 
6589 int os::bind(int fd, struct sockaddr* him, socklen_t len) {
6590    INTERRUPTIBLE_RETURN_INT_NORESTART(::bind(fd, him, len),\
6591                                       os::Solaris::clear_interrupted);
6592 }
6593 
6594 // Get the default path to the core file
6595 // Returns the length of the string
6596 int os::get_core_path(char* buffer, size_t bufferSize) {
6597   const char* p = get_current_directory(buffer, bufferSize);
6598 
6599   if (p == NULL) {
6600     assert(p != NULL, "failed to get current directory");
6601     return 0;
6602   }
6603 
6604   return strlen(buffer);
6605 }
6606 
6607 #ifndef PRODUCT
6608 void TestReserveMemorySpecial_test() {
6609   // No tests available for this platform
6610 }
6611 #endif