1 /*
   2  * Copyright (c) 1997, 2012, 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 "interpreter/interpreter.hpp"
  33 #include "jvm_solaris.h"
  34 #include "memory/allocation.inline.hpp"
  35 #include "memory/filemap.hpp"
  36 #include "mutex_solaris.inline.hpp"
  37 #include "oops/oop.inline.hpp"
  38 #include "os_share_solaris.hpp"
  39 #include "prims/jniFastGetField.hpp"
  40 #include "prims/jvm.h"
  41 #include "prims/jvm_misc.hpp"
  42 #include "runtime/arguments.hpp"
  43 #include "runtime/extendedPC.hpp"
  44 #include "runtime/globals.hpp"
  45 #include "runtime/interfaceSupport.hpp"
  46 #include "runtime/java.hpp"
  47 #include "runtime/javaCalls.hpp"
  48 #include "runtime/mutexLocker.hpp"
  49 #include "runtime/objectMonitor.hpp"
  50 #include "runtime/osThread.hpp"
  51 #include "runtime/perfMemory.hpp"
  52 #include "runtime/sharedRuntime.hpp"
  53 #include "runtime/statSampler.hpp"
  54 #include "runtime/stubRoutines.hpp"
  55 #include "runtime/threadCritical.hpp"
  56 #include "runtime/timer.hpp"
  57 #include "services/attachListener.hpp"
  58 #include "services/runtimeService.hpp"
  59 #include "thread_solaris.inline.hpp"
  60 #include "utilities/decoder.hpp"
  61 #include "utilities/defaultStream.hpp"
  62 #include "utilities/events.hpp"
  63 #include "utilities/growableArray.hpp"
  64 #include "utilities/vmError.hpp"
  65 #ifdef TARGET_ARCH_x86
  66 # include "assembler_x86.inline.hpp"
  67 # include "nativeInst_x86.hpp"
  68 #endif
  69 #ifdef TARGET_ARCH_sparc
  70 # include "assembler_sparc.inline.hpp"
  71 # include "nativeInst_sparc.hpp"
  72 #endif
  73 #ifdef COMPILER1
  74 #include "c1/c1_Runtime1.hpp"
  75 #endif
  76 #ifdef COMPILER2
  77 #include "opto/runtime.hpp"
  78 #endif
  79 
  80 // put OS-includes here
  81 # include <dlfcn.h>
  82 # include <errno.h>
  83 # include <exception>
  84 # include <link.h>
  85 # include <poll.h>
  86 # include <pthread.h>
  87 # include <pwd.h>
  88 # include <schedctl.h>
  89 # include <setjmp.h>
  90 # include <signal.h>
  91 # include <stdio.h>
  92 # include <alloca.h>
  93 # include <sys/filio.h>
  94 # include <sys/ipc.h>
  95 # include <sys/lwp.h>
  96 # include <sys/machelf.h>     // for elf Sym structure used by dladdr1
  97 # include <sys/mman.h>
  98 # include <sys/processor.h>
  99 # include <sys/procset.h>
 100 # include <sys/pset.h>
 101 # include <sys/resource.h>
 102 # include <sys/shm.h>
 103 # include <sys/socket.h>
 104 # include <sys/stat.h>
 105 # include <sys/systeminfo.h>
 106 # include <sys/time.h>
 107 # include <sys/times.h>
 108 # include <sys/types.h>
 109 # include <sys/wait.h>
 110 # include <sys/utsname.h>
 111 # include <thread.h>
 112 # include <unistd.h>
 113 # include <sys/priocntl.h>
 114 # include <sys/rtpriocntl.h>
 115 # include <sys/tspriocntl.h>
 116 # include <sys/iapriocntl.h>
 117 # include <sys/fxpriocntl.h>
 118 # include <sys/loadavg.h>
 119 # include <string.h>
 120 # include <stdio.h>
 121 
 122 # define _STRUCTURED_PROC 1  //  this gets us the new structured proc interfaces of 5.6 & later
 123 # include <sys/procfs.h>     //  see comment in <sys/procfs.h>
 124 
 125 #define MAX_PATH (2 * K)
 126 
 127 // for timer info max values which include all bits
 128 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 129 
 130 #ifdef _GNU_SOURCE
 131 // See bug #6514594
 132 extern "C" int madvise(caddr_t, size_t, int);
 133 extern "C" int memcntl(caddr_t addr, size_t len, int cmd, caddr_t arg,
 134                        int attr, int mask);
 135 #endif //_GNU_SOURCE
 136 
 137 /*
 138   MPSS Changes Start.
 139   The JVM binary needs to be built and run on pre-Solaris 9
 140   systems, but the constants needed by MPSS are only in Solaris 9
 141   header files.  They are textually replicated here to allow
 142   building on earlier systems.  Once building on Solaris 8 is
 143   no longer a requirement, these #defines can be replaced by ordinary
 144   system .h inclusion.
 145 
 146   In earlier versions of the  JDK and Solaris, we used ISM for large pages.
 147   But ISM requires shared memory to achieve this and thus has many caveats.
 148   MPSS is a fully transparent and is a cleaner way to get large pages.
 149   Although we still require keeping ISM for backward compatiblitiy as well as
 150   giving the opportunity to use large pages on older systems it is
 151   recommended that MPSS be used for Solaris 9 and above.
 152 
 153 */
 154 
 155 #ifndef MC_HAT_ADVISE
 156 
 157 struct memcntl_mha {
 158   uint_t          mha_cmd;        /* command(s) */
 159   uint_t          mha_flags;
 160   size_t          mha_pagesize;
 161 };
 162 #define MC_HAT_ADVISE   7       /* advise hat map size */
 163 #define MHA_MAPSIZE_VA  0x1     /* set preferred page size */
 164 #define MAP_ALIGN       0x200   /* addr specifies alignment */
 165 
 166 #endif
 167 // MPSS Changes End.
 168 
 169 
 170 // Here are some liblgrp types from sys/lgrp_user.h to be able to
 171 // compile on older systems without this header file.
 172 
 173 #ifndef MADV_ACCESS_LWP
 174 # define  MADV_ACCESS_LWP         7       /* next LWP to access heavily */
 175 #endif
 176 #ifndef MADV_ACCESS_MANY
 177 # define  MADV_ACCESS_MANY        8       /* many processes to access heavily */
 178 #endif
 179 
 180 #ifndef LGRP_RSRC_CPU
 181 # define LGRP_RSRC_CPU           0       /* CPU resources */
 182 #endif
 183 #ifndef LGRP_RSRC_MEM
 184 # define LGRP_RSRC_MEM           1       /* memory resources */
 185 #endif
 186 
 187 // Some more macros from sys/mman.h that are not present in Solaris 8.
 188 
 189 #ifndef MAX_MEMINFO_CNT
 190 /*
 191  * info_req request type definitions for meminfo
 192  * request types starting with MEMINFO_V are used for Virtual addresses
 193  * and should not be mixed with MEMINFO_PLGRP which is targeted for Physical
 194  * addresses
 195  */
 196 # define MEMINFO_SHIFT           16
 197 # define MEMINFO_MASK            (0xFF << MEMINFO_SHIFT)
 198 # define MEMINFO_VPHYSICAL       (0x01 << MEMINFO_SHIFT) /* get physical addr */
 199 # define MEMINFO_VLGRP           (0x02 << MEMINFO_SHIFT) /* get lgroup */
 200 # define MEMINFO_VPAGESIZE       (0x03 << MEMINFO_SHIFT) /* size of phys page */
 201 # define MEMINFO_VREPLCNT        (0x04 << MEMINFO_SHIFT) /* no. of replica */
 202 # define MEMINFO_VREPL           (0x05 << MEMINFO_SHIFT) /* physical replica */
 203 # define MEMINFO_VREPL_LGRP      (0x06 << MEMINFO_SHIFT) /* lgrp of replica */
 204 # define MEMINFO_PLGRP           (0x07 << MEMINFO_SHIFT) /* lgroup for paddr */
 205 
 206 /* maximum number of addresses meminfo() can process at a time */
 207 # define MAX_MEMINFO_CNT 256
 208 
 209 /* maximum number of request types */
 210 # define MAX_MEMINFO_REQ 31
 211 #endif
 212 
 213 // see thr_setprio(3T) for the basis of these numbers
 214 #define MinimumPriority 0
 215 #define NormalPriority  64
 216 #define MaximumPriority 127
 217 
 218 // Values for ThreadPriorityPolicy == 1
 219 int prio_policy1[CriticalPriority+1] = {
 220   -99999,  0, 16,  32,  48,  64,
 221           80, 96, 112, 124, 127, 127 };
 222 
 223 // System parameters used internally
 224 static clock_t clock_tics_per_sec = 100;
 225 
 226 // Track if we have called enable_extended_FILE_stdio (on Solaris 10u4+)
 227 static bool enabled_extended_FILE_stdio = false;
 228 
 229 // For diagnostics to print a message once. see run_periodic_checks
 230 static bool check_addr0_done = false;
 231 static sigset_t check_signal_done;
 232 static bool check_signals = true;
 233 
 234 address os::Solaris::handler_start;  // start pc of thr_sighndlrinfo
 235 address os::Solaris::handler_end;    // end pc of thr_sighndlrinfo
 236 
 237 address os::Solaris::_main_stack_base = NULL;  // 4352906 workaround
 238 
 239 
 240 // "default" initializers for missing libc APIs
 241 extern "C" {
 242   static int lwp_mutex_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
 243   static int lwp_mutex_destroy(mutex_t *mx)                 { return 0; }
 244 
 245   static int lwp_cond_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
 246   static int lwp_cond_destroy(cond_t *cv)                   { return 0; }
 247 }
 248 
 249 // "default" initializers for pthread-based synchronization
 250 extern "C" {
 251   static int pthread_mutex_default_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
 252   static int pthread_cond_default_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
 253 }
 254 
 255 // Thread Local Storage
 256 // This is common to all Solaris platforms so it is defined here,
 257 // in this common file.
 258 // The declarations are in the os_cpu threadLS*.hpp files.
 259 //
 260 // Static member initialization for TLS
 261 Thread* ThreadLocalStorage::_get_thread_cache[ThreadLocalStorage::_pd_cache_size] = {NULL};
 262 
 263 #ifndef PRODUCT
 264 #define _PCT(n,d)       ((100.0*(double)(n))/(double)(d))
 265 
 266 int ThreadLocalStorage::_tcacheHit = 0;
 267 int ThreadLocalStorage::_tcacheMiss = 0;
 268 
 269 void ThreadLocalStorage::print_statistics() {
 270   int total = _tcacheMiss+_tcacheHit;
 271   tty->print_cr("Thread cache hits %d misses %d total %d percent %f\n",
 272                 _tcacheHit, _tcacheMiss, total, _PCT(_tcacheHit, total));
 273 }
 274 #undef _PCT
 275 #endif // PRODUCT
 276 
 277 Thread* ThreadLocalStorage::get_thread_via_cache_slowly(uintptr_t raw_id,
 278                                                         int index) {
 279   Thread *thread = get_thread_slow();
 280   if (thread != NULL) {
 281     address sp = os::current_stack_pointer();
 282     guarantee(thread->_stack_base == NULL ||
 283               (sp <= thread->_stack_base &&
 284                  sp >= thread->_stack_base - thread->_stack_size) ||
 285                is_error_reported(),
 286               "sp must be inside of selected thread stack");
 287 
 288     thread->set_self_raw_id(raw_id);  // mark for quick retrieval
 289     _get_thread_cache[ index ] = thread;
 290   }
 291   return thread;
 292 }
 293 
 294 
 295 static const double all_zero[ sizeof(Thread) / sizeof(double) + 1 ] = {0};
 296 #define NO_CACHED_THREAD ((Thread*)all_zero)
 297 
 298 void ThreadLocalStorage::pd_set_thread(Thread* thread) {
 299 
 300   // Store the new value before updating the cache to prevent a race
 301   // between get_thread_via_cache_slowly() and this store operation.
 302   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
 303 
 304   // Update thread cache with new thread if setting on thread create,
 305   // or NO_CACHED_THREAD (zeroed) thread if resetting thread on exit.
 306   uintptr_t raw = pd_raw_thread_id();
 307   int ix = pd_cache_index(raw);
 308   _get_thread_cache[ix] = thread == NULL ? NO_CACHED_THREAD : thread;
 309 }
 310 
 311 void ThreadLocalStorage::pd_init() {
 312   for (int i = 0; i < _pd_cache_size; i++) {
 313     _get_thread_cache[i] = NO_CACHED_THREAD;
 314   }
 315 }
 316 
 317 // Invalidate all the caches (happens to be the same as pd_init).
 318 void ThreadLocalStorage::pd_invalidate_all() { pd_init(); }
 319 
 320 #undef NO_CACHED_THREAD
 321 
 322 // END Thread Local Storage
 323 
 324 static inline size_t adjust_stack_size(address base, size_t size) {
 325   if ((ssize_t)size < 0) {
 326     // 4759953: Compensate for ridiculous stack size.
 327     size = max_intx;
 328   }
 329   if (size > (size_t)base) {
 330     // 4812466: Make sure size doesn't allow the stack to wrap the address space.
 331     size = (size_t)base;
 332   }
 333   return size;
 334 }
 335 
 336 static inline stack_t get_stack_info() {
 337   stack_t st;
 338   int retval = thr_stksegment(&st);
 339   st.ss_size = adjust_stack_size((address)st.ss_sp, st.ss_size);
 340   assert(retval == 0, "incorrect return value from thr_stksegment");
 341   assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
 342   assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
 343   return st;
 344 }
 345 
 346 address os::current_stack_base() {
 347   int r = thr_main() ;
 348   guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
 349   bool is_primordial_thread = r;
 350 
 351   // Workaround 4352906, avoid calls to thr_stksegment by
 352   // thr_main after the first one (it looks like we trash
 353   // some data, causing the value for ss_sp to be incorrect).
 354   if (!is_primordial_thread || os::Solaris::_main_stack_base == NULL) {
 355     stack_t st = get_stack_info();
 356     if (is_primordial_thread) {
 357       // cache initial value of stack base
 358       os::Solaris::_main_stack_base = (address)st.ss_sp;
 359     }
 360     return (address)st.ss_sp;
 361   } else {
 362     guarantee(os::Solaris::_main_stack_base != NULL, "Attempt to use null cached stack base");
 363     return os::Solaris::_main_stack_base;
 364   }
 365 }
 366 
 367 size_t os::current_stack_size() {
 368   size_t size;
 369 
 370   int r = thr_main() ;
 371   guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
 372   if(!r) {
 373     size = get_stack_info().ss_size;
 374   } else {
 375     struct rlimit limits;
 376     getrlimit(RLIMIT_STACK, &limits);
 377     size = adjust_stack_size(os::Solaris::_main_stack_base, (size_t)limits.rlim_cur);
 378   }
 379   // base may not be page aligned
 380   address base = current_stack_base();
 381   address bottom = (address)align_size_up((intptr_t)(base - size), os::vm_page_size());;
 382   return (size_t)(base - bottom);
 383 }
 384 
 385 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
 386   return localtime_r(clock, res);
 387 }
 388 
 389 // interruptible infrastructure
 390 
 391 // setup_interruptible saves the thread state before going into an
 392 // interruptible system call.
 393 // The saved state is used to restore the thread to
 394 // its former state whether or not an interrupt is received.
 395 // Used by classloader os::read
 396 // os::restartable_read calls skip this layer and stay in _thread_in_native
 397 
 398 void os::Solaris::setup_interruptible(JavaThread* thread) {
 399 
 400   JavaThreadState thread_state = thread->thread_state();
 401 
 402   assert(thread_state != _thread_blocked, "Coming from the wrong thread");
 403   assert(thread_state != _thread_in_native, "Native threads skip setup_interruptible");
 404   OSThread* osthread = thread->osthread();
 405   osthread->set_saved_interrupt_thread_state(thread_state);
 406   thread->frame_anchor()->make_walkable(thread);
 407   ThreadStateTransition::transition(thread, thread_state, _thread_blocked);
 408 }
 409 
 410 // Version of setup_interruptible() for threads that are already in
 411 // _thread_blocked. Used by os_sleep().
 412 void os::Solaris::setup_interruptible_already_blocked(JavaThread* thread) {
 413   thread->frame_anchor()->make_walkable(thread);
 414 }
 415 
 416 JavaThread* os::Solaris::setup_interruptible() {
 417   JavaThread* thread = (JavaThread*)ThreadLocalStorage::thread();
 418   setup_interruptible(thread);
 419   return thread;
 420 }
 421 
 422 void os::Solaris::try_enable_extended_io() {
 423   typedef int (*enable_extended_FILE_stdio_t)(int, int);
 424 
 425   if (!UseExtendedFileIO) {
 426     return;
 427   }
 428 
 429   enable_extended_FILE_stdio_t enabler =
 430     (enable_extended_FILE_stdio_t) dlsym(RTLD_DEFAULT,
 431                                          "enable_extended_FILE_stdio");
 432   if (enabler) {
 433     enabler(-1, -1);
 434   }
 435 }
 436 
 437 
 438 #ifdef ASSERT
 439 
 440 JavaThread* os::Solaris::setup_interruptible_native() {
 441   JavaThread* thread = (JavaThread*)ThreadLocalStorage::thread();
 442   JavaThreadState thread_state = thread->thread_state();
 443   assert(thread_state == _thread_in_native, "Assumed thread_in_native");
 444   return thread;
 445 }
 446 
 447 void os::Solaris::cleanup_interruptible_native(JavaThread* thread) {
 448   JavaThreadState thread_state = thread->thread_state();
 449   assert(thread_state == _thread_in_native, "Assumed thread_in_native");
 450 }
 451 #endif
 452 
 453 // cleanup_interruptible reverses the effects of setup_interruptible
 454 // setup_interruptible_already_blocked() does not need any cleanup.
 455 
 456 void os::Solaris::cleanup_interruptible(JavaThread* thread) {
 457   OSThread* osthread = thread->osthread();
 458 
 459   ThreadStateTransition::transition(thread, _thread_blocked, osthread->saved_interrupt_thread_state());
 460 }
 461 
 462 // I/O interruption related counters called in _INTERRUPTIBLE
 463 
 464 void os::Solaris::bump_interrupted_before_count() {
 465   RuntimeService::record_interrupted_before_count();
 466 }
 467 
 468 void os::Solaris::bump_interrupted_during_count() {
 469   RuntimeService::record_interrupted_during_count();
 470 }
 471 
 472 static int _processors_online = 0;
 473 
 474          jint os::Solaris::_os_thread_limit = 0;
 475 volatile jint os::Solaris::_os_thread_count = 0;
 476 
 477 julong os::available_memory() {
 478   return Solaris::available_memory();
 479 }
 480 
 481 julong os::Solaris::available_memory() {
 482   return (julong)sysconf(_SC_AVPHYS_PAGES) * os::vm_page_size();
 483 }
 484 
 485 julong os::Solaris::_physical_memory = 0;
 486 
 487 julong os::physical_memory() {
 488    return Solaris::physical_memory();
 489 }
 490 
 491 julong os::allocatable_physical_memory(julong size) {
 492 #ifdef _LP64
 493    return size;
 494 #else
 495    julong result = MIN2(size, (julong)3835*M);
 496    if (!is_allocatable(result)) {
 497      // Memory allocations will be aligned but the alignment
 498      // is not known at this point.  Alignments will
 499      // be at most to LargePageSizeInBytes.  Protect
 500      // allocations from alignments up to illegal
 501      // values. If at this point 2G is illegal.
 502      julong reasonable_size = (julong)2*G - 2 * LargePageSizeInBytes;
 503      result =  MIN2(size, reasonable_size);
 504    }
 505    return result;
 506 #endif
 507 }
 508 
 509 static hrtime_t first_hrtime = 0;
 510 static const hrtime_t hrtime_hz = 1000*1000*1000;
 511 const int LOCK_BUSY = 1;
 512 const int LOCK_FREE = 0;
 513 const int LOCK_INVALID = -1;
 514 static volatile hrtime_t max_hrtime = 0;
 515 static volatile int max_hrtime_lock = LOCK_FREE;     // Update counter with LSB as lock-in-progress
 516 
 517 
 518 void os::Solaris::initialize_system_info() {
 519   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 520   _processors_online = sysconf (_SC_NPROCESSORS_ONLN);
 521   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 522 }
 523 
 524 int os::active_processor_count() {
 525   int online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
 526   pid_t pid = getpid();
 527   psetid_t pset = PS_NONE;
 528   // Are we running in a processor set or is there any processor set around?
 529   if (pset_bind(PS_QUERY, P_PID, pid, &pset) == 0) {
 530     uint_t pset_cpus;
 531     // Query the number of cpus available to us.
 532     if (pset_info(pset, NULL, &pset_cpus, NULL) == 0) {
 533       assert(pset_cpus > 0 && pset_cpus <= online_cpus, "sanity check");
 534       _processors_online = pset_cpus;
 535       return pset_cpus;
 536     }
 537   }
 538   // Otherwise return number of online cpus
 539   return online_cpus;
 540 }
 541 
 542 static bool find_processors_in_pset(psetid_t        pset,
 543                                     processorid_t** id_array,
 544                                     uint_t*         id_length) {
 545   bool result = false;
 546   // Find the number of processors in the processor set.
 547   if (pset_info(pset, NULL, id_length, NULL) == 0) {
 548     // Make up an array to hold their ids.
 549     *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length);
 550     // Fill in the array with their processor ids.
 551     if (pset_info(pset, NULL, id_length, *id_array) == 0) {
 552       result = true;
 553     }
 554   }
 555   return result;
 556 }
 557 
 558 // Callers of find_processors_online() must tolerate imprecise results --
 559 // the system configuration can change asynchronously because of DR
 560 // or explicit psradm operations.
 561 //
 562 // We also need to take care that the loop (below) terminates as the
 563 // number of processors online can change between the _SC_NPROCESSORS_ONLN
 564 // request and the loop that builds the list of processor ids.   Unfortunately
 565 // there's no reliable way to determine the maximum valid processor id,
 566 // so we use a manifest constant, MAX_PROCESSOR_ID, instead.  See p_online
 567 // man pages, which claim the processor id set is "sparse, but
 568 // not too sparse".  MAX_PROCESSOR_ID is used to ensure that we eventually
 569 // exit the loop.
 570 //
 571 // In the future we'll be able to use sysconf(_SC_CPUID_MAX), but that's
 572 // not available on S8.0.
 573 
 574 static bool find_processors_online(processorid_t** id_array,
 575                                    uint*           id_length) {
 576   const processorid_t MAX_PROCESSOR_ID = 100000 ;
 577   // Find the number of processors online.
 578   *id_length = sysconf(_SC_NPROCESSORS_ONLN);
 579   // Make up an array to hold their ids.
 580   *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length);
 581   // Processors need not be numbered consecutively.
 582   long found = 0;
 583   processorid_t next = 0;
 584   while (found < *id_length && next < MAX_PROCESSOR_ID) {
 585     processor_info_t info;
 586     if (processor_info(next, &info) == 0) {
 587       // NB, PI_NOINTR processors are effectively online ...
 588       if (info.pi_state == P_ONLINE || info.pi_state == P_NOINTR) {
 589         (*id_array)[found] = next;
 590         found += 1;
 591       }
 592     }
 593     next += 1;
 594   }
 595   if (found < *id_length) {
 596       // The loop above didn't identify the expected number of processors.
 597       // We could always retry the operation, calling sysconf(_SC_NPROCESSORS_ONLN)
 598       // and re-running the loop, above, but there's no guarantee of progress
 599       // if the system configuration is in flux.  Instead, we just return what
 600       // we've got.  Note that in the worst case find_processors_online() could
 601       // return an empty set.  (As a fall-back in the case of the empty set we
 602       // could just return the ID of the current processor).
 603       *id_length = found ;
 604   }
 605 
 606   return true;
 607 }
 608 
 609 static bool assign_distribution(processorid_t* id_array,
 610                                 uint           id_length,
 611                                 uint*          distribution,
 612                                 uint           distribution_length) {
 613   // We assume we can assign processorid_t's to uint's.
 614   assert(sizeof(processorid_t) == sizeof(uint),
 615          "can't convert processorid_t to uint");
 616   // Quick check to see if we won't succeed.
 617   if (id_length < distribution_length) {
 618     return false;
 619   }
 620   // Assign processor ids to the distribution.
 621   // Try to shuffle processors to distribute work across boards,
 622   // assuming 4 processors per board.
 623   const uint processors_per_board = ProcessDistributionStride;
 624   // Find the maximum processor id.
 625   processorid_t max_id = 0;
 626   for (uint m = 0; m < id_length; m += 1) {
 627     max_id = MAX2(max_id, id_array[m]);
 628   }
 629   // The next id, to limit loops.
 630   const processorid_t limit_id = max_id + 1;
 631   // Make up markers for available processors.
 632   bool* available_id = NEW_C_HEAP_ARRAY(bool, limit_id);
 633   for (uint c = 0; c < limit_id; c += 1) {
 634     available_id[c] = false;
 635   }
 636   for (uint a = 0; a < id_length; a += 1) {
 637     available_id[id_array[a]] = true;
 638   }
 639   // Step by "boards", then by "slot", copying to "assigned".
 640   // NEEDS_CLEANUP: The assignment of processors should be stateful,
 641   //                remembering which processors have been assigned by
 642   //                previous calls, etc., so as to distribute several
 643   //                independent calls of this method.  What we'd like is
 644   //                It would be nice to have an API that let us ask
 645   //                how many processes are bound to a processor,
 646   //                but we don't have that, either.
 647   //                In the short term, "board" is static so that
 648   //                subsequent distributions don't all start at board 0.
 649   static uint board = 0;
 650   uint assigned = 0;
 651   // Until we've found enough processors ....
 652   while (assigned < distribution_length) {
 653     // ... find the next available processor in the board.
 654     for (uint slot = 0; slot < processors_per_board; slot += 1) {
 655       uint try_id = board * processors_per_board + slot;
 656       if ((try_id < limit_id) && (available_id[try_id] == true)) {
 657         distribution[assigned] = try_id;
 658         available_id[try_id] = false;
 659         assigned += 1;
 660         break;
 661       }
 662     }
 663     board += 1;
 664     if (board * processors_per_board + 0 >= limit_id) {
 665       board = 0;
 666     }
 667   }
 668   if (available_id != NULL) {
 669     FREE_C_HEAP_ARRAY(bool, available_id);
 670   }
 671   return true;
 672 }
 673 
 674 void os::set_native_thread_name(const char *name) {
 675   // Not yet implemented.
 676   return;
 677 }
 678 
 679 bool os::distribute_processes(uint length, uint* distribution) {
 680   bool result = false;
 681   // Find the processor id's of all the available CPUs.
 682   processorid_t* id_array  = NULL;
 683   uint           id_length = 0;
 684   // There are some races between querying information and using it,
 685   // since processor sets can change dynamically.
 686   psetid_t pset = PS_NONE;
 687   // Are we running in a processor set?
 688   if ((pset_bind(PS_QUERY, P_PID, P_MYID, &pset) == 0) && pset != PS_NONE) {
 689     result = find_processors_in_pset(pset, &id_array, &id_length);
 690   } else {
 691     result = find_processors_online(&id_array, &id_length);
 692   }
 693   if (result == true) {
 694     if (id_length >= length) {
 695       result = assign_distribution(id_array, id_length, distribution, length);
 696     } else {
 697       result = false;
 698     }
 699   }
 700   if (id_array != NULL) {
 701     FREE_C_HEAP_ARRAY(processorid_t, id_array);
 702   }
 703   return result;
 704 }
 705 
 706 bool os::bind_to_processor(uint processor_id) {
 707   // We assume that a processorid_t can be stored in a uint.
 708   assert(sizeof(uint) == sizeof(processorid_t),
 709          "can't convert uint to processorid_t");
 710   int bind_result =
 711     processor_bind(P_LWPID,                       // bind LWP.
 712                    P_MYID,                        // bind current LWP.
 713                    (processorid_t) processor_id,  // id.
 714                    NULL);                         // don't return old binding.
 715   return (bind_result == 0);
 716 }
 717 
 718 bool os::getenv(const char* name, char* buffer, int len) {
 719   char* val = ::getenv( name );
 720   if ( val == NULL
 721   ||   strlen(val) + 1  >  len ) {
 722     if (len > 0)  buffer[0] = 0; // return a null string
 723     return false;
 724   }
 725   strcpy( buffer, val );
 726   return true;
 727 }
 728 
 729 
 730 // Return true if user is running as root.
 731 
 732 bool os::have_special_privileges() {
 733   static bool init = false;
 734   static bool privileges = false;
 735   if (!init) {
 736     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 737     init = true;
 738   }
 739   return privileges;
 740 }
 741 
 742 
 743 void os::init_system_properties_values() {
 744   char arch[12];
 745   sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 746 
 747   // The next steps are taken in the product version:
 748   //
 749   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 750   // This library should be located at:
 751   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 752   //
 753   // If "/jre/lib/" appears at the right place in the path, then we
 754   // assume libjvm[_g].so is installed in a JDK and we use this path.
 755   //
 756   // Otherwise exit with message: "Could not create the Java virtual machine."
 757   //
 758   // The following extra steps are taken in the debugging version:
 759   //
 760   // If "/jre/lib/" does NOT appear at the right place in the path
 761   // instead of exit check for $JAVA_HOME environment variable.
 762   //
 763   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 764   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 765   // it looks like libjvm[_g].so is installed there
 766   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 767   //
 768   // Otherwise exit.
 769   //
 770   // Important note: if the location of libjvm.so changes this
 771   // code needs to be changed accordingly.
 772 
 773   // The next few definitions allow the code to be verbatim:
 774 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
 775 #define free(p) FREE_C_HEAP_ARRAY(char, p)
 776 #define getenv(n) ::getenv(n)
 777 
 778 #define EXTENSIONS_DIR  "/lib/ext"
 779 #define ENDORSED_DIR    "/lib/endorsed"
 780 #define COMMON_DIR      "/usr/jdk/packages"
 781 
 782   {
 783     /* sysclasspath, java_home, dll_dir */
 784     {
 785         char *home_path;
 786         char *dll_path;
 787         char *pslash;
 788         char buf[MAXPATHLEN];
 789         os::jvm_path(buf, sizeof(buf));
 790 
 791         // Found the full path to libjvm.so.
 792         // Now cut the path to <java_home>/jre if we can.
 793         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 794         pslash = strrchr(buf, '/');
 795         if (pslash != NULL)
 796             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 797         dll_path = malloc(strlen(buf) + 1);
 798         if (dll_path == NULL)
 799             return;
 800         strcpy(dll_path, buf);
 801         Arguments::set_dll_dir(dll_path);
 802 
 803         if (pslash != NULL) {
 804             pslash = strrchr(buf, '/');
 805             if (pslash != NULL) {
 806                 *pslash = '\0';       /* get rid of /<arch> */
 807                 pslash = strrchr(buf, '/');
 808                 if (pslash != NULL)
 809                     *pslash = '\0';   /* get rid of /lib */
 810             }
 811         }
 812 
 813         home_path = malloc(strlen(buf) + 1);
 814         if (home_path == NULL)
 815             return;
 816         strcpy(home_path, buf);
 817         Arguments::set_java_home(home_path);
 818 
 819         if (!set_boot_path('/', ':'))
 820             return;
 821     }
 822 
 823     /*
 824      * Where to look for native libraries
 825      */
 826     {
 827       // Use dlinfo() to determine the correct java.library.path.
 828       //
 829       // If we're launched by the Java launcher, and the user
 830       // does not set java.library.path explicitly on the commandline,
 831       // the Java launcher sets LD_LIBRARY_PATH for us and unsets
 832       // LD_LIBRARY_PATH_32 and LD_LIBRARY_PATH_64.  In this case
 833       // dlinfo returns LD_LIBRARY_PATH + crle settings (including
 834       // /usr/lib), which is exactly what we want.
 835       //
 836       // If the user does set java.library.path, it completely
 837       // overwrites this setting, and always has.
 838       //
 839       // If we're not launched by the Java launcher, we may
 840       // get here with any/all of the LD_LIBRARY_PATH[_32|64]
 841       // settings.  Again, dlinfo does exactly what we want.
 842 
 843       Dl_serinfo     _info, *info = &_info;
 844       Dl_serpath     *path;
 845       char*          library_path;
 846       char           *common_path;
 847       int            i;
 848 
 849       // determine search path count and required buffer size
 850       if (dlinfo(RTLD_SELF, RTLD_DI_SERINFOSIZE, (void *)info) == -1) {
 851         vm_exit_during_initialization("dlinfo SERINFOSIZE request", dlerror());
 852       }
 853 
 854       // allocate new buffer and initialize
 855       info = (Dl_serinfo*)malloc(_info.dls_size);
 856       if (info == NULL) {
 857         vm_exit_out_of_memory(_info.dls_size,
 858                               "init_system_properties_values info");
 859       }
 860       info->dls_size = _info.dls_size;
 861       info->dls_cnt = _info.dls_cnt;
 862 
 863       // obtain search path information
 864       if (dlinfo(RTLD_SELF, RTLD_DI_SERINFO, (void *)info) == -1) {
 865         free(info);
 866         vm_exit_during_initialization("dlinfo SERINFO request", dlerror());
 867       }
 868 
 869       path = &info->dls_serpath[0];
 870 
 871       // Note: Due to a legacy implementation, most of the library path
 872       // is set in the launcher.  This was to accomodate linking restrictions
 873       // on legacy Solaris implementations (which are no longer supported).
 874       // Eventually, all the library path setting will be done here.
 875       //
 876       // However, to prevent the proliferation of improperly built native
 877       // libraries, the new path component /usr/jdk/packages is added here.
 878 
 879       // Determine the actual CPU architecture.
 880       char cpu_arch[12];
 881       sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
 882 #ifdef _LP64
 883       // If we are a 64-bit vm, perform the following translations:
 884       //   sparc   -> sparcv9
 885       //   i386    -> amd64
 886       if (strcmp(cpu_arch, "sparc") == 0)
 887         strcat(cpu_arch, "v9");
 888       else if (strcmp(cpu_arch, "i386") == 0)
 889         strcpy(cpu_arch, "amd64");
 890 #endif
 891 
 892       // Construct the invariant part of ld_library_path. Note that the
 893       // space for the colon and the trailing null are provided by the
 894       // nulls included by the sizeof operator.
 895       size_t bufsize = sizeof(COMMON_DIR) + sizeof("/lib/") + strlen(cpu_arch);
 896       common_path = malloc(bufsize);
 897       if (common_path == NULL) {
 898         free(info);
 899         vm_exit_out_of_memory(bufsize,
 900                               "init_system_properties_values common_path");
 901       }
 902       sprintf(common_path, COMMON_DIR "/lib/%s", cpu_arch);
 903 
 904       // struct size is more than sufficient for the path components obtained
 905       // through the dlinfo() call, so only add additional space for the path
 906       // components explicitly added here.
 907       bufsize = info->dls_size + strlen(common_path);
 908       library_path = malloc(bufsize);
 909       if (library_path == NULL) {
 910         free(info);
 911         free(common_path);
 912         vm_exit_out_of_memory(bufsize,
 913                               "init_system_properties_values library_path");
 914       }
 915       library_path[0] = '\0';
 916 
 917       // Construct the desired Java library path from the linker's library
 918       // search path.
 919       //
 920       // For compatibility, it is optimal that we insert the additional path
 921       // components specific to the Java VM after those components specified
 922       // in LD_LIBRARY_PATH (if any) but before those added by the ld.so
 923       // infrastructure.
 924       if (info->dls_cnt == 0) { // Not sure this can happen, but allow for it
 925         strcpy(library_path, common_path);
 926       } else {
 927         int inserted = 0;
 928         for (i = 0; i < info->dls_cnt; i++, path++) {
 929           uint_t flags = path->dls_flags & LA_SER_MASK;
 930           if (((flags & LA_SER_LIBPATH) == 0) && !inserted) {
 931             strcat(library_path, common_path);
 932             strcat(library_path, os::path_separator());
 933             inserted = 1;
 934           }
 935           strcat(library_path, path->dls_name);
 936           strcat(library_path, os::path_separator());
 937         }
 938         // eliminate trailing path separator
 939         library_path[strlen(library_path)-1] = '\0';
 940       }
 941 
 942       // happens before argument parsing - can't use a trace flag
 943       // tty->print_raw("init_system_properties_values: native lib path: ");
 944       // tty->print_raw_cr(library_path);
 945 
 946       // callee copies into its own buffer
 947       Arguments::set_library_path(library_path);
 948 
 949       free(common_path);
 950       free(library_path);
 951       free(info);
 952     }
 953 
 954     /*
 955      * Extensions directories.
 956      *
 957      * Note that the space for the colon and the trailing null are provided
 958      * by the nulls included by the sizeof operator (so actually one byte more
 959      * than necessary is allocated).
 960      */
 961     {
 962         char *buf = (char *) malloc(strlen(Arguments::get_java_home()) +
 963             sizeof(EXTENSIONS_DIR) + sizeof(COMMON_DIR) +
 964             sizeof(EXTENSIONS_DIR));
 965         sprintf(buf, "%s" EXTENSIONS_DIR ":" COMMON_DIR EXTENSIONS_DIR,
 966             Arguments::get_java_home());
 967         Arguments::set_ext_dirs(buf);
 968     }
 969 
 970     /* Endorsed standards default directory. */
 971     {
 972         char * buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 973         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 974         Arguments::set_endorsed_dirs(buf);
 975     }
 976   }
 977 
 978 #undef malloc
 979 #undef free
 980 #undef getenv
 981 #undef EXTENSIONS_DIR
 982 #undef ENDORSED_DIR
 983 #undef COMMON_DIR
 984 
 985 }
 986 
 987 void os::breakpoint() {
 988   BREAKPOINT;
 989 }
 990 
 991 bool os::obsolete_option(const JavaVMOption *option)
 992 {
 993   if (!strncmp(option->optionString, "-Xt", 3)) {
 994     return true;
 995   } else if (!strncmp(option->optionString, "-Xtm", 4)) {
 996     return true;
 997   } else if (!strncmp(option->optionString, "-Xverifyheap", 12)) {
 998     return true;
 999   } else if (!strncmp(option->optionString, "-Xmaxjitcodesize", 16)) {
1000     return true;
1001   }
1002   return false;
1003 }
1004 
1005 bool os::Solaris::valid_stack_address(Thread* thread, address sp) {
1006   address  stackStart  = (address)thread->stack_base();
1007   address  stackEnd    = (address)(stackStart - (address)thread->stack_size());
1008   if (sp < stackStart && sp >= stackEnd ) return true;
1009   return false;
1010 }
1011 
1012 extern "C" void breakpoint() {
1013   // use debugger to set breakpoint here
1014 }
1015 
1016 // Returns an estimate of the current stack pointer. Result must be guaranteed to
1017 // point into the calling threads stack, and be no lower than the current stack
1018 // pointer.
1019 address os::current_stack_pointer() {
1020   volatile int dummy;
1021   address sp = (address)&dummy + 8;     // %%%% need to confirm if this is right
1022   return sp;
1023 }
1024 
1025 static thread_t main_thread;
1026 
1027 // Thread start routine for all new Java threads
1028 extern "C" void* java_start(void* thread_addr) {
1029   // Try to randomize the cache line index of hot stack frames.
1030   // This helps when threads of the same stack traces evict each other's
1031   // cache lines. The threads can be either from the same JVM instance, or
1032   // from different JVM instances. The benefit is especially true for
1033   // processors with hyperthreading technology.
1034   static int counter = 0;
1035   int pid = os::current_process_id();
1036   alloca(((pid ^ counter++) & 7) * 128);
1037 
1038   int prio;
1039   Thread* thread = (Thread*)thread_addr;
1040   OSThread* osthr = thread->osthread();
1041 
1042   osthr->set_lwp_id( _lwp_self() );  // Store lwp in case we are bound
1043   thread->_schedctl = (void *) schedctl_init () ;
1044 
1045   if (UseNUMA) {
1046     int lgrp_id = os::numa_get_group_id();
1047     if (lgrp_id != -1) {
1048       thread->set_lgrp_id(lgrp_id);
1049     }
1050   }
1051 
1052   // If the creator called set priority before we started,
1053   // we need to call set_native_priority now that we have an lwp.
1054   // We used to get the priority from thr_getprio (we called
1055   // thr_setprio way back in create_thread) and pass it to
1056   // set_native_priority, but Solaris scales the priority
1057   // in java_to_os_priority, so when we read it back here,
1058   // we pass trash to set_native_priority instead of what's
1059   // in java_to_os_priority. So we save the native priority
1060   // in the osThread and recall it here.
1061 
1062   if ( osthr->thread_id() != -1 ) {
1063     if ( UseThreadPriorities ) {
1064       int prio = osthr->native_priority();
1065       if (ThreadPriorityVerbose) {
1066         tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is "
1067                       INTPTR_FORMAT ", setting priority: %d\n",
1068                       osthr->thread_id(), osthr->lwp_id(), prio);
1069       }
1070       os::set_native_priority(thread, prio);
1071     }
1072   } else if (ThreadPriorityVerbose) {
1073     warning("Can't set priority in _start routine, thread id hasn't been set\n");
1074   }
1075 
1076   assert(osthr->get_state() == RUNNABLE, "invalid os thread state");
1077 
1078   // initialize signal mask for this thread
1079   os::Solaris::hotspot_sigmask(thread);
1080 
1081   thread->run();
1082 
1083   // One less thread is executing
1084   // When the VMThread gets here, the main thread may have already exited
1085   // which frees the CodeHeap containing the Atomic::dec code
1086   if (thread != VMThread::vm_thread() && VMThread::vm_thread() != NULL) {
1087     Atomic::dec(&os::Solaris::_os_thread_count);
1088   }
1089 
1090   if (UseDetachedThreads) {
1091     thr_exit(NULL);
1092     ShouldNotReachHere();
1093   }
1094   return NULL;
1095 }
1096 
1097 static OSThread* create_os_thread(Thread* thread, thread_t thread_id) {
1098   // Allocate the OSThread object
1099   OSThread* osthread = new OSThread(NULL, NULL);
1100   if (osthread == NULL) return NULL;
1101 
1102   // Store info on the Solaris thread into the OSThread
1103   osthread->set_thread_id(thread_id);
1104   osthread->set_lwp_id(_lwp_self());
1105   thread->_schedctl = (void *) schedctl_init () ;
1106 
1107   if (UseNUMA) {
1108     int lgrp_id = os::numa_get_group_id();
1109     if (lgrp_id != -1) {
1110       thread->set_lgrp_id(lgrp_id);
1111     }
1112   }
1113 
1114   if ( ThreadPriorityVerbose ) {
1115     tty->print_cr("In create_os_thread, Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT "\n",
1116                   osthread->thread_id(), osthread->lwp_id() );
1117   }
1118 
1119   // Initial thread state is INITIALIZED, not SUSPENDED
1120   osthread->set_state(INITIALIZED);
1121 
1122   return osthread;
1123 }
1124 
1125 void os::Solaris::hotspot_sigmask(Thread* thread) {
1126 
1127   //Save caller's signal mask
1128   sigset_t sigmask;
1129   thr_sigsetmask(SIG_SETMASK, NULL, &sigmask);
1130   OSThread *osthread = thread->osthread();
1131   osthread->set_caller_sigmask(sigmask);
1132 
1133   thr_sigsetmask(SIG_UNBLOCK, os::Solaris::unblocked_signals(), NULL);
1134   if (!ReduceSignalUsage) {
1135     if (thread->is_VM_thread()) {
1136       // Only the VM thread handles BREAK_SIGNAL ...
1137       thr_sigsetmask(SIG_UNBLOCK, vm_signals(), NULL);
1138     } else {
1139       // ... all other threads block BREAK_SIGNAL
1140       assert(!sigismember(vm_signals(), SIGINT), "SIGINT should not be blocked");
1141       thr_sigsetmask(SIG_BLOCK, vm_signals(), NULL);
1142     }
1143   }
1144 }
1145 
1146 bool os::create_attached_thread(JavaThread* thread) {
1147 #ifdef ASSERT
1148   thread->verify_not_published();
1149 #endif
1150   OSThread* osthread = create_os_thread(thread, thr_self());
1151   if (osthread == NULL) {
1152      return false;
1153   }
1154 
1155   // Initial thread state is RUNNABLE
1156   osthread->set_state(RUNNABLE);
1157   thread->set_osthread(osthread);
1158 
1159   // initialize signal mask for this thread
1160   // and save the caller's signal mask
1161   os::Solaris::hotspot_sigmask(thread);
1162 
1163   return true;
1164 }
1165 
1166 bool os::create_main_thread(JavaThread* thread) {
1167 #ifdef ASSERT
1168   thread->verify_not_published();
1169 #endif
1170   if (_starting_thread == NULL) {
1171     _starting_thread = create_os_thread(thread, main_thread);
1172      if (_starting_thread == NULL) {
1173         return false;
1174      }
1175   }
1176 
1177   // The primodial thread is runnable from the start
1178   _starting_thread->set_state(RUNNABLE);
1179 
1180   thread->set_osthread(_starting_thread);
1181 
1182   // initialize signal mask for this thread
1183   // and save the caller's signal mask
1184   os::Solaris::hotspot_sigmask(thread);
1185 
1186   return true;
1187 }
1188 
1189 // _T2_libthread is true if we believe we are running with the newer
1190 // SunSoft lwp/libthread.so (2.8 patch, 2.9 default)
1191 bool os::Solaris::_T2_libthread = false;
1192 
1193 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
1194   // Allocate the OSThread object
1195   OSThread* osthread = new OSThread(NULL, NULL);
1196   if (osthread == NULL) {
1197     return false;
1198   }
1199 
1200   if ( ThreadPriorityVerbose ) {
1201     char *thrtyp;
1202     switch ( thr_type ) {
1203       case vm_thread:
1204         thrtyp = (char *)"vm";
1205         break;
1206       case cgc_thread:
1207         thrtyp = (char *)"cgc";
1208         break;
1209       case pgc_thread:
1210         thrtyp = (char *)"pgc";
1211         break;
1212       case java_thread:
1213         thrtyp = (char *)"java";
1214         break;
1215       case compiler_thread:
1216         thrtyp = (char *)"compiler";
1217         break;
1218       case watcher_thread:
1219         thrtyp = (char *)"watcher";
1220         break;
1221       default:
1222         thrtyp = (char *)"unknown";
1223         break;
1224     }
1225     tty->print_cr("In create_thread, creating a %s thread\n", thrtyp);
1226   }
1227 
1228   // Calculate stack size if it's not specified by caller.
1229   if (stack_size == 0) {
1230     // The default stack size 1M (2M for LP64).
1231     stack_size = (BytesPerWord >> 2) * K * K;
1232 
1233     switch (thr_type) {
1234     case os::java_thread:
1235       // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
1236       if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
1237       break;
1238     case os::compiler_thread:
1239       if (CompilerThreadStackSize > 0) {
1240         stack_size = (size_t)(CompilerThreadStackSize * K);
1241         break;
1242       } // else fall through:
1243         // use VMThreadStackSize if CompilerThreadStackSize is not defined
1244     case os::vm_thread:
1245     case os::pgc_thread:
1246     case os::cgc_thread:
1247     case os::watcher_thread:
1248       if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
1249       break;
1250     }
1251   }
1252   stack_size = MAX2(stack_size, os::Solaris::min_stack_allowed);
1253 
1254   // Initial state is ALLOCATED but not INITIALIZED
1255   osthread->set_state(ALLOCATED);
1256 
1257   if (os::Solaris::_os_thread_count > os::Solaris::_os_thread_limit) {
1258     // We got lots of threads. Check if we still have some address space left.
1259     // Need to be at least 5Mb of unreserved address space. We do check by
1260     // trying to reserve some.
1261     const size_t VirtualMemoryBangSize = 20*K*K;
1262     char* mem = os::reserve_memory(VirtualMemoryBangSize);
1263     if (mem == NULL) {
1264       delete osthread;
1265       return false;
1266     } else {
1267       // Release the memory again
1268       os::release_memory(mem, VirtualMemoryBangSize);
1269     }
1270   }
1271 
1272   // Setup osthread because the child thread may need it.
1273   thread->set_osthread(osthread);
1274 
1275   // Create the Solaris thread
1276   // explicit THR_BOUND for T2_libthread case in case
1277   // that assumption is not accurate, but our alternate signal stack
1278   // handling is based on it which must have bound threads
1279   thread_t tid = 0;
1280   long     flags = (UseDetachedThreads ? THR_DETACHED : 0) | THR_SUSPENDED
1281                    | ((UseBoundThreads || os::Solaris::T2_libthread() ||
1282                        (thr_type == vm_thread) ||
1283                        (thr_type == cgc_thread) ||
1284                        (thr_type == pgc_thread) ||
1285                        (thr_type == compiler_thread && BackgroundCompilation)) ?
1286                       THR_BOUND : 0);
1287   int      status;
1288 
1289   // 4376845 -- libthread/kernel don't provide enough LWPs to utilize all CPUs.
1290   //
1291   // On multiprocessors systems, libthread sometimes under-provisions our
1292   // process with LWPs.  On a 30-way systems, for instance, we could have
1293   // 50 user-level threads in ready state and only 2 or 3 LWPs assigned
1294   // to our process.  This can result in under utilization of PEs.
1295   // I suspect the problem is related to libthread's LWP
1296   // pool management and to the kernel's SIGBLOCKING "last LWP parked"
1297   // upcall policy.
1298   //
1299   // The following code is palliative -- it attempts to ensure that our
1300   // process has sufficient LWPs to take advantage of multiple PEs.
1301   // Proper long-term cures include using user-level threads bound to LWPs
1302   // (THR_BOUND) or using LWP-based synchronization.  Note that there is a
1303   // slight timing window with respect to sampling _os_thread_count, but
1304   // the race is benign.  Also, we should periodically recompute
1305   // _processors_online as the min of SC_NPROCESSORS_ONLN and the
1306   // the number of PEs in our partition.  You might be tempted to use
1307   // THR_NEW_LWP here, but I'd recommend against it as that could
1308   // result in undesirable growth of the libthread's LWP pool.
1309   // The fix below isn't sufficient; for instance, it doesn't take into count
1310   // LWPs parked on IO.  It does, however, help certain CPU-bound benchmarks.
1311   //
1312   // Some pathologies this scheme doesn't handle:
1313   // *  Threads can block, releasing the LWPs.  The LWPs can age out.
1314   //    When a large number of threads become ready again there aren't
1315   //    enough LWPs available to service them.  This can occur when the
1316   //    number of ready threads oscillates.
1317   // *  LWPs/Threads park on IO, thus taking the LWP out of circulation.
1318   //
1319   // Finally, we should call thr_setconcurrency() periodically to refresh
1320   // the LWP pool and thwart the LWP age-out mechanism.
1321   // The "+3" term provides a little slop -- we want to slightly overprovision.
1322 
1323   if (AdjustConcurrency && os::Solaris::_os_thread_count < (_processors_online+3)) {
1324     if (!(flags & THR_BOUND)) {
1325       thr_setconcurrency (os::Solaris::_os_thread_count);       // avoid starvation
1326     }
1327   }
1328   // Although this doesn't hurt, we should warn of undefined behavior
1329   // when using unbound T1 threads with schedctl().  This should never
1330   // happen, as the compiler and VM threads are always created bound
1331   DEBUG_ONLY(
1332       if ((VMThreadHintNoPreempt || CompilerThreadHintNoPreempt) &&
1333           (!os::Solaris::T2_libthread() && (!(flags & THR_BOUND))) &&
1334           ((thr_type == vm_thread) || (thr_type == cgc_thread) ||
1335            (thr_type == pgc_thread) || (thr_type == compiler_thread && BackgroundCompilation))) {
1336          warning("schedctl behavior undefined when Compiler/VM/GC Threads are Unbound");
1337       }
1338   );
1339 
1340 
1341   // Mark that we don't have an lwp or thread id yet.
1342   // In case we attempt to set the priority before the thread starts.
1343   osthread->set_lwp_id(-1);
1344   osthread->set_thread_id(-1);
1345 
1346   status = thr_create(NULL, stack_size, java_start, thread, flags, &tid);
1347   if (status != 0) {
1348     if (PrintMiscellaneous && (Verbose || WizardMode)) {
1349       perror("os::create_thread");
1350     }
1351     thread->set_osthread(NULL);
1352     // Need to clean up stuff we've allocated so far
1353     delete osthread;
1354     return false;
1355   }
1356 
1357   Atomic::inc(&os::Solaris::_os_thread_count);
1358 
1359   // Store info on the Solaris thread into the OSThread
1360   osthread->set_thread_id(tid);
1361 
1362   // Remember that we created this thread so we can set priority on it
1363   osthread->set_vm_created();
1364 
1365   // Set the default thread priority.  If using bound threads, setting
1366   // lwp priority will be delayed until thread start.
1367   set_native_priority(thread,
1368                       DefaultThreadPriority == -1 ?
1369                         java_to_os_priority[NormPriority] :
1370                         DefaultThreadPriority);
1371 
1372   // Initial thread state is INITIALIZED, not SUSPENDED
1373   osthread->set_state(INITIALIZED);
1374 
1375   // The thread is returned suspended (in state INITIALIZED), and is started higher up in the call chain
1376   return true;
1377 }
1378 
1379 /* defined for >= Solaris 10. This allows builds on earlier versions
1380  *  of Solaris to take advantage of the newly reserved Solaris JVM signals
1381  *  With SIGJVM1, SIGJVM2, INTERRUPT_SIGNAL is SIGJVM1, ASYNC_SIGNAL is SIGJVM2
1382  *  and -XX:+UseAltSigs does nothing since these should have no conflict
1383  */
1384 #if !defined(SIGJVM1)
1385 #define SIGJVM1 39
1386 #define SIGJVM2 40
1387 #endif
1388 
1389 debug_only(static bool signal_sets_initialized = false);
1390 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
1391 int os::Solaris::_SIGinterrupt = INTERRUPT_SIGNAL;
1392 int os::Solaris::_SIGasync = ASYNC_SIGNAL;
1393 
1394 bool os::Solaris::is_sig_ignored(int sig) {
1395       struct sigaction oact;
1396       sigaction(sig, (struct sigaction*)NULL, &oact);
1397       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
1398                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
1399       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
1400            return true;
1401       else
1402            return false;
1403 }
1404 
1405 // Note: SIGRTMIN is a macro that calls sysconf() so it will
1406 // dynamically detect SIGRTMIN value for the system at runtime, not buildtime
1407 static bool isJVM1available() {
1408   return SIGJVM1 < SIGRTMIN;
1409 }
1410 
1411 void os::Solaris::signal_sets_init() {
1412   // Should also have an assertion stating we are still single-threaded.
1413   assert(!signal_sets_initialized, "Already initialized");
1414   // Fill in signals that are necessarily unblocked for all threads in
1415   // the VM. Currently, we unblock the following signals:
1416   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
1417   //                         by -Xrs (=ReduceSignalUsage));
1418   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
1419   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
1420   // the dispositions or masks wrt these signals.
1421   // Programs embedding the VM that want to use the above signals for their
1422   // own purposes must, at this time, use the "-Xrs" option to prevent
1423   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
1424   // (See bug 4345157, and other related bugs).
1425   // In reality, though, unblocking these signals is really a nop, since
1426   // these signals are not blocked by default.
1427   sigemptyset(&unblocked_sigs);
1428   sigemptyset(&allowdebug_blocked_sigs);
1429   sigaddset(&unblocked_sigs, SIGILL);
1430   sigaddset(&unblocked_sigs, SIGSEGV);
1431   sigaddset(&unblocked_sigs, SIGBUS);
1432   sigaddset(&unblocked_sigs, SIGFPE);
1433 
1434   if (isJVM1available) {
1435     os::Solaris::set_SIGinterrupt(SIGJVM1);
1436     os::Solaris::set_SIGasync(SIGJVM2);
1437   } else if (UseAltSigs) {
1438     os::Solaris::set_SIGinterrupt(ALT_INTERRUPT_SIGNAL);
1439     os::Solaris::set_SIGasync(ALT_ASYNC_SIGNAL);
1440   } else {
1441     os::Solaris::set_SIGinterrupt(INTERRUPT_SIGNAL);
1442     os::Solaris::set_SIGasync(ASYNC_SIGNAL);
1443   }
1444 
1445   sigaddset(&unblocked_sigs, os::Solaris::SIGinterrupt());
1446   sigaddset(&unblocked_sigs, os::Solaris::SIGasync());
1447 
1448   if (!ReduceSignalUsage) {
1449    if (!os::Solaris::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
1450       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
1451       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
1452    }
1453    if (!os::Solaris::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
1454       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
1455       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
1456    }
1457    if (!os::Solaris::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
1458       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
1459       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
1460    }
1461   }
1462   // Fill in signals that are blocked by all but the VM thread.
1463   sigemptyset(&vm_sigs);
1464   if (!ReduceSignalUsage)
1465     sigaddset(&vm_sigs, BREAK_SIGNAL);
1466   debug_only(signal_sets_initialized = true);
1467 
1468   // For diagnostics only used in run_periodic_checks
1469   sigemptyset(&check_signal_done);
1470 }
1471 
1472 // These are signals that are unblocked while a thread is running Java.
1473 // (For some reason, they get blocked by default.)
1474 sigset_t* os::Solaris::unblocked_signals() {
1475   assert(signal_sets_initialized, "Not initialized");
1476   return &unblocked_sigs;
1477 }
1478 
1479 // These are the signals that are blocked while a (non-VM) thread is
1480 // running Java. Only the VM thread handles these signals.
1481 sigset_t* os::Solaris::vm_signals() {
1482   assert(signal_sets_initialized, "Not initialized");
1483   return &vm_sigs;
1484 }
1485 
1486 // These are signals that are blocked during cond_wait to allow debugger in
1487 sigset_t* os::Solaris::allowdebug_blocked_signals() {
1488   assert(signal_sets_initialized, "Not initialized");
1489   return &allowdebug_blocked_sigs;
1490 }
1491 
1492 
1493 void _handle_uncaught_cxx_exception() {
1494   VMError err("An uncaught C++ exception");
1495   err.report_and_die();
1496 }
1497 
1498 
1499 // First crack at OS-specific initialization, from inside the new thread.
1500 void os::initialize_thread() {
1501   int r = thr_main() ;
1502   guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
1503   if (r) {
1504     JavaThread* jt = (JavaThread *)Thread::current();
1505     assert(jt != NULL,"Sanity check");
1506     size_t stack_size;
1507     address base = jt->stack_base();
1508     if (Arguments::created_by_java_launcher()) {
1509       // Use 2MB to allow for Solaris 7 64 bit mode.
1510       stack_size = JavaThread::stack_size_at_create() == 0
1511         ? 2048*K : JavaThread::stack_size_at_create();
1512 
1513       // There are rare cases when we may have already used more than
1514       // the basic stack size allotment before this method is invoked.
1515       // Attempt to allow for a normally sized java_stack.
1516       size_t current_stack_offset = (size_t)(base - (address)&stack_size);
1517       stack_size += ReservedSpace::page_align_size_down(current_stack_offset);
1518     } else {
1519       // 6269555: If we were not created by a Java launcher, i.e. if we are
1520       // running embedded in a native application, treat the primordial thread
1521       // as much like a native attached thread as possible.  This means using
1522       // the current stack size from thr_stksegment(), unless it is too large
1523       // to reliably setup guard pages.  A reasonable max size is 8MB.
1524       size_t current_size = current_stack_size();
1525       // This should never happen, but just in case....
1526       if (current_size == 0) current_size = 2 * K * K;
1527       stack_size = current_size > (8 * K * K) ? (8 * K * K) : current_size;
1528     }
1529     address bottom = (address)align_size_up((intptr_t)(base - stack_size), os::vm_page_size());;
1530     stack_size = (size_t)(base - bottom);
1531 
1532     assert(stack_size > 0, "Stack size calculation problem");
1533 
1534     if (stack_size > jt->stack_size()) {
1535       NOT_PRODUCT(
1536         struct rlimit limits;
1537         getrlimit(RLIMIT_STACK, &limits);
1538         size_t size = adjust_stack_size(base, (size_t)limits.rlim_cur);
1539         assert(size >= jt->stack_size(), "Stack size problem in main thread");
1540       )
1541       tty->print_cr(
1542         "Stack size of %d Kb exceeds current limit of %d Kb.\n"
1543         "(Stack sizes are rounded up to a multiple of the system page size.)\n"
1544         "See limit(1) to increase the stack size limit.",
1545         stack_size / K, jt->stack_size() / K);
1546       vm_exit(1);
1547     }
1548     assert(jt->stack_size() >= stack_size,
1549           "Attempt to map more stack than was allocated");
1550     jt->set_stack_size(stack_size);
1551   }
1552 
1553    // 5/22/01: Right now alternate signal stacks do not handle
1554    // throwing stack overflow exceptions, see bug 4463178
1555    // Until a fix is found for this, T2 will NOT imply alternate signal
1556    // stacks.
1557    // If using T2 libthread threads, install an alternate signal stack.
1558    // Because alternate stacks associate with LWPs on Solaris,
1559    // see sigaltstack(2), if using UNBOUND threads, or if UseBoundThreads
1560    // we prefer to explicitly stack bang.
1561    // If not using T2 libthread, but using UseBoundThreads any threads
1562    // (primordial thread, jni_attachCurrentThread) we do not create,
1563    // probably are not bound, therefore they can not have an alternate
1564    // signal stack. Since our stack banging code is generated and
1565    // is shared across threads, all threads must be bound to allow
1566    // using alternate signal stacks.  The alternative is to interpose
1567    // on _lwp_create to associate an alt sig stack with each LWP,
1568    // and this could be a problem when the JVM is embedded.
1569    // We would prefer to use alternate signal stacks with T2
1570    // Since there is currently no accurate way to detect T2
1571    // we do not. Assuming T2 when running T1 causes sig 11s or assertions
1572    // on installing alternate signal stacks
1573 
1574 
1575    // 05/09/03: removed alternate signal stack support for Solaris
1576    // The alternate signal stack mechanism is no longer needed to
1577    // handle stack overflow. This is now handled by allocating
1578    // guard pages (red zone) and stackbanging.
1579    // Initially the alternate signal stack mechanism was removed because
1580    // it did not work with T1 llibthread. Alternate
1581    // signal stacks MUST have all threads bound to lwps. Applications
1582    // can create their own threads and attach them without their being
1583    // bound under T1. This is frequently the case for the primordial thread.
1584    // If we were ever to reenable this mechanism we would need to
1585    // use the dynamic check for T2 libthread.
1586 
1587   os::Solaris::init_thread_fpu_state();
1588   std::set_terminate(_handle_uncaught_cxx_exception);
1589 }
1590 
1591 
1592 
1593 // Free Solaris resources related to the OSThread
1594 void os::free_thread(OSThread* osthread) {
1595   assert(osthread != NULL, "os::free_thread but osthread not set");
1596 
1597 
1598   // We are told to free resources of the argument thread,
1599   // but we can only really operate on the current thread.
1600   // The main thread must take the VMThread down synchronously
1601   // before the main thread exits and frees up CodeHeap
1602   guarantee((Thread::current()->osthread() == osthread
1603      || (osthread == VMThread::vm_thread()->osthread())), "os::free_thread but not current thread");
1604   if (Thread::current()->osthread() == osthread) {
1605     // Restore caller's signal mask
1606     sigset_t sigmask = osthread->caller_sigmask();
1607     thr_sigsetmask(SIG_SETMASK, &sigmask, NULL);
1608   }
1609   delete osthread;
1610 }
1611 
1612 void os::pd_start_thread(Thread* thread) {
1613   int status = thr_continue(thread->osthread()->thread_id());
1614   assert_status(status == 0, status, "thr_continue failed");
1615 }
1616 
1617 
1618 intx os::current_thread_id() {
1619   return (intx)thr_self();
1620 }
1621 
1622 static pid_t _initial_pid = 0;
1623 
1624 int os::current_process_id() {
1625   return (int)(_initial_pid ? _initial_pid : getpid());
1626 }
1627 
1628 int os::allocate_thread_local_storage() {
1629   // %%%       in Win32 this allocates a memory segment pointed to by a
1630   //           register.  Dan Stein can implement a similar feature in
1631   //           Solaris.  Alternatively, the VM can do the same thing
1632   //           explicitly: malloc some storage and keep the pointer in a
1633   //           register (which is part of the thread's context) (or keep it
1634   //           in TLS).
1635   // %%%       In current versions of Solaris, thr_self and TSD can
1636   //           be accessed via short sequences of displaced indirections.
1637   //           The value of thr_self is available as %g7(36).
1638   //           The value of thr_getspecific(k) is stored in %g7(12)(4)(k*4-4),
1639   //           assuming that the current thread already has a value bound to k.
1640   //           It may be worth experimenting with such access patterns,
1641   //           and later having the parameters formally exported from a Solaris
1642   //           interface.  I think, however, that it will be faster to
1643   //           maintain the invariant that %g2 always contains the
1644   //           JavaThread in Java code, and have stubs simply
1645   //           treat %g2 as a caller-save register, preserving it in a %lN.
1646   thread_key_t tk;
1647   if (thr_keycreate( &tk, NULL ) )
1648     fatal(err_msg("os::allocate_thread_local_storage: thr_keycreate failed "
1649                   "(%s)", strerror(errno)));
1650   return int(tk);
1651 }
1652 
1653 void os::free_thread_local_storage(int index) {
1654   // %%% don't think we need anything here
1655   // if ( pthread_key_delete((pthread_key_t) tk) )
1656   //   fatal("os::free_thread_local_storage: pthread_key_delete failed");
1657 }
1658 
1659 #define SMALLINT 32   // libthread allocate for tsd_common is a version specific
1660                       // small number - point is NO swap space available
1661 void os::thread_local_storage_at_put(int index, void* value) {
1662   // %%% this is used only in threadLocalStorage.cpp
1663   if (thr_setspecific((thread_key_t)index, value)) {
1664     if (errno == ENOMEM) {
1665        vm_exit_out_of_memory(SMALLINT, "thr_setspecific: out of swap space");
1666     } else {
1667       fatal(err_msg("os::thread_local_storage_at_put: thr_setspecific failed "
1668                     "(%s)", strerror(errno)));
1669     }
1670   } else {
1671       ThreadLocalStorage::set_thread_in_slot ((Thread *) value) ;
1672   }
1673 }
1674 
1675 // This function could be called before TLS is initialized, for example, when
1676 // VM receives an async signal or when VM causes a fatal error during
1677 // initialization. Return NULL if thr_getspecific() fails.
1678 void* os::thread_local_storage_at(int index) {
1679   // %%% this is used only in threadLocalStorage.cpp
1680   void* r = NULL;
1681   return thr_getspecific((thread_key_t)index, &r) != 0 ? NULL : r;
1682 }
1683 
1684 
1685 // gethrtime can move backwards if read from one cpu and then a different cpu
1686 // getTimeNanos is guaranteed to not move backward on Solaris
1687 // local spinloop created as faster for a CAS on an int than
1688 // a CAS on a 64bit jlong. Also Atomic::cmpxchg for jlong is not
1689 // supported on sparc v8 or pre supports_cx8 intel boxes.
1690 // oldgetTimeNanos for systems which do not support CAS on 64bit jlong
1691 // i.e. sparc v8 and pre supports_cx8 (i486) intel boxes
1692 inline hrtime_t oldgetTimeNanos() {
1693   int gotlock = LOCK_INVALID;
1694   hrtime_t newtime = gethrtime();
1695 
1696   for (;;) {
1697 // grab lock for max_hrtime
1698     int curlock = max_hrtime_lock;
1699     if (curlock & LOCK_BUSY)  continue;
1700     if (gotlock = Atomic::cmpxchg(LOCK_BUSY, &max_hrtime_lock, LOCK_FREE) != LOCK_FREE) continue;
1701     if (newtime > max_hrtime) {
1702       max_hrtime = newtime;
1703     } else {
1704       newtime = max_hrtime;
1705     }
1706     // release lock
1707     max_hrtime_lock = LOCK_FREE;
1708     return newtime;
1709   }
1710 }
1711 // gethrtime can move backwards if read from one cpu and then a different cpu
1712 // getTimeNanos is guaranteed to not move backward on Solaris
1713 inline hrtime_t getTimeNanos() {
1714   if (VM_Version::supports_cx8()) {
1715     const hrtime_t now = gethrtime();
1716     // Use atomic long load since 32-bit x86 uses 2 registers to keep long.
1717     const hrtime_t prev = Atomic::load((volatile jlong*)&max_hrtime);
1718     if (now <= prev)  return prev;   // same or retrograde time;
1719     const hrtime_t obsv = Atomic::cmpxchg(now, (volatile jlong*)&max_hrtime, prev);
1720     assert(obsv >= prev, "invariant");   // Monotonicity
1721     // If the CAS succeeded then we're done and return "now".
1722     // If the CAS failed and the observed value "obs" is >= now then
1723     // we should return "obs".  If the CAS failed and now > obs > prv then
1724     // some other thread raced this thread and installed a new value, in which case
1725     // we could either (a) retry the entire operation, (b) retry trying to install now
1726     // or (c) just return obs.  We use (c).   No loop is required although in some cases
1727     // we might discard a higher "now" value in deference to a slightly lower but freshly
1728     // installed obs value.   That's entirely benign -- it admits no new orderings compared
1729     // to (a) or (b) -- and greatly reduces coherence traffic.
1730     // We might also condition (c) on the magnitude of the delta between obs and now.
1731     // Avoiding excessive CAS operations to hot RW locations is critical.
1732     // See http://blogs.sun.com/dave/entry/cas_and_cache_trivia_invalidate
1733     return (prev == obsv) ? now : obsv ;
1734   } else {
1735     return oldgetTimeNanos();
1736   }
1737 }
1738 
1739 // Time since start-up in seconds to a fine granularity.
1740 // Used by VMSelfDestructTimer and the MemProfiler.
1741 double os::elapsedTime() {
1742   return (double)(getTimeNanos() - first_hrtime) / (double)hrtime_hz;
1743 }
1744 
1745 jlong os::elapsed_counter() {
1746   return (jlong)(getTimeNanos() - first_hrtime);
1747 }
1748 
1749 jlong os::elapsed_frequency() {
1750    return hrtime_hz;
1751 }
1752 
1753 // Return the real, user, and system times in seconds from an
1754 // arbitrary fixed point in the past.
1755 bool os::getTimesSecs(double* process_real_time,
1756                   double* process_user_time,
1757                   double* process_system_time) {
1758   struct tms ticks;
1759   clock_t real_ticks = times(&ticks);
1760 
1761   if (real_ticks == (clock_t) (-1)) {
1762     return false;
1763   } else {
1764     double ticks_per_second = (double) clock_tics_per_sec;
1765     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1766     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1767     // For consistency return the real time from getTimeNanos()
1768     // converted to seconds.
1769     *process_real_time = ((double) getTimeNanos()) / ((double) NANOUNITS);
1770 
1771     return true;
1772   }
1773 }
1774 
1775 bool os::supports_vtime() { return true; }
1776 
1777 bool os::enable_vtime() {
1778   int fd = ::open("/proc/self/ctl", O_WRONLY);
1779   if (fd == -1)
1780     return false;
1781 
1782   long cmd[] = { PCSET, PR_MSACCT };
1783   int res = ::write(fd, cmd, sizeof(long) * 2);
1784   ::close(fd);
1785   if (res != sizeof(long) * 2)
1786     return false;
1787 
1788   return true;
1789 }
1790 
1791 bool os::vtime_enabled() {
1792   int fd = ::open("/proc/self/status", O_RDONLY);
1793   if (fd == -1)
1794     return false;
1795 
1796   pstatus_t status;
1797   int res = os::read(fd, (void*) &status, sizeof(pstatus_t));
1798   ::close(fd);
1799   if (res != sizeof(pstatus_t))
1800     return false;
1801 
1802   return status.pr_flags & PR_MSACCT;
1803 }
1804 
1805 double os::elapsedVTime() {
1806   return (double)gethrvtime() / (double)hrtime_hz;
1807 }
1808 
1809 // Used internally for comparisons only
1810 // getTimeMillis guaranteed to not move backwards on Solaris
1811 jlong getTimeMillis() {
1812   jlong nanotime = getTimeNanos();
1813   return (jlong)(nanotime / NANOSECS_PER_MILLISEC);
1814 }
1815 
1816 // Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1817 jlong os::javaTimeMillis() {
1818   timeval t;
1819   if (gettimeofday( &t, NULL) == -1)
1820     fatal(err_msg("os::javaTimeMillis: gettimeofday (%s)", strerror(errno)));
1821   return jlong(t.tv_sec) * 1000  +  jlong(t.tv_usec) / 1000;
1822 }
1823 
1824 jlong os::javaTimeNanos() {
1825   return (jlong)getTimeNanos();
1826 }
1827 
1828 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1829   info_ptr->max_value = ALL_64_BITS;      // gethrtime() uses all 64 bits
1830   info_ptr->may_skip_backward = false;    // not subject to resetting or drifting
1831   info_ptr->may_skip_forward = false;     // not subject to resetting or drifting
1832   info_ptr->kind = JVMTI_TIMER_ELAPSED;   // elapsed not CPU time
1833 }
1834 
1835 char * os::local_time_string(char *buf, size_t buflen) {
1836   struct tm t;
1837   time_t long_time;
1838   time(&long_time);
1839   localtime_r(&long_time, &t);
1840   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1841                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1842                t.tm_hour, t.tm_min, t.tm_sec);
1843   return buf;
1844 }
1845 
1846 // Note: os::shutdown() might be called very early during initialization, or
1847 // called from signal handler. Before adding something to os::shutdown(), make
1848 // sure it is async-safe and can handle partially initialized VM.
1849 void os::shutdown() {
1850 
1851   // allow PerfMemory to attempt cleanup of any persistent resources
1852   perfMemory_exit();
1853 
1854   // needs to remove object in file system
1855   AttachListener::abort();
1856 
1857   // flush buffered output, finish log files
1858   ostream_abort();
1859 
1860   // Check for abort hook
1861   abort_hook_t abort_hook = Arguments::abort_hook();
1862   if (abort_hook != NULL) {
1863     abort_hook();
1864   }
1865 }
1866 
1867 // Note: os::abort() might be called very early during initialization, or
1868 // called from signal handler. Before adding something to os::abort(), make
1869 // sure it is async-safe and can handle partially initialized VM.
1870 void os::abort(bool dump_core) {
1871   os::shutdown();
1872   if (dump_core) {
1873 #ifndef PRODUCT
1874     fdStream out(defaultStream::output_fd());
1875     out.print_raw("Current thread is ");
1876     char buf[16];
1877     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1878     out.print_raw_cr(buf);
1879     out.print_raw_cr("Dumping core ...");
1880 #endif
1881     ::abort(); // dump core (for debugging)
1882   }
1883 
1884   ::exit(1);
1885 }
1886 
1887 // Die immediately, no exit hook, no abort hook, no cleanup.
1888 void os::die() {
1889   _exit(-1);
1890 }
1891 
1892 // unused
1893 void os::set_error_file(const char *logfile) {}
1894 
1895 // DLL functions
1896 
1897 const char* os::dll_file_extension() { return ".so"; }
1898 
1899 // This must be hard coded because it's the system's temporary
1900 // directory not the java application's temp directory, ala java.io.tmpdir.
1901 const char* os::get_temp_directory() { return "/tmp"; }
1902 
1903 static bool file_exists(const char* filename) {
1904   struct stat statbuf;
1905   if (filename == NULL || strlen(filename) == 0) {
1906     return false;
1907   }
1908   return os::stat(filename, &statbuf) == 0;
1909 }
1910 
1911 void os::dll_build_name(char* buffer, size_t buflen,
1912                         const char* pname, const char* fname) {
1913   const size_t pnamelen = pname ? strlen(pname) : 0;
1914 
1915   // Quietly truncate on buffer overflow.  Should be an error.
1916   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1917     *buffer = '\0';
1918     return;
1919   }
1920 
1921   if (pnamelen == 0) {
1922     snprintf(buffer, buflen, "lib%s.so", fname);
1923   } else if (strchr(pname, *os::path_separator()) != NULL) {
1924     int n;
1925     char** pelements = split_path(pname, &n);
1926     for (int i = 0 ; i < n ; i++) {
1927       // really shouldn't be NULL but what the heck, check can't hurt
1928       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1929         continue; // skip the empty path values
1930       }
1931       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1932       if (file_exists(buffer)) {
1933         break;
1934       }
1935     }
1936     // release the storage
1937     for (int i = 0 ; i < n ; i++) {
1938       if (pelements[i] != NULL) {
1939         FREE_C_HEAP_ARRAY(char, pelements[i]);
1940       }
1941     }
1942     if (pelements != NULL) {
1943       FREE_C_HEAP_ARRAY(char*, pelements);
1944     }
1945   } else {
1946     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1947   }
1948 }
1949 
1950 const char* os::get_current_directory(char *buf, int buflen) {
1951   return getcwd(buf, buflen);
1952 }
1953 
1954 // check if addr is inside libjvm[_g].so
1955 bool os::address_is_in_vm(address addr) {
1956   static address libjvm_base_addr;
1957   Dl_info dlinfo;
1958 
1959   if (libjvm_base_addr == NULL) {
1960     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1961     libjvm_base_addr = (address)dlinfo.dli_fbase;
1962     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1963   }
1964 
1965   if (dladdr((void *)addr, &dlinfo)) {
1966     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1967   }
1968 
1969   return false;
1970 }
1971 
1972 typedef int (*dladdr1_func_type) (void *, Dl_info *, void **, int);
1973 static dladdr1_func_type dladdr1_func = NULL;
1974 
1975 bool os::dll_address_to_function_name(address addr, char *buf,
1976                                       int buflen, int * offset) {
1977   Dl_info dlinfo;
1978 
1979   // dladdr1_func was initialized in os::init()
1980   if (dladdr1_func){
1981       // yes, we have dladdr1
1982 
1983       // Support for dladdr1 is checked at runtime; it may be
1984       // available even if the vm is built on a machine that does
1985       // not have dladdr1 support.  Make sure there is a value for
1986       // RTLD_DL_SYMENT.
1987       #ifndef RTLD_DL_SYMENT
1988       #define RTLD_DL_SYMENT 1
1989       #endif
1990 #ifdef _LP64
1991       Elf64_Sym * info;
1992 #else
1993       Elf32_Sym * info;
1994 #endif
1995       if (dladdr1_func((void *)addr, &dlinfo, (void **)&info,
1996                        RTLD_DL_SYMENT)) {
1997         if ((char *)dlinfo.dli_saddr + info->st_size > (char *)addr) {
1998           if (buf != NULL) {
1999             if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen))
2000               jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
2001             }
2002             if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
2003             return true;
2004         }
2005       }
2006       if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
2007         if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
2008            buf, buflen, offset, dlinfo.dli_fname)) {
2009           return true;
2010         }
2011       }
2012       if (buf != NULL) buf[0] = '\0';
2013       if (offset != NULL) *offset  = -1;
2014       return false;
2015   } else {
2016       // no, only dladdr is available
2017       if (dladdr((void *)addr, &dlinfo)) {
2018         if (buf != NULL) {
2019           if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen))
2020             jio_snprintf(buf, buflen, dlinfo.dli_sname);
2021         }
2022         if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
2023         return true;
2024       } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
2025         if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
2026           buf, buflen, offset, dlinfo.dli_fname)) {
2027           return true;
2028         }
2029       }
2030       if (buf != NULL) buf[0] = '\0';
2031       if (offset != NULL) *offset  = -1;
2032       return false;
2033   }
2034 }
2035 
2036 bool os::dll_address_to_library_name(address addr, char* buf,
2037                                      int buflen, int* offset) {
2038   Dl_info dlinfo;
2039 
2040   if (dladdr((void*)addr, &dlinfo)){
2041      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
2042      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
2043      return true;
2044   } else {
2045      if (buf) buf[0] = '\0';
2046      if (offset) *offset = -1;
2047      return false;
2048   }
2049 }
2050 
2051 // Prints the names and full paths of all opened dynamic libraries
2052 // for current process
2053 void os::print_dll_info(outputStream * st) {
2054     Dl_info dli;
2055     void *handle;
2056     Link_map *map;
2057     Link_map *p;
2058 
2059     st->print_cr("Dynamic libraries:"); st->flush();
2060 
2061     if (!dladdr(CAST_FROM_FN_PTR(void *, os::print_dll_info), &dli)) {
2062         st->print_cr("Error: Cannot print dynamic libraries.");
2063         return;
2064     }
2065     handle = dlopen(dli.dli_fname, RTLD_LAZY);
2066     if (handle == NULL) {
2067         st->print_cr("Error: Cannot print dynamic libraries.");
2068         return;
2069     }
2070     dlinfo(handle, RTLD_DI_LINKMAP, &map);
2071     if (map == NULL) {
2072         st->print_cr("Error: Cannot print dynamic libraries.");
2073         return;
2074     }
2075 
2076     while (map->l_prev != NULL)
2077         map = map->l_prev;
2078 
2079     while (map != NULL) {
2080         st->print_cr(PTR_FORMAT " \t%s", map->l_addr, map->l_name);
2081         map = map->l_next;
2082     }
2083 
2084     dlclose(handle);
2085 }
2086 
2087   // Loads .dll/.so and
2088   // in case of error it checks if .dll/.so was built for the
2089   // same architecture as Hotspot is running on
2090 
2091 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
2092 {
2093   void * result= ::dlopen(filename, RTLD_LAZY);
2094   if (result != NULL) {
2095     // Successful loading
2096     return result;
2097   }
2098 
2099   Elf32_Ehdr elf_head;
2100 
2101   // Read system error message into ebuf
2102   // It may or may not be overwritten below
2103   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2104   ebuf[ebuflen-1]='\0';
2105   int diag_msg_max_length=ebuflen-strlen(ebuf);
2106   char* diag_msg_buf=ebuf+strlen(ebuf);
2107 
2108   if (diag_msg_max_length==0) {
2109     // No more space in ebuf for additional diagnostics message
2110     return NULL;
2111   }
2112 
2113 
2114   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
2115 
2116   if (file_descriptor < 0) {
2117     // Can't open library, report dlerror() message
2118     return NULL;
2119   }
2120 
2121   bool failed_to_read_elf_head=
2122     (sizeof(elf_head)!=
2123         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
2124 
2125   ::close(file_descriptor);
2126   if (failed_to_read_elf_head) {
2127     // file i/o error - report dlerror() msg
2128     return NULL;
2129   }
2130 
2131   typedef struct {
2132     Elf32_Half  code;         // Actual value as defined in elf.h
2133     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
2134     char        elf_class;    // 32 or 64 bit
2135     char        endianess;    // MSB or LSB
2136     char*       name;         // String representation
2137   } arch_t;
2138 
2139   static const arch_t arch_array[]={
2140     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2141     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2142     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
2143     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
2144     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2145     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2146     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
2147     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
2148     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
2149     {EM_ARM,         EM_ARM,     ELFCLASS32, ELFDATA2LSB, (char*)"ARM 32"}
2150   };
2151 
2152   #if  (defined IA32)
2153     static  Elf32_Half running_arch_code=EM_386;
2154   #elif   (defined AMD64)
2155     static  Elf32_Half running_arch_code=EM_X86_64;
2156   #elif  (defined IA64)
2157     static  Elf32_Half running_arch_code=EM_IA_64;
2158   #elif  (defined __sparc) && (defined _LP64)
2159     static  Elf32_Half running_arch_code=EM_SPARCV9;
2160   #elif  (defined __sparc) && (!defined _LP64)
2161     static  Elf32_Half running_arch_code=EM_SPARC;
2162   #elif  (defined __powerpc64__)
2163     static  Elf32_Half running_arch_code=EM_PPC64;
2164   #elif  (defined __powerpc__)
2165     static  Elf32_Half running_arch_code=EM_PPC;
2166   #elif (defined ARM)
2167     static  Elf32_Half running_arch_code=EM_ARM;
2168   #else
2169     #error Method os::dll_load requires that one of following is defined:\
2170          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, ARM
2171   #endif
2172 
2173   // Identify compatability class for VM's architecture and library's architecture
2174   // Obtain string descriptions for architectures
2175 
2176   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2177   int running_arch_index=-1;
2178 
2179   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2180     if (running_arch_code == arch_array[i].code) {
2181       running_arch_index    = i;
2182     }
2183     if (lib_arch.code == arch_array[i].code) {
2184       lib_arch.compat_class = arch_array[i].compat_class;
2185       lib_arch.name         = arch_array[i].name;
2186     }
2187   }
2188 
2189   assert(running_arch_index != -1,
2190     "Didn't find running architecture code (running_arch_code) in arch_array");
2191   if (running_arch_index == -1) {
2192     // Even though running architecture detection failed
2193     // we may still continue with reporting dlerror() message
2194     return NULL;
2195   }
2196 
2197   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2198     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2199     return NULL;
2200   }
2201 
2202   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2203     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2204     return NULL;
2205   }
2206 
2207   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2208     if ( lib_arch.name!=NULL ) {
2209       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2210         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2211         lib_arch.name, arch_array[running_arch_index].name);
2212     } else {
2213       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2214       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2215         lib_arch.code,
2216         arch_array[running_arch_index].name);
2217     }
2218   }
2219 
2220   return NULL;
2221 }
2222 
2223 void* os::dll_lookup(void* handle, const char* name) {
2224   return dlsym(handle, name);
2225 }
2226 
2227 int os::stat(const char *path, struct stat *sbuf) {
2228   char pathbuf[MAX_PATH];
2229   if (strlen(path) > MAX_PATH - 1) {
2230     errno = ENAMETOOLONG;
2231     return -1;
2232   }
2233   os::native_path(strcpy(pathbuf, path));
2234   return ::stat(pathbuf, sbuf);
2235 }
2236 
2237 static bool _print_ascii_file(const char* filename, outputStream* st) {
2238   int fd = ::open(filename, O_RDONLY);
2239   if (fd == -1) {
2240      return false;
2241   }
2242 
2243   char buf[32];
2244   int bytes;
2245   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2246     st->print_raw(buf, bytes);
2247   }
2248 
2249   ::close(fd);
2250 
2251   return true;
2252 }
2253 
2254 void os::print_os_info(outputStream* st) {
2255   st->print("OS:");
2256 
2257   if (!_print_ascii_file("/etc/release", st)) {
2258     st->print("Solaris");
2259   }
2260   st->cr();
2261 
2262   // kernel
2263   st->print("uname:");
2264   struct utsname name;
2265   uname(&name);
2266   st->print(name.sysname); st->print(" ");
2267   st->print(name.release); st->print(" ");
2268   st->print(name.version); st->print(" ");
2269   st->print(name.machine);
2270 
2271   // libthread
2272   if (os::Solaris::T2_libthread()) st->print("  (T2 libthread)");
2273   else st->print("  (T1 libthread)");
2274   st->cr();
2275 
2276   // rlimit
2277   st->print("rlimit:");
2278   struct rlimit rlim;
2279 
2280   st->print(" STACK ");
2281   getrlimit(RLIMIT_STACK, &rlim);
2282   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2283   else st->print("%uk", rlim.rlim_cur >> 10);
2284 
2285   st->print(", CORE ");
2286   getrlimit(RLIMIT_CORE, &rlim);
2287   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2288   else st->print("%uk", rlim.rlim_cur >> 10);
2289 
2290   st->print(", NOFILE ");
2291   getrlimit(RLIMIT_NOFILE, &rlim);
2292   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2293   else st->print("%d", rlim.rlim_cur);
2294 
2295   st->print(", AS ");
2296   getrlimit(RLIMIT_AS, &rlim);
2297   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2298   else st->print("%uk", rlim.rlim_cur >> 10);
2299   st->cr();
2300 
2301   // load average
2302   st->print("load average:");
2303   double loadavg[3];
2304   os::loadavg(loadavg, 3);
2305   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2306   st->cr();
2307 }
2308 
2309 
2310 static bool check_addr0(outputStream* st) {
2311   jboolean status = false;
2312   int fd = ::open("/proc/self/map",O_RDONLY);
2313   if (fd >= 0) {
2314     prmap_t p;
2315     while(::read(fd, &p, sizeof(p)) > 0) {
2316       if (p.pr_vaddr == 0x0) {
2317         st->print("Warning: Address: 0x%x, Size: %dK, ",p.pr_vaddr, p.pr_size/1024, p.pr_mapname);
2318         st->print("Mapped file: %s, ", p.pr_mapname[0] == '\0' ? "None" : p.pr_mapname);
2319         st->print("Access:");
2320         st->print("%s",(p.pr_mflags & MA_READ)  ? "r" : "-");
2321         st->print("%s",(p.pr_mflags & MA_WRITE) ? "w" : "-");
2322         st->print("%s",(p.pr_mflags & MA_EXEC)  ? "x" : "-");
2323         st->cr();
2324         status = true;
2325       }
2326       ::close(fd);
2327     }
2328   }
2329   return status;
2330 }
2331 
2332 void os::pd_print_cpu_info(outputStream* st) {
2333   // Nothing to do for now.
2334 }
2335 
2336 void os::print_memory_info(outputStream* st) {
2337   st->print("Memory:");
2338   st->print(" %dk page", os::vm_page_size()>>10);
2339   st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
2340   st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
2341   st->cr();
2342   (void) check_addr0(st);
2343 }
2344 
2345 // Taken from /usr/include/sys/machsig.h  Supposed to be architecture specific
2346 // but they're the same for all the solaris architectures that we support.
2347 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2348                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2349                           "ILL_COPROC", "ILL_BADSTK" };
2350 
2351 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2352                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2353                           "FPE_FLTINV", "FPE_FLTSUB" };
2354 
2355 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2356 
2357 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2358 
2359 void os::print_siginfo(outputStream* st, void* siginfo) {
2360   st->print("siginfo:");
2361 
2362   const int buflen = 100;
2363   char buf[buflen];
2364   siginfo_t *si = (siginfo_t*)siginfo;
2365   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2366   char *err = strerror(si->si_errno);
2367   if (si->si_errno != 0 && err != NULL) {
2368     st->print("si_errno=%s", err);
2369   } else {
2370     st->print("si_errno=%d", si->si_errno);
2371   }
2372   const int c = si->si_code;
2373   assert(c > 0, "unexpected si_code");
2374   switch (si->si_signo) {
2375   case SIGILL:
2376     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2377     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2378     break;
2379   case SIGFPE:
2380     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2381     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2382     break;
2383   case SIGSEGV:
2384     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2385     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2386     break;
2387   case SIGBUS:
2388     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2389     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2390     break;
2391   default:
2392     st->print(", si_code=%d", si->si_code);
2393     // no si_addr
2394   }
2395 
2396   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2397       UseSharedSpaces) {
2398     FileMapInfo* mapinfo = FileMapInfo::current_info();
2399     if (mapinfo->is_in_shared_space(si->si_addr)) {
2400       st->print("\n\nError accessing class data sharing archive."   \
2401                 " Mapped file inaccessible during execution, "      \
2402                 " possible disk/network problem.");
2403     }
2404   }
2405   st->cr();
2406 }
2407 
2408 // Moved from whole group, because we need them here for diagnostic
2409 // prints.
2410 #define OLDMAXSIGNUM 32
2411 static int Maxsignum = 0;
2412 static int *ourSigFlags = NULL;
2413 
2414 extern "C" void sigINTRHandler(int, siginfo_t*, void*);
2415 
2416 int os::Solaris::get_our_sigflags(int sig) {
2417   assert(ourSigFlags!=NULL, "signal data structure not initialized");
2418   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2419   return ourSigFlags[sig];
2420 }
2421 
2422 void os::Solaris::set_our_sigflags(int sig, int flags) {
2423   assert(ourSigFlags!=NULL, "signal data structure not initialized");
2424   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2425   ourSigFlags[sig] = flags;
2426 }
2427 
2428 
2429 static const char* get_signal_handler_name(address handler,
2430                                            char* buf, int buflen) {
2431   int offset;
2432   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
2433   if (found) {
2434     // skip directory names
2435     const char *p1, *p2;
2436     p1 = buf;
2437     size_t len = strlen(os::file_separator());
2438     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
2439     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
2440   } else {
2441     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
2442   }
2443   return buf;
2444 }
2445 
2446 static void print_signal_handler(outputStream* st, int sig,
2447                                   char* buf, size_t buflen) {
2448   struct sigaction sa;
2449 
2450   sigaction(sig, NULL, &sa);
2451 
2452   st->print("%s: ", os::exception_name(sig, buf, buflen));
2453 
2454   address handler = (sa.sa_flags & SA_SIGINFO)
2455                   ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
2456                   : CAST_FROM_FN_PTR(address, sa.sa_handler);
2457 
2458   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
2459     st->print("SIG_DFL");
2460   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
2461     st->print("SIG_IGN");
2462   } else {
2463     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
2464   }
2465 
2466   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
2467 
2468   address rh = VMError::get_resetted_sighandler(sig);
2469   // May be, handler was resetted by VMError?
2470   if(rh != NULL) {
2471     handler = rh;
2472     sa.sa_flags = VMError::get_resetted_sigflags(sig);
2473   }
2474 
2475   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
2476 
2477   // Check: is it our handler?
2478   if(handler == CAST_FROM_FN_PTR(address, signalHandler) ||
2479      handler == CAST_FROM_FN_PTR(address, sigINTRHandler)) {
2480     // It is our signal handler
2481     // check for flags
2482     if(sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
2483       st->print(
2484         ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
2485         os::Solaris::get_our_sigflags(sig));
2486     }
2487   }
2488   st->cr();
2489 }
2490 
2491 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2492   st->print_cr("Signal Handlers:");
2493   print_signal_handler(st, SIGSEGV, buf, buflen);
2494   print_signal_handler(st, SIGBUS , buf, buflen);
2495   print_signal_handler(st, SIGFPE , buf, buflen);
2496   print_signal_handler(st, SIGPIPE, buf, buflen);
2497   print_signal_handler(st, SIGXFSZ, buf, buflen);
2498   print_signal_handler(st, SIGILL , buf, buflen);
2499   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2500   print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
2501   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2502   print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
2503   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2504   print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
2505   print_signal_handler(st, os::Solaris::SIGinterrupt(), buf, buflen);
2506   print_signal_handler(st, os::Solaris::SIGasync(), buf, buflen);
2507 }
2508 
2509 static char saved_jvm_path[MAXPATHLEN] = { 0 };
2510 
2511 // Find the full path to the current module, libjvm.so or libjvm_g.so
2512 void os::jvm_path(char *buf, jint buflen) {
2513   // Error checking.
2514   if (buflen < MAXPATHLEN) {
2515     assert(false, "must use a large-enough buffer");
2516     buf[0] = '\0';
2517     return;
2518   }
2519   // Lazy resolve the path to current module.
2520   if (saved_jvm_path[0] != 0) {
2521     strcpy(buf, saved_jvm_path);
2522     return;
2523   }
2524 
2525   Dl_info dlinfo;
2526   int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
2527   assert(ret != 0, "cannot locate libjvm");
2528   realpath((char *)dlinfo.dli_fname, buf);
2529 
2530   if (Arguments::created_by_gamma_launcher()) {
2531     // Support for the gamma launcher.  Typical value for buf is
2532     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2533     // the right place in the string, then assume we are installed in a JDK and
2534     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2535     // up the path so it looks like libjvm.so is installed there (append a
2536     // fake suffix hotspot/libjvm.so).
2537     const char *p = buf + strlen(buf) - 1;
2538     for (int count = 0; p > buf && count < 5; ++count) {
2539       for (--p; p > buf && *p != '/'; --p)
2540         /* empty */ ;
2541     }
2542 
2543     if (strncmp(p, "/jre/lib/", 9) != 0) {
2544       // Look for JAVA_HOME in the environment.
2545       char* java_home_var = ::getenv("JAVA_HOME");
2546       if (java_home_var != NULL && java_home_var[0] != 0) {
2547         char cpu_arch[12];
2548         char* jrelib_p;
2549         int   len;
2550         sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
2551 #ifdef _LP64
2552         // If we are on sparc running a 64-bit vm, look in jre/lib/sparcv9.
2553         if (strcmp(cpu_arch, "sparc") == 0) {
2554           strcat(cpu_arch, "v9");
2555         } else if (strcmp(cpu_arch, "i386") == 0) {
2556           strcpy(cpu_arch, "amd64");
2557         }
2558 #endif
2559         // Check the current module name "libjvm.so" or "libjvm_g.so".
2560         p = strrchr(buf, '/');
2561         assert(strstr(p, "/libjvm") == p, "invalid library name");
2562         p = strstr(p, "_g") ? "_g" : "";
2563 
2564         realpath(java_home_var, buf);
2565         // determine if this is a legacy image or modules image
2566         // modules image doesn't have "jre" subdirectory
2567         len = strlen(buf);
2568         jrelib_p = buf + len;
2569         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2570         if (0 != access(buf, F_OK)) {
2571           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2572         }
2573 
2574         if (0 == access(buf, F_OK)) {
2575           // Use current module name "libjvm[_g].so" instead of
2576           // "libjvm"debug_only("_g")".so" since for fastdebug version
2577           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2578           len = strlen(buf);
2579           snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2580         } else {
2581           // Go back to path of .so
2582           realpath((char *)dlinfo.dli_fname, buf);
2583         }
2584       }
2585     }
2586   }
2587 
2588   strcpy(saved_jvm_path, buf);
2589 }
2590 
2591 
2592 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2593   // no prefix required, not even "_"
2594 }
2595 
2596 
2597 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2598   // no suffix required
2599 }
2600 
2601 // This method is a copy of JDK's sysGetLastErrorString
2602 // from src/solaris/hpi/src/system_md.c
2603 
2604 size_t os::lasterror(char *buf, size_t len) {
2605 
2606   if (errno == 0)  return 0;
2607 
2608   const char *s = ::strerror(errno);
2609   size_t n = ::strlen(s);
2610   if (n >= len) {
2611     n = len - 1;
2612   }
2613   ::strncpy(buf, s, n);
2614   buf[n] = '\0';
2615   return n;
2616 }
2617 
2618 
2619 // sun.misc.Signal
2620 
2621 extern "C" {
2622   static void UserHandler(int sig, void *siginfo, void *context) {
2623     // Ctrl-C is pressed during error reporting, likely because the error
2624     // handler fails to abort. Let VM die immediately.
2625     if (sig == SIGINT && is_error_reported()) {
2626        os::die();
2627     }
2628 
2629     os::signal_notify(sig);
2630     // We do not need to reinstate the signal handler each time...
2631   }
2632 }
2633 
2634 void* os::user_handler() {
2635   return CAST_FROM_FN_PTR(void*, UserHandler);
2636 }
2637 
2638 extern "C" {
2639   typedef void (*sa_handler_t)(int);
2640   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2641 }
2642 
2643 void* os::signal(int signal_number, void* handler) {
2644   struct sigaction sigAct, oldSigAct;
2645   sigfillset(&(sigAct.sa_mask));
2646   sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2647   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2648 
2649   if (sigaction(signal_number, &sigAct, &oldSigAct))
2650     // -1 means registration failed
2651     return (void *)-1;
2652 
2653   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2654 }
2655 
2656 void os::signal_raise(int signal_number) {
2657   raise(signal_number);
2658 }
2659 
2660 /*
2661  * The following code is moved from os.cpp for making this
2662  * code platform specific, which it is by its very nature.
2663  */
2664 
2665 // a counter for each possible signal value
2666 static int Sigexit = 0;
2667 static int Maxlibjsigsigs;
2668 static jint *pending_signals = NULL;
2669 static int *preinstalled_sigs = NULL;
2670 static struct sigaction *chainedsigactions = NULL;
2671 static sema_t sig_sem;
2672 typedef int (*version_getting_t)();
2673 version_getting_t os::Solaris::get_libjsig_version = NULL;
2674 static int libjsigversion = NULL;
2675 
2676 int os::sigexitnum_pd() {
2677   assert(Sigexit > 0, "signal memory not yet initialized");
2678   return Sigexit;
2679 }
2680 
2681 void os::Solaris::init_signal_mem() {
2682   // Initialize signal structures
2683   Maxsignum = SIGRTMAX;
2684   Sigexit = Maxsignum+1;
2685   assert(Maxsignum >0, "Unable to obtain max signal number");
2686 
2687   Maxlibjsigsigs = Maxsignum;
2688 
2689   // pending_signals has one int per signal
2690   // The additional signal is for SIGEXIT - exit signal to signal_thread
2691   pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1));
2692   memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2693 
2694   if (UseSignalChaining) {
2695      chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2696        * (Maxsignum + 1));
2697      memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2698      preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1));
2699      memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2700   }
2701   ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1 ));
2702   memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2703 }
2704 
2705 void os::signal_init_pd() {
2706   int ret;
2707 
2708   ret = ::sema_init(&sig_sem, 0, NULL, NULL);
2709   assert(ret == 0, "sema_init() failed");
2710 }
2711 
2712 void os::signal_notify(int signal_number) {
2713   int ret;
2714 
2715   Atomic::inc(&pending_signals[signal_number]);
2716   ret = ::sema_post(&sig_sem);
2717   assert(ret == 0, "sema_post() failed");
2718 }
2719 
2720 static int check_pending_signals(bool wait_for_signal) {
2721   int ret;
2722   while (true) {
2723     for (int i = 0; i < Sigexit + 1; i++) {
2724       jint n = pending_signals[i];
2725       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2726         return i;
2727       }
2728     }
2729     if (!wait_for_signal) {
2730       return -1;
2731     }
2732     JavaThread *thread = JavaThread::current();
2733     ThreadBlockInVM tbivm(thread);
2734 
2735     bool threadIsSuspended;
2736     do {
2737       thread->set_suspend_equivalent();
2738       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2739       while((ret = ::sema_wait(&sig_sem)) == EINTR)
2740           ;
2741       assert(ret == 0, "sema_wait() failed");
2742 
2743       // were we externally suspended while we were waiting?
2744       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2745       if (threadIsSuspended) {
2746         //
2747         // The semaphore has been incremented, but while we were waiting
2748         // another thread suspended us. We don't want to continue running
2749         // while suspended because that would surprise the thread that
2750         // suspended us.
2751         //
2752         ret = ::sema_post(&sig_sem);
2753         assert(ret == 0, "sema_post() failed");
2754 
2755         thread->java_suspend_self();
2756       }
2757     } while (threadIsSuspended);
2758   }
2759 }
2760 
2761 int os::signal_lookup() {
2762   return check_pending_signals(false);
2763 }
2764 
2765 int os::signal_wait() {
2766   return check_pending_signals(true);
2767 }
2768 
2769 ////////////////////////////////////////////////////////////////////////////////
2770 // Virtual Memory
2771 
2772 static int page_size = -1;
2773 
2774 // The mmap MAP_ALIGN flag is supported on Solaris 9 and later.  init_2() will
2775 // clear this var if support is not available.
2776 static bool has_map_align = true;
2777 
2778 int os::vm_page_size() {
2779   assert(page_size != -1, "must call os::init");
2780   return page_size;
2781 }
2782 
2783 // Solaris allocates memory by pages.
2784 int os::vm_allocation_granularity() {
2785   assert(page_size != -1, "must call os::init");
2786   return page_size;
2787 }
2788 
2789 bool os::commit_memory(char* addr, size_t bytes, bool exec) {
2790   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2791   size_t size = bytes;
2792   char *res = Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2793   if (res != NULL) {
2794     if (UseNUMAInterleaving) {
2795       numa_make_global(addr, bytes);
2796     }
2797     return true;
2798   }
2799   return false;
2800 }
2801 
2802 bool os::commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2803                        bool exec) {
2804   if (commit_memory(addr, bytes, exec)) {
2805     if (UseMPSS && alignment_hint > (size_t)vm_page_size()) {
2806       // If the large page size has been set and the VM
2807       // is using large pages, use the large page size
2808       // if it is smaller than the alignment hint. This is
2809       // a case where the VM wants to use a larger alignment size
2810       // for its own reasons but still want to use large pages
2811       // (which is what matters to setting the mpss range.
2812       size_t page_size = 0;
2813       if (large_page_size() < alignment_hint) {
2814         assert(UseLargePages, "Expected to be here for large page use only");
2815         page_size = large_page_size();
2816       } else {
2817         // If the alignment hint is less than the large page
2818         // size, the VM wants a particular alignment (thus the hint)
2819         // for internal reasons.  Try to set the mpss range using
2820         // the alignment_hint.
2821         page_size = alignment_hint;
2822       }
2823       // Since this is a hint, ignore any failures.
2824       (void)Solaris::set_mpss_range(addr, bytes, page_size);
2825     }
2826     return true;
2827   }
2828   return false;
2829 }
2830 
2831 // Uncommit the pages in a specified region.
2832 void os::free_memory(char* addr, size_t bytes, size_t alignment_hint) {
2833   if (madvise(addr, bytes, MADV_FREE) < 0) {
2834     debug_only(warning("MADV_FREE failed."));
2835     return;
2836   }
2837 }
2838 
2839 bool os::create_stack_guard_pages(char* addr, size_t size) {
2840   return os::commit_memory(addr, size);
2841 }
2842 
2843 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2844   return os::uncommit_memory(addr, size);
2845 }
2846 
2847 // Change the page size in a given range.
2848 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2849   assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2850   assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2851   if (UseLargePages && UseMPSS) {
2852     Solaris::set_mpss_range(addr, bytes, alignment_hint);
2853   }
2854 }
2855 
2856 // Tell the OS to make the range local to the first-touching LWP
2857 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2858   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2859   if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2860     debug_only(warning("MADV_ACCESS_LWP failed."));
2861   }
2862 }
2863 
2864 // Tell the OS that this range would be accessed from different LWPs.
2865 void os::numa_make_global(char *addr, size_t bytes) {
2866   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2867   if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2868     debug_only(warning("MADV_ACCESS_MANY failed."));
2869   }
2870 }
2871 
2872 // Get the number of the locality groups.
2873 size_t os::numa_get_groups_num() {
2874   size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2875   return n != -1 ? n : 1;
2876 }
2877 
2878 // Get a list of leaf locality groups. A leaf lgroup is group that
2879 // doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2880 // board. An LWP is assigned to one of these groups upon creation.
2881 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2882    if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2883      ids[0] = 0;
2884      return 1;
2885    }
2886    int result_size = 0, top = 1, bottom = 0, cur = 0;
2887    for (int k = 0; k < size; k++) {
2888      int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2889                                     (Solaris::lgrp_id_t*)&ids[top], size - top);
2890      if (r == -1) {
2891        ids[0] = 0;
2892        return 1;
2893      }
2894      if (!r) {
2895        // That's a leaf node.
2896        assert (bottom <= cur, "Sanity check");
2897        // Check if the node has memory
2898        if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2899                                    NULL, 0, LGRP_RSRC_MEM) > 0) {
2900          ids[bottom++] = ids[cur];
2901        }
2902      }
2903      top += r;
2904      cur++;
2905    }
2906    if (bottom == 0) {
2907      // Handle a situation, when the OS reports no memory available.
2908      // Assume UMA architecture.
2909      ids[0] = 0;
2910      return 1;
2911    }
2912    return bottom;
2913 }
2914 
2915 // Detect the topology change. Typically happens during CPU plugging-unplugging.
2916 bool os::numa_topology_changed() {
2917   int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2918   if (is_stale != -1 && is_stale) {
2919     Solaris::lgrp_fini(Solaris::lgrp_cookie());
2920     Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2921     assert(c != 0, "Failure to initialize LGRP API");
2922     Solaris::set_lgrp_cookie(c);
2923     return true;
2924   }
2925   return false;
2926 }
2927 
2928 // Get the group id of the current LWP.
2929 int os::numa_get_group_id() {
2930   int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2931   if (lgrp_id == -1) {
2932     return 0;
2933   }
2934   const int size = os::numa_get_groups_num();
2935   int *ids = (int*)alloca(size * sizeof(int));
2936 
2937   // Get the ids of all lgroups with memory; r is the count.
2938   int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2939                                   (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2940   if (r <= 0) {
2941     return 0;
2942   }
2943   return ids[os::random() % r];
2944 }
2945 
2946 // Request information about the page.
2947 bool os::get_page_info(char *start, page_info* info) {
2948   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2949   uint64_t addr = (uintptr_t)start;
2950   uint64_t outdata[2];
2951   uint_t validity = 0;
2952 
2953   if (os::Solaris::meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2954     return false;
2955   }
2956 
2957   info->size = 0;
2958   info->lgrp_id = -1;
2959 
2960   if ((validity & 1) != 0) {
2961     if ((validity & 2) != 0) {
2962       info->lgrp_id = outdata[0];
2963     }
2964     if ((validity & 4) != 0) {
2965       info->size = outdata[1];
2966     }
2967     return true;
2968   }
2969   return false;
2970 }
2971 
2972 // Scan the pages from start to end until a page different than
2973 // the one described in the info parameter is encountered.
2974 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2975   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2976   const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2977   uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT];
2978   uint_t validity[MAX_MEMINFO_CNT];
2979 
2980   size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2981   uint64_t p = (uint64_t)start;
2982   while (p < (uint64_t)end) {
2983     addrs[0] = p;
2984     size_t addrs_count = 1;
2985     while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] < (uint64_t)end) {
2986       addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2987       addrs_count++;
2988     }
2989 
2990     if (os::Solaris::meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2991       return NULL;
2992     }
2993 
2994     size_t i = 0;
2995     for (; i < addrs_count; i++) {
2996       if ((validity[i] & 1) != 0) {
2997         if ((validity[i] & 4) != 0) {
2998           if (outdata[types * i + 1] != page_expected->size) {
2999             break;
3000           }
3001         } else
3002           if (page_expected->size != 0) {
3003             break;
3004           }
3005 
3006         if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
3007           if (outdata[types * i] != page_expected->lgrp_id) {
3008             break;
3009           }
3010         }
3011       } else {
3012         return NULL;
3013       }
3014     }
3015 
3016     if (i != addrs_count) {
3017       if ((validity[i] & 2) != 0) {
3018         page_found->lgrp_id = outdata[types * i];
3019       } else {
3020         page_found->lgrp_id = -1;
3021       }
3022       if ((validity[i] & 4) != 0) {
3023         page_found->size = outdata[types * i + 1];
3024       } else {
3025         page_found->size = 0;
3026       }
3027       return (char*)addrs[i];
3028     }
3029 
3030     p = addrs[addrs_count - 1] + page_size;
3031   }
3032   return end;
3033 }
3034 
3035 bool os::uncommit_memory(char* addr, size_t bytes) {
3036   size_t size = bytes;
3037   // Map uncommitted pages PROT_NONE so we fail early if we touch an
3038   // uncommitted page. Otherwise, the read/write might succeed if we
3039   // have enough swap space to back the physical page.
3040   return
3041     NULL != Solaris::mmap_chunk(addr, size,
3042                                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
3043                                 PROT_NONE);
3044 }
3045 
3046 char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
3047   char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
3048 
3049   if (b == MAP_FAILED) {
3050     return NULL;
3051   }
3052   return b;
3053 }
3054 
3055 char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes, size_t alignment_hint, bool fixed) {
3056   char* addr = requested_addr;
3057   int flags = MAP_PRIVATE | MAP_NORESERVE;
3058 
3059   assert(!(fixed && (alignment_hint > 0)), "alignment hint meaningless with fixed mmap");
3060 
3061   if (fixed) {
3062     flags |= MAP_FIXED;
3063   } else if (has_map_align && (alignment_hint > (size_t) vm_page_size())) {
3064     flags |= MAP_ALIGN;
3065     addr = (char*) alignment_hint;
3066   }
3067 
3068   // Map uncommitted pages PROT_NONE so we fail early if we touch an
3069   // uncommitted page. Otherwise, the read/write might succeed if we
3070   // have enough swap space to back the physical page.
3071   return mmap_chunk(addr, bytes, flags, PROT_NONE);
3072 }
3073 
3074 char* os::reserve_memory(size_t bytes, char* requested_addr, size_t alignment_hint) {
3075   char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint, (requested_addr != NULL));
3076 
3077   guarantee(requested_addr == NULL || requested_addr == addr,
3078             "OS failed to return requested mmap address.");
3079   return addr;
3080 }
3081 
3082 // Reserve memory at an arbitrary address, only if that area is
3083 // available (and not reserved for something else).
3084 
3085 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3086   const int max_tries = 10;
3087   char* base[max_tries];
3088   size_t size[max_tries];
3089 
3090   // Solaris adds a gap between mmap'ed regions.  The size of the gap
3091   // is dependent on the requested size and the MMU.  Our initial gap
3092   // value here is just a guess and will be corrected later.
3093   bool had_top_overlap = false;
3094   bool have_adjusted_gap = false;
3095   size_t gap = 0x400000;
3096 
3097   // Assert only that the size is a multiple of the page size, since
3098   // that's all that mmap requires, and since that's all we really know
3099   // about at this low abstraction level.  If we need higher alignment,
3100   // we can either pass an alignment to this method or verify alignment
3101   // in one of the methods further up the call chain.  See bug 5044738.
3102   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3103 
3104   // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
3105   // Give it a try, if the kernel honors the hint we can return immediately.
3106   char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
3107   volatile int err = errno;
3108   if (addr == requested_addr) {
3109     return addr;
3110   } else if (addr != NULL) {
3111     unmap_memory(addr, bytes);
3112   }
3113 
3114   if (PrintMiscellaneous && Verbose) {
3115     char buf[256];
3116     buf[0] = '\0';
3117     if (addr == NULL) {
3118       jio_snprintf(buf, sizeof(buf), ": %s", strerror(err));
3119     }
3120     warning("attempt_reserve_memory_at: couldn't reserve " SIZE_FORMAT " bytes at "
3121             PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
3122             "%s", bytes, requested_addr, addr, buf);
3123   }
3124 
3125   // Address hint method didn't work.  Fall back to the old method.
3126   // In theory, once SNV becomes our oldest supported platform, this
3127   // code will no longer be needed.
3128   //
3129   // Repeatedly allocate blocks until the block is allocated at the
3130   // right spot. Give up after max_tries.
3131   int i;
3132   for (i = 0; i < max_tries; ++i) {
3133     base[i] = reserve_memory(bytes);
3134 
3135     if (base[i] != NULL) {
3136       // Is this the block we wanted?
3137       if (base[i] == requested_addr) {
3138         size[i] = bytes;
3139         break;
3140       }
3141 
3142       // check that the gap value is right
3143       if (had_top_overlap && !have_adjusted_gap) {
3144         size_t actual_gap = base[i-1] - base[i] - bytes;
3145         if (gap != actual_gap) {
3146           // adjust the gap value and retry the last 2 allocations
3147           assert(i > 0, "gap adjustment code problem");
3148           have_adjusted_gap = true;  // adjust the gap only once, just in case
3149           gap = actual_gap;
3150           if (PrintMiscellaneous && Verbose) {
3151             warning("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
3152           }
3153           unmap_memory(base[i], bytes);
3154           unmap_memory(base[i-1], size[i-1]);
3155           i-=2;
3156           continue;
3157         }
3158       }
3159 
3160       // Does this overlap the block we wanted? Give back the overlapped
3161       // parts and try again.
3162       //
3163       // There is still a bug in this code: if top_overlap == bytes,
3164       // the overlap is offset from requested region by the value of gap.
3165       // In this case giving back the overlapped part will not work,
3166       // because we'll give back the entire block at base[i] and
3167       // therefore the subsequent allocation will not generate a new gap.
3168       // This could be fixed with a new algorithm that used larger
3169       // or variable size chunks to find the requested region -
3170       // but such a change would introduce additional complications.
3171       // It's rare enough that the planets align for this bug,
3172       // so we'll just wait for a fix for 6204603/5003415 which
3173       // will provide a mmap flag to allow us to avoid this business.
3174 
3175       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3176       if (top_overlap >= 0 && top_overlap < bytes) {
3177         had_top_overlap = true;
3178         unmap_memory(base[i], top_overlap);
3179         base[i] += top_overlap;
3180         size[i] = bytes - top_overlap;
3181       } else {
3182         size_t bottom_overlap = base[i] + bytes - requested_addr;
3183         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3184           if (PrintMiscellaneous && Verbose && bottom_overlap == 0) {
3185             warning("attempt_reserve_memory_at: possible alignment bug");
3186           }
3187           unmap_memory(requested_addr, bottom_overlap);
3188           size[i] = bytes - bottom_overlap;
3189         } else {
3190           size[i] = bytes;
3191         }
3192       }
3193     }
3194   }
3195 
3196   // Give back the unused reserved pieces.
3197 
3198   for (int j = 0; j < i; ++j) {
3199     if (base[j] != NULL) {
3200       unmap_memory(base[j], size[j]);
3201     }
3202   }
3203 
3204   return (i < max_tries) ? requested_addr : NULL;
3205 }
3206 
3207 bool os::release_memory(char* addr, size_t bytes) {
3208   size_t size = bytes;
3209   return munmap(addr, size) == 0;
3210 }
3211 
3212 static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
3213   assert(addr == (char*)align_size_down((uintptr_t)addr, os::vm_page_size()),
3214          "addr must be page aligned");
3215   int retVal = mprotect(addr, bytes, prot);
3216   return retVal == 0;
3217 }
3218 
3219 // Protect memory (Used to pass readonly pages through
3220 // JNI GetArray<type>Elements with empty arrays.)
3221 // Also, used for serialization page and for compressed oops null pointer
3222 // checking.
3223 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3224                         bool is_committed) {
3225   unsigned int p = 0;
3226   switch (prot) {
3227   case MEM_PROT_NONE: p = PROT_NONE; break;
3228   case MEM_PROT_READ: p = PROT_READ; break;
3229   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3230   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3231   default:
3232     ShouldNotReachHere();
3233   }
3234   // is_committed is unused.
3235   return solaris_mprotect(addr, bytes, p);
3236 }
3237 
3238 // guard_memory and unguard_memory only happens within stack guard pages.
3239 // Since ISM pertains only to the heap, guard and unguard memory should not
3240 /// happen with an ISM region.
3241 bool os::guard_memory(char* addr, size_t bytes) {
3242   return solaris_mprotect(addr, bytes, PROT_NONE);
3243 }
3244 
3245 bool os::unguard_memory(char* addr, size_t bytes) {
3246   return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
3247 }
3248 
3249 // Large page support
3250 
3251 // UseLargePages is the master flag to enable/disable large page memory.
3252 // UseMPSS and UseISM are supported for compatibility reasons. Their combined
3253 // effects can be described in the following table:
3254 //
3255 // UseLargePages UseMPSS UseISM
3256 //    false         *       *   => UseLargePages is the master switch, turning
3257 //                                 it off will turn off both UseMPSS and
3258 //                                 UseISM. VM will not use large page memory
3259 //                                 regardless the settings of UseMPSS/UseISM.
3260 //     true      false    false => Unless future Solaris provides other
3261 //                                 mechanism to use large page memory, this
3262 //                                 combination is equivalent to -UseLargePages,
3263 //                                 VM will not use large page memory
3264 //     true      true     false => JVM will use MPSS for large page memory.
3265 //                                 This is the default behavior.
3266 //     true      false    true  => JVM will use ISM for large page memory.
3267 //     true      true     true  => JVM will use ISM if it is available.
3268 //                                 Otherwise, JVM will fall back to MPSS.
3269 //                                 Becaues ISM is now available on all
3270 //                                 supported Solaris versions, this combination
3271 //                                 is equivalent to +UseISM -UseMPSS.
3272 
3273 static size_t _large_page_size = 0;
3274 
3275 bool os::Solaris::ism_sanity_check(bool warn, size_t * page_size) {
3276   // x86 uses either 2M or 4M page, depending on whether PAE (Physical Address
3277   // Extensions) mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. Sparc
3278   // can support multiple page sizes.
3279 
3280   // Don't bother to probe page size because getpagesizes() comes with MPSS.
3281   // ISM is only recommended on old Solaris where there is no MPSS support.
3282   // Simply choose a conservative value as default.
3283   *page_size = LargePageSizeInBytes ? LargePageSizeInBytes :
3284                SPARC_ONLY(4 * M) IA32_ONLY(4 * M) AMD64_ONLY(2 * M)
3285                ARM_ONLY(2 * M);
3286 
3287   // ISM is available on all supported Solaris versions
3288   return true;
3289 }
3290 
3291 // Insertion sort for small arrays (descending order).
3292 static void insertion_sort_descending(size_t* array, int len) {
3293   for (int i = 0; i < len; i++) {
3294     size_t val = array[i];
3295     for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
3296       size_t tmp = array[key];
3297       array[key] = array[key - 1];
3298       array[key - 1] = tmp;
3299     }
3300   }
3301 }
3302 
3303 bool os::Solaris::mpss_sanity_check(bool warn, size_t * page_size) {
3304   const unsigned int usable_count = VM_Version::page_size_count();
3305   if (usable_count == 1) {
3306     return false;
3307   }
3308 
3309   // Find the right getpagesizes interface.  When solaris 11 is the minimum
3310   // build platform, getpagesizes() (without the '2') can be called directly.
3311   typedef int (*gps_t)(size_t[], int);
3312   gps_t gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes2"));
3313   if (gps_func == NULL) {
3314     gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes"));
3315     if (gps_func == NULL) {
3316       if (warn) {
3317         warning("MPSS is not supported by the operating system.");
3318       }
3319       return false;
3320     }
3321   }
3322 
3323   // Fill the array of page sizes.
3324   int n = (*gps_func)(_page_sizes, page_sizes_max);
3325   assert(n > 0, "Solaris bug?");
3326 
3327   if (n == page_sizes_max) {
3328     // Add a sentinel value (necessary only if the array was completely filled
3329     // since it is static (zeroed at initialization)).
3330     _page_sizes[--n] = 0;
3331     DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
3332   }
3333   assert(_page_sizes[n] == 0, "missing sentinel");
3334   trace_page_sizes("available page sizes", _page_sizes, n);
3335 
3336   if (n == 1) return false;     // Only one page size available.
3337 
3338   // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
3339   // select up to usable_count elements.  First sort the array, find the first
3340   // acceptable value, then copy the usable sizes to the top of the array and
3341   // trim the rest.  Make sure to include the default page size :-).
3342   //
3343   // A better policy could get rid of the 4M limit by taking the sizes of the
3344   // important VM memory regions (java heap and possibly the code cache) into
3345   // account.
3346   insertion_sort_descending(_page_sizes, n);
3347   const size_t size_limit =
3348     FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
3349   int beg;
3350   for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */ ;
3351   const int end = MIN2((int)usable_count, n) - 1;
3352   for (int cur = 0; cur < end; ++cur, ++beg) {
3353     _page_sizes[cur] = _page_sizes[beg];
3354   }
3355   _page_sizes[end] = vm_page_size();
3356   _page_sizes[end + 1] = 0;
3357 
3358   if (_page_sizes[end] > _page_sizes[end - 1]) {
3359     // Default page size is not the smallest; sort again.
3360     insertion_sort_descending(_page_sizes, end + 1);
3361   }
3362   *page_size = _page_sizes[0];
3363 
3364   trace_page_sizes("usable page sizes", _page_sizes, end + 1);
3365   return true;
3366 }
3367 
3368 void os::large_page_init() {
3369   if (!UseLargePages) {
3370     UseISM = false;
3371     UseMPSS = false;
3372     return;
3373   }
3374 
3375   // print a warning if any large page related flag is specified on command line
3376   bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
3377                          !FLAG_IS_DEFAULT(UseISM)               ||
3378                          !FLAG_IS_DEFAULT(UseMPSS)              ||
3379                          !FLAG_IS_DEFAULT(LargePageSizeInBytes);
3380   UseISM = UseISM &&
3381            Solaris::ism_sanity_check(warn_on_failure, &_large_page_size);
3382   if (UseISM) {
3383     // ISM disables MPSS to be compatible with old JDK behavior
3384     UseMPSS = false;
3385     _page_sizes[0] = _large_page_size;
3386     _page_sizes[1] = vm_page_size();
3387   }
3388 
3389   UseMPSS = UseMPSS &&
3390             Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
3391 
3392   UseLargePages = UseISM || UseMPSS;
3393 }
3394 
3395 bool os::Solaris::set_mpss_range(caddr_t start, size_t bytes, size_t align) {
3396   // Signal to OS that we want large pages for addresses
3397   // from addr, addr + bytes
3398   struct memcntl_mha mpss_struct;
3399   mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
3400   mpss_struct.mha_pagesize = align;
3401   mpss_struct.mha_flags = 0;
3402   if (memcntl(start, bytes, MC_HAT_ADVISE,
3403               (caddr_t) &mpss_struct, 0, 0) < 0) {
3404     debug_only(warning("Attempt to use MPSS failed."));
3405     return false;
3406   }
3407   return true;
3408 }
3409 
3410 char* os::reserve_memory_special(size_t size, char* addr, bool exec) {
3411   // "exec" is passed in but not used.  Creating the shared image for
3412   // the code cache doesn't have an SHM_X executable permission to check.
3413   assert(UseLargePages && UseISM, "only for ISM large pages");
3414 
3415   char* retAddr = NULL;
3416   int shmid;
3417   key_t ismKey;
3418 
3419   bool warn_on_failure = UseISM &&
3420                         (!FLAG_IS_DEFAULT(UseLargePages)         ||
3421                          !FLAG_IS_DEFAULT(UseISM)                ||
3422                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3423                         );
3424   char msg[128];
3425 
3426   ismKey = IPC_PRIVATE;
3427 
3428   // Create a large shared memory region to attach to based on size.
3429   // Currently, size is the total size of the heap
3430   shmid = shmget(ismKey, size, SHM_R | SHM_W | IPC_CREAT);
3431   if (shmid == -1){
3432      if (warn_on_failure) {
3433        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3434        warning(msg);
3435      }
3436      return NULL;
3437   }
3438 
3439   // Attach to the region
3440   retAddr = (char *) shmat(shmid, 0, SHM_SHARE_MMU | SHM_R | SHM_W);
3441   int err = errno;
3442 
3443   // Remove shmid. If shmat() is successful, the actual shared memory segment
3444   // will be deleted when it's detached by shmdt() or when the process
3445   // terminates. If shmat() is not successful this will remove the shared
3446   // segment immediately.
3447   shmctl(shmid, IPC_RMID, NULL);
3448 
3449   if (retAddr == (char *) -1) {
3450     if (warn_on_failure) {
3451       jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3452       warning(msg);
3453     }
3454     return NULL;
3455   }
3456   if ((retAddr != NULL) && UseNUMAInterleaving) {
3457     numa_make_global(retAddr, size);
3458   }
3459   return retAddr;
3460 }
3461 
3462 bool os::release_memory_special(char* base, size_t bytes) {
3463   // detaching the SHM segment will also delete it, see reserve_memory_special()
3464   int rslt = shmdt(base);
3465   return rslt == 0;
3466 }
3467 
3468 size_t os::large_page_size() {
3469   return _large_page_size;
3470 }
3471 
3472 // MPSS allows application to commit large page memory on demand; with ISM
3473 // the entire memory region must be allocated as shared memory.
3474 bool os::can_commit_large_page_memory() {
3475   return UseISM ? false : true;
3476 }
3477 
3478 bool os::can_execute_large_page_memory() {
3479   return UseISM ? false : true;
3480 }
3481 
3482 static int os_sleep(jlong millis, bool interruptible) {
3483   const jlong limit = INT_MAX;
3484   jlong prevtime;
3485   int res;
3486 
3487   while (millis > limit) {
3488     if ((res = os_sleep(limit, interruptible)) != OS_OK)
3489       return res;
3490     millis -= limit;
3491   }
3492 
3493   // Restart interrupted polls with new parameters until the proper delay
3494   // has been completed.
3495 
3496   prevtime = getTimeMillis();
3497 
3498   while (millis > 0) {
3499     jlong newtime;
3500 
3501     if (!interruptible) {
3502       // Following assert fails for os::yield_all:
3503       // assert(!thread->is_Java_thread(), "must not be java thread");
3504       res = poll(NULL, 0, millis);
3505     } else {
3506       JavaThread *jt = JavaThread::current();
3507 
3508       INTERRUPTIBLE_NORESTART_VM_ALWAYS(poll(NULL, 0, millis), res, jt,
3509         os::Solaris::clear_interrupted);
3510     }
3511 
3512     // INTERRUPTIBLE_NORESTART_VM_ALWAYS returns res == OS_INTRPT for
3513     // thread.Interrupt.
3514 
3515     // See c/r 6751923. Poll can return 0 before time
3516     // has elapsed if time is set via clock_settime (as NTP does).
3517     // res == 0 if poll timed out (see man poll RETURN VALUES)
3518     // using the logic below checks that we really did
3519     // sleep at least "millis" if not we'll sleep again.
3520     if( ( res == 0 ) || ((res == OS_ERR) && (errno == EINTR))) {
3521       newtime = getTimeMillis();
3522       assert(newtime >= prevtime, "time moving backwards");
3523     /* Doing prevtime and newtime in microseconds doesn't help precision,
3524        and trying to round up to avoid lost milliseconds can result in a
3525        too-short delay. */
3526       millis -= newtime - prevtime;
3527       if(millis <= 0)
3528         return OS_OK;
3529       prevtime = newtime;
3530     } else
3531       return res;
3532   }
3533 
3534   return OS_OK;
3535 }
3536 
3537 // Read calls from inside the vm need to perform state transitions
3538 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3539   INTERRUPTIBLE_RETURN_INT_VM(::read(fd, buf, nBytes), os::Solaris::clear_interrupted);
3540 }
3541 
3542 size_t os::restartable_read(int fd, void *buf, unsigned int nBytes) {
3543   INTERRUPTIBLE_RETURN_INT(::read(fd, buf, nBytes), os::Solaris::clear_interrupted);
3544 }
3545 
3546 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3547   assert(thread == Thread::current(),  "thread consistency check");
3548 
3549   // TODO-FIXME: this should be removed.
3550   // On Solaris machines (especially 2.5.1) we found that sometimes the VM gets into a live lock
3551   // situation with a JavaThread being starved out of a lwp. The kernel doesn't seem to generate
3552   // a SIGWAITING signal which would enable the threads library to create a new lwp for the starving
3553   // thread. We suspect that because the Watcher thread keeps waking up at periodic intervals the kernel
3554   // is fooled into believing that the system is making progress. In the code below we block the
3555   // the watcher thread while safepoint is in progress so that it would not appear as though the
3556   // system is making progress.
3557   if (!Solaris::T2_libthread() &&
3558       thread->is_Watcher_thread() && SafepointSynchronize::is_synchronizing() && !Arguments::has_profile()) {
3559     // We now try to acquire the threads lock. Since this lock is held by the VM thread during
3560     // the entire safepoint, the watcher thread will  line up here during the safepoint.
3561     Threads_lock->lock_without_safepoint_check();
3562     Threads_lock->unlock();
3563   }
3564 
3565   if (thread->is_Java_thread()) {
3566     // This is a JavaThread so we honor the _thread_blocked protocol
3567     // even for sleeps of 0 milliseconds. This was originally done
3568     // as a workaround for bug 4338139. However, now we also do it
3569     // to honor the suspend-equivalent protocol.
3570 
3571     JavaThread *jt = (JavaThread *) thread;
3572     ThreadBlockInVM tbivm(jt);
3573 
3574     jt->set_suspend_equivalent();
3575     // cleared by handle_special_suspend_equivalent_condition() or
3576     // java_suspend_self() via check_and_wait_while_suspended()
3577 
3578     int ret_code;
3579     if (millis <= 0) {
3580       thr_yield();
3581       ret_code = 0;
3582     } else {
3583       // The original sleep() implementation did not create an
3584       // OSThreadWaitState helper for sleeps of 0 milliseconds.
3585       // I'm preserving that decision for now.
3586       OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3587 
3588       ret_code = os_sleep(millis, interruptible);
3589     }
3590 
3591     // were we externally suspended while we were waiting?
3592     jt->check_and_wait_while_suspended();
3593 
3594     return ret_code;
3595   }
3596 
3597   // non-JavaThread from this point on:
3598 
3599   if (millis <= 0) {
3600     thr_yield();
3601     return 0;
3602   }
3603 
3604   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3605 
3606   return os_sleep(millis, interruptible);
3607 }
3608 
3609 int os::naked_sleep() {
3610   // %% make the sleep time an integer flag. for now use 1 millisec.
3611   return os_sleep(1, false);
3612 }
3613 
3614 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3615 void os::infinite_sleep() {
3616   while (true) {    // sleep forever ...
3617     ::sleep(100);   // ... 100 seconds at a time
3618   }
3619 }
3620 
3621 // Used to convert frequent JVM_Yield() to nops
3622 bool os::dont_yield() {
3623   if (DontYieldALot) {
3624     static hrtime_t last_time = 0;
3625     hrtime_t diff = getTimeNanos() - last_time;
3626 
3627     if (diff < DontYieldALotInterval * 1000000)
3628       return true;
3629 
3630     last_time += diff;
3631 
3632     return false;
3633   }
3634   else {
3635     return false;
3636   }
3637 }
3638 
3639 // Caveat: Solaris os::yield() causes a thread-state transition whereas
3640 // the linux and win32 implementations do not.  This should be checked.
3641 
3642 void os::yield() {
3643   // Yields to all threads with same or greater priority
3644   os::sleep(Thread::current(), 0, false);
3645 }
3646 
3647 // Note that yield semantics are defined by the scheduling class to which
3648 // the thread currently belongs.  Typically, yield will _not yield to
3649 // other equal or higher priority threads that reside on the dispatch queues
3650 // of other CPUs.
3651 
3652 os::YieldResult os::NakedYield() { thr_yield(); return os::YIELD_UNKNOWN; }
3653 
3654 
3655 // On Solaris we found that yield_all doesn't always yield to all other threads.
3656 // There have been cases where there is a thread ready to execute but it doesn't
3657 // get an lwp as the VM thread continues to spin with sleeps of 1 millisecond.
3658 // The 1 millisecond wait doesn't seem long enough for the kernel to issue a
3659 // SIGWAITING signal which will cause a new lwp to be created. So we count the
3660 // number of times yield_all is called in the one loop and increase the sleep
3661 // time after 8 attempts. If this fails too we increase the concurrency level
3662 // so that the starving thread would get an lwp
3663 
3664 void os::yield_all(int attempts) {
3665   // Yields to all threads, including threads with lower priorities
3666   if (attempts == 0) {
3667     os::sleep(Thread::current(), 1, false);
3668   } else {
3669     int iterations = attempts % 30;
3670     if (iterations == 0 && !os::Solaris::T2_libthread()) {
3671       // thr_setconcurrency and _getconcurrency make sense only under T1.
3672       int noofLWPS = thr_getconcurrency();
3673       if (noofLWPS < (Threads::number_of_threads() + 2)) {
3674         thr_setconcurrency(thr_getconcurrency() + 1);
3675       }
3676     } else if (iterations < 25) {
3677       os::sleep(Thread::current(), 1, false);
3678     } else {
3679       os::sleep(Thread::current(), 10, false);
3680     }
3681   }
3682 }
3683 
3684 // Called from the tight loops to possibly influence time-sharing heuristics
3685 void os::loop_breaker(int attempts) {
3686   os::yield_all(attempts);
3687 }
3688 
3689 
3690 // Interface for setting lwp priorities.  If we are using T2 libthread,
3691 // which forces the use of BoundThreads or we manually set UseBoundThreads,
3692 // all of our threads will be assigned to real lwp's.  Using the thr_setprio
3693 // function is meaningless in this mode so we must adjust the real lwp's priority
3694 // The routines below implement the getting and setting of lwp priorities.
3695 //
3696 // Note: There are three priority scales used on Solaris.  Java priotities
3697 //       which range from 1 to 10, libthread "thr_setprio" scale which range
3698 //       from 0 to 127, and the current scheduling class of the process we
3699 //       are running in.  This is typically from -60 to +60.
3700 //       The setting of the lwp priorities in done after a call to thr_setprio
3701 //       so Java priorities are mapped to libthread priorities and we map from
3702 //       the latter to lwp priorities.  We don't keep priorities stored in
3703 //       Java priorities since some of our worker threads want to set priorities
3704 //       higher than all Java threads.
3705 //
3706 // For related information:
3707 // (1)  man -s 2 priocntl
3708 // (2)  man -s 4 priocntl
3709 // (3)  man dispadmin
3710 // =    librt.so
3711 // =    libthread/common/rtsched.c - thrp_setlwpprio().
3712 // =    ps -cL <pid> ... to validate priority.
3713 // =    sched_get_priority_min and _max
3714 //              pthread_create
3715 //              sched_setparam
3716 //              pthread_setschedparam
3717 //
3718 // Assumptions:
3719 // +    We assume that all threads in the process belong to the same
3720 //              scheduling class.   IE. an homogenous process.
3721 // +    Must be root or in IA group to change change "interactive" attribute.
3722 //              Priocntl() will fail silently.  The only indication of failure is when
3723 //              we read-back the value and notice that it hasn't changed.
3724 // +    Interactive threads enter the runq at the head, non-interactive at the tail.
3725 // +    For RT, change timeslice as well.  Invariant:
3726 //              constant "priority integral"
3727 //              Konst == TimeSlice * (60-Priority)
3728 //              Given a priority, compute appropriate timeslice.
3729 // +    Higher numerical values have higher priority.
3730 
3731 // sched class attributes
3732 typedef struct {
3733         int   schedPolicy;              // classID
3734         int   maxPrio;
3735         int   minPrio;
3736 } SchedInfo;
3737 
3738 
3739 static SchedInfo tsLimits, iaLimits, rtLimits, fxLimits;
3740 
3741 #ifdef ASSERT
3742 static int  ReadBackValidate = 1;
3743 #endif
3744 static int  myClass     = 0;
3745 static int  myMin       = 0;
3746 static int  myMax       = 0;
3747 static int  myCur       = 0;
3748 static bool priocntl_enable = false;
3749 
3750 static const int criticalPrio = 60; // FX/60 is critical thread class/priority on T4
3751 static int java_MaxPriority_to_os_priority = 0; // Saved mapping
3752 
3753 // Call the version of priocntl suitable for all supported versions
3754 // of Solaris. We need to call through this wrapper so that we can
3755 // build on Solaris 9 and run on Solaris 8, 9 and 10.
3756 //
3757 // This code should be removed if we ever stop supporting Solaris 8
3758 // and earlier releases.
3759 
3760 static long priocntl_stub(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg);
3761 typedef long (*priocntl_type)(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg);
3762 static priocntl_type priocntl_ptr = priocntl_stub;
3763 
3764 // Stub to set the value of the real pointer, and then call the real
3765 // function.
3766 
3767 static long priocntl_stub(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg) {
3768   // Try Solaris 8- name only.
3769   priocntl_type tmp = (priocntl_type)dlsym(RTLD_DEFAULT, "__priocntl");
3770   guarantee(tmp != NULL, "priocntl function not found.");
3771   priocntl_ptr = tmp;
3772   return (*priocntl_ptr)(PC_VERSION, idtype, id, cmd, arg);
3773 }
3774 
3775 
3776 // lwp_priocntl_init
3777 //
3778 // Try to determine the priority scale for our process.
3779 //
3780 // Return errno or 0 if OK.
3781 //
3782 static
3783 int     lwp_priocntl_init ()
3784 {
3785   int rslt;
3786   pcinfo_t ClassInfo;
3787   pcparms_t ParmInfo;
3788   int i;
3789 
3790   if (!UseThreadPriorities) return 0;
3791 
3792   // We are using Bound threads, we need to determine our priority ranges
3793   if (os::Solaris::T2_libthread() || UseBoundThreads) {
3794     // If ThreadPriorityPolicy is 1, switch tables
3795     if (ThreadPriorityPolicy == 1) {
3796       for (i = 0 ; i < CriticalPriority+1; i++)
3797         os::java_to_os_priority[i] = prio_policy1[i];
3798     }
3799     if (UseCriticalJavaThreadPriority) {
3800       // MaxPriority always maps to the FX scheduling class and criticalPrio.
3801       // See set_native_priority() and set_lwp_class_and_priority().
3802       // Save original MaxPriority mapping in case attempt to
3803       // use critical priority fails.
3804       java_MaxPriority_to_os_priority = os::java_to_os_priority[MaxPriority];
3805       // Set negative to distinguish from other priorities
3806       os::java_to_os_priority[MaxPriority] = -criticalPrio;
3807     }
3808   }
3809   // Not using Bound Threads, set to ThreadPolicy 1
3810   else {
3811     for ( i = 0 ; i < CriticalPriority+1; i++ ) {
3812       os::java_to_os_priority[i] = prio_policy1[i];
3813     }
3814     return 0;
3815   }
3816 
3817   // Get IDs for a set of well-known scheduling classes.
3818   // TODO-FIXME: GETCLINFO returns the current # of classes in the
3819   // the system.  We should have a loop that iterates over the
3820   // classID values, which are known to be "small" integers.
3821 
3822   strcpy(ClassInfo.pc_clname, "TS");
3823   ClassInfo.pc_cid = -1;
3824   rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3825   if (rslt < 0) return errno;
3826   assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3827   tsLimits.schedPolicy = ClassInfo.pc_cid;
3828   tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3829   tsLimits.minPrio = -tsLimits.maxPrio;
3830 
3831   strcpy(ClassInfo.pc_clname, "IA");
3832   ClassInfo.pc_cid = -1;
3833   rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3834   if (rslt < 0) return errno;
3835   assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3836   iaLimits.schedPolicy = ClassInfo.pc_cid;
3837   iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3838   iaLimits.minPrio = -iaLimits.maxPrio;
3839 
3840   strcpy(ClassInfo.pc_clname, "RT");
3841   ClassInfo.pc_cid = -1;
3842   rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3843   if (rslt < 0) return errno;
3844   assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3845   rtLimits.schedPolicy = ClassInfo.pc_cid;
3846   rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3847   rtLimits.minPrio = 0;
3848 
3849   strcpy(ClassInfo.pc_clname, "FX");
3850   ClassInfo.pc_cid = -1;
3851   rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3852   if (rslt < 0) return errno;
3853   assert(ClassInfo.pc_cid != -1, "cid for FX class is -1");
3854   fxLimits.schedPolicy = ClassInfo.pc_cid;
3855   fxLimits.maxPrio = ((fxinfo_t*)ClassInfo.pc_clinfo)->fx_maxupri;
3856   fxLimits.minPrio = 0;
3857 
3858   // Query our "current" scheduling class.
3859   // This will normally be IA, TS or, rarely, FX or RT.
3860   memset(&ParmInfo, 0, sizeof(ParmInfo));
3861   ParmInfo.pc_cid = PC_CLNULL;
3862   rslt = (*priocntl_ptr) (PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3863   if (rslt < 0) return errno;
3864   myClass = ParmInfo.pc_cid;
3865 
3866   // We now know our scheduling classId, get specific information
3867   // about the class.
3868   ClassInfo.pc_cid = myClass;
3869   ClassInfo.pc_clname[0] = 0;
3870   rslt = (*priocntl_ptr) (PC_VERSION, (idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo);
3871   if (rslt < 0) return errno;
3872 
3873   if (ThreadPriorityVerbose) {
3874     tty->print_cr("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3875   }
3876 
3877   memset(&ParmInfo, 0, sizeof(pcparms_t));
3878   ParmInfo.pc_cid = PC_CLNULL;
3879   rslt = (*priocntl_ptr)(PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3880   if (rslt < 0) return errno;
3881 
3882   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3883     myMin = rtLimits.minPrio;
3884     myMax = rtLimits.maxPrio;
3885   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3886     iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3887     myMin = iaLimits.minPrio;
3888     myMax = iaLimits.maxPrio;
3889     myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3890   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3891     tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3892     myMin = tsLimits.minPrio;
3893     myMax = tsLimits.maxPrio;
3894     myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3895   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3896     fxparms_t *fxInfo = (fxparms_t*)ParmInfo.pc_clparms;
3897     myMin = fxLimits.minPrio;
3898     myMax = fxLimits.maxPrio;
3899     myMax = MIN2(myMax, (int)fxInfo->fx_uprilim);       // clamp - restrict
3900   } else {
3901     // No clue - punt
3902     if (ThreadPriorityVerbose)
3903       tty->print_cr ("Unknown scheduling class: %s ... \n", ClassInfo.pc_clname);
3904     return EINVAL;      // no clue, punt
3905   }
3906 
3907   if (ThreadPriorityVerbose) {
3908     tty->print_cr ("Thread priority Range: [%d..%d]\n", myMin, myMax);
3909   }
3910 
3911   priocntl_enable = true;  // Enable changing priorities
3912   return 0;
3913 }
3914 
3915 #define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3916 #define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3917 #define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3918 #define FXPRI(x)        ((fxparms_t *)((x).pc_clparms))
3919 
3920 
3921 // scale_to_lwp_priority
3922 //
3923 // Convert from the libthread "thr_setprio" scale to our current
3924 // lwp scheduling class scale.
3925 //
3926 static
3927 int     scale_to_lwp_priority (int rMin, int rMax, int x)
3928 {
3929   int v;
3930 
3931   if (x == 127) return rMax;            // avoid round-down
3932     v = (((x*(rMax-rMin)))/128)+rMin;
3933   return v;
3934 }
3935 
3936 
3937 // set_lwp_class_and_priority
3938 //
3939 // Set the class and priority of the lwp.  This call should only
3940 // be made when using bound threads (T2 threads are bound by default).
3941 //
3942 int set_lwp_class_and_priority(int ThreadID, int lwpid,
3943                                int newPrio, int new_class, bool scale) {
3944   int rslt;
3945   int Actual, Expected, prv;
3946   pcparms_t ParmInfo;                   // for GET-SET
3947 #ifdef ASSERT
3948   pcparms_t ReadBack;                   // for readback
3949 #endif
3950 
3951   // Set priority via PC_GETPARMS, update, PC_SETPARMS
3952   // Query current values.
3953   // TODO: accelerate this by eliminating the PC_GETPARMS call.
3954   // Cache "pcparms_t" in global ParmCache.
3955   // TODO: elide set-to-same-value
3956 
3957   // If something went wrong on init, don't change priorities.
3958   if ( !priocntl_enable ) {
3959     if (ThreadPriorityVerbose)
3960       tty->print_cr("Trying to set priority but init failed, ignoring");
3961     return EINVAL;
3962   }
3963 
3964   // If lwp hasn't started yet, just return
3965   // the _start routine will call us again.
3966   if ( lwpid <= 0 ) {
3967     if (ThreadPriorityVerbose) {
3968       tty->print_cr ("deferring the set_lwp_class_and_priority of thread "
3969                      INTPTR_FORMAT " to %d, lwpid not set",
3970                      ThreadID, newPrio);
3971     }
3972     return 0;
3973   }
3974 
3975   if (ThreadPriorityVerbose) {
3976     tty->print_cr ("set_lwp_class_and_priority("
3977                    INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3978                    ThreadID, lwpid, newPrio);
3979   }
3980 
3981   memset(&ParmInfo, 0, sizeof(pcparms_t));
3982   ParmInfo.pc_cid = PC_CLNULL;
3983   rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3984   if (rslt < 0) return errno;
3985 
3986   int cur_class = ParmInfo.pc_cid;
3987   ParmInfo.pc_cid = (id_t)new_class;
3988 
3989   if (new_class == rtLimits.schedPolicy) {
3990     rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3991     rtInfo->rt_pri     = scale ? scale_to_lwp_priority(rtLimits.minPrio,
3992                                                        rtLimits.maxPrio, newPrio)
3993                                : newPrio;
3994     rtInfo->rt_tqsecs  = RT_NOCHANGE;
3995     rtInfo->rt_tqnsecs = RT_NOCHANGE;
3996     if (ThreadPriorityVerbose) {
3997       tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3998     }
3999   } else if (new_class == iaLimits.schedPolicy) {
4000     iaparms_t* iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
4001     int maxClamped     = MIN2(iaLimits.maxPrio,
4002                               cur_class == new_class
4003                                 ? (int)iaInfo->ia_uprilim : iaLimits.maxPrio);
4004     iaInfo->ia_upri    = scale ? scale_to_lwp_priority(iaLimits.minPrio,
4005                                                        maxClamped, newPrio)
4006                                : newPrio;
4007     iaInfo->ia_uprilim = cur_class == new_class
4008                            ? IA_NOCHANGE : (pri_t)iaLimits.maxPrio;
4009     iaInfo->ia_mode    = IA_NOCHANGE;
4010     iaInfo->ia_nice    = cur_class == new_class ? IA_NOCHANGE : NZERO;
4011     if (ThreadPriorityVerbose) {
4012       tty->print_cr("IA: [%d...%d] %d->%d\n",
4013                     iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
4014     }
4015   } else if (new_class == tsLimits.schedPolicy) {
4016     tsparms_t* tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
4017     int maxClamped     = MIN2(tsLimits.maxPrio,
4018                               cur_class == new_class
4019                                 ? (int)tsInfo->ts_uprilim : tsLimits.maxPrio);
4020     tsInfo->ts_upri    = scale ? scale_to_lwp_priority(tsLimits.minPrio,
4021                                                        maxClamped, newPrio)
4022                                : newPrio;
4023     tsInfo->ts_uprilim = cur_class == new_class
4024                            ? TS_NOCHANGE : (pri_t)tsLimits.maxPrio;
4025     if (ThreadPriorityVerbose) {
4026       tty->print_cr("TS: [%d...%d] %d->%d\n",
4027                     tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
4028     }
4029   } else if (new_class == fxLimits.schedPolicy) {
4030     fxparms_t* fxInfo  = (fxparms_t*)ParmInfo.pc_clparms;
4031     int maxClamped     = MIN2(fxLimits.maxPrio,
4032                               cur_class == new_class
4033                                 ? (int)fxInfo->fx_uprilim : fxLimits.maxPrio);
4034     fxInfo->fx_upri    = scale ? scale_to_lwp_priority(fxLimits.minPrio,
4035                                                        maxClamped, newPrio)
4036                                : newPrio;
4037     fxInfo->fx_uprilim = cur_class == new_class
4038                            ? FX_NOCHANGE : (pri_t)fxLimits.maxPrio;
4039     fxInfo->fx_tqsecs  = FX_NOCHANGE;
4040     fxInfo->fx_tqnsecs = FX_NOCHANGE;
4041     if (ThreadPriorityVerbose) {
4042       tty->print_cr("FX: [%d...%d] %d->%d\n",
4043                     fxLimits.minPrio, maxClamped, newPrio, fxInfo->fx_upri);
4044     }
4045   } else {
4046     if (ThreadPriorityVerbose) {
4047       tty->print_cr("Unknown new scheduling class %d\n", new_class);
4048     }
4049     return EINVAL;    // no clue, punt
4050   }
4051 
4052   rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
4053   if (ThreadPriorityVerbose && rslt) {
4054     tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
4055   }
4056   if (rslt < 0) return errno;
4057 
4058 #ifdef ASSERT
4059   // Sanity check: read back what we just attempted to set.
4060   // In theory it could have changed in the interim ...
4061   //
4062   // The priocntl system call is tricky.
4063   // Sometimes it'll validate the priority value argument and
4064   // return EINVAL if unhappy.  At other times it fails silently.
4065   // Readbacks are prudent.
4066 
4067   if (!ReadBackValidate) return 0;
4068 
4069   memset(&ReadBack, 0, sizeof(pcparms_t));
4070   ReadBack.pc_cid = PC_CLNULL;
4071   rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
4072   assert(rslt >= 0, "priocntl failed");
4073   Actual = Expected = 0xBAD;
4074   assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
4075   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
4076     Actual   = RTPRI(ReadBack)->rt_pri;
4077     Expected = RTPRI(ParmInfo)->rt_pri;
4078   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
4079     Actual   = IAPRI(ReadBack)->ia_upri;
4080     Expected = IAPRI(ParmInfo)->ia_upri;
4081   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
4082     Actual   = TSPRI(ReadBack)->ts_upri;
4083     Expected = TSPRI(ParmInfo)->ts_upri;
4084   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
4085     Actual   = FXPRI(ReadBack)->fx_upri;
4086     Expected = FXPRI(ParmInfo)->fx_upri;
4087   } else {
4088     if (ThreadPriorityVerbose) {
4089       tty->print_cr("set_lwp_class_and_priority: unexpected class in readback: %d\n",
4090                     ParmInfo.pc_cid);
4091     }
4092   }
4093 
4094   if (Actual != Expected) {
4095     if (ThreadPriorityVerbose) {
4096       tty->print_cr ("set_lwp_class_and_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
4097                      lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
4098     }
4099   }
4100 #endif
4101 
4102   return 0;
4103 }
4104 
4105 // Solaris only gives access to 128 real priorities at a time,
4106 // so we expand Java's ten to fill this range.  This would be better
4107 // if we dynamically adjusted relative priorities.
4108 //
4109 // The ThreadPriorityPolicy option allows us to select 2 different
4110 // priority scales.
4111 //
4112 // ThreadPriorityPolicy=0
4113 // Since the Solaris' default priority is MaximumPriority, we do not
4114 // set a priority lower than Max unless a priority lower than
4115 // NormPriority is requested.
4116 //
4117 // ThreadPriorityPolicy=1
4118 // This mode causes the priority table to get filled with
4119 // linear values.  NormPriority get's mapped to 50% of the
4120 // Maximum priority an so on.  This will cause VM threads
4121 // to get unfair treatment against other Solaris processes
4122 // which do not explicitly alter their thread priorities.
4123 //
4124 
4125 int os::java_to_os_priority[CriticalPriority + 1] = {
4126   -99999,         // 0 Entry should never be used
4127 
4128   0,              // 1 MinPriority
4129   32,             // 2
4130   64,             // 3
4131 
4132   96,             // 4
4133   127,            // 5 NormPriority
4134   127,            // 6
4135 
4136   127,            // 7
4137   127,            // 8
4138   127,            // 9 NearMaxPriority
4139 
4140   127,            // 10 MaxPriority
4141 
4142   -criticalPrio   // 11 CriticalPriority
4143 };
4144 
4145 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4146   OSThread* osthread = thread->osthread();
4147 
4148   // Save requested priority in case the thread hasn't been started
4149   osthread->set_native_priority(newpri);
4150 
4151   // Check for critical priority request
4152   bool fxcritical = false;
4153   if (newpri == -criticalPrio) {
4154     fxcritical = true;
4155     newpri = criticalPrio;
4156   }
4157 
4158   assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
4159   if (!UseThreadPriorities) return OS_OK;
4160 
4161   int status = 0;
4162 
4163   if (!fxcritical) {
4164     // Use thr_setprio only if we have a priority that thr_setprio understands
4165     status = thr_setprio(thread->osthread()->thread_id(), newpri);
4166   }
4167 
4168   if (os::Solaris::T2_libthread() ||
4169       (UseBoundThreads && osthread->is_vm_created())) {
4170     int lwp_status =
4171       set_lwp_class_and_priority(osthread->thread_id(),
4172                                  osthread->lwp_id(),
4173                                  newpri,
4174                                  fxcritical ? fxLimits.schedPolicy : myClass,
4175                                  !fxcritical);
4176     if (lwp_status != 0 && fxcritical) {
4177       // Try again, this time without changing the scheduling class
4178       newpri = java_MaxPriority_to_os_priority;
4179       lwp_status = set_lwp_class_and_priority(osthread->thread_id(),
4180                                               osthread->lwp_id(),
4181                                               newpri, myClass, false);
4182     }
4183     status |= lwp_status;
4184   }
4185   return (status == 0) ? OS_OK : OS_ERR;
4186 }
4187 
4188 
4189 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4190   int p;
4191   if ( !UseThreadPriorities ) {
4192     *priority_ptr = NormalPriority;
4193     return OS_OK;
4194   }
4195   int status = thr_getprio(thread->osthread()->thread_id(), &p);
4196   if (status != 0) {
4197     return OS_ERR;
4198   }
4199   *priority_ptr = p;
4200   return OS_OK;
4201 }
4202 
4203 
4204 // Hint to the underlying OS that a task switch would not be good.
4205 // Void return because it's a hint and can fail.
4206 void os::hint_no_preempt() {
4207   schedctl_start(schedctl_init());
4208 }
4209 
4210 void os::interrupt(Thread* thread) {
4211   assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
4212 
4213   OSThread* osthread = thread->osthread();
4214 
4215   int isInterrupted = osthread->interrupted();
4216   if (!isInterrupted) {
4217       osthread->set_interrupted(true);
4218       OrderAccess::fence();
4219       // os::sleep() is implemented with either poll (NULL,0,timeout) or
4220       // by parking on _SleepEvent.  If the former, thr_kill will unwedge
4221       // the sleeper by SIGINTR, otherwise the unpark() will wake the sleeper.
4222       ParkEvent * const slp = thread->_SleepEvent ;
4223       if (slp != NULL) slp->unpark() ;
4224   }
4225 
4226   // For JSR166:  unpark after setting status but before thr_kill -dl
4227   if (thread->is_Java_thread()) {
4228     ((JavaThread*)thread)->parker()->unpark();
4229   }
4230 
4231   // Handle interruptible wait() ...
4232   ParkEvent * const ev = thread->_ParkEvent ;
4233   if (ev != NULL) ev->unpark() ;
4234 
4235   // When events are used everywhere for os::sleep, then this thr_kill
4236   // will only be needed if UseVMInterruptibleIO is true.
4237 
4238   if (!isInterrupted) {
4239     int status = thr_kill(osthread->thread_id(), os::Solaris::SIGinterrupt());
4240     assert_status(status == 0, status, "thr_kill");
4241 
4242     // Bump thread interruption counter
4243     RuntimeService::record_thread_interrupt_signaled_count();
4244   }
4245 }
4246 
4247 
4248 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4249   assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
4250 
4251   OSThread* osthread = thread->osthread();
4252 
4253   bool res = osthread->interrupted();
4254 
4255   // NOTE that since there is no "lock" around these two operations,
4256   // there is the possibility that the interrupted flag will be
4257   // "false" but that the interrupt event will be set. This is
4258   // intentional. The effect of this is that Object.wait() will appear
4259   // to have a spurious wakeup, which is not harmful, and the
4260   // possibility is so rare that it is not worth the added complexity
4261   // to add yet another lock. It has also been recommended not to put
4262   // the interrupted flag into the os::Solaris::Event structure,
4263   // because it hides the issue.
4264   if (res && clear_interrupted) {
4265     osthread->set_interrupted(false);
4266   }
4267   return res;
4268 }
4269 
4270 
4271 void os::print_statistics() {
4272 }
4273 
4274 int os::message_box(const char* title, const char* message) {
4275   int i;
4276   fdStream err(defaultStream::error_fd());
4277   for (i = 0; i < 78; i++) err.print_raw("=");
4278   err.cr();
4279   err.print_raw_cr(title);
4280   for (i = 0; i < 78; i++) err.print_raw("-");
4281   err.cr();
4282   err.print_raw_cr(message);
4283   for (i = 0; i < 78; i++) err.print_raw("=");
4284   err.cr();
4285 
4286   char buf[16];
4287   // Prevent process from exiting upon "read error" without consuming all CPU
4288   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4289 
4290   return buf[0] == 'y' || buf[0] == 'Y';
4291 }
4292 
4293 // A lightweight implementation that does not suspend the target thread and
4294 // thus returns only a hint. Used for profiling only!
4295 ExtendedPC os::get_thread_pc(Thread* thread) {
4296   // Make sure that it is called by the watcher and the Threads lock is owned.
4297   assert(Thread::current()->is_Watcher_thread(), "Must be watcher and own Threads_lock");
4298   // For now, is only used to profile the VM Thread
4299   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4300   ExtendedPC epc;
4301 
4302   GetThreadPC_Callback  cb(ProfileVM_lock);
4303   OSThread *osthread = thread->osthread();
4304   const int time_to_wait = 400; // 400ms wait for initial response
4305   int status = cb.interrupt(thread, time_to_wait);
4306 
4307   if (cb.is_done() ) {
4308     epc = cb.addr();
4309   } else {
4310     DEBUG_ONLY(tty->print_cr("Failed to get pc for thread: %d got %d status",
4311                               osthread->thread_id(), status););
4312     // epc is already NULL
4313   }
4314   return epc;
4315 }
4316 
4317 
4318 // This does not do anything on Solaris. This is basically a hook for being
4319 // able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
4320 void os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, JavaCallArguments* args, Thread* thread) {
4321   f(value, method, args, thread);
4322 }
4323 
4324 // This routine may be used by user applications as a "hook" to catch signals.
4325 // The user-defined signal handler must pass unrecognized signals to this
4326 // routine, and if it returns true (non-zero), then the signal handler must
4327 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4328 // routine will never retun false (zero), but instead will execute a VM panic
4329 // routine kill the process.
4330 //
4331 // If this routine returns false, it is OK to call it again.  This allows
4332 // the user-defined signal handler to perform checks either before or after
4333 // the VM performs its own checks.  Naturally, the user code would be making
4334 // a serious error if it tried to handle an exception (such as a null check
4335 // or breakpoint) that the VM was generating for its own correct operation.
4336 //
4337 // This routine may recognize any of the following kinds of signals:
4338 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
4339 // os::Solaris::SIGasync
4340 // It should be consulted by handlers for any of those signals.
4341 // It explicitly does not recognize os::Solaris::SIGinterrupt
4342 //
4343 // The caller of this routine must pass in the three arguments supplied
4344 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4345 // field of the structure passed to sigaction().  This routine assumes that
4346 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4347 //
4348 // Note that the VM will print warnings if it detects conflicting signal
4349 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4350 //
4351 extern "C" JNIEXPORT int
4352 JVM_handle_solaris_signal(int signo, siginfo_t* siginfo, void* ucontext,
4353                           int abort_if_unrecognized);
4354 
4355 
4356 void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
4357   JVM_handle_solaris_signal(sig, info, ucVoid, true);
4358 }
4359 
4360 /* Do not delete - if guarantee is ever removed,  a signal handler (even empty)
4361    is needed to provoke threads blocked on IO to return an EINTR
4362    Note: this explicitly does NOT call JVM_handle_solaris_signal and
4363    does NOT participate in signal chaining due to requirement for
4364    NOT setting SA_RESTART to make EINTR work. */
4365 extern "C" void sigINTRHandler(int sig, siginfo_t* info, void* ucVoid) {
4366    if (UseSignalChaining) {
4367       struct sigaction *actp = os::Solaris::get_chained_signal_action(sig);
4368       if (actp && actp->sa_handler) {
4369         vm_exit_during_initialization("Signal chaining detected for VM interrupt signal, try -XX:+UseAltSigs");
4370       }
4371    }
4372 }
4373 
4374 // This boolean allows users to forward their own non-matching signals
4375 // to JVM_handle_solaris_signal, harmlessly.
4376 bool os::Solaris::signal_handlers_are_installed = false;
4377 
4378 // For signal-chaining
4379 bool os::Solaris::libjsig_is_loaded = false;
4380 typedef struct sigaction *(*get_signal_t)(int);
4381 get_signal_t os::Solaris::get_signal_action = NULL;
4382 
4383 struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
4384   struct sigaction *actp = NULL;
4385 
4386   if ((libjsig_is_loaded)  && (sig <= Maxlibjsigsigs)) {
4387     // Retrieve the old signal handler from libjsig
4388     actp = (*get_signal_action)(sig);
4389   }
4390   if (actp == NULL) {
4391     // Retrieve the preinstalled signal handler from jvm
4392     actp = get_preinstalled_handler(sig);
4393   }
4394 
4395   return actp;
4396 }
4397 
4398 static bool call_chained_handler(struct sigaction *actp, int sig,
4399                                  siginfo_t *siginfo, void *context) {
4400   // Call the old signal handler
4401   if (actp->sa_handler == SIG_DFL) {
4402     // It's more reasonable to let jvm treat it as an unexpected exception
4403     // instead of taking the default action.
4404     return false;
4405   } else if (actp->sa_handler != SIG_IGN) {
4406     if ((actp->sa_flags & SA_NODEFER) == 0) {
4407       // automaticlly block the signal
4408       sigaddset(&(actp->sa_mask), sig);
4409     }
4410 
4411     sa_handler_t hand;
4412     sa_sigaction_t sa;
4413     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4414     // retrieve the chained handler
4415     if (siginfo_flag_set) {
4416       sa = actp->sa_sigaction;
4417     } else {
4418       hand = actp->sa_handler;
4419     }
4420 
4421     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4422       actp->sa_handler = SIG_DFL;
4423     }
4424 
4425     // try to honor the signal mask
4426     sigset_t oset;
4427     thr_sigsetmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4428 
4429     // call into the chained handler
4430     if (siginfo_flag_set) {
4431       (*sa)(sig, siginfo, context);
4432     } else {
4433       (*hand)(sig);
4434     }
4435 
4436     // restore the signal mask
4437     thr_sigsetmask(SIG_SETMASK, &oset, 0);
4438   }
4439   // Tell jvm's signal handler the signal is taken care of.
4440   return true;
4441 }
4442 
4443 bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4444   bool chained = false;
4445   // signal-chaining
4446   if (UseSignalChaining) {
4447     struct sigaction *actp = get_chained_signal_action(sig);
4448     if (actp != NULL) {
4449       chained = call_chained_handler(actp, sig, siginfo, context);
4450     }
4451   }
4452   return chained;
4453 }
4454 
4455 struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
4456   assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4457   if (preinstalled_sigs[sig] != 0) {
4458     return &chainedsigactions[sig];
4459   }
4460   return NULL;
4461 }
4462 
4463 void os::Solaris::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4464 
4465   assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
4466   assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4467   chainedsigactions[sig] = oldAct;
4468   preinstalled_sigs[sig] = 1;
4469 }
4470 
4471 void os::Solaris::set_signal_handler(int sig, bool set_installed, bool oktochain) {
4472   // Check for overwrite.
4473   struct sigaction oldAct;
4474   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4475   void* oldhand = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4476                                       : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4477   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4478       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4479       oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
4480     if (AllowUserSignalHandlers || !set_installed) {
4481       // Do not overwrite; user takes responsibility to forward to us.
4482       return;
4483     } else if (UseSignalChaining) {
4484       if (oktochain) {
4485         // save the old handler in jvm
4486         save_preinstalled_handler(sig, oldAct);
4487       } else {
4488         vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal, try -XX:+UseAltSigs.");
4489       }
4490       // libjsig also interposes the sigaction() call below and saves the
4491       // old sigaction on it own.
4492     } else {
4493       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4494                     "%#lx for signal %d.", (long)oldhand, sig));
4495     }
4496   }
4497 
4498   struct sigaction sigAct;
4499   sigfillset(&(sigAct.sa_mask));
4500   sigAct.sa_handler = SIG_DFL;
4501 
4502   sigAct.sa_sigaction = signalHandler;
4503   // Handle SIGSEGV on alternate signal stack if
4504   // not using stack banging
4505   if (!UseStackBanging && sig == SIGSEGV) {
4506     sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
4507   // Interruptible i/o requires SA_RESTART cleared so EINTR
4508   // is returned instead of restarting system calls
4509   } else if (sig == os::Solaris::SIGinterrupt()) {
4510     sigemptyset(&sigAct.sa_mask);
4511     sigAct.sa_handler = NULL;
4512     sigAct.sa_flags = SA_SIGINFO;
4513     sigAct.sa_sigaction = sigINTRHandler;
4514   } else {
4515     sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
4516   }
4517   os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
4518 
4519   sigaction(sig, &sigAct, &oldAct);
4520 
4521   void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4522                                        : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4523   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4524 }
4525 
4526 
4527 #define DO_SIGNAL_CHECK(sig) \
4528   if (!sigismember(&check_signal_done, sig)) \
4529     os::Solaris::check_signal_handler(sig)
4530 
4531 // This method is a periodic task to check for misbehaving JNI applications
4532 // under CheckJNI, we can add any periodic checks here
4533 
4534 void os::run_periodic_checks() {
4535   // A big source of grief is hijacking virt. addr 0x0 on Solaris,
4536   // thereby preventing a NULL checks.
4537   if(!check_addr0_done) check_addr0_done = check_addr0(tty);
4538 
4539   if (check_signals == false) return;
4540 
4541   // SEGV and BUS if overridden could potentially prevent
4542   // generation of hs*.log in the event of a crash, debugging
4543   // such a case can be very challenging, so we absolutely
4544   // check for the following for a good measure:
4545   DO_SIGNAL_CHECK(SIGSEGV);
4546   DO_SIGNAL_CHECK(SIGILL);
4547   DO_SIGNAL_CHECK(SIGFPE);
4548   DO_SIGNAL_CHECK(SIGBUS);
4549   DO_SIGNAL_CHECK(SIGPIPE);
4550   DO_SIGNAL_CHECK(SIGXFSZ);
4551 
4552   // ReduceSignalUsage allows the user to override these handlers
4553   // see comments at the very top and jvm_solaris.h
4554   if (!ReduceSignalUsage) {
4555     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4556     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4557     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4558     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4559   }
4560 
4561   // See comments above for using JVM1/JVM2 and UseAltSigs
4562   DO_SIGNAL_CHECK(os::Solaris::SIGinterrupt());
4563   DO_SIGNAL_CHECK(os::Solaris::SIGasync());
4564 
4565 }
4566 
4567 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4568 
4569 static os_sigaction_t os_sigaction = NULL;
4570 
4571 void os::Solaris::check_signal_handler(int sig) {
4572   char buf[O_BUFLEN];
4573   address jvmHandler = NULL;
4574 
4575   struct sigaction act;
4576   if (os_sigaction == NULL) {
4577     // only trust the default sigaction, in case it has been interposed
4578     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4579     if (os_sigaction == NULL) return;
4580   }
4581 
4582   os_sigaction(sig, (struct sigaction*)NULL, &act);
4583 
4584   address thisHandler = (act.sa_flags & SA_SIGINFO)
4585     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4586     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4587 
4588 
4589   switch(sig) {
4590     case SIGSEGV:
4591     case SIGBUS:
4592     case SIGFPE:
4593     case SIGPIPE:
4594     case SIGXFSZ:
4595     case SIGILL:
4596       jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4597       break;
4598 
4599     case SHUTDOWN1_SIGNAL:
4600     case SHUTDOWN2_SIGNAL:
4601     case SHUTDOWN3_SIGNAL:
4602     case BREAK_SIGNAL:
4603       jvmHandler = (address)user_handler();
4604       break;
4605 
4606     default:
4607       int intrsig = os::Solaris::SIGinterrupt();
4608       int asynsig = os::Solaris::SIGasync();
4609 
4610       if (sig == intrsig) {
4611         jvmHandler = CAST_FROM_FN_PTR(address, sigINTRHandler);
4612       } else if (sig == asynsig) {
4613         jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4614       } else {
4615         return;
4616       }
4617       break;
4618   }
4619 
4620 
4621   if (thisHandler != jvmHandler) {
4622     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4623     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4624     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4625     // No need to check this sig any longer
4626     sigaddset(&check_signal_done, sig);
4627   } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
4628     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4629     tty->print("expected:" PTR32_FORMAT, os::Solaris::get_our_sigflags(sig));
4630     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4631     // No need to check this sig any longer
4632     sigaddset(&check_signal_done, sig);
4633   }
4634 
4635   // Print all the signal handler state
4636   if (sigismember(&check_signal_done, sig)) {
4637     print_signal_handlers(tty, buf, O_BUFLEN);
4638   }
4639 
4640 }
4641 
4642 void os::Solaris::install_signal_handlers() {
4643   bool libjsigdone = false;
4644   signal_handlers_are_installed = true;
4645 
4646   // signal-chaining
4647   typedef void (*signal_setting_t)();
4648   signal_setting_t begin_signal_setting = NULL;
4649   signal_setting_t end_signal_setting = NULL;
4650   begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4651                                         dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4652   if (begin_signal_setting != NULL) {
4653     end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4654                                         dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4655     get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4656                                        dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4657     get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
4658                                          dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
4659     libjsig_is_loaded = true;
4660     if (os::Solaris::get_libjsig_version != NULL) {
4661       libjsigversion =  (*os::Solaris::get_libjsig_version)();
4662     }
4663     assert(UseSignalChaining, "should enable signal-chaining");
4664   }
4665   if (libjsig_is_loaded) {
4666     // Tell libjsig jvm is setting signal handlers
4667     (*begin_signal_setting)();
4668   }
4669 
4670   set_signal_handler(SIGSEGV, true, true);
4671   set_signal_handler(SIGPIPE, true, true);
4672   set_signal_handler(SIGXFSZ, true, true);
4673   set_signal_handler(SIGBUS, true, true);
4674   set_signal_handler(SIGILL, true, true);
4675   set_signal_handler(SIGFPE, true, true);
4676 
4677 
4678   if (os::Solaris::SIGinterrupt() > OLDMAXSIGNUM || os::Solaris::SIGasync() > OLDMAXSIGNUM) {
4679 
4680     // Pre-1.4.1 Libjsig limited to signal chaining signals <= 32 so
4681     // can not register overridable signals which might be > 32
4682     if (libjsig_is_loaded && libjsigversion <= JSIG_VERSION_1_4_1) {
4683     // Tell libjsig jvm has finished setting signal handlers
4684       (*end_signal_setting)();
4685       libjsigdone = true;
4686     }
4687   }
4688 
4689   // Never ok to chain our SIGinterrupt
4690   set_signal_handler(os::Solaris::SIGinterrupt(), true, false);
4691   set_signal_handler(os::Solaris::SIGasync(), true, true);
4692 
4693   if (libjsig_is_loaded && !libjsigdone) {
4694     // Tell libjsig jvm finishes setting signal handlers
4695     (*end_signal_setting)();
4696   }
4697 
4698   // We don't activate signal checker if libjsig is in place, we trust ourselves
4699   // and if UserSignalHandler is installed all bets are off.
4700   // Log that signal checking is off only if -verbose:jni is specified.
4701   if (CheckJNICalls) {
4702     if (libjsig_is_loaded) {
4703       if (PrintJNIResolving) {
4704         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4705       }
4706       check_signals = false;
4707     }
4708     if (AllowUserSignalHandlers) {
4709       if (PrintJNIResolving) {
4710         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4711       }
4712       check_signals = false;
4713     }
4714   }
4715 }
4716 
4717 
4718 void report_error(const char* file_name, int line_no, const char* title, const char* format, ...);
4719 
4720 const char * signames[] = {
4721   "SIG0",
4722   "SIGHUP", "SIGINT", "SIGQUIT", "SIGILL", "SIGTRAP",
4723   "SIGABRT", "SIGEMT", "SIGFPE", "SIGKILL", "SIGBUS",
4724   "SIGSEGV", "SIGSYS", "SIGPIPE", "SIGALRM", "SIGTERM",
4725   "SIGUSR1", "SIGUSR2", "SIGCLD", "SIGPWR", "SIGWINCH",
4726   "SIGURG", "SIGPOLL", "SIGSTOP", "SIGTSTP", "SIGCONT",
4727   "SIGTTIN", "SIGTTOU", "SIGVTALRM", "SIGPROF", "SIGXCPU",
4728   "SIGXFSZ", "SIGWAITING", "SIGLWP", "SIGFREEZE", "SIGTHAW",
4729   "SIGCANCEL", "SIGLOST"
4730 };
4731 
4732 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4733   if (0 < exception_code && exception_code <= SIGRTMAX) {
4734     // signal
4735     if (exception_code < sizeof(signames)/sizeof(const char*)) {
4736        jio_snprintf(buf, size, "%s", signames[exception_code]);
4737     } else {
4738        jio_snprintf(buf, size, "SIG%d", exception_code);
4739     }
4740     return buf;
4741   } else {
4742     return NULL;
4743   }
4744 }
4745 
4746 // (Static) wrappers for the new libthread API
4747 int_fnP_thread_t_iP_uP_stack_tP_gregset_t os::Solaris::_thr_getstate;
4748 int_fnP_thread_t_i_gregset_t os::Solaris::_thr_setstate;
4749 int_fnP_thread_t_i os::Solaris::_thr_setmutator;
4750 int_fnP_thread_t os::Solaris::_thr_suspend_mutator;
4751 int_fnP_thread_t os::Solaris::_thr_continue_mutator;
4752 
4753 // (Static) wrapper for getisax(2) call.
4754 os::Solaris::getisax_func_t os::Solaris::_getisax = 0;
4755 
4756 // (Static) wrappers for the liblgrp API
4757 os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4758 os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4759 os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4760 os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4761 os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4762 os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4763 os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4764 os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4765 os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4766 
4767 // (Static) wrapper for meminfo() call.
4768 os::Solaris::meminfo_func_t os::Solaris::_meminfo = 0;
4769 
4770 static address resolve_symbol_lazy(const char* name) {
4771   address addr = (address) dlsym(RTLD_DEFAULT, name);
4772   if(addr == NULL) {
4773     // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4774     addr = (address) dlsym(RTLD_NEXT, name);
4775   }
4776   return addr;
4777 }
4778 
4779 static address resolve_symbol(const char* name) {
4780   address addr = resolve_symbol_lazy(name);
4781   if(addr == NULL) {
4782     fatal(dlerror());
4783   }
4784   return addr;
4785 }
4786 
4787 
4788 
4789 // isT2_libthread()
4790 //
4791 // Routine to determine if we are currently using the new T2 libthread.
4792 //
4793 // We determine if we are using T2 by reading /proc/self/lstatus and
4794 // looking for a thread with the ASLWP bit set.  If we find this status
4795 // bit set, we must assume that we are NOT using T2.  The T2 team
4796 // has approved this algorithm.
4797 //
4798 // We need to determine if we are running with the new T2 libthread
4799 // since setting native thread priorities is handled differently
4800 // when using this library.  All threads created using T2 are bound
4801 // threads. Calling thr_setprio is meaningless in this case.
4802 //
4803 bool isT2_libthread() {
4804   static prheader_t * lwpArray = NULL;
4805   static int lwpSize = 0;
4806   static int lwpFile = -1;
4807   lwpstatus_t * that;
4808   char lwpName [128];
4809   bool isT2 = false;
4810 
4811 #define ADR(x)  ((uintptr_t)(x))
4812 #define LWPINDEX(ary,ix)   ((lwpstatus_t *)(((ary)->pr_entsize * (ix)) + (ADR((ary) + 1))))
4813 
4814   lwpFile = ::open("/proc/self/lstatus", O_RDONLY, 0);
4815   if (lwpFile < 0) {
4816       if (ThreadPriorityVerbose) warning ("Couldn't open /proc/self/lstatus\n");
4817       return false;
4818   }
4819   lwpSize = 16*1024;
4820   for (;;) {
4821     ::lseek64 (lwpFile, 0, SEEK_SET);
4822     lwpArray = (prheader_t *)NEW_C_HEAP_ARRAY(char, lwpSize);
4823     if (::read(lwpFile, lwpArray, lwpSize) < 0) {
4824       if (ThreadPriorityVerbose) warning("Error reading /proc/self/lstatus\n");
4825       break;
4826     }
4827     if ((lwpArray->pr_nent * lwpArray->pr_entsize) <= lwpSize) {
4828        // We got a good snapshot - now iterate over the list.
4829       int aslwpcount = 0;
4830       for (int i = 0; i < lwpArray->pr_nent; i++ ) {
4831         that = LWPINDEX(lwpArray,i);
4832         if (that->pr_flags & PR_ASLWP) {
4833           aslwpcount++;
4834         }
4835       }
4836       if (aslwpcount == 0) isT2 = true;
4837       break;
4838     }
4839     lwpSize = lwpArray->pr_nent * lwpArray->pr_entsize;
4840     FREE_C_HEAP_ARRAY(char, lwpArray);  // retry.
4841   }
4842 
4843   FREE_C_HEAP_ARRAY(char, lwpArray);
4844   ::close (lwpFile);
4845   if (ThreadPriorityVerbose) {
4846     if (isT2) tty->print_cr("We are running with a T2 libthread\n");
4847     else tty->print_cr("We are not running with a T2 libthread\n");
4848   }
4849   return isT2;
4850 }
4851 
4852 
4853 void os::Solaris::libthread_init() {
4854   address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4855 
4856   // Determine if we are running with the new T2 libthread
4857   os::Solaris::set_T2_libthread(isT2_libthread());
4858 
4859   lwp_priocntl_init();
4860 
4861   // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4862   if(func == NULL) {
4863     func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4864     // Guarantee that this VM is running on an new enough OS (5.6 or
4865     // later) that it will have a new enough libthread.so.
4866     guarantee(func != NULL, "libthread.so is too old.");
4867   }
4868 
4869   // Initialize the new libthread getstate API wrappers
4870   func = resolve_symbol("thr_getstate");
4871   os::Solaris::set_thr_getstate(CAST_TO_FN_PTR(int_fnP_thread_t_iP_uP_stack_tP_gregset_t, func));
4872 
4873   func = resolve_symbol("thr_setstate");
4874   os::Solaris::set_thr_setstate(CAST_TO_FN_PTR(int_fnP_thread_t_i_gregset_t, func));
4875 
4876   func = resolve_symbol("thr_setmutator");
4877   os::Solaris::set_thr_setmutator(CAST_TO_FN_PTR(int_fnP_thread_t_i, func));
4878 
4879   func = resolve_symbol("thr_suspend_mutator");
4880   os::Solaris::set_thr_suspend_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4881 
4882   func = resolve_symbol("thr_continue_mutator");
4883   os::Solaris::set_thr_continue_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4884 
4885   int size;
4886   void (*handler_info_func)(address *, int *);
4887   handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4888   handler_info_func(&handler_start, &size);
4889   handler_end = handler_start + size;
4890 }
4891 
4892 
4893 int_fnP_mutex_tP os::Solaris::_mutex_lock;
4894 int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4895 int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4896 int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4897 int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4898 int os::Solaris::_mutex_scope = USYNC_THREAD;
4899 
4900 int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4901 int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4902 int_fnP_cond_tP os::Solaris::_cond_signal;
4903 int_fnP_cond_tP os::Solaris::_cond_broadcast;
4904 int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4905 int_fnP_cond_tP os::Solaris::_cond_destroy;
4906 int os::Solaris::_cond_scope = USYNC_THREAD;
4907 
4908 void os::Solaris::synchronization_init() {
4909   if(UseLWPSynchronization) {
4910     os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4911     os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4912     os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4913     os::Solaris::set_mutex_init(lwp_mutex_init);
4914     os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4915     os::Solaris::set_mutex_scope(USYNC_THREAD);
4916 
4917     os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4918     os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4919     os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4920     os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4921     os::Solaris::set_cond_init(lwp_cond_init);
4922     os::Solaris::set_cond_destroy(lwp_cond_destroy);
4923     os::Solaris::set_cond_scope(USYNC_THREAD);
4924   }
4925   else {
4926     os::Solaris::set_mutex_scope(USYNC_THREAD);
4927     os::Solaris::set_cond_scope(USYNC_THREAD);
4928 
4929     if(UsePthreads) {
4930       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4931       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4932       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4933       os::Solaris::set_mutex_init(pthread_mutex_default_init);
4934       os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4935 
4936       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4937       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4938       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4939       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4940       os::Solaris::set_cond_init(pthread_cond_default_init);
4941       os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4942     }
4943     else {
4944       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4945       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4946       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4947       os::Solaris::set_mutex_init(::mutex_init);
4948       os::Solaris::set_mutex_destroy(::mutex_destroy);
4949 
4950       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4951       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4952       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4953       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4954       os::Solaris::set_cond_init(::cond_init);
4955       os::Solaris::set_cond_destroy(::cond_destroy);
4956     }
4957   }
4958 }
4959 
4960 bool os::Solaris::liblgrp_init() {
4961   void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4962   if (handle != NULL) {
4963     os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4964     os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4965     os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4966     os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4967     os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4968     os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4969     os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4970     os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4971                                        dlsym(handle, "lgrp_cookie_stale")));
4972 
4973     lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4974     set_lgrp_cookie(c);
4975     return true;
4976   }
4977   return false;
4978 }
4979 
4980 void os::Solaris::misc_sym_init() {
4981   address func;
4982 
4983   // getisax
4984   func = resolve_symbol_lazy("getisax");
4985   if (func != NULL) {
4986     os::Solaris::_getisax = CAST_TO_FN_PTR(getisax_func_t, func);
4987   }
4988 
4989   // meminfo
4990   func = resolve_symbol_lazy("meminfo");
4991   if (func != NULL) {
4992     os::Solaris::set_meminfo(CAST_TO_FN_PTR(meminfo_func_t, func));
4993   }
4994 }
4995 
4996 uint_t os::Solaris::getisax(uint32_t* array, uint_t n) {
4997   assert(_getisax != NULL, "_getisax not set");
4998   return _getisax(array, n);
4999 }
5000 
5001 // Symbol doesn't exist in Solaris 8 pset.h
5002 #ifndef PS_MYID
5003 #define PS_MYID -3
5004 #endif
5005 
5006 // int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
5007 typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
5008 static pset_getloadavg_type pset_getloadavg_ptr = NULL;
5009 
5010 void init_pset_getloadavg_ptr(void) {
5011   pset_getloadavg_ptr =
5012     (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
5013   if (PrintMiscellaneous && Verbose && pset_getloadavg_ptr == NULL) {
5014     warning("pset_getloadavg function not found");
5015   }
5016 }
5017 
5018 int os::Solaris::_dev_zero_fd = -1;
5019 
5020 // this is called _before_ the global arguments have been parsed
5021 void os::init(void) {
5022   _initial_pid = getpid();
5023 
5024   max_hrtime = first_hrtime = gethrtime();
5025 
5026   init_random(1234567);
5027 
5028   page_size = sysconf(_SC_PAGESIZE);
5029   if (page_size == -1)
5030     fatal(err_msg("os_solaris.cpp: os::init: sysconf failed (%s)",
5031                   strerror(errno)));
5032   init_page_sizes((size_t) page_size);
5033 
5034   Solaris::initialize_system_info();
5035 
5036   // Initialize misc. symbols as soon as possible, so we can use them
5037   // if we need them.
5038   Solaris::misc_sym_init();
5039 
5040   int fd = ::open("/dev/zero", O_RDWR);
5041   if (fd < 0) {
5042     fatal(err_msg("os::init: cannot open /dev/zero (%s)", strerror(errno)));
5043   } else {
5044     Solaris::set_dev_zero_fd(fd);
5045 
5046     // Close on exec, child won't inherit.
5047     fcntl(fd, F_SETFD, FD_CLOEXEC);
5048   }
5049 
5050   clock_tics_per_sec = CLK_TCK;
5051 
5052   // check if dladdr1() exists; dladdr1 can provide more information than
5053   // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
5054   // and is available on linker patches for 5.7 and 5.8.
5055   // libdl.so must have been loaded, this call is just an entry lookup
5056   void * hdl = dlopen("libdl.so", RTLD_NOW);
5057   if (hdl)
5058     dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
5059 
5060   // (Solaris only) this switches to calls that actually do locking.
5061   ThreadCritical::initialize();
5062 
5063   main_thread = thr_self();
5064 
5065   // Constant minimum stack size allowed. It must be at least
5066   // the minimum of what the OS supports (thr_min_stack()), and
5067   // enough to allow the thread to get to user bytecode execution.
5068   Solaris::min_stack_allowed = MAX2(thr_min_stack(), Solaris::min_stack_allowed);
5069   // If the pagesize of the VM is greater than 8K determine the appropriate
5070   // number of initial guard pages.  The user can change this with the
5071   // command line arguments, if needed.
5072   if (vm_page_size() > 8*K) {
5073     StackYellowPages = 1;
5074     StackRedPages = 1;
5075     StackShadowPages = round_to((StackShadowPages*8*K), vm_page_size()) / vm_page_size();
5076   }
5077 }
5078 
5079 // To install functions for atexit system call
5080 extern "C" {
5081   static void perfMemory_exit_helper() {
5082     perfMemory_exit();
5083   }
5084 }
5085 
5086 // this is called _after_ the global arguments have been parsed
5087 jint os::init_2(void) {
5088   // try to enable extended file IO ASAP, see 6431278
5089   os::Solaris::try_enable_extended_io();
5090 
5091   // Allocate a single page and mark it as readable for safepoint polling.  Also
5092   // use this first mmap call to check support for MAP_ALIGN.
5093   address polling_page = (address)Solaris::mmap_chunk((char*)page_size,
5094                                                       page_size,
5095                                                       MAP_PRIVATE | MAP_ALIGN,
5096                                                       PROT_READ);
5097   if (polling_page == NULL) {
5098     has_map_align = false;
5099     polling_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE,
5100                                                 PROT_READ);
5101   }
5102 
5103   os::set_polling_page(polling_page);
5104 
5105 #ifndef PRODUCT
5106   if( Verbose && PrintMiscellaneous )
5107     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
5108 #endif
5109 
5110   if (!UseMembar) {
5111     address mem_serialize_page = (address)Solaris::mmap_chunk( NULL, page_size, MAP_PRIVATE, PROT_READ | PROT_WRITE );
5112     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
5113     os::set_memory_serialize_page( mem_serialize_page );
5114 
5115 #ifndef PRODUCT
5116     if(Verbose && PrintMiscellaneous)
5117       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
5118 #endif
5119 }
5120 
5121   os::large_page_init();
5122 
5123   // Check minimum allowable stack size for thread creation and to initialize
5124   // the java system classes, including StackOverflowError - depends on page
5125   // size.  Add a page for compiler2 recursion in main thread.
5126   // Add in 2*BytesPerWord times page size to account for VM stack during
5127   // class initialization depending on 32 or 64 bit VM.
5128   os::Solaris::min_stack_allowed = MAX2(os::Solaris::min_stack_allowed,
5129             (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
5130                     2*BytesPerWord COMPILER2_PRESENT(+1)) * page_size);
5131 
5132   size_t threadStackSizeInBytes = ThreadStackSize * K;
5133   if (threadStackSizeInBytes != 0 &&
5134     threadStackSizeInBytes < os::Solaris::min_stack_allowed) {
5135     tty->print_cr("\nThe stack size specified is too small, Specify at least %dk",
5136                   os::Solaris::min_stack_allowed/K);
5137     return JNI_ERR;
5138   }
5139 
5140   // For 64kbps there will be a 64kb page size, which makes
5141   // the usable default stack size quite a bit less.  Increase the
5142   // stack for 64kb (or any > than 8kb) pages, this increases
5143   // virtual memory fragmentation (since we're not creating the
5144   // stack on a power of 2 boundary.  The real fix for this
5145   // should be to fix the guard page mechanism.
5146 
5147   if (vm_page_size() > 8*K) {
5148       threadStackSizeInBytes = (threadStackSizeInBytes != 0)
5149          ? threadStackSizeInBytes +
5150            ((StackYellowPages + StackRedPages) * vm_page_size())
5151          : 0;
5152       ThreadStackSize = threadStackSizeInBytes/K;
5153   }
5154 
5155   // Make the stack size a multiple of the page size so that
5156   // the yellow/red zones can be guarded.
5157   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
5158         vm_page_size()));
5159 
5160   Solaris::libthread_init();
5161 
5162   if (UseNUMA) {
5163     if (!Solaris::liblgrp_init()) {
5164       UseNUMA = false;
5165     } else {
5166       size_t lgrp_limit = os::numa_get_groups_num();
5167       int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit);
5168       size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
5169       FREE_C_HEAP_ARRAY(int, lgrp_ids);
5170       if (lgrp_num < 2) {
5171         // There's only one locality group, disable NUMA.
5172         UseNUMA = false;
5173       }
5174     }
5175     // ISM is not compatible with the NUMA allocator - it always allocates
5176     // pages round-robin across the lgroups.
5177     if (UseNUMA && UseLargePages && UseISM) {
5178       if (!FLAG_IS_DEFAULT(UseNUMA)) {
5179         if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseISM)) {
5180           UseLargePages = false;
5181         } else {
5182           warning("UseNUMA is not compatible with ISM large pages, disabling NUMA allocator");
5183           UseNUMA = false;
5184         }
5185       } else {
5186         UseNUMA = false;
5187       }
5188     }
5189     if (!UseNUMA && ForceNUMA) {
5190       UseNUMA = true;
5191     }
5192   }
5193 
5194   Solaris::signal_sets_init();
5195   Solaris::init_signal_mem();
5196   Solaris::install_signal_handlers();
5197 
5198   if (libjsigversion < JSIG_VERSION_1_4_1) {
5199     Maxlibjsigsigs = OLDMAXSIGNUM;
5200   }
5201 
5202   // initialize synchronization primitives to use either thread or
5203   // lwp synchronization (controlled by UseLWPSynchronization)
5204   Solaris::synchronization_init();
5205 
5206   if (MaxFDLimit) {
5207     // set the number of file descriptors to max. print out error
5208     // if getrlimit/setrlimit fails but continue regardless.
5209     struct rlimit nbr_files;
5210     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5211     if (status != 0) {
5212       if (PrintMiscellaneous && (Verbose || WizardMode))
5213         perror("os::init_2 getrlimit failed");
5214     } else {
5215       nbr_files.rlim_cur = nbr_files.rlim_max;
5216       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5217       if (status != 0) {
5218         if (PrintMiscellaneous && (Verbose || WizardMode))
5219           perror("os::init_2 setrlimit failed");
5220       }
5221     }
5222   }
5223 
5224   // Calculate theoretical max. size of Threads to guard gainst
5225   // artifical out-of-memory situations, where all available address-
5226   // space has been reserved by thread stacks. Default stack size is 1Mb.
5227   size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
5228     JavaThread::stack_size_at_create() : (1*K*K);
5229   assert(pre_thread_stack_size != 0, "Must have a stack");
5230   // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
5231   // we should start doing Virtual Memory banging. Currently when the threads will
5232   // have used all but 200Mb of space.
5233   size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
5234   Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
5235 
5236   // at-exit methods are called in the reverse order of their registration.
5237   // In Solaris 7 and earlier, atexit functions are called on return from
5238   // main or as a result of a call to exit(3C). There can be only 32 of
5239   // these functions registered and atexit() does not set errno. In Solaris
5240   // 8 and later, there is no limit to the number of functions registered
5241   // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
5242   // functions are called upon dlclose(3DL) in addition to return from main
5243   // and exit(3C).
5244 
5245   if (PerfAllowAtExitRegistration) {
5246     // only register atexit functions if PerfAllowAtExitRegistration is set.
5247     // atexit functions can be delayed until process exit time, which
5248     // can be problematic for embedded VM situations. Embedded VMs should
5249     // call DestroyJavaVM() to assure that VM resources are released.
5250 
5251     // note: perfMemory_exit_helper atexit function may be removed in
5252     // the future if the appropriate cleanup code can be added to the
5253     // VM_Exit VMOperation's doit method.
5254     if (atexit(perfMemory_exit_helper) != 0) {
5255       warning("os::init2 atexit(perfMemory_exit_helper) failed");
5256     }
5257   }
5258 
5259   // Init pset_loadavg function pointer
5260   init_pset_getloadavg_ptr();
5261 
5262   return JNI_OK;
5263 }
5264 
5265 void os::init_3(void) {
5266   return;
5267 }
5268 
5269 // Mark the polling page as unreadable
5270 void os::make_polling_page_unreadable(void) {
5271   if( mprotect((char *)_polling_page, page_size, PROT_NONE) != 0 )
5272     fatal("Could not disable polling page");
5273 };
5274 
5275 // Mark the polling page as readable
5276 void os::make_polling_page_readable(void) {
5277   if( mprotect((char *)_polling_page, page_size, PROT_READ) != 0 )
5278     fatal("Could not enable polling page");
5279 };
5280 
5281 // OS interface.
5282 
5283 bool os::check_heap(bool force) { return true; }
5284 
5285 typedef int (*vsnprintf_t)(char* buf, size_t count, const char* fmt, va_list argptr);
5286 static vsnprintf_t sol_vsnprintf = NULL;
5287 
5288 int local_vsnprintf(char* buf, size_t count, const char* fmt, va_list argptr) {
5289   if (!sol_vsnprintf) {
5290     //search  for the named symbol in the objects that were loaded after libjvm
5291     void* where = RTLD_NEXT;
5292     if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5293         sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5294     if (!sol_vsnprintf){
5295       //search  for the named symbol in the objects that were loaded before libjvm
5296       where = RTLD_DEFAULT;
5297       if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5298         sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5299       assert(sol_vsnprintf != NULL, "vsnprintf not found");
5300     }
5301   }
5302   return (*sol_vsnprintf)(buf, count, fmt, argptr);
5303 }
5304 
5305 
5306 // Is a (classpath) directory empty?
5307 bool os::dir_is_empty(const char* path) {
5308   DIR *dir = NULL;
5309   struct dirent *ptr;
5310 
5311   dir = opendir(path);
5312   if (dir == NULL) return true;
5313 
5314   /* Scan the directory */
5315   bool result = true;
5316   char buf[sizeof(struct dirent) + MAX_PATH];
5317   struct dirent *dbuf = (struct dirent *) buf;
5318   while (result && (ptr = readdir(dir, dbuf)) != NULL) {
5319     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5320       result = false;
5321     }
5322   }
5323   closedir(dir);
5324   return result;
5325 }
5326 
5327 // This code originates from JDK's sysOpen and open64_w
5328 // from src/solaris/hpi/src/system_md.c
5329 
5330 #ifndef O_DELETE
5331 #define O_DELETE 0x10000
5332 #endif
5333 
5334 // Open a file. Unlink the file immediately after open returns
5335 // if the specified oflag has the O_DELETE flag set.
5336 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5337 
5338 int os::open(const char *path, int oflag, int mode) {
5339   if (strlen(path) > MAX_PATH - 1) {
5340     errno = ENAMETOOLONG;
5341     return -1;
5342   }
5343   int fd;
5344   int o_delete = (oflag & O_DELETE);
5345   oflag = oflag & ~O_DELETE;
5346 
5347   fd = ::open64(path, oflag, mode);
5348   if (fd == -1) return -1;
5349 
5350   //If the open succeeded, the file might still be a directory
5351   {
5352     struct stat64 buf64;
5353     int ret = ::fstat64(fd, &buf64);
5354     int st_mode = buf64.st_mode;
5355 
5356     if (ret != -1) {
5357       if ((st_mode & S_IFMT) == S_IFDIR) {
5358         errno = EISDIR;
5359         ::close(fd);
5360         return -1;
5361       }
5362     } else {
5363       ::close(fd);
5364       return -1;
5365     }
5366   }
5367     /*
5368      * 32-bit Solaris systems suffer from:
5369      *
5370      * - an historical default soft limit of 256 per-process file
5371      *   descriptors that is too low for many Java programs.
5372      *
5373      * - a design flaw where file descriptors created using stdio
5374      *   fopen must be less than 256, _even_ when the first limit above
5375      *   has been raised.  This can cause calls to fopen (but not calls to
5376      *   open, for example) to fail mysteriously, perhaps in 3rd party
5377      *   native code (although the JDK itself uses fopen).  One can hardly
5378      *   criticize them for using this most standard of all functions.
5379      *
5380      * We attempt to make everything work anyways by:
5381      *
5382      * - raising the soft limit on per-process file descriptors beyond
5383      *   256
5384      *
5385      * - As of Solaris 10u4, we can request that Solaris raise the 256
5386      *   stdio fopen limit by calling function enable_extended_FILE_stdio.
5387      *   This is done in init_2 and recorded in enabled_extended_FILE_stdio
5388      *
5389      * - If we are stuck on an old (pre 10u4) Solaris system, we can
5390      *   workaround the bug by remapping non-stdio file descriptors below
5391      *   256 to ones beyond 256, which is done below.
5392      *
5393      * See:
5394      * 1085341: 32-bit stdio routines should support file descriptors >255
5395      * 6533291: Work around 32-bit Solaris stdio limit of 256 open files
5396      * 6431278: Netbeans crash on 32 bit Solaris: need to call
5397      *          enable_extended_FILE_stdio() in VM initialisation
5398      * Giri Mandalika's blog
5399      * http://technopark02.blogspot.com/2005_05_01_archive.html
5400      */
5401 #ifndef  _LP64
5402      if ((!enabled_extended_FILE_stdio) && fd < 256) {
5403          int newfd = ::fcntl(fd, F_DUPFD, 256);
5404          if (newfd != -1) {
5405              ::close(fd);
5406              fd = newfd;
5407          }
5408      }
5409 #endif // 32-bit Solaris
5410     /*
5411      * All file descriptors that are opened in the JVM and not
5412      * specifically destined for a subprocess should have the
5413      * close-on-exec flag set.  If we don't set it, then careless 3rd
5414      * party native code might fork and exec without closing all
5415      * appropriate file descriptors (e.g. as we do in closeDescriptors in
5416      * UNIXProcess.c), and this in turn might:
5417      *
5418      * - cause end-of-file to fail to be detected on some file
5419      *   descriptors, resulting in mysterious hangs, or
5420      *
5421      * - might cause an fopen in the subprocess to fail on a system
5422      *   suffering from bug 1085341.
5423      *
5424      * (Yes, the default setting of the close-on-exec flag is a Unix
5425      * design flaw)
5426      *
5427      * See:
5428      * 1085341: 32-bit stdio routines should support file descriptors >255
5429      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5430      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5431      */
5432 #ifdef FD_CLOEXEC
5433     {
5434         int flags = ::fcntl(fd, F_GETFD);
5435         if (flags != -1)
5436             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5437     }
5438 #endif
5439 
5440   if (o_delete != 0) {
5441     ::unlink(path);
5442   }
5443   return fd;
5444 }
5445 
5446 // create binary file, rewriting existing file if required
5447 int os::create_binary_file(const char* path, bool rewrite_existing) {
5448   int oflags = O_WRONLY | O_CREAT;
5449   if (!rewrite_existing) {
5450     oflags |= O_EXCL;
5451   }
5452   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5453 }
5454 
5455 // return current position of file pointer
5456 jlong os::current_file_offset(int fd) {
5457   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5458 }
5459 
5460 // move file pointer to the specified offset
5461 jlong os::seek_to_file_offset(int fd, jlong offset) {
5462   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5463 }
5464 
5465 jlong os::lseek(int fd, jlong offset, int whence) {
5466   return (jlong) ::lseek64(fd, offset, whence);
5467 }
5468 
5469 char * os::native_path(char *path) {
5470   return path;
5471 }
5472 
5473 int os::ftruncate(int fd, jlong length) {
5474   return ::ftruncate64(fd, length);
5475 }
5476 
5477 int os::fsync(int fd)  {
5478   RESTARTABLE_RETURN_INT(::fsync(fd));
5479 }
5480 
5481 int os::available(int fd, jlong *bytes) {
5482   jlong cur, end;
5483   int mode;
5484   struct stat64 buf64;
5485 
5486   if (::fstat64(fd, &buf64) >= 0) {
5487     mode = buf64.st_mode;
5488     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5489       /*
5490       * XXX: is the following call interruptible? If so, this might
5491       * need to go through the INTERRUPT_IO() wrapper as for other
5492       * blocking, interruptible calls in this file.
5493       */
5494       int n,ioctl_return;
5495 
5496       INTERRUPTIBLE(::ioctl(fd, FIONREAD, &n),ioctl_return,os::Solaris::clear_interrupted);
5497       if (ioctl_return>= 0) {
5498           *bytes = n;
5499         return 1;
5500       }
5501     }
5502   }
5503   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5504     return 0;
5505   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5506     return 0;
5507   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5508     return 0;
5509   }
5510   *bytes = end - cur;
5511   return 1;
5512 }
5513 
5514 // Map a block of memory.
5515 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
5516                      char *addr, size_t bytes, bool read_only,
5517                      bool allow_exec) {
5518   int prot;
5519   int flags;
5520 
5521   if (read_only) {
5522     prot = PROT_READ;
5523     flags = MAP_SHARED;
5524   } else {
5525     prot = PROT_READ | PROT_WRITE;
5526     flags = MAP_PRIVATE;
5527   }
5528 
5529   if (allow_exec) {
5530     prot |= PROT_EXEC;
5531   }
5532 
5533   if (addr != NULL) {
5534     flags |= MAP_FIXED;
5535   }
5536 
5537   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5538                                      fd, file_offset);
5539   if (mapped_address == MAP_FAILED) {
5540     return NULL;
5541   }
5542   return mapped_address;
5543 }
5544 
5545 
5546 // Remap a block of memory.
5547 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
5548                        char *addr, size_t bytes, bool read_only,
5549                        bool allow_exec) {
5550   // same as map_memory() on this OS
5551   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5552                         allow_exec);
5553 }
5554 
5555 
5556 // Unmap a block of memory.
5557 bool os::unmap_memory(char* addr, size_t bytes) {
5558   return munmap(addr, bytes) == 0;
5559 }
5560 
5561 void os::pause() {
5562   char filename[MAX_PATH];
5563   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5564     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5565   } else {
5566     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5567   }
5568 
5569   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5570   if (fd != -1) {
5571     struct stat buf;
5572     ::close(fd);
5573     while (::stat(filename, &buf) == 0) {
5574       (void)::poll(NULL, 0, 100);
5575     }
5576   } else {
5577     jio_fprintf(stderr,
5578       "Could not open pause file '%s', continuing immediately.\n", filename);
5579   }
5580 }
5581 
5582 #ifndef PRODUCT
5583 #ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5584 // Turn this on if you need to trace synch operations.
5585 // Set RECORD_SYNCH_LIMIT to a large-enough value,
5586 // and call record_synch_enable and record_synch_disable
5587 // around the computation of interest.
5588 
5589 void record_synch(char* name, bool returning);  // defined below
5590 
5591 class RecordSynch {
5592   char* _name;
5593  public:
5594   RecordSynch(char* name) :_name(name)
5595                  { record_synch(_name, false); }
5596   ~RecordSynch() { record_synch(_name,   true);  }
5597 };
5598 
5599 #define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
5600 extern "C" ret name params {                                    \
5601   typedef ret name##_t params;                                  \
5602   static name##_t* implem = NULL;                               \
5603   static int callcount = 0;                                     \
5604   if (implem == NULL) {                                         \
5605     implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
5606     if (implem == NULL)  fatal(dlerror());                      \
5607   }                                                             \
5608   ++callcount;                                                  \
5609   RecordSynch _rs(#name);                                       \
5610   inner;                                                        \
5611   return implem args;                                           \
5612 }
5613 // in dbx, examine callcounts this way:
5614 // for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
5615 
5616 #define CHECK_POINTER_OK(p) \
5617   (Universe::perm_gen() == NULL || !Universe::is_reserved_heap((oop)(p)))
5618 #define CHECK_MU \
5619   if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
5620 #define CHECK_CV \
5621   if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
5622 #define CHECK_P(p) \
5623   if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
5624 
5625 #define CHECK_MUTEX(mutex_op) \
5626 CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
5627 
5628 CHECK_MUTEX(   mutex_lock)
5629 CHECK_MUTEX(  _mutex_lock)
5630 CHECK_MUTEX( mutex_unlock)
5631 CHECK_MUTEX(_mutex_unlock)
5632 CHECK_MUTEX( mutex_trylock)
5633 CHECK_MUTEX(_mutex_trylock)
5634 
5635 #define CHECK_COND(cond_op) \
5636 CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU;CHECK_CV);
5637 
5638 CHECK_COND( cond_wait);
5639 CHECK_COND(_cond_wait);
5640 CHECK_COND(_cond_wait_cancel);
5641 
5642 #define CHECK_COND2(cond_op) \
5643 CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU;CHECK_CV);
5644 
5645 CHECK_COND2( cond_timedwait);
5646 CHECK_COND2(_cond_timedwait);
5647 CHECK_COND2(_cond_timedwait_cancel);
5648 
5649 // do the _lwp_* versions too
5650 #define mutex_t lwp_mutex_t
5651 #define cond_t  lwp_cond_t
5652 CHECK_MUTEX(  _lwp_mutex_lock)
5653 CHECK_MUTEX(  _lwp_mutex_unlock)
5654 CHECK_MUTEX(  _lwp_mutex_trylock)
5655 CHECK_MUTEX( __lwp_mutex_lock)
5656 CHECK_MUTEX( __lwp_mutex_unlock)
5657 CHECK_MUTEX( __lwp_mutex_trylock)
5658 CHECK_MUTEX(___lwp_mutex_lock)
5659 CHECK_MUTEX(___lwp_mutex_unlock)
5660 
5661 CHECK_COND(  _lwp_cond_wait);
5662 CHECK_COND( __lwp_cond_wait);
5663 CHECK_COND(___lwp_cond_wait);
5664 
5665 CHECK_COND2(  _lwp_cond_timedwait);
5666 CHECK_COND2( __lwp_cond_timedwait);
5667 #undef mutex_t
5668 #undef cond_t
5669 
5670 CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5671 CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5672 CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5673 CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5674 CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5675 CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5676 CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5677 CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5678 
5679 
5680 // recording machinery:
5681 
5682 enum { RECORD_SYNCH_LIMIT = 200 };
5683 char* record_synch_name[RECORD_SYNCH_LIMIT];
5684 void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
5685 bool record_synch_returning[RECORD_SYNCH_LIMIT];
5686 thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
5687 int record_synch_count = 0;
5688 bool record_synch_enabled = false;
5689 
5690 // in dbx, examine recorded data this way:
5691 // for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
5692 
5693 void record_synch(char* name, bool returning) {
5694   if (record_synch_enabled) {
5695     if (record_synch_count < RECORD_SYNCH_LIMIT) {
5696       record_synch_name[record_synch_count] = name;
5697       record_synch_returning[record_synch_count] = returning;
5698       record_synch_thread[record_synch_count] = thr_self();
5699       record_synch_arg0ptr[record_synch_count] = &name;
5700       record_synch_count++;
5701     }
5702     // put more checking code here:
5703     // ...
5704   }
5705 }
5706 
5707 void record_synch_enable() {
5708   // start collecting trace data, if not already doing so
5709   if (!record_synch_enabled)  record_synch_count = 0;
5710   record_synch_enabled = true;
5711 }
5712 
5713 void record_synch_disable() {
5714   // stop collecting trace data
5715   record_synch_enabled = false;
5716 }
5717 
5718 #endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5719 #endif // PRODUCT
5720 
5721 const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5722 const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
5723                                (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5724 
5725 
5726 // JVMTI & JVM monitoring and management support
5727 // The thread_cpu_time() and current_thread_cpu_time() are only
5728 // supported if is_thread_cpu_time_supported() returns true.
5729 // They are not supported on Solaris T1.
5730 
5731 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5732 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5733 // of a thread.
5734 //
5735 // current_thread_cpu_time() and thread_cpu_time(Thread *)
5736 // returns the fast estimate available on the platform.
5737 
5738 // hrtime_t gethrvtime() return value includes
5739 // user time but does not include system time
5740 jlong os::current_thread_cpu_time() {
5741   return (jlong) gethrvtime();
5742 }
5743 
5744 jlong os::thread_cpu_time(Thread *thread) {
5745   // return user level CPU time only to be consistent with
5746   // what current_thread_cpu_time returns.
5747   // thread_cpu_time_info() must be changed if this changes
5748   return os::thread_cpu_time(thread, false /* user time only */);
5749 }
5750 
5751 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5752   if (user_sys_cpu_time) {
5753     return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5754   } else {
5755     return os::current_thread_cpu_time();
5756   }
5757 }
5758 
5759 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5760   char proc_name[64];
5761   int count;
5762   prusage_t prusage;
5763   jlong lwp_time;
5764   int fd;
5765 
5766   sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
5767                      getpid(),
5768                      thread->osthread()->lwp_id());
5769   fd = ::open(proc_name, O_RDONLY);
5770   if ( fd == -1 ) return -1;
5771 
5772   do {
5773     count = ::pread(fd,
5774                   (void *)&prusage.pr_utime,
5775                   thr_time_size,
5776                   thr_time_off);
5777   } while (count < 0 && errno == EINTR);
5778   ::close(fd);
5779   if ( count < 0 ) return -1;
5780 
5781   if (user_sys_cpu_time) {
5782     // user + system CPU time
5783     lwp_time = (((jlong)prusage.pr_stime.tv_sec +
5784                  (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
5785                  (jlong)prusage.pr_stime.tv_nsec +
5786                  (jlong)prusage.pr_utime.tv_nsec;
5787   } else {
5788     // user level CPU time only
5789     lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
5790                 (jlong)prusage.pr_utime.tv_nsec;
5791   }
5792 
5793   return(lwp_time);
5794 }
5795 
5796 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5797   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5798   info_ptr->may_skip_backward = false;    // elapsed time not wall time
5799   info_ptr->may_skip_forward = false;     // elapsed time not wall time
5800   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5801 }
5802 
5803 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5804   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5805   info_ptr->may_skip_backward = false;    // elapsed time not wall time
5806   info_ptr->may_skip_forward = false;     // elapsed time not wall time
5807   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5808 }
5809 
5810 bool os::is_thread_cpu_time_supported() {
5811   if ( os::Solaris::T2_libthread() || UseBoundThreads ) {
5812     return true;
5813   } else {
5814     return false;
5815   }
5816 }
5817 
5818 // System loadavg support.  Returns -1 if load average cannot be obtained.
5819 // Return the load average for our processor set if the primitive exists
5820 // (Solaris 9 and later).  Otherwise just return system wide loadavg.
5821 int os::loadavg(double loadavg[], int nelem) {
5822   if (pset_getloadavg_ptr != NULL) {
5823     return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
5824   } else {
5825     return ::getloadavg(loadavg, nelem);
5826   }
5827 }
5828 
5829 //---------------------------------------------------------------------------------
5830 
5831 static address same_page(address x, address y) {
5832   intptr_t page_bits = -os::vm_page_size();
5833   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
5834     return x;
5835   else if (x > y)
5836     return (address)(intptr_t(y) | ~page_bits) + 1;
5837   else
5838     return (address)(intptr_t(y) & page_bits);
5839 }
5840 
5841 bool os::find(address addr, outputStream* st) {
5842   Dl_info dlinfo;
5843   memset(&dlinfo, 0, sizeof(dlinfo));
5844   if (dladdr(addr, &dlinfo)) {
5845 #ifdef _LP64
5846     st->print("0x%016lx: ", addr);
5847 #else
5848     st->print("0x%08x: ", addr);
5849 #endif
5850     if (dlinfo.dli_sname != NULL)
5851       st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
5852     else if (dlinfo.dli_fname)
5853       st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
5854     else
5855       st->print("<absolute address>");
5856     if (dlinfo.dli_fname)  st->print(" in %s", dlinfo.dli_fname);
5857 #ifdef _LP64
5858     if (dlinfo.dli_fbase)  st->print(" at 0x%016lx", dlinfo.dli_fbase);
5859 #else
5860     if (dlinfo.dli_fbase)  st->print(" at 0x%08x", dlinfo.dli_fbase);
5861 #endif
5862     st->cr();
5863 
5864     if (Verbose) {
5865       // decode some bytes around the PC
5866       address begin = same_page(addr-40, addr);
5867       address end   = same_page(addr+40, addr);
5868       address       lowest = (address) dlinfo.dli_sname;
5869       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5870       if (begin < lowest)  begin = lowest;
5871       Dl_info dlinfo2;
5872       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
5873           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5874         end = (address) dlinfo2.dli_saddr;
5875       Disassembler::decode(begin, end, st);
5876     }
5877     return true;
5878   }
5879   return false;
5880 }
5881 
5882 // Following function has been added to support HotSparc's libjvm.so running
5883 // under Solaris production JDK 1.2.2 / 1.3.0.  These came from
5884 // src/solaris/hpi/native_threads in the EVM codebase.
5885 //
5886 // NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
5887 // libraries and should thus be removed. We will leave it behind for a while
5888 // until we no longer want to able to run on top of 1.3.0 Solaris production
5889 // JDK. See 4341971.
5890 
5891 #define STACK_SLACK 0x800
5892 
5893 extern "C" {
5894   intptr_t sysThreadAvailableStackWithSlack() {
5895     stack_t st;
5896     intptr_t retval, stack_top;
5897     retval = thr_stksegment(&st);
5898     assert(retval == 0, "incorrect return value from thr_stksegment");
5899     assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
5900     assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
5901     stack_top=(intptr_t)st.ss_sp-st.ss_size;
5902     return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
5903   }
5904 }
5905 
5906 // Just to get the Kernel build to link on solaris for testing.
5907 
5908 extern "C" {
5909 class ASGCT_CallTrace;
5910 void AsyncGetCallTrace(ASGCT_CallTrace *trace, jint depth, void* ucontext)
5911   KERNEL_RETURN;
5912 }
5913 
5914 
5915 // ObjectMonitor park-unpark infrastructure ...
5916 //
5917 // We implement Solaris and Linux PlatformEvents with the
5918 // obvious condvar-mutex-flag triple.
5919 // Another alternative that works quite well is pipes:
5920 // Each PlatformEvent consists of a pipe-pair.
5921 // The thread associated with the PlatformEvent
5922 // calls park(), which reads from the input end of the pipe.
5923 // Unpark() writes into the other end of the pipe.
5924 // The write-side of the pipe must be set NDELAY.
5925 // Unfortunately pipes consume a large # of handles.
5926 // Native solaris lwp_park() and lwp_unpark() work nicely, too.
5927 // Using pipes for the 1st few threads might be workable, however.
5928 //
5929 // park() is permitted to return spuriously.
5930 // Callers of park() should wrap the call to park() in
5931 // an appropriate loop.  A litmus test for the correct
5932 // usage of park is the following: if park() were modified
5933 // to immediately return 0 your code should still work,
5934 // albeit degenerating to a spin loop.
5935 //
5936 // An interesting optimization for park() is to use a trylock()
5937 // to attempt to acquire the mutex.  If the trylock() fails
5938 // then we know that a concurrent unpark() operation is in-progress.
5939 // in that case the park() code could simply set _count to 0
5940 // and return immediately.  The subsequent park() operation *might*
5941 // return immediately.  That's harmless as the caller of park() is
5942 // expected to loop.  By using trylock() we will have avoided a
5943 // avoided a context switch caused by contention on the per-thread mutex.
5944 //
5945 // TODO-FIXME:
5946 // 1.  Reconcile Doug's JSR166 j.u.c park-unpark with the
5947 //     objectmonitor implementation.
5948 // 2.  Collapse the JSR166 parker event, and the
5949 //     objectmonitor ParkEvent into a single "Event" construct.
5950 // 3.  In park() and unpark() add:
5951 //     assert (Thread::current() == AssociatedWith).
5952 // 4.  add spurious wakeup injection on a -XX:EarlyParkReturn=N switch.
5953 //     1-out-of-N park() operations will return immediately.
5954 //
5955 // _Event transitions in park()
5956 //   -1 => -1 : illegal
5957 //    1 =>  0 : pass - return immediately
5958 //    0 => -1 : block
5959 //
5960 // _Event serves as a restricted-range semaphore.
5961 //
5962 // Another possible encoding of _Event would be with
5963 // explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
5964 //
5965 // TODO-FIXME: add DTRACE probes for:
5966 // 1.   Tx parks
5967 // 2.   Ty unparks Tx
5968 // 3.   Tx resumes from park
5969 
5970 
5971 // value determined through experimentation
5972 #define ROUNDINGFIX 11
5973 
5974 // utility to compute the abstime argument to timedwait.
5975 // TODO-FIXME: switch from compute_abstime() to unpackTime().
5976 
5977 static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
5978   // millis is the relative timeout time
5979   // abstime will be the absolute timeout time
5980   if (millis < 0)  millis = 0;
5981   struct timeval now;
5982   int status = gettimeofday(&now, NULL);
5983   assert(status == 0, "gettimeofday");
5984   jlong seconds = millis / 1000;
5985   jlong max_wait_period;
5986 
5987   if (UseLWPSynchronization) {
5988     // forward port of fix for 4275818 (not sleeping long enough)
5989     // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
5990     // _lwp_cond_timedwait() used a round_down algorithm rather
5991     // than a round_up. For millis less than our roundfactor
5992     // it rounded down to 0 which doesn't meet the spec.
5993     // For millis > roundfactor we may return a bit sooner, but
5994     // since we can not accurately identify the patch level and
5995     // this has already been fixed in Solaris 9 and 8 we will
5996     // leave it alone rather than always rounding down.
5997 
5998     if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
5999        // It appears that when we go directly through Solaris _lwp_cond_timedwait()
6000            // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
6001            max_wait_period = 21000000;
6002   } else {
6003     max_wait_period = 50000000;
6004   }
6005   millis %= 1000;
6006   if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
6007      seconds = max_wait_period;
6008   }
6009   abstime->tv_sec = now.tv_sec  + seconds;
6010   long       usec = now.tv_usec + millis * 1000;
6011   if (usec >= 1000000) {
6012     abstime->tv_sec += 1;
6013     usec -= 1000000;
6014   }
6015   abstime->tv_nsec = usec * 1000;
6016   return abstime;
6017 }
6018 
6019 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
6020 // Conceptually TryPark() should be equivalent to park(0).
6021 
6022 int os::PlatformEvent::TryPark() {
6023   for (;;) {
6024     const int v = _Event ;
6025     guarantee ((v == 0) || (v == 1), "invariant") ;
6026     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
6027   }
6028 }
6029 
6030 void os::PlatformEvent::park() {           // AKA: down()
6031   // Invariant: Only the thread associated with the Event/PlatformEvent
6032   // may call park().
6033   int v ;
6034   for (;;) {
6035       v = _Event ;
6036       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6037   }
6038   guarantee (v >= 0, "invariant") ;
6039   if (v == 0) {
6040      // Do this the hard way by blocking ...
6041      // See http://monaco.sfbay/detail.jsf?cr=5094058.
6042      // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
6043      // Only for SPARC >= V8PlusA
6044 #if defined(__sparc) && defined(COMPILER2)
6045      if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6046 #endif
6047      int status = os::Solaris::mutex_lock(_mutex);
6048      assert_status(status == 0, status,  "mutex_lock");
6049      guarantee (_nParked == 0, "invariant") ;
6050      ++ _nParked ;
6051      while (_Event < 0) {
6052         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
6053         // Treat this the same as if the wait was interrupted
6054         // With usr/lib/lwp going to kernel, always handle ETIME
6055         status = os::Solaris::cond_wait(_cond, _mutex);
6056         if (status == ETIME) status = EINTR ;
6057         assert_status(status == 0 || status == EINTR, status, "cond_wait");
6058      }
6059      -- _nParked ;
6060      _Event = 0 ;
6061      status = os::Solaris::mutex_unlock(_mutex);
6062      assert_status(status == 0, status, "mutex_unlock");
6063   }
6064 }
6065 
6066 int os::PlatformEvent::park(jlong millis) {
6067   guarantee (_nParked == 0, "invariant") ;
6068   int v ;
6069   for (;;) {
6070       v = _Event ;
6071       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6072   }
6073   guarantee (v >= 0, "invariant") ;
6074   if (v != 0) return OS_OK ;
6075 
6076   int ret = OS_TIMEOUT;
6077   timestruc_t abst;
6078   compute_abstime (&abst, millis);
6079 
6080   // See http://monaco.sfbay/detail.jsf?cr=5094058.
6081   // For Solaris SPARC set fprs.FEF=0 prior to parking.
6082   // Only for SPARC >= V8PlusA
6083 #if defined(__sparc) && defined(COMPILER2)
6084  if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6085 #endif
6086   int status = os::Solaris::mutex_lock(_mutex);
6087   assert_status(status == 0, status, "mutex_lock");
6088   guarantee (_nParked == 0, "invariant") ;
6089   ++ _nParked ;
6090   while (_Event < 0) {
6091      int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
6092      assert_status(status == 0 || status == EINTR ||
6093                    status == ETIME || status == ETIMEDOUT,
6094                    status, "cond_timedwait");
6095      if (!FilterSpuriousWakeups) break ;                // previous semantics
6096      if (status == ETIME || status == ETIMEDOUT) break ;
6097      // We consume and ignore EINTR and spurious wakeups.
6098   }
6099   -- _nParked ;
6100   if (_Event >= 0) ret = OS_OK ;
6101   _Event = 0 ;
6102   status = os::Solaris::mutex_unlock(_mutex);
6103   assert_status(status == 0, status, "mutex_unlock");
6104   return ret;
6105 }
6106 
6107 void os::PlatformEvent::unpark() {
6108   int v, AnyWaiters;
6109 
6110   // Increment _Event.
6111   // Another acceptable implementation would be to simply swap 1
6112   // into _Event:
6113   //   if (Swap (&_Event, 1) < 0) {
6114   //      mutex_lock (_mutex) ; AnyWaiters = nParked; mutex_unlock (_mutex) ;
6115   //      if (AnyWaiters) cond_signal (_cond) ;
6116   //   }
6117 
6118   for (;;) {
6119     v = _Event ;
6120     if (v > 0) {
6121        // The LD of _Event could have reordered or be satisfied
6122        // by a read-aside from this processor's write buffer.
6123        // To avoid problems execute a barrier and then
6124        // ratify the value.  A degenerate CAS() would also work.
6125        // Viz., CAS (v+0, &_Event, v) == v).
6126        OrderAccess::fence() ;
6127        if (_Event == v) return ;
6128        continue ;
6129     }
6130     if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
6131   }
6132 
6133   // If the thread associated with the event was parked, wake it.
6134   if (v < 0) {
6135      int status ;
6136      // Wait for the thread assoc with the PlatformEvent to vacate.
6137      status = os::Solaris::mutex_lock(_mutex);
6138      assert_status(status == 0, status, "mutex_lock");
6139      AnyWaiters = _nParked ;
6140      status = os::Solaris::mutex_unlock(_mutex);
6141      assert_status(status == 0, status, "mutex_unlock");
6142      guarantee (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
6143      if (AnyWaiters != 0) {
6144        // We intentional signal *after* dropping the lock
6145        // to avoid a common class of futile wakeups.
6146        status = os::Solaris::cond_signal(_cond);
6147        assert_status(status == 0, status, "cond_signal");
6148      }
6149   }
6150 }
6151 
6152 // JSR166
6153 // -------------------------------------------------------
6154 
6155 /*
6156  * The solaris and linux implementations of park/unpark are fairly
6157  * conservative for now, but can be improved. They currently use a
6158  * mutex/condvar pair, plus _counter.
6159  * Park decrements _counter if > 0, else does a condvar wait.  Unpark
6160  * sets count to 1 and signals condvar.  Only one thread ever waits
6161  * on the condvar. Contention seen when trying to park implies that someone
6162  * is unparking you, so don't wait. And spurious returns are fine, so there
6163  * is no need to track notifications.
6164  */
6165 
6166 #define MAX_SECS 100000000
6167 /*
6168  * This code is common to linux and solaris and will be moved to a
6169  * common place in dolphin.
6170  *
6171  * The passed in time value is either a relative time in nanoseconds
6172  * or an absolute time in milliseconds. Either way it has to be unpacked
6173  * into suitable seconds and nanoseconds components and stored in the
6174  * given timespec structure.
6175  * Given time is a 64-bit value and the time_t used in the timespec is only
6176  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
6177  * overflow if times way in the future are given. Further on Solaris versions
6178  * prior to 10 there is a restriction (see cond_timedwait) that the specified
6179  * number of seconds, in abstime, is less than current_time  + 100,000,000.
6180  * As it will be 28 years before "now + 100000000" will overflow we can
6181  * ignore overflow and just impose a hard-limit on seconds using the value
6182  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
6183  * years from "now".
6184  */
6185 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
6186   assert (time > 0, "convertTime");
6187 
6188   struct timeval now;
6189   int status = gettimeofday(&now, NULL);
6190   assert(status == 0, "gettimeofday");
6191 
6192   time_t max_secs = now.tv_sec + MAX_SECS;
6193 
6194   if (isAbsolute) {
6195     jlong secs = time / 1000;
6196     if (secs > max_secs) {
6197       absTime->tv_sec = max_secs;
6198     }
6199     else {
6200       absTime->tv_sec = secs;
6201     }
6202     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
6203   }
6204   else {
6205     jlong secs = time / NANOSECS_PER_SEC;
6206     if (secs >= MAX_SECS) {
6207       absTime->tv_sec = max_secs;
6208       absTime->tv_nsec = 0;
6209     }
6210     else {
6211       absTime->tv_sec = now.tv_sec + secs;
6212       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
6213       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6214         absTime->tv_nsec -= NANOSECS_PER_SEC;
6215         ++absTime->tv_sec; // note: this must be <= max_secs
6216       }
6217     }
6218   }
6219   assert(absTime->tv_sec >= 0, "tv_sec < 0");
6220   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
6221   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
6222   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
6223 }
6224 
6225 void Parker::park(bool isAbsolute, jlong time) {
6226 
6227   // Optional fast-path check:
6228   // Return immediately if a permit is available.
6229   if (_counter > 0) {
6230       _counter = 0 ;
6231       OrderAccess::fence();
6232       return ;
6233   }
6234 
6235   // Optional fast-exit: Check interrupt before trying to wait
6236   Thread* thread = Thread::current();
6237   assert(thread->is_Java_thread(), "Must be JavaThread");
6238   JavaThread *jt = (JavaThread *)thread;
6239   if (Thread::is_interrupted(thread, false)) {
6240     return;
6241   }
6242 
6243   // First, demultiplex/decode time arguments
6244   timespec absTime;
6245   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
6246     return;
6247   }
6248   if (time > 0) {
6249     // Warning: this code might be exposed to the old Solaris time
6250     // round-down bugs.  Grep "roundingFix" for details.
6251     unpackTime(&absTime, isAbsolute, time);
6252   }
6253 
6254   // Enter safepoint region
6255   // Beware of deadlocks such as 6317397.
6256   // The per-thread Parker:: _mutex is a classic leaf-lock.
6257   // In particular a thread must never block on the Threads_lock while
6258   // holding the Parker:: mutex.  If safepoints are pending both the
6259   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6260   ThreadBlockInVM tbivm(jt);
6261 
6262   // Don't wait if cannot get lock since interference arises from
6263   // unblocking.  Also. check interrupt before trying wait
6264   if (Thread::is_interrupted(thread, false) ||
6265       os::Solaris::mutex_trylock(_mutex) != 0) {
6266     return;
6267   }
6268 
6269   int status ;
6270 
6271   if (_counter > 0)  { // no wait needed
6272     _counter = 0;
6273     status = os::Solaris::mutex_unlock(_mutex);
6274     assert (status == 0, "invariant") ;
6275     OrderAccess::fence();
6276     return;
6277   }
6278 
6279 #ifdef ASSERT
6280   // Don't catch signals while blocked; let the running threads have the signals.
6281   // (This allows a debugger to break into the running thread.)
6282   sigset_t oldsigs;
6283   sigset_t* allowdebug_blocked = os::Solaris::allowdebug_blocked_signals();
6284   thr_sigsetmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6285 #endif
6286 
6287   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6288   jt->set_suspend_equivalent();
6289   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6290 
6291   // Do this the hard way by blocking ...
6292   // See http://monaco.sfbay/detail.jsf?cr=5094058.
6293   // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
6294   // Only for SPARC >= V8PlusA
6295 #if defined(__sparc) && defined(COMPILER2)
6296   if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6297 #endif
6298 
6299   if (time == 0) {
6300     status = os::Solaris::cond_wait (_cond, _mutex) ;
6301   } else {
6302     status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
6303   }
6304   // Note that an untimed cond_wait() can sometimes return ETIME on older
6305   // versions of the Solaris.
6306   assert_status(status == 0 || status == EINTR ||
6307                 status == ETIME || status == ETIMEDOUT,
6308                 status, "cond_timedwait");
6309 
6310 #ifdef ASSERT
6311   thr_sigsetmask(SIG_SETMASK, &oldsigs, NULL);
6312 #endif
6313   _counter = 0 ;
6314   status = os::Solaris::mutex_unlock(_mutex);
6315   assert_status(status == 0, status, "mutex_unlock") ;
6316 
6317   // If externally suspended while waiting, re-suspend
6318   if (jt->handle_special_suspend_equivalent_condition()) {
6319     jt->java_suspend_self();
6320   }
6321   OrderAccess::fence();
6322 }
6323 
6324 void Parker::unpark() {
6325   int s, status ;
6326   status = os::Solaris::mutex_lock (_mutex) ;
6327   assert (status == 0, "invariant") ;
6328   s = _counter;
6329   _counter = 1;
6330   status = os::Solaris::mutex_unlock (_mutex) ;
6331   assert (status == 0, "invariant") ;
6332 
6333   if (s < 1) {
6334     status = os::Solaris::cond_signal (_cond) ;
6335     assert (status == 0, "invariant") ;
6336   }
6337 }
6338 
6339 extern char** environ;
6340 
6341 // Run the specified command in a separate process. Return its exit value,
6342 // or -1 on failure (e.g. can't fork a new process).
6343 // Unlike system(), this function can be called from signal handler. It
6344 // doesn't block SIGINT et al.
6345 int os::fork_and_exec(char* cmd) {
6346   char * argv[4];
6347   argv[0] = (char *)"sh";
6348   argv[1] = (char *)"-c";
6349   argv[2] = cmd;
6350   argv[3] = NULL;
6351 
6352   // fork is async-safe, fork1 is not so can't use in signal handler
6353   pid_t pid;
6354   Thread* t = ThreadLocalStorage::get_thread_slow();
6355   if (t != NULL && t->is_inside_signal_handler()) {
6356     pid = fork();
6357   } else {
6358     pid = fork1();
6359   }
6360 
6361   if (pid < 0) {
6362     // fork failed
6363     warning("fork failed: %s", strerror(errno));
6364     return -1;
6365 
6366   } else if (pid == 0) {
6367     // child process
6368 
6369     // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
6370     execve("/usr/bin/sh", argv, environ);
6371 
6372     // execve failed
6373     _exit(-1);
6374 
6375   } else  {
6376     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6377     // care about the actual exit code, for now.
6378 
6379     int status;
6380 
6381     // Wait for the child process to exit.  This returns immediately if
6382     // the child has already exited. */
6383     while (waitpid(pid, &status, 0) < 0) {
6384         switch (errno) {
6385         case ECHILD: return 0;
6386         case EINTR: break;
6387         default: return -1;
6388         }
6389     }
6390 
6391     if (WIFEXITED(status)) {
6392        // The child exited normally; get its exit code.
6393        return WEXITSTATUS(status);
6394     } else if (WIFSIGNALED(status)) {
6395        // The child exited because of a signal
6396        // The best value to return is 0x80 + signal number,
6397        // because that is what all Unix shells do, and because
6398        // it allows callers to distinguish between process exit and
6399        // process death by signal.
6400        return 0x80 + WTERMSIG(status);
6401     } else {
6402        // Unknown exit code; pass it through
6403        return status;
6404     }
6405   }
6406 }
6407 
6408 // is_headless_jre()
6409 //
6410 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6411 // in order to report if we are running in a headless jre
6412 //
6413 // Since JDK8 xawt/libmawt.so was moved into the same directory
6414 // as libawt.so, and renamed libawt_xawt.so
6415 //
6416 bool os::is_headless_jre() {
6417     struct stat statbuf;
6418     char buf[MAXPATHLEN];
6419     char libmawtpath[MAXPATHLEN];
6420     const char *xawtstr  = "/xawt/libmawt.so";
6421     const char *new_xawtstr = "/libawt_xawt.so";
6422     char *p;
6423 
6424     // Get path to libjvm.so
6425     os::jvm_path(buf, sizeof(buf));
6426 
6427     // Get rid of libjvm.so
6428     p = strrchr(buf, '/');
6429     if (p == NULL) return false;
6430     else *p = '\0';
6431 
6432     // Get rid of client or server
6433     p = strrchr(buf, '/');
6434     if (p == NULL) return false;
6435     else *p = '\0';
6436 
6437     // check xawt/libmawt.so
6438     strcpy(libmawtpath, buf);
6439     strcat(libmawtpath, xawtstr);
6440     if (::stat(libmawtpath, &statbuf) == 0) return false;
6441 
6442     // check libawt_xawt.so
6443     strcpy(libmawtpath, buf);
6444     strcat(libmawtpath, new_xawtstr);
6445     if (::stat(libmawtpath, &statbuf) == 0) return false;
6446 
6447     return true;
6448 }
6449 
6450 size_t os::write(int fd, const void *buf, unsigned int nBytes) {
6451   INTERRUPTIBLE_RETURN_INT(::write(fd, buf, nBytes), os::Solaris::clear_interrupted);
6452 }
6453 
6454 int os::close(int fd) {
6455   RESTARTABLE_RETURN_INT(::close(fd));
6456 }
6457 
6458 int os::socket_close(int fd) {
6459   RESTARTABLE_RETURN_INT(::close(fd));
6460 }
6461 
6462 int os::recv(int fd, char* buf, size_t nBytes, uint flags) {
6463   INTERRUPTIBLE_RETURN_INT((int)::recv(fd, buf, nBytes, flags), os::Solaris::clear_interrupted);
6464 }
6465 
6466 int os::send(int fd, char* buf, size_t nBytes, uint flags) {
6467   INTERRUPTIBLE_RETURN_INT((int)::send(fd, buf, nBytes, flags), os::Solaris::clear_interrupted);
6468 }
6469 
6470 int os::raw_send(int fd, char* buf, size_t nBytes, uint flags) {
6471   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
6472 }
6473 
6474 // As both poll and select can be interrupted by signals, we have to be
6475 // prepared to restart the system call after updating the timeout, unless
6476 // a poll() is done with timeout == -1, in which case we repeat with this
6477 // "wait forever" value.
6478 
6479 int os::timeout(int fd, long timeout) {
6480   int res;
6481   struct timeval t;
6482   julong prevtime, newtime;
6483   static const char* aNull = 0;
6484   struct pollfd pfd;
6485   pfd.fd = fd;
6486   pfd.events = POLLIN;
6487 
6488   gettimeofday(&t, &aNull);
6489   prevtime = ((julong)t.tv_sec * 1000)  +  t.tv_usec / 1000;
6490 
6491   for(;;) {
6492     INTERRUPTIBLE_NORESTART(::poll(&pfd, 1, timeout), res, os::Solaris::clear_interrupted);
6493     if(res == OS_ERR && errno == EINTR) {
6494         if(timeout != -1) {
6495           gettimeofday(&t, &aNull);
6496           newtime = ((julong)t.tv_sec * 1000)  +  t.tv_usec /1000;
6497           timeout -= newtime - prevtime;
6498           if(timeout <= 0)
6499             return OS_OK;
6500           prevtime = newtime;
6501         }
6502     } else return res;
6503   }
6504 }
6505 
6506 int os::connect(int fd, struct sockaddr *him, socklen_t len) {
6507   int _result;
6508   INTERRUPTIBLE_NORESTART(::connect(fd, him, len), _result,\
6509                           os::Solaris::clear_interrupted);
6510 
6511   // Depending on when thread interruption is reset, _result could be
6512   // one of two values when errno == EINTR
6513 
6514   if (((_result == OS_INTRPT) || (_result == OS_ERR))
6515       && (errno == EINTR)) {
6516      /* restarting a connect() changes its errno semantics */
6517      INTERRUPTIBLE(::connect(fd, him, len), _result,\
6518                    os::Solaris::clear_interrupted);
6519      /* undo these changes */
6520      if (_result == OS_ERR) {
6521        if (errno == EALREADY) {
6522          errno = EINPROGRESS; /* fall through */
6523        } else if (errno == EISCONN) {
6524          errno = 0;
6525          return OS_OK;
6526        }
6527      }
6528    }
6529    return _result;
6530  }
6531 
6532 int os::accept(int fd, struct sockaddr* him, socklen_t* len) {
6533   if (fd < 0) {
6534     return OS_ERR;
6535   }
6536   INTERRUPTIBLE_RETURN_INT((int)::accept(fd, him, len),\
6537                            os::Solaris::clear_interrupted);
6538 }
6539 
6540 int os::recvfrom(int fd, char* buf, size_t nBytes, uint flags,
6541                  sockaddr* from, socklen_t* fromlen) {
6542   INTERRUPTIBLE_RETURN_INT((int)::recvfrom(fd, buf, nBytes, flags, from, fromlen),\
6543                            os::Solaris::clear_interrupted);
6544 }
6545 
6546 int os::sendto(int fd, char* buf, size_t len, uint flags,
6547                struct sockaddr* to, socklen_t tolen) {
6548   INTERRUPTIBLE_RETURN_INT((int)::sendto(fd, buf, len, flags, to, tolen),\
6549                            os::Solaris::clear_interrupted);
6550 }
6551 
6552 int os::socket_available(int fd, jint *pbytes) {
6553   if (fd < 0) {
6554     return OS_OK;
6555   }
6556   int ret;
6557   RESTARTABLE(::ioctl(fd, FIONREAD, pbytes), ret);
6558   // note: ioctl can return 0 when successful, JVM_SocketAvailable
6559   // is expected to return 0 on failure and 1 on success to the jdk.
6560   return (ret == OS_ERR) ? 0 : 1;
6561 }
6562 
6563 int os::bind(int fd, struct sockaddr* him, socklen_t len) {
6564    INTERRUPTIBLE_RETURN_INT_NORESTART(::bind(fd, him, len),\
6565                                       os::Solaris::clear_interrupted);
6566 }