1 /*
   2  * Copyright (c) 1997, 2009, 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 // do not include  precompiled  header file
  26 # include "incls/_os_solaris.cpp.incl"
  27 
  28 // put OS-includes here
  29 # include <dlfcn.h>
  30 # include <errno.h>
  31 # include <link.h>
  32 # include <poll.h>
  33 # include <pthread.h>
  34 # include <pwd.h>
  35 # include <schedctl.h>
  36 # include <setjmp.h>
  37 # include <signal.h>
  38 # include <stdio.h>
  39 # include <alloca.h>
  40 # include <sys/filio.h>
  41 # include <sys/ipc.h>
  42 # include <sys/lwp.h>
  43 # include <sys/machelf.h>     // for elf Sym structure used by dladdr1
  44 # include <sys/mman.h>
  45 # include <sys/processor.h>
  46 # include <sys/procset.h>
  47 # include <sys/pset.h>
  48 # include <sys/resource.h>
  49 # include <sys/shm.h>
  50 # include <sys/socket.h>
  51 # include <sys/stat.h>
  52 # include <sys/systeminfo.h>
  53 # include <sys/time.h>
  54 # include <sys/times.h>
  55 # include <sys/types.h>
  56 # include <sys/wait.h>
  57 # include <sys/utsname.h>
  58 # include <thread.h>
  59 # include <unistd.h>
  60 # include <sys/priocntl.h>
  61 # include <sys/rtpriocntl.h>
  62 # include <sys/tspriocntl.h>
  63 # include <sys/iapriocntl.h>
  64 # include <sys/loadavg.h>
  65 # include <string.h>
  66 
  67 # define _STRUCTURED_PROC 1  //  this gets us the new structured proc interfaces of 5.6 & later
  68 # include <sys/procfs.h>     //  see comment in <sys/procfs.h>
  69 
  70 #define MAX_PATH (2 * K)
  71 
  72 // for timer info max values which include all bits
  73 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
  74 
  75 #ifdef _GNU_SOURCE
  76 // See bug #6514594
  77 extern "C" int madvise(caddr_t, size_t, int);
  78 extern "C"  int memcntl(caddr_t addr, size_t len, int cmd, caddr_t  arg,
  79      int attr, int mask);
  80 #endif //_GNU_SOURCE
  81 
  82 /*
  83   MPSS Changes Start.
  84   The JVM binary needs to be built and run on pre-Solaris 9
  85   systems, but the constants needed by MPSS are only in Solaris 9
  86   header files.  They are textually replicated here to allow
  87   building on earlier systems.  Once building on Solaris 8 is
  88   no longer a requirement, these #defines can be replaced by ordinary
  89   system .h inclusion.
  90 
  91   In earlier versions of the  JDK and Solaris, we used ISM for large pages.
  92   But ISM requires shared memory to achieve this and thus has many caveats.
  93   MPSS is a fully transparent and is a cleaner way to get large pages.
  94   Although we still require keeping ISM for backward compatiblitiy as well as
  95   giving the opportunity to use large pages on older systems it is
  96   recommended that MPSS be used for Solaris 9 and above.
  97 
  98 */
  99 
 100 #ifndef MC_HAT_ADVISE
 101 
 102 struct memcntl_mha {
 103   uint_t          mha_cmd;        /* command(s) */
 104   uint_t          mha_flags;
 105   size_t          mha_pagesize;
 106 };
 107 #define MC_HAT_ADVISE   7       /* advise hat map size */
 108 #define MHA_MAPSIZE_VA  0x1     /* set preferred page size */
 109 #define MAP_ALIGN       0x200   /* addr specifies alignment */
 110 
 111 #endif
 112 // MPSS Changes End.
 113 
 114 
 115 // Here are some liblgrp types from sys/lgrp_user.h to be able to
 116 // compile on older systems without this header file.
 117 
 118 #ifndef MADV_ACCESS_LWP
 119 # define  MADV_ACCESS_LWP         7       /* next LWP to access heavily */
 120 #endif
 121 #ifndef MADV_ACCESS_MANY
 122 # define  MADV_ACCESS_MANY        8       /* many processes to access heavily */
 123 #endif
 124 
 125 #ifndef LGRP_RSRC_CPU
 126 # define LGRP_RSRC_CPU           0       /* CPU resources */
 127 #endif
 128 #ifndef LGRP_RSRC_MEM
 129 # define LGRP_RSRC_MEM           1       /* memory resources */
 130 #endif
 131 
 132 // Some more macros from sys/mman.h that are not present in Solaris 8.
 133 
 134 #ifndef MAX_MEMINFO_CNT
 135 /*
 136  * info_req request type definitions for meminfo
 137  * request types starting with MEMINFO_V are used for Virtual addresses
 138  * and should not be mixed with MEMINFO_PLGRP which is targeted for Physical
 139  * addresses
 140  */
 141 # define MEMINFO_SHIFT           16
 142 # define MEMINFO_MASK            (0xFF << MEMINFO_SHIFT)
 143 # define MEMINFO_VPHYSICAL       (0x01 << MEMINFO_SHIFT) /* get physical addr */
 144 # define MEMINFO_VLGRP           (0x02 << MEMINFO_SHIFT) /* get lgroup */
 145 # define MEMINFO_VPAGESIZE       (0x03 << MEMINFO_SHIFT) /* size of phys page */
 146 # define MEMINFO_VREPLCNT        (0x04 << MEMINFO_SHIFT) /* no. of replica */
 147 # define MEMINFO_VREPL           (0x05 << MEMINFO_SHIFT) /* physical replica */
 148 # define MEMINFO_VREPL_LGRP      (0x06 << MEMINFO_SHIFT) /* lgrp of replica */
 149 # define MEMINFO_PLGRP           (0x07 << MEMINFO_SHIFT) /* lgroup for paddr */
 150 
 151 /* maximum number of addresses meminfo() can process at a time */
 152 # define MAX_MEMINFO_CNT 256
 153 
 154 /* maximum number of request types */
 155 # define MAX_MEMINFO_REQ 31
 156 #endif
 157 
 158 // see thr_setprio(3T) for the basis of these numbers
 159 #define MinimumPriority 0
 160 #define NormalPriority  64
 161 #define MaximumPriority 127
 162 
 163 // Values for ThreadPriorityPolicy == 1
 164 int prio_policy1[MaxPriority+1] = { -99999, 0, 16, 32, 48, 64,
 165                                         80, 96, 112, 124, 127 };
 166 
 167 // System parameters used internally
 168 static clock_t clock_tics_per_sec = 100;
 169 
 170 // For diagnostics to print a message once. see run_periodic_checks
 171 static bool check_addr0_done = false;
 172 static sigset_t check_signal_done;
 173 static bool check_signals = true;
 174 
 175 address os::Solaris::handler_start;  // start pc of thr_sighndlrinfo
 176 address os::Solaris::handler_end;    // end pc of thr_sighndlrinfo
 177 
 178 address os::Solaris::_main_stack_base = NULL;  // 4352906 workaround
 179 
 180 
 181 // "default" initializers for missing libc APIs
 182 extern "C" {
 183   static int lwp_mutex_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
 184   static int lwp_mutex_destroy(mutex_t *mx)                 { return 0; }
 185 
 186   static int lwp_cond_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
 187   static int lwp_cond_destroy(cond_t *cv)                   { return 0; }
 188 }
 189 
 190 // "default" initializers for pthread-based synchronization
 191 extern "C" {
 192   static int pthread_mutex_default_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
 193   static int pthread_cond_default_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
 194 }
 195 
 196 // Thread Local Storage
 197 // This is common to all Solaris platforms so it is defined here,
 198 // in this common file.
 199 // The declarations are in the os_cpu threadLS*.hpp files.
 200 //
 201 // Static member initialization for TLS
 202 Thread* ThreadLocalStorage::_get_thread_cache[ThreadLocalStorage::_pd_cache_size] = {NULL};
 203 
 204 #ifndef PRODUCT
 205 #define _PCT(n,d)       ((100.0*(double)(n))/(double)(d))
 206 
 207 int ThreadLocalStorage::_tcacheHit = 0;
 208 int ThreadLocalStorage::_tcacheMiss = 0;
 209 
 210 void ThreadLocalStorage::print_statistics() {
 211   int total = _tcacheMiss+_tcacheHit;
 212   tty->print_cr("Thread cache hits %d misses %d total %d percent %f\n",
 213                 _tcacheHit, _tcacheMiss, total, _PCT(_tcacheHit, total));
 214 }
 215 #undef _PCT
 216 #endif // PRODUCT
 217 
 218 Thread* ThreadLocalStorage::get_thread_via_cache_slowly(uintptr_t raw_id,
 219                                                         int index) {
 220   Thread *thread = get_thread_slow();
 221   if (thread != NULL) {
 222     address sp = os::current_stack_pointer();
 223     guarantee(thread->_stack_base == NULL ||
 224               (sp <= thread->_stack_base &&
 225                  sp >= thread->_stack_base - thread->_stack_size) ||
 226                is_error_reported(),
 227               "sp must be inside of selected thread stack");
 228 
 229     thread->_self_raw_id = raw_id;  // mark for quick retrieval
 230     _get_thread_cache[ index ] = thread;
 231   }
 232   return thread;
 233 }
 234 
 235 
 236 static const double all_zero[ sizeof(Thread) / sizeof(double) + 1 ] = {0};
 237 #define NO_CACHED_THREAD ((Thread*)all_zero)
 238 
 239 void ThreadLocalStorage::pd_set_thread(Thread* thread) {
 240 
 241   // Store the new value before updating the cache to prevent a race
 242   // between get_thread_via_cache_slowly() and this store operation.
 243   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
 244 
 245   // Update thread cache with new thread if setting on thread create,
 246   // or NO_CACHED_THREAD (zeroed) thread if resetting thread on exit.
 247   uintptr_t raw = pd_raw_thread_id();
 248   int ix = pd_cache_index(raw);
 249   _get_thread_cache[ix] = thread == NULL ? NO_CACHED_THREAD : thread;
 250 }
 251 
 252 void ThreadLocalStorage::pd_init() {
 253   for (int i = 0; i < _pd_cache_size; i++) {
 254     _get_thread_cache[i] = NO_CACHED_THREAD;
 255   }
 256 }
 257 
 258 // Invalidate all the caches (happens to be the same as pd_init).
 259 void ThreadLocalStorage::pd_invalidate_all() { pd_init(); }
 260 
 261 #undef NO_CACHED_THREAD
 262 
 263 // END Thread Local Storage
 264 
 265 static inline size_t adjust_stack_size(address base, size_t size) {
 266   if ((ssize_t)size < 0) {
 267     // 4759953: Compensate for ridiculous stack size.
 268     size = max_intx;
 269   }
 270   if (size > (size_t)base) {
 271     // 4812466: Make sure size doesn't allow the stack to wrap the address space.
 272     size = (size_t)base;
 273   }
 274   return size;
 275 }
 276 
 277 static inline stack_t get_stack_info() {
 278   stack_t st;
 279   int retval = thr_stksegment(&st);
 280   st.ss_size = adjust_stack_size((address)st.ss_sp, st.ss_size);
 281   assert(retval == 0, "incorrect return value from thr_stksegment");
 282   assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
 283   assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
 284   return st;
 285 }
 286 
 287 address os::current_stack_base() {
 288   int r = thr_main() ;
 289   guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
 290   bool is_primordial_thread = r;
 291 
 292   // Workaround 4352906, avoid calls to thr_stksegment by
 293   // thr_main after the first one (it looks like we trash
 294   // some data, causing the value for ss_sp to be incorrect).
 295   if (!is_primordial_thread || os::Solaris::_main_stack_base == NULL) {
 296     stack_t st = get_stack_info();
 297     if (is_primordial_thread) {
 298       // cache initial value of stack base
 299       os::Solaris::_main_stack_base = (address)st.ss_sp;
 300     }
 301     return (address)st.ss_sp;
 302   } else {
 303     guarantee(os::Solaris::_main_stack_base != NULL, "Attempt to use null cached stack base");
 304     return os::Solaris::_main_stack_base;
 305   }
 306 }
 307 
 308 size_t os::current_stack_size() {
 309   size_t size;
 310 
 311   int r = thr_main() ;
 312   guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
 313   if(!r) {
 314     size = get_stack_info().ss_size;
 315   } else {
 316     struct rlimit limits;
 317     getrlimit(RLIMIT_STACK, &limits);
 318     size = adjust_stack_size(os::Solaris::_main_stack_base, (size_t)limits.rlim_cur);
 319   }
 320   // base may not be page aligned
 321   address base = current_stack_base();
 322   address bottom = (address)align_size_up((intptr_t)(base - size), os::vm_page_size());;
 323   return (size_t)(base - bottom);
 324 }
 325 
 326 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
 327   return localtime_r(clock, res);
 328 }
 329 
 330 // interruptible infrastructure
 331 
 332 // setup_interruptible saves the thread state before going into an
 333 // interruptible system call.
 334 // The saved state is used to restore the thread to
 335 // its former state whether or not an interrupt is received.
 336 // Used by classloader os::read
 337 // hpi calls skip this layer and stay in _thread_in_native
 338 
 339 void os::Solaris::setup_interruptible(JavaThread* thread) {
 340 
 341   JavaThreadState thread_state = thread->thread_state();
 342 
 343   assert(thread_state != _thread_blocked, "Coming from the wrong thread");
 344   assert(thread_state != _thread_in_native, "Native threads skip setup_interruptible");
 345   OSThread* osthread = thread->osthread();
 346   osthread->set_saved_interrupt_thread_state(thread_state);
 347   thread->frame_anchor()->make_walkable(thread);
 348   ThreadStateTransition::transition(thread, thread_state, _thread_blocked);
 349 }
 350 
 351 // Version of setup_interruptible() for threads that are already in
 352 // _thread_blocked. Used by os_sleep().
 353 void os::Solaris::setup_interruptible_already_blocked(JavaThread* thread) {
 354   thread->frame_anchor()->make_walkable(thread);
 355 }
 356 
 357 JavaThread* os::Solaris::setup_interruptible() {
 358   JavaThread* thread = (JavaThread*)ThreadLocalStorage::thread();
 359   setup_interruptible(thread);
 360   return thread;
 361 }
 362 
 363 void os::Solaris::try_enable_extended_io() {
 364   typedef int (*enable_extended_FILE_stdio_t)(int, int);
 365 
 366   if (!UseExtendedFileIO) {
 367     return;
 368   }
 369 
 370   enable_extended_FILE_stdio_t enabler =
 371     (enable_extended_FILE_stdio_t) dlsym(RTLD_DEFAULT,
 372                                          "enable_extended_FILE_stdio");
 373   if (enabler) {
 374     enabler(-1, -1);
 375   }
 376 }
 377 
 378 
 379 #ifdef ASSERT
 380 
 381 JavaThread* os::Solaris::setup_interruptible_native() {
 382   JavaThread* thread = (JavaThread*)ThreadLocalStorage::thread();
 383   JavaThreadState thread_state = thread->thread_state();
 384   assert(thread_state == _thread_in_native, "Assumed thread_in_native");
 385   return thread;
 386 }
 387 
 388 void os::Solaris::cleanup_interruptible_native(JavaThread* thread) {
 389   JavaThreadState thread_state = thread->thread_state();
 390   assert(thread_state == _thread_in_native, "Assumed thread_in_native");
 391 }
 392 #endif
 393 
 394 // cleanup_interruptible reverses the effects of setup_interruptible
 395 // setup_interruptible_already_blocked() does not need any cleanup.
 396 
 397 void os::Solaris::cleanup_interruptible(JavaThread* thread) {
 398   OSThread* osthread = thread->osthread();
 399 
 400   ThreadStateTransition::transition(thread, _thread_blocked, osthread->saved_interrupt_thread_state());
 401 }
 402 
 403 // I/O interruption related counters called in _INTERRUPTIBLE
 404 
 405 void os::Solaris::bump_interrupted_before_count() {
 406   RuntimeService::record_interrupted_before_count();
 407 }
 408 
 409 void os::Solaris::bump_interrupted_during_count() {
 410   RuntimeService::record_interrupted_during_count();
 411 }
 412 
 413 static int _processors_online = 0;
 414 
 415          jint os::Solaris::_os_thread_limit = 0;
 416 volatile jint os::Solaris::_os_thread_count = 0;
 417 
 418 julong os::available_memory() {
 419   return Solaris::available_memory();
 420 }
 421 
 422 julong os::Solaris::available_memory() {
 423   return (julong)sysconf(_SC_AVPHYS_PAGES) * os::vm_page_size();
 424 }
 425 
 426 julong os::Solaris::_physical_memory = 0;
 427 
 428 julong os::physical_memory() {
 429    return Solaris::physical_memory();
 430 }
 431 
 432 julong os::allocatable_physical_memory(julong size) {
 433 #ifdef _LP64
 434    return size;
 435 #else
 436    julong result = MIN2(size, (julong)3835*M);
 437    if (!is_allocatable(result)) {
 438      // Memory allocations will be aligned but the alignment
 439      // is not known at this point.  Alignments will
 440      // be at most to LargePageSizeInBytes.  Protect
 441      // allocations from alignments up to illegal
 442      // values. If at this point 2G is illegal.
 443      julong reasonable_size = (julong)2*G - 2 * LargePageSizeInBytes;
 444      result =  MIN2(size, reasonable_size);
 445    }
 446    return result;
 447 #endif
 448 }
 449 
 450 static hrtime_t first_hrtime = 0;
 451 static const hrtime_t hrtime_hz = 1000*1000*1000;
 452 const int LOCK_BUSY = 1;
 453 const int LOCK_FREE = 0;
 454 const int LOCK_INVALID = -1;
 455 static volatile hrtime_t max_hrtime = 0;
 456 static volatile int max_hrtime_lock = LOCK_FREE;     // Update counter with LSB as lock-in-progress
 457 
 458 
 459 void os::Solaris::initialize_system_info() {
 460   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 461   _processors_online = sysconf (_SC_NPROCESSORS_ONLN);
 462   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 463 }
 464 
 465 int os::active_processor_count() {
 466   int online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
 467   pid_t pid = getpid();
 468   psetid_t pset = PS_NONE;
 469   // Are we running in a processor set or is there any processor set around?
 470   if (pset_bind(PS_QUERY, P_PID, pid, &pset) == 0) {
 471     uint_t pset_cpus;
 472     // Query the number of cpus available to us.
 473     if (pset_info(pset, NULL, &pset_cpus, NULL) == 0) {
 474       assert(pset_cpus > 0 && pset_cpus <= online_cpus, "sanity check");
 475       _processors_online = pset_cpus;
 476       return pset_cpus;
 477     }
 478   }
 479   // Otherwise return number of online cpus
 480   return online_cpus;
 481 }
 482 
 483 static bool find_processors_in_pset(psetid_t        pset,
 484                                     processorid_t** id_array,
 485                                     uint_t*         id_length) {
 486   bool result = false;
 487   // Find the number of processors in the processor set.
 488   if (pset_info(pset, NULL, id_length, NULL) == 0) {
 489     // Make up an array to hold their ids.
 490     *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length);
 491     // Fill in the array with their processor ids.
 492     if (pset_info(pset, NULL, id_length, *id_array) == 0) {
 493       result = true;
 494     }
 495   }
 496   return result;
 497 }
 498 
 499 // Callers of find_processors_online() must tolerate imprecise results --
 500 // the system configuration can change asynchronously because of DR
 501 // or explicit psradm operations.
 502 //
 503 // We also need to take care that the loop (below) terminates as the
 504 // number of processors online can change between the _SC_NPROCESSORS_ONLN
 505 // request and the loop that builds the list of processor ids.   Unfortunately
 506 // there's no reliable way to determine the maximum valid processor id,
 507 // so we use a manifest constant, MAX_PROCESSOR_ID, instead.  See p_online
 508 // man pages, which claim the processor id set is "sparse, but
 509 // not too sparse".  MAX_PROCESSOR_ID is used to ensure that we eventually
 510 // exit the loop.
 511 //
 512 // In the future we'll be able to use sysconf(_SC_CPUID_MAX), but that's
 513 // not available on S8.0.
 514 
 515 static bool find_processors_online(processorid_t** id_array,
 516                                    uint*           id_length) {
 517   const processorid_t MAX_PROCESSOR_ID = 100000 ;
 518   // Find the number of processors online.
 519   *id_length = sysconf(_SC_NPROCESSORS_ONLN);
 520   // Make up an array to hold their ids.
 521   *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length);
 522   // Processors need not be numbered consecutively.
 523   long found = 0;
 524   processorid_t next = 0;
 525   while (found < *id_length && next < MAX_PROCESSOR_ID) {
 526     processor_info_t info;
 527     if (processor_info(next, &info) == 0) {
 528       // NB, PI_NOINTR processors are effectively online ...
 529       if (info.pi_state == P_ONLINE || info.pi_state == P_NOINTR) {
 530         (*id_array)[found] = next;
 531         found += 1;
 532       }
 533     }
 534     next += 1;
 535   }
 536   if (found < *id_length) {
 537       // The loop above didn't identify the expected number of processors.
 538       // We could always retry the operation, calling sysconf(_SC_NPROCESSORS_ONLN)
 539       // and re-running the loop, above, but there's no guarantee of progress
 540       // if the system configuration is in flux.  Instead, we just return what
 541       // we've got.  Note that in the worst case find_processors_online() could
 542       // return an empty set.  (As a fall-back in the case of the empty set we
 543       // could just return the ID of the current processor).
 544       *id_length = found ;
 545   }
 546 
 547   return true;
 548 }
 549 
 550 static bool assign_distribution(processorid_t* id_array,
 551                                 uint           id_length,
 552                                 uint*          distribution,
 553                                 uint           distribution_length) {
 554   // We assume we can assign processorid_t's to uint's.
 555   assert(sizeof(processorid_t) == sizeof(uint),
 556          "can't convert processorid_t to uint");
 557   // Quick check to see if we won't succeed.
 558   if (id_length < distribution_length) {
 559     return false;
 560   }
 561   // Assign processor ids to the distribution.
 562   // Try to shuffle processors to distribute work across boards,
 563   // assuming 4 processors per board.
 564   const uint processors_per_board = ProcessDistributionStride;
 565   // Find the maximum processor id.
 566   processorid_t max_id = 0;
 567   for (uint m = 0; m < id_length; m += 1) {
 568     max_id = MAX2(max_id, id_array[m]);
 569   }
 570   // The next id, to limit loops.
 571   const processorid_t limit_id = max_id + 1;
 572   // Make up markers for available processors.
 573   bool* available_id = NEW_C_HEAP_ARRAY(bool, limit_id);
 574   for (uint c = 0; c < limit_id; c += 1) {
 575     available_id[c] = false;
 576   }
 577   for (uint a = 0; a < id_length; a += 1) {
 578     available_id[id_array[a]] = true;
 579   }
 580   // Step by "boards", then by "slot", copying to "assigned".
 581   // NEEDS_CLEANUP: The assignment of processors should be stateful,
 582   //                remembering which processors have been assigned by
 583   //                previous calls, etc., so as to distribute several
 584   //                independent calls of this method.  What we'd like is
 585   //                It would be nice to have an API that let us ask
 586   //                how many processes are bound to a processor,
 587   //                but we don't have that, either.
 588   //                In the short term, "board" is static so that
 589   //                subsequent distributions don't all start at board 0.
 590   static uint board = 0;
 591   uint assigned = 0;
 592   // Until we've found enough processors ....
 593   while (assigned < distribution_length) {
 594     // ... find the next available processor in the board.
 595     for (uint slot = 0; slot < processors_per_board; slot += 1) {
 596       uint try_id = board * processors_per_board + slot;
 597       if ((try_id < limit_id) && (available_id[try_id] == true)) {
 598         distribution[assigned] = try_id;
 599         available_id[try_id] = false;
 600         assigned += 1;
 601         break;
 602       }
 603     }
 604     board += 1;
 605     if (board * processors_per_board + 0 >= limit_id) {
 606       board = 0;
 607     }
 608   }
 609   if (available_id != NULL) {
 610     FREE_C_HEAP_ARRAY(bool, available_id);
 611   }
 612   return true;
 613 }
 614 
 615 bool os::distribute_processes(uint length, uint* distribution) {
 616   bool result = false;
 617   // Find the processor id's of all the available CPUs.
 618   processorid_t* id_array  = NULL;
 619   uint           id_length = 0;
 620   // There are some races between querying information and using it,
 621   // since processor sets can change dynamically.
 622   psetid_t pset = PS_NONE;
 623   // Are we running in a processor set?
 624   if ((pset_bind(PS_QUERY, P_PID, P_MYID, &pset) == 0) && pset != PS_NONE) {
 625     result = find_processors_in_pset(pset, &id_array, &id_length);
 626   } else {
 627     result = find_processors_online(&id_array, &id_length);
 628   }
 629   if (result == true) {
 630     if (id_length >= length) {
 631       result = assign_distribution(id_array, id_length, distribution, length);
 632     } else {
 633       result = false;
 634     }
 635   }
 636   if (id_array != NULL) {
 637     FREE_C_HEAP_ARRAY(processorid_t, id_array);
 638   }
 639   return result;
 640 }
 641 
 642 bool os::bind_to_processor(uint processor_id) {
 643   // We assume that a processorid_t can be stored in a uint.
 644   assert(sizeof(uint) == sizeof(processorid_t),
 645          "can't convert uint to processorid_t");
 646   int bind_result =
 647     processor_bind(P_LWPID,                       // bind LWP.
 648                    P_MYID,                        // bind current LWP.
 649                    (processorid_t) processor_id,  // id.
 650                    NULL);                         // don't return old binding.
 651   return (bind_result == 0);
 652 }
 653 
 654 bool os::getenv(const char* name, char* buffer, int len) {
 655   char* val = ::getenv( name );
 656   if ( val == NULL
 657   ||   strlen(val) + 1  >  len ) {
 658     if (len > 0)  buffer[0] = 0; // return a null string
 659     return false;
 660   }
 661   strcpy( buffer, val );
 662   return true;
 663 }
 664 
 665 
 666 // Return true if user is running as root.
 667 
 668 bool os::have_special_privileges() {
 669   static bool init = false;
 670   static bool privileges = false;
 671   if (!init) {
 672     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 673     init = true;
 674   }
 675   return privileges;
 676 }
 677 
 678 
 679 void os::init_system_properties_values() {
 680   char arch[12];
 681   sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 682 
 683   // The next steps are taken in the product version:
 684   //
 685   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 686   // This library should be located at:
 687   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 688   //
 689   // If "/jre/lib/" appears at the right place in the path, then we
 690   // assume libjvm[_g].so is installed in a JDK and we use this path.
 691   //
 692   // Otherwise exit with message: "Could not create the Java virtual machine."
 693   //
 694   // The following extra steps are taken in the debugging version:
 695   //
 696   // If "/jre/lib/" does NOT appear at the right place in the path
 697   // instead of exit check for $JAVA_HOME environment variable.
 698   //
 699   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 700   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 701   // it looks like libjvm[_g].so is installed there
 702   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 703   //
 704   // Otherwise exit.
 705   //
 706   // Important note: if the location of libjvm.so changes this
 707   // code needs to be changed accordingly.
 708 
 709   // The next few definitions allow the code to be verbatim:
 710 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
 711 #define free(p) FREE_C_HEAP_ARRAY(char, p)
 712 #define getenv(n) ::getenv(n)
 713 
 714 #define EXTENSIONS_DIR  "/lib/ext"
 715 #define ENDORSED_DIR    "/lib/endorsed"
 716 #define COMMON_DIR      "/usr/jdk/packages"
 717 
 718   {
 719     /* sysclasspath, java_home, dll_dir */
 720     {
 721         char *home_path;
 722         char *dll_path;
 723         char *pslash;
 724         char buf[MAXPATHLEN];
 725         os::jvm_path(buf, sizeof(buf));
 726 
 727         // Found the full path to libjvm.so.
 728         // Now cut the path to <java_home>/jre if we can.
 729         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 730         pslash = strrchr(buf, '/');
 731         if (pslash != NULL)
 732             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 733         dll_path = malloc(strlen(buf) + 1);
 734         if (dll_path == NULL)
 735             return;
 736         strcpy(dll_path, buf);
 737         Arguments::set_dll_dir(dll_path);
 738 
 739         if (pslash != NULL) {
 740             pslash = strrchr(buf, '/');
 741             if (pslash != NULL) {
 742                 *pslash = '\0';       /* get rid of /<arch> */
 743                 pslash = strrchr(buf, '/');
 744                 if (pslash != NULL)
 745                     *pslash = '\0';   /* get rid of /lib */
 746             }
 747         }
 748 
 749         home_path = malloc(strlen(buf) + 1);
 750         if (home_path == NULL)
 751             return;
 752         strcpy(home_path, buf);
 753         Arguments::set_java_home(home_path);
 754 
 755         if (!set_boot_path('/', ':'))
 756             return;
 757     }
 758 
 759     /*
 760      * Where to look for native libraries
 761      */
 762     {
 763       // Use dlinfo() to determine the correct java.library.path.
 764       //
 765       // If we're launched by the Java launcher, and the user
 766       // does not set java.library.path explicitly on the commandline,
 767       // the Java launcher sets LD_LIBRARY_PATH for us and unsets
 768       // LD_LIBRARY_PATH_32 and LD_LIBRARY_PATH_64.  In this case
 769       // dlinfo returns LD_LIBRARY_PATH + crle settings (including
 770       // /usr/lib), which is exactly what we want.
 771       //
 772       // If the user does set java.library.path, it completely
 773       // overwrites this setting, and always has.
 774       //
 775       // If we're not launched by the Java launcher, we may
 776       // get here with any/all of the LD_LIBRARY_PATH[_32|64]
 777       // settings.  Again, dlinfo does exactly what we want.
 778 
 779       Dl_serinfo     _info, *info = &_info;
 780       Dl_serpath     *path;
 781       char*          library_path;
 782       char           *common_path;
 783       int            i;
 784 
 785       // determine search path count and required buffer size
 786       if (dlinfo(RTLD_SELF, RTLD_DI_SERINFOSIZE, (void *)info) == -1) {
 787         vm_exit_during_initialization("dlinfo SERINFOSIZE request", dlerror());
 788       }
 789 
 790       // allocate new buffer and initialize
 791       info = (Dl_serinfo*)malloc(_info.dls_size);
 792       if (info == NULL) {
 793         vm_exit_out_of_memory(_info.dls_size,
 794                               "init_system_properties_values info");
 795       }
 796       info->dls_size = _info.dls_size;
 797       info->dls_cnt = _info.dls_cnt;
 798 
 799       // obtain search path information
 800       if (dlinfo(RTLD_SELF, RTLD_DI_SERINFO, (void *)info) == -1) {
 801         free(info);
 802         vm_exit_during_initialization("dlinfo SERINFO request", dlerror());
 803       }
 804 
 805       path = &info->dls_serpath[0];
 806 
 807       // Note: Due to a legacy implementation, most of the library path
 808       // is set in the launcher.  This was to accomodate linking restrictions
 809       // on legacy Solaris implementations (which are no longer supported).
 810       // Eventually, all the library path setting will be done here.
 811       //
 812       // However, to prevent the proliferation of improperly built native
 813       // libraries, the new path component /usr/jdk/packages is added here.
 814 
 815       // Determine the actual CPU architecture.
 816       char cpu_arch[12];
 817       sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
 818 #ifdef _LP64
 819       // If we are a 64-bit vm, perform the following translations:
 820       //   sparc   -> sparcv9
 821       //   i386    -> amd64
 822       if (strcmp(cpu_arch, "sparc") == 0)
 823         strcat(cpu_arch, "v9");
 824       else if (strcmp(cpu_arch, "i386") == 0)
 825         strcpy(cpu_arch, "amd64");
 826 #endif
 827 
 828       // Construct the invariant part of ld_library_path. Note that the
 829       // space for the colon and the trailing null are provided by the
 830       // nulls included by the sizeof operator.
 831       size_t bufsize = sizeof(COMMON_DIR) + sizeof("/lib/") + strlen(cpu_arch);
 832       common_path = malloc(bufsize);
 833       if (common_path == NULL) {
 834         free(info);
 835         vm_exit_out_of_memory(bufsize,
 836                               "init_system_properties_values common_path");
 837       }
 838       sprintf(common_path, COMMON_DIR "/lib/%s", cpu_arch);
 839 
 840       // struct size is more than sufficient for the path components obtained
 841       // through the dlinfo() call, so only add additional space for the path
 842       // components explicitly added here.
 843       bufsize = info->dls_size + strlen(common_path);
 844       library_path = malloc(bufsize);
 845       if (library_path == NULL) {
 846         free(info);
 847         free(common_path);
 848         vm_exit_out_of_memory(bufsize,
 849                               "init_system_properties_values library_path");
 850       }
 851       library_path[0] = '\0';
 852 
 853       // Construct the desired Java library path from the linker's library
 854       // search path.
 855       //
 856       // For compatibility, it is optimal that we insert the additional path
 857       // components specific to the Java VM after those components specified
 858       // in LD_LIBRARY_PATH (if any) but before those added by the ld.so
 859       // infrastructure.
 860       if (info->dls_cnt == 0) { // Not sure this can happen, but allow for it
 861         strcpy(library_path, common_path);
 862       } else {
 863         int inserted = 0;
 864         for (i = 0; i < info->dls_cnt; i++, path++) {
 865           uint_t flags = path->dls_flags & LA_SER_MASK;
 866           if (((flags & LA_SER_LIBPATH) == 0) && !inserted) {
 867             strcat(library_path, common_path);
 868             strcat(library_path, os::path_separator());
 869             inserted = 1;
 870           }
 871           strcat(library_path, path->dls_name);
 872           strcat(library_path, os::path_separator());
 873         }
 874         // eliminate trailing path separator
 875         library_path[strlen(library_path)-1] = '\0';
 876       }
 877 
 878       // happens before argument parsing - can't use a trace flag
 879       // tty->print_raw("init_system_properties_values: native lib path: ");
 880       // tty->print_raw_cr(library_path);
 881 
 882       // callee copies into its own buffer
 883       Arguments::set_library_path(library_path);
 884 
 885       free(common_path);
 886       free(library_path);
 887       free(info);
 888     }
 889 
 890     /*
 891      * Extensions directories.
 892      *
 893      * Note that the space for the colon and the trailing null are provided
 894      * by the nulls included by the sizeof operator (so actually one byte more
 895      * than necessary is allocated).
 896      */
 897     {
 898         char *buf = (char *) malloc(strlen(Arguments::get_java_home()) +
 899             sizeof(EXTENSIONS_DIR) + sizeof(COMMON_DIR) +
 900             sizeof(EXTENSIONS_DIR));
 901         sprintf(buf, "%s" EXTENSIONS_DIR ":" COMMON_DIR EXTENSIONS_DIR,
 902             Arguments::get_java_home());
 903         Arguments::set_ext_dirs(buf);
 904     }
 905 
 906     /* Endorsed standards default directory. */
 907     {
 908         char * buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 909         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 910         Arguments::set_endorsed_dirs(buf);
 911     }
 912   }
 913 
 914 #undef malloc
 915 #undef free
 916 #undef getenv
 917 #undef EXTENSIONS_DIR
 918 #undef ENDORSED_DIR
 919 #undef COMMON_DIR
 920 
 921 }
 922 
 923 void os::breakpoint() {
 924   BREAKPOINT;
 925 }
 926 
 927 bool os::obsolete_option(const JavaVMOption *option)
 928 {
 929   if (!strncmp(option->optionString, "-Xt", 3)) {
 930     return true;
 931   } else if (!strncmp(option->optionString, "-Xtm", 4)) {
 932     return true;
 933   } else if (!strncmp(option->optionString, "-Xverifyheap", 12)) {
 934     return true;
 935   } else if (!strncmp(option->optionString, "-Xmaxjitcodesize", 16)) {
 936     return true;
 937   }
 938   return false;
 939 }
 940 
 941 bool os::Solaris::valid_stack_address(Thread* thread, address sp) {
 942   address  stackStart  = (address)thread->stack_base();
 943   address  stackEnd    = (address)(stackStart - (address)thread->stack_size());
 944   if (sp < stackStart && sp >= stackEnd ) return true;
 945   return false;
 946 }
 947 
 948 extern "C" void breakpoint() {
 949   // use debugger to set breakpoint here
 950 }
 951 
 952 // Returns an estimate of the current stack pointer. Result must be guaranteed to
 953 // point into the calling threads stack, and be no lower than the current stack
 954 // pointer.
 955 address os::current_stack_pointer() {
 956   volatile int dummy;
 957   address sp = (address)&dummy + 8;     // %%%% need to confirm if this is right
 958   return sp;
 959 }
 960 
 961 static thread_t main_thread;
 962 
 963 // Thread start routine for all new Java threads
 964 extern "C" void* java_start(void* thread_addr) {
 965   // Try to randomize the cache line index of hot stack frames.
 966   // This helps when threads of the same stack traces evict each other's
 967   // cache lines. The threads can be either from the same JVM instance, or
 968   // from different JVM instances. The benefit is especially true for
 969   // processors with hyperthreading technology.
 970   static int counter = 0;
 971   int pid = os::current_process_id();
 972   alloca(((pid ^ counter++) & 7) * 128);
 973 
 974   int prio;
 975   Thread* thread = (Thread*)thread_addr;
 976   OSThread* osthr = thread->osthread();
 977 
 978   osthr->set_lwp_id( _lwp_self() );  // Store lwp in case we are bound
 979   thread->_schedctl = (void *) schedctl_init () ;
 980 
 981   if (UseNUMA) {
 982     int lgrp_id = os::numa_get_group_id();
 983     if (lgrp_id != -1) {
 984       thread->set_lgrp_id(lgrp_id);
 985     }
 986   }
 987 
 988   // If the creator called set priority before we started,
 989   // we need to call set priority now that we have an lwp.
 990   // Get the priority from libthread and set the priority
 991   // for the new Solaris lwp.
 992   if ( osthr->thread_id() != -1 ) {
 993     if ( UseThreadPriorities ) {
 994       thr_getprio(osthr->thread_id(), &prio);
 995       if (ThreadPriorityVerbose) {
 996         tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT ", setting priority: %d\n",
 997                       osthr->thread_id(), osthr->lwp_id(), prio );
 998       }
 999       os::set_native_priority(thread, prio);
1000     }
1001   } else if (ThreadPriorityVerbose) {
1002     warning("Can't set priority in _start routine, thread id hasn't been set\n");
1003   }
1004 
1005   assert(osthr->get_state() == RUNNABLE, "invalid os thread state");
1006 
1007   // initialize signal mask for this thread
1008   os::Solaris::hotspot_sigmask(thread);
1009 
1010   thread->run();
1011 
1012   // One less thread is executing
1013   // When the VMThread gets here, the main thread may have already exited
1014   // which frees the CodeHeap containing the Atomic::dec code
1015   if (thread != VMThread::vm_thread() && VMThread::vm_thread() != NULL) {
1016     Atomic::dec(&os::Solaris::_os_thread_count);
1017   }
1018 
1019   if (UseDetachedThreads) {
1020     thr_exit(NULL);
1021     ShouldNotReachHere();
1022   }
1023   return NULL;
1024 }
1025 
1026 static OSThread* create_os_thread(Thread* thread, thread_t thread_id) {
1027   // Allocate the OSThread object
1028   OSThread* osthread = new OSThread(NULL, NULL);
1029   if (osthread == NULL) return NULL;
1030 
1031   // Store info on the Solaris thread into the OSThread
1032   osthread->set_thread_id(thread_id);
1033   osthread->set_lwp_id(_lwp_self());
1034   thread->_schedctl = (void *) schedctl_init () ;
1035 
1036   if (UseNUMA) {
1037     int lgrp_id = os::numa_get_group_id();
1038     if (lgrp_id != -1) {
1039       thread->set_lgrp_id(lgrp_id);
1040     }
1041   }
1042 
1043   if ( ThreadPriorityVerbose ) {
1044     tty->print_cr("In create_os_thread, Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT "\n",
1045                   osthread->thread_id(), osthread->lwp_id() );
1046   }
1047 
1048   // Initial thread state is INITIALIZED, not SUSPENDED
1049   osthread->set_state(INITIALIZED);
1050 
1051   return osthread;
1052 }
1053 
1054 void os::Solaris::hotspot_sigmask(Thread* thread) {
1055 
1056   //Save caller's signal mask
1057   sigset_t sigmask;
1058   thr_sigsetmask(SIG_SETMASK, NULL, &sigmask);
1059   OSThread *osthread = thread->osthread();
1060   osthread->set_caller_sigmask(sigmask);
1061 
1062   thr_sigsetmask(SIG_UNBLOCK, os::Solaris::unblocked_signals(), NULL);
1063   if (!ReduceSignalUsage) {
1064     if (thread->is_VM_thread()) {
1065       // Only the VM thread handles BREAK_SIGNAL ...
1066       thr_sigsetmask(SIG_UNBLOCK, vm_signals(), NULL);
1067     } else {
1068       // ... all other threads block BREAK_SIGNAL
1069       assert(!sigismember(vm_signals(), SIGINT), "SIGINT should not be blocked");
1070       thr_sigsetmask(SIG_BLOCK, vm_signals(), NULL);
1071     }
1072   }
1073 }
1074 
1075 bool os::create_attached_thread(JavaThread* thread) {
1076 #ifdef ASSERT
1077   thread->verify_not_published();
1078 #endif
1079   OSThread* osthread = create_os_thread(thread, thr_self());
1080   if (osthread == NULL) {
1081      return false;
1082   }
1083 
1084   // Initial thread state is RUNNABLE
1085   osthread->set_state(RUNNABLE);
1086   thread->set_osthread(osthread);
1087 
1088   // initialize signal mask for this thread
1089   // and save the caller's signal mask
1090   os::Solaris::hotspot_sigmask(thread);
1091 
1092   return true;
1093 }
1094 
1095 bool os::create_main_thread(JavaThread* thread) {
1096 #ifdef ASSERT
1097   thread->verify_not_published();
1098 #endif
1099   if (_starting_thread == NULL) {
1100     _starting_thread = create_os_thread(thread, main_thread);
1101      if (_starting_thread == NULL) {
1102         return false;
1103      }
1104   }
1105 
1106   // The primodial thread is runnable from the start
1107   _starting_thread->set_state(RUNNABLE);
1108 
1109   thread->set_osthread(_starting_thread);
1110 
1111   // initialize signal mask for this thread
1112   // and save the caller's signal mask
1113   os::Solaris::hotspot_sigmask(thread);
1114 
1115   return true;
1116 }
1117 
1118 // _T2_libthread is true if we believe we are running with the newer
1119 // SunSoft lwp/libthread.so (2.8 patch, 2.9 default)
1120 bool os::Solaris::_T2_libthread = false;
1121 
1122 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
1123   // Allocate the OSThread object
1124   OSThread* osthread = new OSThread(NULL, NULL);
1125   if (osthread == NULL) {
1126     return false;
1127   }
1128 
1129   if ( ThreadPriorityVerbose ) {
1130     char *thrtyp;
1131     switch ( thr_type ) {
1132       case vm_thread:
1133         thrtyp = (char *)"vm";
1134         break;
1135       case cgc_thread:
1136         thrtyp = (char *)"cgc";
1137         break;
1138       case pgc_thread:
1139         thrtyp = (char *)"pgc";
1140         break;
1141       case java_thread:
1142         thrtyp = (char *)"java";
1143         break;
1144       case compiler_thread:
1145         thrtyp = (char *)"compiler";
1146         break;
1147       case watcher_thread:
1148         thrtyp = (char *)"watcher";
1149         break;
1150       default:
1151         thrtyp = (char *)"unknown";
1152         break;
1153     }
1154     tty->print_cr("In create_thread, creating a %s thread\n", thrtyp);
1155   }
1156 
1157   // Calculate stack size if it's not specified by caller.
1158   if (stack_size == 0) {
1159     // The default stack size 1M (2M for LP64).
1160     stack_size = (BytesPerWord >> 2) * K * K;
1161 
1162     switch (thr_type) {
1163     case os::java_thread:
1164       // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
1165       if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
1166       break;
1167     case os::compiler_thread:
1168       if (CompilerThreadStackSize > 0) {
1169         stack_size = (size_t)(CompilerThreadStackSize * K);
1170         break;
1171       } // else fall through:
1172         // use VMThreadStackSize if CompilerThreadStackSize is not defined
1173     case os::vm_thread:
1174     case os::pgc_thread:
1175     case os::cgc_thread:
1176     case os::watcher_thread:
1177       if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
1178       break;
1179     }
1180   }
1181   stack_size = MAX2(stack_size, os::Solaris::min_stack_allowed);
1182 
1183   // Initial state is ALLOCATED but not INITIALIZED
1184   osthread->set_state(ALLOCATED);
1185 
1186   if (os::Solaris::_os_thread_count > os::Solaris::_os_thread_limit) {
1187     // We got lots of threads. Check if we still have some address space left.
1188     // Need to be at least 5Mb of unreserved address space. We do check by
1189     // trying to reserve some.
1190     const size_t VirtualMemoryBangSize = 20*K*K;
1191     char* mem = os::reserve_memory(VirtualMemoryBangSize);
1192     if (mem == NULL) {
1193       delete osthread;
1194       return false;
1195     } else {
1196       // Release the memory again
1197       os::release_memory(mem, VirtualMemoryBangSize);
1198     }
1199   }
1200 
1201   // Setup osthread because the child thread may need it.
1202   thread->set_osthread(osthread);
1203 
1204   // Create the Solaris thread
1205   // explicit THR_BOUND for T2_libthread case in case
1206   // that assumption is not accurate, but our alternate signal stack
1207   // handling is based on it which must have bound threads
1208   thread_t tid = 0;
1209   long     flags = (UseDetachedThreads ? THR_DETACHED : 0) | THR_SUSPENDED
1210                    | ((UseBoundThreads || os::Solaris::T2_libthread() ||
1211                        (thr_type == vm_thread) ||
1212                        (thr_type == cgc_thread) ||
1213                        (thr_type == pgc_thread) ||
1214                        (thr_type == compiler_thread && BackgroundCompilation)) ?
1215                       THR_BOUND : 0);
1216   int      status;
1217 
1218   // 4376845 -- libthread/kernel don't provide enough LWPs to utilize all CPUs.
1219   //
1220   // On multiprocessors systems, libthread sometimes under-provisions our
1221   // process with LWPs.  On a 30-way systems, for instance, we could have
1222   // 50 user-level threads in ready state and only 2 or 3 LWPs assigned
1223   // to our process.  This can result in under utilization of PEs.
1224   // I suspect the problem is related to libthread's LWP
1225   // pool management and to the kernel's SIGBLOCKING "last LWP parked"
1226   // upcall policy.
1227   //
1228   // The following code is palliative -- it attempts to ensure that our
1229   // process has sufficient LWPs to take advantage of multiple PEs.
1230   // Proper long-term cures include using user-level threads bound to LWPs
1231   // (THR_BOUND) or using LWP-based synchronization.  Note that there is a
1232   // slight timing window with respect to sampling _os_thread_count, but
1233   // the race is benign.  Also, we should periodically recompute
1234   // _processors_online as the min of SC_NPROCESSORS_ONLN and the
1235   // the number of PEs in our partition.  You might be tempted to use
1236   // THR_NEW_LWP here, but I'd recommend against it as that could
1237   // result in undesirable growth of the libthread's LWP pool.
1238   // The fix below isn't sufficient; for instance, it doesn't take into count
1239   // LWPs parked on IO.  It does, however, help certain CPU-bound benchmarks.
1240   //
1241   // Some pathologies this scheme doesn't handle:
1242   // *  Threads can block, releasing the LWPs.  The LWPs can age out.
1243   //    When a large number of threads become ready again there aren't
1244   //    enough LWPs available to service them.  This can occur when the
1245   //    number of ready threads oscillates.
1246   // *  LWPs/Threads park on IO, thus taking the LWP out of circulation.
1247   //
1248   // Finally, we should call thr_setconcurrency() periodically to refresh
1249   // the LWP pool and thwart the LWP age-out mechanism.
1250   // The "+3" term provides a little slop -- we want to slightly overprovision.
1251 
1252   if (AdjustConcurrency && os::Solaris::_os_thread_count < (_processors_online+3)) {
1253     if (!(flags & THR_BOUND)) {
1254       thr_setconcurrency (os::Solaris::_os_thread_count);       // avoid starvation
1255     }
1256   }
1257   // Although this doesn't hurt, we should warn of undefined behavior
1258   // when using unbound T1 threads with schedctl().  This should never
1259   // happen, as the compiler and VM threads are always created bound
1260   DEBUG_ONLY(
1261       if ((VMThreadHintNoPreempt || CompilerThreadHintNoPreempt) &&
1262           (!os::Solaris::T2_libthread() && (!(flags & THR_BOUND))) &&
1263           ((thr_type == vm_thread) || (thr_type == cgc_thread) ||
1264            (thr_type == pgc_thread) || (thr_type == compiler_thread && BackgroundCompilation))) {
1265          warning("schedctl behavior undefined when Compiler/VM/GC Threads are Unbound");
1266       }
1267   );
1268 
1269 
1270   // Mark that we don't have an lwp or thread id yet.
1271   // In case we attempt to set the priority before the thread starts.
1272   osthread->set_lwp_id(-1);
1273   osthread->set_thread_id(-1);
1274 
1275   status = thr_create(NULL, stack_size, java_start, thread, flags, &tid);
1276   if (status != 0) {
1277     if (PrintMiscellaneous && (Verbose || WizardMode)) {
1278       perror("os::create_thread");
1279     }
1280     thread->set_osthread(NULL);
1281     // Need to clean up stuff we've allocated so far
1282     delete osthread;
1283     return false;
1284   }
1285 
1286   Atomic::inc(&os::Solaris::_os_thread_count);
1287 
1288   // Store info on the Solaris thread into the OSThread
1289   osthread->set_thread_id(tid);
1290 
1291   // Remember that we created this thread so we can set priority on it
1292   osthread->set_vm_created();
1293 
1294   // Set the default thread priority otherwise use NormalPriority
1295 
1296   if ( UseThreadPriorities ) {
1297      thr_setprio(tid, (DefaultThreadPriority == -1) ?
1298                         java_to_os_priority[NormPriority] :
1299                         DefaultThreadPriority);
1300   }
1301 
1302   // Initial thread state is INITIALIZED, not SUSPENDED
1303   osthread->set_state(INITIALIZED);
1304 
1305   // The thread is returned suspended (in state INITIALIZED), and is started higher up in the call chain
1306   return true;
1307 }
1308 
1309 /* defined for >= Solaris 10. This allows builds on earlier versions
1310  *  of Solaris to take advantage of the newly reserved Solaris JVM signals
1311  *  With SIGJVM1, SIGJVM2, INTERRUPT_SIGNAL is SIGJVM1, ASYNC_SIGNAL is SIGJVM2
1312  *  and -XX:+UseAltSigs does nothing since these should have no conflict
1313  */
1314 #if !defined(SIGJVM1)
1315 #define SIGJVM1 39
1316 #define SIGJVM2 40
1317 #endif
1318 
1319 debug_only(static bool signal_sets_initialized = false);
1320 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
1321 int os::Solaris::_SIGinterrupt = INTERRUPT_SIGNAL;
1322 int os::Solaris::_SIGasync = ASYNC_SIGNAL;
1323 
1324 bool os::Solaris::is_sig_ignored(int sig) {
1325       struct sigaction oact;
1326       sigaction(sig, (struct sigaction*)NULL, &oact);
1327       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
1328                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
1329       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
1330            return true;
1331       else
1332            return false;
1333 }
1334 
1335 // Note: SIGRTMIN is a macro that calls sysconf() so it will
1336 // dynamically detect SIGRTMIN value for the system at runtime, not buildtime
1337 static bool isJVM1available() {
1338   return SIGJVM1 < SIGRTMIN;
1339 }
1340 
1341 void os::Solaris::signal_sets_init() {
1342   // Should also have an assertion stating we are still single-threaded.
1343   assert(!signal_sets_initialized, "Already initialized");
1344   // Fill in signals that are necessarily unblocked for all threads in
1345   // the VM. Currently, we unblock the following signals:
1346   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
1347   //                         by -Xrs (=ReduceSignalUsage));
1348   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
1349   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
1350   // the dispositions or masks wrt these signals.
1351   // Programs embedding the VM that want to use the above signals for their
1352   // own purposes must, at this time, use the "-Xrs" option to prevent
1353   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
1354   // (See bug 4345157, and other related bugs).
1355   // In reality, though, unblocking these signals is really a nop, since
1356   // these signals are not blocked by default.
1357   sigemptyset(&unblocked_sigs);
1358   sigemptyset(&allowdebug_blocked_sigs);
1359   sigaddset(&unblocked_sigs, SIGILL);
1360   sigaddset(&unblocked_sigs, SIGSEGV);
1361   sigaddset(&unblocked_sigs, SIGBUS);
1362   sigaddset(&unblocked_sigs, SIGFPE);
1363 
1364   if (isJVM1available) {
1365     os::Solaris::set_SIGinterrupt(SIGJVM1);
1366     os::Solaris::set_SIGasync(SIGJVM2);
1367   } else if (UseAltSigs) {
1368     os::Solaris::set_SIGinterrupt(ALT_INTERRUPT_SIGNAL);
1369     os::Solaris::set_SIGasync(ALT_ASYNC_SIGNAL);
1370   } else {
1371     os::Solaris::set_SIGinterrupt(INTERRUPT_SIGNAL);
1372     os::Solaris::set_SIGasync(ASYNC_SIGNAL);
1373   }
1374 
1375   sigaddset(&unblocked_sigs, os::Solaris::SIGinterrupt());
1376   sigaddset(&unblocked_sigs, os::Solaris::SIGasync());
1377 
1378   if (!ReduceSignalUsage) {
1379    if (!os::Solaris::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
1380       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
1381       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
1382    }
1383    if (!os::Solaris::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
1384       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
1385       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
1386    }
1387    if (!os::Solaris::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
1388       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
1389       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
1390    }
1391   }
1392   // Fill in signals that are blocked by all but the VM thread.
1393   sigemptyset(&vm_sigs);
1394   if (!ReduceSignalUsage)
1395     sigaddset(&vm_sigs, BREAK_SIGNAL);
1396   debug_only(signal_sets_initialized = true);
1397 
1398   // For diagnostics only used in run_periodic_checks
1399   sigemptyset(&check_signal_done);
1400 }
1401 
1402 // These are signals that are unblocked while a thread is running Java.
1403 // (For some reason, they get blocked by default.)
1404 sigset_t* os::Solaris::unblocked_signals() {
1405   assert(signal_sets_initialized, "Not initialized");
1406   return &unblocked_sigs;
1407 }
1408 
1409 // These are the signals that are blocked while a (non-VM) thread is
1410 // running Java. Only the VM thread handles these signals.
1411 sigset_t* os::Solaris::vm_signals() {
1412   assert(signal_sets_initialized, "Not initialized");
1413   return &vm_sigs;
1414 }
1415 
1416 // These are signals that are blocked during cond_wait to allow debugger in
1417 sigset_t* os::Solaris::allowdebug_blocked_signals() {
1418   assert(signal_sets_initialized, "Not initialized");
1419   return &allowdebug_blocked_sigs;
1420 }
1421 
1422 // First crack at OS-specific initialization, from inside the new thread.
1423 void os::initialize_thread() {
1424   int r = thr_main() ;
1425   guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
1426   if (r) {
1427     JavaThread* jt = (JavaThread *)Thread::current();
1428     assert(jt != NULL,"Sanity check");
1429     size_t stack_size;
1430     address base = jt->stack_base();
1431     if (Arguments::created_by_java_launcher()) {
1432       // Use 2MB to allow for Solaris 7 64 bit mode.
1433       stack_size = JavaThread::stack_size_at_create() == 0
1434         ? 2048*K : JavaThread::stack_size_at_create();
1435 
1436       // There are rare cases when we may have already used more than
1437       // the basic stack size allotment before this method is invoked.
1438       // Attempt to allow for a normally sized java_stack.
1439       size_t current_stack_offset = (size_t)(base - (address)&stack_size);
1440       stack_size += ReservedSpace::page_align_size_down(current_stack_offset);
1441     } else {
1442       // 6269555: If we were not created by a Java launcher, i.e. if we are
1443       // running embedded in a native application, treat the primordial thread
1444       // as much like a native attached thread as possible.  This means using
1445       // the current stack size from thr_stksegment(), unless it is too large
1446       // to reliably setup guard pages.  A reasonable max size is 8MB.
1447       size_t current_size = current_stack_size();
1448       // This should never happen, but just in case....
1449       if (current_size == 0) current_size = 2 * K * K;
1450       stack_size = current_size > (8 * K * K) ? (8 * K * K) : current_size;
1451     }
1452     address bottom = (address)align_size_up((intptr_t)(base - stack_size), os::vm_page_size());;
1453     stack_size = (size_t)(base - bottom);
1454 
1455     assert(stack_size > 0, "Stack size calculation problem");
1456 
1457     if (stack_size > jt->stack_size()) {
1458       NOT_PRODUCT(
1459         struct rlimit limits;
1460         getrlimit(RLIMIT_STACK, &limits);
1461         size_t size = adjust_stack_size(base, (size_t)limits.rlim_cur);
1462         assert(size >= jt->stack_size(), "Stack size problem in main thread");
1463       )
1464       tty->print_cr(
1465         "Stack size of %d Kb exceeds current limit of %d Kb.\n"
1466         "(Stack sizes are rounded up to a multiple of the system page size.)\n"
1467         "See limit(1) to increase the stack size limit.",
1468         stack_size / K, jt->stack_size() / K);
1469       vm_exit(1);
1470     }
1471     assert(jt->stack_size() >= stack_size,
1472           "Attempt to map more stack than was allocated");
1473     jt->set_stack_size(stack_size);
1474   }
1475 
1476    // 5/22/01: Right now alternate signal stacks do not handle
1477    // throwing stack overflow exceptions, see bug 4463178
1478    // Until a fix is found for this, T2 will NOT imply alternate signal
1479    // stacks.
1480    // If using T2 libthread threads, install an alternate signal stack.
1481    // Because alternate stacks associate with LWPs on Solaris,
1482    // see sigaltstack(2), if using UNBOUND threads, or if UseBoundThreads
1483    // we prefer to explicitly stack bang.
1484    // If not using T2 libthread, but using UseBoundThreads any threads
1485    // (primordial thread, jni_attachCurrentThread) we do not create,
1486    // probably are not bound, therefore they can not have an alternate
1487    // signal stack. Since our stack banging code is generated and
1488    // is shared across threads, all threads must be bound to allow
1489    // using alternate signal stacks.  The alternative is to interpose
1490    // on _lwp_create to associate an alt sig stack with each LWP,
1491    // and this could be a problem when the JVM is embedded.
1492    // We would prefer to use alternate signal stacks with T2
1493    // Since there is currently no accurate way to detect T2
1494    // we do not. Assuming T2 when running T1 causes sig 11s or assertions
1495    // on installing alternate signal stacks
1496 
1497 
1498    // 05/09/03: removed alternate signal stack support for Solaris
1499    // The alternate signal stack mechanism is no longer needed to
1500    // handle stack overflow. This is now handled by allocating
1501    // guard pages (red zone) and stackbanging.
1502    // Initially the alternate signal stack mechanism was removed because
1503    // it did not work with T1 llibthread. Alternate
1504    // signal stacks MUST have all threads bound to lwps. Applications
1505    // can create their own threads and attach them without their being
1506    // bound under T1. This is frequently the case for the primordial thread.
1507    // If we were ever to reenable this mechanism we would need to
1508    // use the dynamic check for T2 libthread.
1509 
1510   os::Solaris::init_thread_fpu_state();
1511 }
1512 
1513 
1514 
1515 // Free Solaris resources related to the OSThread
1516 void os::free_thread(OSThread* osthread) {
1517   assert(osthread != NULL, "os::free_thread but osthread not set");
1518 
1519 
1520   // We are told to free resources of the argument thread,
1521   // but we can only really operate on the current thread.
1522   // The main thread must take the VMThread down synchronously
1523   // before the main thread exits and frees up CodeHeap
1524   guarantee((Thread::current()->osthread() == osthread
1525      || (osthread == VMThread::vm_thread()->osthread())), "os::free_thread but not current thread");
1526   if (Thread::current()->osthread() == osthread) {
1527     // Restore caller's signal mask
1528     sigset_t sigmask = osthread->caller_sigmask();
1529     thr_sigsetmask(SIG_SETMASK, &sigmask, NULL);
1530   }
1531   delete osthread;
1532 }
1533 
1534 void os::pd_start_thread(Thread* thread) {
1535   int status = thr_continue(thread->osthread()->thread_id());
1536   assert_status(status == 0, status, "thr_continue failed");
1537 }
1538 
1539 
1540 intx os::current_thread_id() {
1541   return (intx)thr_self();
1542 }
1543 
1544 static pid_t _initial_pid = 0;
1545 
1546 int os::current_process_id() {
1547   return (int)(_initial_pid ? _initial_pid : getpid());
1548 }
1549 
1550 int os::allocate_thread_local_storage() {
1551   // %%%       in Win32 this allocates a memory segment pointed to by a
1552   //           register.  Dan Stein can implement a similar feature in
1553   //           Solaris.  Alternatively, the VM can do the same thing
1554   //           explicitly: malloc some storage and keep the pointer in a
1555   //           register (which is part of the thread's context) (or keep it
1556   //           in TLS).
1557   // %%%       In current versions of Solaris, thr_self and TSD can
1558   //           be accessed via short sequences of displaced indirections.
1559   //           The value of thr_self is available as %g7(36).
1560   //           The value of thr_getspecific(k) is stored in %g7(12)(4)(k*4-4),
1561   //           assuming that the current thread already has a value bound to k.
1562   //           It may be worth experimenting with such access patterns,
1563   //           and later having the parameters formally exported from a Solaris
1564   //           interface.  I think, however, that it will be faster to
1565   //           maintain the invariant that %g2 always contains the
1566   //           JavaThread in Java code, and have stubs simply
1567   //           treat %g2 as a caller-save register, preserving it in a %lN.
1568   thread_key_t tk;
1569   if (thr_keycreate( &tk, NULL ) )
1570     fatal(err_msg("os::allocate_thread_local_storage: thr_keycreate failed "
1571                   "(%s)", strerror(errno)));
1572   return int(tk);
1573 }
1574 
1575 void os::free_thread_local_storage(int index) {
1576   // %%% don't think we need anything here
1577   // if ( pthread_key_delete((pthread_key_t) tk) )
1578   //   fatal("os::free_thread_local_storage: pthread_key_delete failed");
1579 }
1580 
1581 #define SMALLINT 32   // libthread allocate for tsd_common is a version specific
1582                       // small number - point is NO swap space available
1583 void os::thread_local_storage_at_put(int index, void* value) {
1584   // %%% this is used only in threadLocalStorage.cpp
1585   if (thr_setspecific((thread_key_t)index, value)) {
1586     if (errno == ENOMEM) {
1587        vm_exit_out_of_memory(SMALLINT, "thr_setspecific: out of swap space");
1588     } else {
1589       fatal(err_msg("os::thread_local_storage_at_put: thr_setspecific failed "
1590                     "(%s)", strerror(errno)));
1591     }
1592   } else {
1593       ThreadLocalStorage::set_thread_in_slot ((Thread *) value) ;
1594   }
1595 }
1596 
1597 // This function could be called before TLS is initialized, for example, when
1598 // VM receives an async signal or when VM causes a fatal error during
1599 // initialization. Return NULL if thr_getspecific() fails.
1600 void* os::thread_local_storage_at(int index) {
1601   // %%% this is used only in threadLocalStorage.cpp
1602   void* r = NULL;
1603   return thr_getspecific((thread_key_t)index, &r) != 0 ? NULL : r;
1604 }
1605 
1606 
1607 const int NANOSECS_PER_MILLISECS = 1000000;
1608 // gethrtime can move backwards if read from one cpu and then a different cpu
1609 // getTimeNanos is guaranteed to not move backward on Solaris
1610 // local spinloop created as faster for a CAS on an int than
1611 // a CAS on a 64bit jlong. Also Atomic::cmpxchg for jlong is not
1612 // supported on sparc v8 or pre supports_cx8 intel boxes.
1613 // oldgetTimeNanos for systems which do not support CAS on 64bit jlong
1614 // i.e. sparc v8 and pre supports_cx8 (i486) intel boxes
1615 inline hrtime_t oldgetTimeNanos() {
1616   int gotlock = LOCK_INVALID;
1617   hrtime_t newtime = gethrtime();
1618 
1619   for (;;) {
1620 // grab lock for max_hrtime
1621     int curlock = max_hrtime_lock;
1622     if (curlock & LOCK_BUSY)  continue;
1623     if (gotlock = Atomic::cmpxchg(LOCK_BUSY, &max_hrtime_lock, LOCK_FREE) != LOCK_FREE) continue;
1624     if (newtime > max_hrtime) {
1625       max_hrtime = newtime;
1626     } else {
1627       newtime = max_hrtime;
1628     }
1629     // release lock
1630     max_hrtime_lock = LOCK_FREE;
1631     return newtime;
1632   }
1633 }
1634 // gethrtime can move backwards if read from one cpu and then a different cpu
1635 // getTimeNanos is guaranteed to not move backward on Solaris
1636 inline hrtime_t getTimeNanos() {
1637   if (VM_Version::supports_cx8()) {
1638     const hrtime_t now = gethrtime();
1639     // Use atomic long load since 32-bit x86 uses 2 registers to keep long.
1640     const hrtime_t prev = Atomic::load((volatile jlong*)&max_hrtime);
1641     if (now <= prev)  return prev;   // same or retrograde time;
1642     const hrtime_t obsv = Atomic::cmpxchg(now, (volatile jlong*)&max_hrtime, prev);
1643     assert(obsv >= prev, "invariant");   // Monotonicity
1644     // If the CAS succeeded then we're done and return "now".
1645     // If the CAS failed and the observed value "obs" is >= now then
1646     // we should return "obs".  If the CAS failed and now > obs > prv then
1647     // some other thread raced this thread and installed a new value, in which case
1648     // we could either (a) retry the entire operation, (b) retry trying to install now
1649     // or (c) just return obs.  We use (c).   No loop is required although in some cases
1650     // we might discard a higher "now" value in deference to a slightly lower but freshly
1651     // installed obs value.   That's entirely benign -- it admits no new orderings compared
1652     // to (a) or (b) -- and greatly reduces coherence traffic.
1653     // We might also condition (c) on the magnitude of the delta between obs and now.
1654     // Avoiding excessive CAS operations to hot RW locations is critical.
1655     // See http://blogs.sun.com/dave/entry/cas_and_cache_trivia_invalidate
1656     return (prev == obsv) ? now : obsv ;
1657   } else {
1658     return oldgetTimeNanos();
1659   }
1660 }
1661 
1662 // Time since start-up in seconds to a fine granularity.
1663 // Used by VMSelfDestructTimer and the MemProfiler.
1664 double os::elapsedTime() {
1665   return (double)(getTimeNanos() - first_hrtime) / (double)hrtime_hz;
1666 }
1667 
1668 jlong os::elapsed_counter() {
1669   return (jlong)(getTimeNanos() - first_hrtime);
1670 }
1671 
1672 jlong os::elapsed_frequency() {
1673    return hrtime_hz;
1674 }
1675 
1676 // Return the real, user, and system times in seconds from an
1677 // arbitrary fixed point in the past.
1678 bool os::getTimesSecs(double* process_real_time,
1679                   double* process_user_time,
1680                   double* process_system_time) {
1681   struct tms ticks;
1682   clock_t real_ticks = times(&ticks);
1683 
1684   if (real_ticks == (clock_t) (-1)) {
1685     return false;
1686   } else {
1687     double ticks_per_second = (double) clock_tics_per_sec;
1688     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1689     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1690     // For consistency return the real time from getTimeNanos()
1691     // converted to seconds.
1692     *process_real_time = ((double) getTimeNanos()) / ((double) NANOUNITS);
1693 
1694     return true;
1695   }
1696 }
1697 
1698 bool os::supports_vtime() { return true; }
1699 
1700 bool os::enable_vtime() {
1701   int fd = open("/proc/self/ctl", O_WRONLY);
1702   if (fd == -1)
1703     return false;
1704 
1705   long cmd[] = { PCSET, PR_MSACCT };
1706   int res = write(fd, cmd, sizeof(long) * 2);
1707   close(fd);
1708   if (res != sizeof(long) * 2)
1709     return false;
1710 
1711   return true;
1712 }
1713 
1714 bool os::vtime_enabled() {
1715   int fd = open("/proc/self/status", O_RDONLY);
1716   if (fd == -1)
1717     return false;
1718 
1719   pstatus_t status;
1720   int res = read(fd, (void*) &status, sizeof(pstatus_t));
1721   close(fd);
1722   if (res != sizeof(pstatus_t))
1723     return false;
1724 
1725   return status.pr_flags & PR_MSACCT;
1726 }
1727 
1728 double os::elapsedVTime() {
1729   return (double)gethrvtime() / (double)hrtime_hz;
1730 }
1731 
1732 // Used internally for comparisons only
1733 // getTimeMillis guaranteed to not move backwards on Solaris
1734 jlong getTimeMillis() {
1735   jlong nanotime = getTimeNanos();
1736   return (jlong)(nanotime / NANOSECS_PER_MILLISECS);
1737 }
1738 
1739 // Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1740 jlong os::javaTimeMillis() {
1741   timeval t;
1742   if (gettimeofday( &t, NULL) == -1)
1743     fatal(err_msg("os::javaTimeMillis: gettimeofday (%s)", strerror(errno)));
1744   return jlong(t.tv_sec) * 1000  +  jlong(t.tv_usec) / 1000;
1745 }
1746 
1747 jlong os::javaTimeNanos() {
1748   return (jlong)getTimeNanos();
1749 }
1750 
1751 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1752   info_ptr->max_value = ALL_64_BITS;      // gethrtime() uses all 64 bits
1753   info_ptr->may_skip_backward = false;    // not subject to resetting or drifting
1754   info_ptr->may_skip_forward = false;     // not subject to resetting or drifting
1755   info_ptr->kind = JVMTI_TIMER_ELAPSED;   // elapsed not CPU time
1756 }
1757 
1758 char * os::local_time_string(char *buf, size_t buflen) {
1759   struct tm t;
1760   time_t long_time;
1761   time(&long_time);
1762   localtime_r(&long_time, &t);
1763   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1764                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1765                t.tm_hour, t.tm_min, t.tm_sec);
1766   return buf;
1767 }
1768 
1769 // Note: os::shutdown() might be called very early during initialization, or
1770 // called from signal handler. Before adding something to os::shutdown(), make
1771 // sure it is async-safe and can handle partially initialized VM.
1772 void os::shutdown() {
1773 
1774   // allow PerfMemory to attempt cleanup of any persistent resources
1775   perfMemory_exit();
1776 
1777   // needs to remove object in file system
1778   AttachListener::abort();
1779 
1780   // flush buffered output, finish log files
1781   ostream_abort();
1782 
1783   // Check for abort hook
1784   abort_hook_t abort_hook = Arguments::abort_hook();
1785   if (abort_hook != NULL) {
1786     abort_hook();
1787   }
1788 }
1789 
1790 // Note: os::abort() might be called very early during initialization, or
1791 // called from signal handler. Before adding something to os::abort(), make
1792 // sure it is async-safe and can handle partially initialized VM.
1793 void os::abort(bool dump_core) {
1794   os::shutdown();
1795   if (dump_core) {
1796 #ifndef PRODUCT
1797     fdStream out(defaultStream::output_fd());
1798     out.print_raw("Current thread is ");
1799     char buf[16];
1800     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1801     out.print_raw_cr(buf);
1802     out.print_raw_cr("Dumping core ...");
1803 #endif
1804     ::abort(); // dump core (for debugging)
1805   }
1806 
1807   ::exit(1);
1808 }
1809 
1810 // Die immediately, no exit hook, no abort hook, no cleanup.
1811 void os::die() {
1812   _exit(-1);
1813 }
1814 
1815 // unused
1816 void os::set_error_file(const char *logfile) {}
1817 
1818 // DLL functions
1819 
1820 const char* os::dll_file_extension() { return ".so"; }
1821 
1822 const char* os::get_temp_directory() {
1823   const char *prop = Arguments::get_property("java.io.tmpdir");
1824   return prop == NULL ? "/tmp" : prop;
1825 }
1826 
1827 static bool file_exists(const char* filename) {
1828   struct stat statbuf;
1829   if (filename == NULL || strlen(filename) == 0) {
1830     return false;
1831   }
1832   return os::stat(filename, &statbuf) == 0;
1833 }
1834 
1835 void os::dll_build_name(char* buffer, size_t buflen,
1836                         const char* pname, const char* fname) {
1837   // Copied from libhpi
1838   const size_t pnamelen = pname ? strlen(pname) : 0;
1839 
1840   // Quietly truncate on buffer overflow.  Should be an error.
1841   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1842       *buffer = '\0';
1843       return;
1844   }
1845 
1846   if (pnamelen == 0) {
1847     snprintf(buffer, buflen, "lib%s.so", fname);
1848   } else if (strchr(pname, *os::path_separator()) != NULL) {
1849     int n;
1850     char** pelements = split_path(pname, &n);
1851     for (int i = 0 ; i < n ; i++) {
1852       // really shouldn't be NULL but what the heck, check can't hurt
1853       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1854         continue; // skip the empty path values
1855       }
1856       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1857       if (file_exists(buffer)) {
1858         break;
1859       }
1860     }
1861     // release the storage
1862     for (int i = 0 ; i < n ; i++) {
1863       if (pelements[i] != NULL) {
1864         FREE_C_HEAP_ARRAY(char, pelements[i]);
1865       }
1866     }
1867     if (pelements != NULL) {
1868       FREE_C_HEAP_ARRAY(char*, pelements);
1869     }
1870   } else {
1871     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1872   }
1873 }
1874 
1875 const char* os::get_current_directory(char *buf, int buflen) {
1876   return getcwd(buf, buflen);
1877 }
1878 
1879 // check if addr is inside libjvm[_g].so
1880 bool os::address_is_in_vm(address addr) {
1881   static address libjvm_base_addr;
1882   Dl_info dlinfo;
1883 
1884   if (libjvm_base_addr == NULL) {
1885     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1886     libjvm_base_addr = (address)dlinfo.dli_fbase;
1887     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1888   }
1889 
1890   if (dladdr((void *)addr, &dlinfo)) {
1891     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1892   }
1893 
1894   return false;
1895 }
1896 
1897 typedef int (*dladdr1_func_type) (void *, Dl_info *, void **, int);
1898 static dladdr1_func_type dladdr1_func = NULL;
1899 
1900 bool os::dll_address_to_function_name(address addr, char *buf,
1901                                       int buflen, int * offset) {
1902   Dl_info dlinfo;
1903 
1904   // dladdr1_func was initialized in os::init()
1905   if (dladdr1_func){
1906       // yes, we have dladdr1
1907 
1908       // Support for dladdr1 is checked at runtime; it may be
1909       // available even if the vm is built on a machine that does
1910       // not have dladdr1 support.  Make sure there is a value for
1911       // RTLD_DL_SYMENT.
1912       #ifndef RTLD_DL_SYMENT
1913       #define RTLD_DL_SYMENT 1
1914       #endif
1915       Sym * info;
1916       if (dladdr1_func((void *)addr, &dlinfo, (void **)&info,
1917                        RTLD_DL_SYMENT)) {
1918           if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1919           if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1920 
1921           // check if the returned symbol really covers addr
1922           return ((char *)dlinfo.dli_saddr + info->st_size > (char *)addr);
1923       } else {
1924           if (buf) buf[0] = '\0';
1925           if (offset) *offset  = -1;
1926           return false;
1927       }
1928   } else {
1929       // no, only dladdr is available
1930       if(dladdr((void *)addr, &dlinfo)) {
1931           if (buf) jio_snprintf(buf, buflen, dlinfo.dli_sname);
1932           if (offset) *offset = addr - (address)dlinfo.dli_saddr;
1933           return true;
1934       } else {
1935           if (buf) buf[0] = '\0';
1936           if (offset) *offset  = -1;
1937           return false;
1938       }
1939   }
1940 }
1941 
1942 bool os::dll_address_to_library_name(address addr, char* buf,
1943                                      int buflen, int* offset) {
1944   Dl_info dlinfo;
1945 
1946   if (dladdr((void*)addr, &dlinfo)){
1947      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1948      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1949      return true;
1950   } else {
1951      if (buf) buf[0] = '\0';
1952      if (offset) *offset = -1;
1953      return false;
1954   }
1955 }
1956 
1957 // Prints the names and full paths of all opened dynamic libraries
1958 // for current process
1959 void os::print_dll_info(outputStream * st) {
1960     Dl_info dli;
1961     void *handle;
1962     Link_map *map;
1963     Link_map *p;
1964 
1965     st->print_cr("Dynamic libraries:"); st->flush();
1966 
1967     if (!dladdr(CAST_FROM_FN_PTR(void *, os::print_dll_info), &dli)) {
1968         st->print_cr("Error: Cannot print dynamic libraries.");
1969         return;
1970     }
1971     handle = dlopen(dli.dli_fname, RTLD_LAZY);
1972     if (handle == NULL) {
1973         st->print_cr("Error: Cannot print dynamic libraries.");
1974         return;
1975     }
1976     dlinfo(handle, RTLD_DI_LINKMAP, &map);
1977     if (map == NULL) {
1978         st->print_cr("Error: Cannot print dynamic libraries.");
1979         return;
1980     }
1981 
1982     while (map->l_prev != NULL)
1983         map = map->l_prev;
1984 
1985     while (map != NULL) {
1986         st->print_cr(PTR_FORMAT " \t%s", map->l_addr, map->l_name);
1987         map = map->l_next;
1988     }
1989 
1990     dlclose(handle);
1991 }
1992 
1993   // Loads .dll/.so and
1994   // in case of error it checks if .dll/.so was built for the
1995   // same architecture as Hotspot is running on
1996 
1997 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1998 {
1999   void * result= ::dlopen(filename, RTLD_LAZY);
2000   if (result != NULL) {
2001     // Successful loading
2002     return result;
2003   }
2004 
2005   Elf32_Ehdr elf_head;
2006 
2007   // Read system error message into ebuf
2008   // It may or may not be overwritten below
2009   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2010   ebuf[ebuflen-1]='\0';
2011   int diag_msg_max_length=ebuflen-strlen(ebuf);
2012   char* diag_msg_buf=ebuf+strlen(ebuf);
2013 
2014   if (diag_msg_max_length==0) {
2015     // No more space in ebuf for additional diagnostics message
2016     return NULL;
2017   }
2018 
2019 
2020   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
2021 
2022   if (file_descriptor < 0) {
2023     // Can't open library, report dlerror() message
2024     return NULL;
2025   }
2026 
2027   bool failed_to_read_elf_head=
2028     (sizeof(elf_head)!=
2029         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
2030 
2031   ::close(file_descriptor);
2032   if (failed_to_read_elf_head) {
2033     // file i/o error - report dlerror() msg
2034     return NULL;
2035   }
2036 
2037   typedef struct {
2038     Elf32_Half  code;         // Actual value as defined in elf.h
2039     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
2040     char        elf_class;    // 32 or 64 bit
2041     char        endianess;    // MSB or LSB
2042     char*       name;         // String representation
2043   } arch_t;
2044 
2045   static const arch_t arch_array[]={
2046     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2047     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2048     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
2049     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
2050     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2051     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2052     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
2053     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
2054     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"}
2055   };
2056 
2057   #if  (defined IA32)
2058     static  Elf32_Half running_arch_code=EM_386;
2059   #elif   (defined AMD64)
2060     static  Elf32_Half running_arch_code=EM_X86_64;
2061   #elif  (defined IA64)
2062     static  Elf32_Half running_arch_code=EM_IA_64;
2063   #elif  (defined __sparc) && (defined _LP64)
2064     static  Elf32_Half running_arch_code=EM_SPARCV9;
2065   #elif  (defined __sparc) && (!defined _LP64)
2066     static  Elf32_Half running_arch_code=EM_SPARC;
2067   #elif  (defined __powerpc64__)
2068     static  Elf32_Half running_arch_code=EM_PPC64;
2069   #elif  (defined __powerpc__)
2070     static  Elf32_Half running_arch_code=EM_PPC;
2071   #else
2072     #error Method os::dll_load requires that one of following is defined:\
2073          IA32, AMD64, IA64, __sparc, __powerpc__
2074   #endif
2075 
2076   // Identify compatability class for VM's architecture and library's architecture
2077   // Obtain string descriptions for architectures
2078 
2079   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2080   int running_arch_index=-1;
2081 
2082   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2083     if (running_arch_code == arch_array[i].code) {
2084       running_arch_index    = i;
2085     }
2086     if (lib_arch.code == arch_array[i].code) {
2087       lib_arch.compat_class = arch_array[i].compat_class;
2088       lib_arch.name         = arch_array[i].name;
2089     }
2090   }
2091 
2092   assert(running_arch_index != -1,
2093     "Didn't find running architecture code (running_arch_code) in arch_array");
2094   if (running_arch_index == -1) {
2095     // Even though running architecture detection failed
2096     // we may still continue with reporting dlerror() message
2097     return NULL;
2098   }
2099 
2100   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2101     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2102     return NULL;
2103   }
2104 
2105   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2106     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2107     return NULL;
2108   }
2109 
2110   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2111     if ( lib_arch.name!=NULL ) {
2112       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2113         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2114         lib_arch.name, arch_array[running_arch_index].name);
2115     } else {
2116       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2117       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2118         lib_arch.code,
2119         arch_array[running_arch_index].name);
2120     }
2121   }
2122 
2123   return NULL;
2124 }
2125 
2126 void* os::dll_lookup(void* handle, const char* name) {
2127   return dlsym(handle, name);
2128 }
2129 
2130 
2131 bool _print_ascii_file(const char* filename, outputStream* st) {
2132   int fd = open(filename, O_RDONLY);
2133   if (fd == -1) {
2134      return false;
2135   }
2136 
2137   char buf[32];
2138   int bytes;
2139   while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
2140     st->print_raw(buf, bytes);
2141   }
2142 
2143   close(fd);
2144 
2145   return true;
2146 }
2147 
2148 void os::print_os_info(outputStream* st) {
2149   st->print("OS:");
2150 
2151   if (!_print_ascii_file("/etc/release", st)) {
2152     st->print("Solaris");
2153   }
2154   st->cr();
2155 
2156   // kernel
2157   st->print("uname:");
2158   struct utsname name;
2159   uname(&name);
2160   st->print(name.sysname); st->print(" ");
2161   st->print(name.release); st->print(" ");
2162   st->print(name.version); st->print(" ");
2163   st->print(name.machine);
2164 
2165   // libthread
2166   if (os::Solaris::T2_libthread()) st->print("  (T2 libthread)");
2167   else st->print("  (T1 libthread)");
2168   st->cr();
2169 
2170   // rlimit
2171   st->print("rlimit:");
2172   struct rlimit rlim;
2173 
2174   st->print(" STACK ");
2175   getrlimit(RLIMIT_STACK, &rlim);
2176   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2177   else st->print("%uk", rlim.rlim_cur >> 10);
2178 
2179   st->print(", CORE ");
2180   getrlimit(RLIMIT_CORE, &rlim);
2181   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2182   else st->print("%uk", rlim.rlim_cur >> 10);
2183 
2184   st->print(", NOFILE ");
2185   getrlimit(RLIMIT_NOFILE, &rlim);
2186   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2187   else st->print("%d", rlim.rlim_cur);
2188 
2189   st->print(", AS ");
2190   getrlimit(RLIMIT_AS, &rlim);
2191   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2192   else st->print("%uk", rlim.rlim_cur >> 10);
2193   st->cr();
2194 
2195   // load average
2196   st->print("load average:");
2197   double loadavg[3];
2198   os::loadavg(loadavg, 3);
2199   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2200   st->cr();
2201 }
2202 
2203 
2204 static bool check_addr0(outputStream* st) {
2205   jboolean status = false;
2206   int fd = open("/proc/self/map",O_RDONLY);
2207   if (fd >= 0) {
2208     prmap_t p;
2209     while(read(fd, &p, sizeof(p)) > 0) {
2210       if (p.pr_vaddr == 0x0) {
2211         st->print("Warning: Address: 0x%x, Size: %dK, ",p.pr_vaddr, p.pr_size/1024, p.pr_mapname);
2212         st->print("Mapped file: %s, ", p.pr_mapname[0] == '\0' ? "None" : p.pr_mapname);
2213         st->print("Access:");
2214         st->print("%s",(p.pr_mflags & MA_READ)  ? "r" : "-");
2215         st->print("%s",(p.pr_mflags & MA_WRITE) ? "w" : "-");
2216         st->print("%s",(p.pr_mflags & MA_EXEC)  ? "x" : "-");
2217         st->cr();
2218         status = true;
2219       }
2220       close(fd);
2221     }
2222   }
2223   return status;
2224 }
2225 
2226 void os::print_memory_info(outputStream* st) {
2227   st->print("Memory:");
2228   st->print(" %dk page", os::vm_page_size()>>10);
2229   st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
2230   st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
2231   st->cr();
2232   (void) check_addr0(st);
2233 }
2234 
2235 // Taken from /usr/include/sys/machsig.h  Supposed to be architecture specific
2236 // but they're the same for all the solaris architectures that we support.
2237 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2238                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2239                           "ILL_COPROC", "ILL_BADSTK" };
2240 
2241 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2242                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2243                           "FPE_FLTINV", "FPE_FLTSUB" };
2244 
2245 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2246 
2247 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2248 
2249 void os::print_siginfo(outputStream* st, void* siginfo) {
2250   st->print("siginfo:");
2251 
2252   const int buflen = 100;
2253   char buf[buflen];
2254   siginfo_t *si = (siginfo_t*)siginfo;
2255   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2256   char *err = strerror(si->si_errno);
2257   if (si->si_errno != 0 && err != NULL) {
2258     st->print("si_errno=%s", err);
2259   } else {
2260     st->print("si_errno=%d", si->si_errno);
2261   }
2262   const int c = si->si_code;
2263   assert(c > 0, "unexpected si_code");
2264   switch (si->si_signo) {
2265   case SIGILL:
2266     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2267     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2268     break;
2269   case SIGFPE:
2270     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2271     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2272     break;
2273   case SIGSEGV:
2274     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2275     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2276     break;
2277   case SIGBUS:
2278     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2279     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2280     break;
2281   default:
2282     st->print(", si_code=%d", si->si_code);
2283     // no si_addr
2284   }
2285 
2286   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2287       UseSharedSpaces) {
2288     FileMapInfo* mapinfo = FileMapInfo::current_info();
2289     if (mapinfo->is_in_shared_space(si->si_addr)) {
2290       st->print("\n\nError accessing class data sharing archive."   \
2291                 " Mapped file inaccessible during execution, "      \
2292                 " possible disk/network problem.");
2293     }
2294   }
2295   st->cr();
2296 }
2297 
2298 // Moved from whole group, because we need them here for diagnostic
2299 // prints.
2300 #define OLDMAXSIGNUM 32
2301 static int Maxsignum = 0;
2302 static int *ourSigFlags = NULL;
2303 
2304 extern "C" void sigINTRHandler(int, siginfo_t*, void*);
2305 
2306 int os::Solaris::get_our_sigflags(int sig) {
2307   assert(ourSigFlags!=NULL, "signal data structure not initialized");
2308   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2309   return ourSigFlags[sig];
2310 }
2311 
2312 void os::Solaris::set_our_sigflags(int sig, int flags) {
2313   assert(ourSigFlags!=NULL, "signal data structure not initialized");
2314   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2315   ourSigFlags[sig] = flags;
2316 }
2317 
2318 
2319 static const char* get_signal_handler_name(address handler,
2320                                            char* buf, int buflen) {
2321   int offset;
2322   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
2323   if (found) {
2324     // skip directory names
2325     const char *p1, *p2;
2326     p1 = buf;
2327     size_t len = strlen(os::file_separator());
2328     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
2329     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
2330   } else {
2331     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
2332   }
2333   return buf;
2334 }
2335 
2336 static void print_signal_handler(outputStream* st, int sig,
2337                                   char* buf, size_t buflen) {
2338   struct sigaction sa;
2339 
2340   sigaction(sig, NULL, &sa);
2341 
2342   st->print("%s: ", os::exception_name(sig, buf, buflen));
2343 
2344   address handler = (sa.sa_flags & SA_SIGINFO)
2345                   ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
2346                   : CAST_FROM_FN_PTR(address, sa.sa_handler);
2347 
2348   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
2349     st->print("SIG_DFL");
2350   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
2351     st->print("SIG_IGN");
2352   } else {
2353     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
2354   }
2355 
2356   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
2357 
2358   address rh = VMError::get_resetted_sighandler(sig);
2359   // May be, handler was resetted by VMError?
2360   if(rh != NULL) {
2361     handler = rh;
2362     sa.sa_flags = VMError::get_resetted_sigflags(sig);
2363   }
2364 
2365   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
2366 
2367   // Check: is it our handler?
2368   if(handler == CAST_FROM_FN_PTR(address, signalHandler) ||
2369      handler == CAST_FROM_FN_PTR(address, sigINTRHandler)) {
2370     // It is our signal handler
2371     // check for flags
2372     if(sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
2373       st->print(
2374         ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
2375         os::Solaris::get_our_sigflags(sig));
2376     }
2377   }
2378   st->cr();
2379 }
2380 
2381 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2382   st->print_cr("Signal Handlers:");
2383   print_signal_handler(st, SIGSEGV, buf, buflen);
2384   print_signal_handler(st, SIGBUS , buf, buflen);
2385   print_signal_handler(st, SIGFPE , buf, buflen);
2386   print_signal_handler(st, SIGPIPE, buf, buflen);
2387   print_signal_handler(st, SIGXFSZ, buf, buflen);
2388   print_signal_handler(st, SIGILL , buf, buflen);
2389   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2390   print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
2391   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2392   print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
2393   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2394   print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
2395   print_signal_handler(st, os::Solaris::SIGinterrupt(), buf, buflen);
2396   print_signal_handler(st, os::Solaris::SIGasync(), buf, buflen);
2397 }
2398 
2399 static char saved_jvm_path[MAXPATHLEN] = { 0 };
2400 
2401 // Find the full path to the current module, libjvm.so or libjvm_g.so
2402 void os::jvm_path(char *buf, jint buflen) {
2403   // Error checking.
2404   if (buflen < MAXPATHLEN) {
2405     assert(false, "must use a large-enough buffer");
2406     buf[0] = '\0';
2407     return;
2408   }
2409   // Lazy resolve the path to current module.
2410   if (saved_jvm_path[0] != 0) {
2411     strcpy(buf, saved_jvm_path);
2412     return;
2413   }
2414 
2415   Dl_info dlinfo;
2416   int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
2417   assert(ret != 0, "cannot locate libjvm");
2418   realpath((char *)dlinfo.dli_fname, buf);
2419 
2420   if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2421     // Support for the gamma launcher.  Typical value for buf is
2422     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2423     // the right place in the string, then assume we are installed in a JDK and
2424     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2425     // up the path so it looks like libjvm.so is installed there (append a
2426     // fake suffix hotspot/libjvm.so).
2427     const char *p = buf + strlen(buf) - 1;
2428     for (int count = 0; p > buf && count < 5; ++count) {
2429       for (--p; p > buf && *p != '/'; --p)
2430         /* empty */ ;
2431     }
2432 
2433     if (strncmp(p, "/jre/lib/", 9) != 0) {
2434       // Look for JAVA_HOME in the environment.
2435       char* java_home_var = ::getenv("JAVA_HOME");
2436       if (java_home_var != NULL && java_home_var[0] != 0) {
2437         char cpu_arch[12];
2438         sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
2439 #ifdef _LP64
2440         // If we are on sparc running a 64-bit vm, look in jre/lib/sparcv9.
2441         if (strcmp(cpu_arch, "sparc") == 0) {
2442           strcat(cpu_arch, "v9");
2443         } else if (strcmp(cpu_arch, "i386") == 0) {
2444           strcpy(cpu_arch, "amd64");
2445         }
2446 #endif
2447         // Check the current module name "libjvm.so" or "libjvm_g.so".
2448         p = strrchr(buf, '/');
2449         assert(strstr(p, "/libjvm") == p, "invalid library name");
2450         p = strstr(p, "_g") ? "_g" : "";
2451 
2452         realpath(java_home_var, buf);
2453         sprintf(buf + strlen(buf), "/jre/lib/%s", cpu_arch);
2454         if (0 == access(buf, F_OK)) {
2455           // Use current module name "libjvm[_g].so" instead of
2456           // "libjvm"debug_only("_g")".so" since for fastdebug version
2457           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2458           // It is used when we are choosing the HPI library's name
2459           // "libhpi[_g].so" in hpi::initialize_get_interface().
2460           sprintf(buf + strlen(buf), "/hotspot/libjvm%s.so", p);
2461         } else {
2462           // Go back to path of .so
2463           realpath((char *)dlinfo.dli_fname, buf);
2464         }
2465       }
2466     }
2467   }
2468 
2469   strcpy(saved_jvm_path, buf);
2470 }
2471 
2472 
2473 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2474   // no prefix required, not even "_"
2475 }
2476 
2477 
2478 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2479   // no suffix required
2480 }
2481 
2482 
2483 // sun.misc.Signal
2484 
2485 extern "C" {
2486   static void UserHandler(int sig, void *siginfo, void *context) {
2487     // Ctrl-C is pressed during error reporting, likely because the error
2488     // handler fails to abort. Let VM die immediately.
2489     if (sig == SIGINT && is_error_reported()) {
2490        os::die();
2491     }
2492 
2493     os::signal_notify(sig);
2494     // We do not need to reinstate the signal handler each time...
2495   }
2496 }
2497 
2498 void* os::user_handler() {
2499   return CAST_FROM_FN_PTR(void*, UserHandler);
2500 }
2501 
2502 extern "C" {
2503   typedef void (*sa_handler_t)(int);
2504   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2505 }
2506 
2507 void* os::signal(int signal_number, void* handler) {
2508   struct sigaction sigAct, oldSigAct;
2509   sigfillset(&(sigAct.sa_mask));
2510   sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2511   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2512 
2513   if (sigaction(signal_number, &sigAct, &oldSigAct))
2514     // -1 means registration failed
2515     return (void *)-1;
2516 
2517   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2518 }
2519 
2520 void os::signal_raise(int signal_number) {
2521   raise(signal_number);
2522 }
2523 
2524 /*
2525  * The following code is moved from os.cpp for making this
2526  * code platform specific, which it is by its very nature.
2527  */
2528 
2529 // a counter for each possible signal value
2530 static int Sigexit = 0;
2531 static int Maxlibjsigsigs;
2532 static jint *pending_signals = NULL;
2533 static int *preinstalled_sigs = NULL;
2534 static struct sigaction *chainedsigactions = NULL;
2535 static sema_t sig_sem;
2536 typedef int (*version_getting_t)();
2537 version_getting_t os::Solaris::get_libjsig_version = NULL;
2538 static int libjsigversion = NULL;
2539 
2540 int os::sigexitnum_pd() {
2541   assert(Sigexit > 0, "signal memory not yet initialized");
2542   return Sigexit;
2543 }
2544 
2545 void os::Solaris::init_signal_mem() {
2546   // Initialize signal structures
2547   Maxsignum = SIGRTMAX;
2548   Sigexit = Maxsignum+1;
2549   assert(Maxsignum >0, "Unable to obtain max signal number");
2550 
2551   Maxlibjsigsigs = Maxsignum;
2552 
2553   // pending_signals has one int per signal
2554   // The additional signal is for SIGEXIT - exit signal to signal_thread
2555   pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1));
2556   memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2557 
2558   if (UseSignalChaining) {
2559      chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2560        * (Maxsignum + 1));
2561      memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2562      preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1));
2563      memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2564   }
2565   ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1 ));
2566   memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2567 }
2568 
2569 void os::signal_init_pd() {
2570   int ret;
2571 
2572   ret = ::sema_init(&sig_sem, 0, NULL, NULL);
2573   assert(ret == 0, "sema_init() failed");
2574 }
2575 
2576 void os::signal_notify(int signal_number) {
2577   int ret;
2578 
2579   Atomic::inc(&pending_signals[signal_number]);
2580   ret = ::sema_post(&sig_sem);
2581   assert(ret == 0, "sema_post() failed");
2582 }
2583 
2584 static int check_pending_signals(bool wait_for_signal) {
2585   int ret;
2586   while (true) {
2587     for (int i = 0; i < Sigexit + 1; i++) {
2588       jint n = pending_signals[i];
2589       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2590         return i;
2591       }
2592     }
2593     if (!wait_for_signal) {
2594       return -1;
2595     }
2596     JavaThread *thread = JavaThread::current();
2597     ThreadBlockInVM tbivm(thread);
2598 
2599     bool threadIsSuspended;
2600     do {
2601       thread->set_suspend_equivalent();
2602       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2603       while((ret = ::sema_wait(&sig_sem)) == EINTR)
2604           ;
2605       assert(ret == 0, "sema_wait() failed");
2606 
2607       // were we externally suspended while we were waiting?
2608       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2609       if (threadIsSuspended) {
2610         //
2611         // The semaphore has been incremented, but while we were waiting
2612         // another thread suspended us. We don't want to continue running
2613         // while suspended because that would surprise the thread that
2614         // suspended us.
2615         //
2616         ret = ::sema_post(&sig_sem);
2617         assert(ret == 0, "sema_post() failed");
2618 
2619         thread->java_suspend_self();
2620       }
2621     } while (threadIsSuspended);
2622   }
2623 }
2624 
2625 int os::signal_lookup() {
2626   return check_pending_signals(false);
2627 }
2628 
2629 int os::signal_wait() {
2630   return check_pending_signals(true);
2631 }
2632 
2633 ////////////////////////////////////////////////////////////////////////////////
2634 // Virtual Memory
2635 
2636 static int page_size = -1;
2637 
2638 // The mmap MAP_ALIGN flag is supported on Solaris 9 and later.  init_2() will
2639 // clear this var if support is not available.
2640 static bool has_map_align = true;
2641 
2642 int os::vm_page_size() {
2643   assert(page_size != -1, "must call os::init");
2644   return page_size;
2645 }
2646 
2647 // Solaris allocates memory by pages.
2648 int os::vm_allocation_granularity() {
2649   assert(page_size != -1, "must call os::init");
2650   return page_size;
2651 }
2652 
2653 bool os::commit_memory(char* addr, size_t bytes, bool exec) {
2654   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2655   size_t size = bytes;
2656   return
2657      NULL != Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2658 }
2659 
2660 bool os::commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2661                        bool exec) {
2662   if (commit_memory(addr, bytes, exec)) {
2663     if (UseMPSS && alignment_hint > (size_t)vm_page_size()) {
2664       // If the large page size has been set and the VM
2665       // is using large pages, use the large page size
2666       // if it is smaller than the alignment hint. This is
2667       // a case where the VM wants to use a larger alignment size
2668       // for its own reasons but still want to use large pages
2669       // (which is what matters to setting the mpss range.
2670       size_t page_size = 0;
2671       if (large_page_size() < alignment_hint) {
2672         assert(UseLargePages, "Expected to be here for large page use only");
2673         page_size = large_page_size();
2674       } else {
2675         // If the alignment hint is less than the large page
2676         // size, the VM wants a particular alignment (thus the hint)
2677         // for internal reasons.  Try to set the mpss range using
2678         // the alignment_hint.
2679         page_size = alignment_hint;
2680       }
2681       // Since this is a hint, ignore any failures.
2682       (void)Solaris::set_mpss_range(addr, bytes, page_size);
2683     }
2684     return true;
2685   }
2686   return false;
2687 }
2688 
2689 // Uncommit the pages in a specified region.
2690 void os::free_memory(char* addr, size_t bytes) {
2691   if (madvise(addr, bytes, MADV_FREE) < 0) {
2692     debug_only(warning("MADV_FREE failed."));
2693     return;
2694   }
2695 }
2696 
2697 bool os::create_stack_guard_pages(char* addr, size_t size) {
2698   return os::commit_memory(addr, size);
2699 }
2700 
2701 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2702   return os::uncommit_memory(addr, size);
2703 }
2704 
2705 // Change the page size in a given range.
2706 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2707   assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2708   assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2709   Solaris::set_mpss_range(addr, bytes, alignment_hint);
2710 }
2711 
2712 // Tell the OS to make the range local to the first-touching LWP
2713 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2714   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2715   if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2716     debug_only(warning("MADV_ACCESS_LWP failed."));
2717   }
2718 }
2719 
2720 // Tell the OS that this range would be accessed from different LWPs.
2721 void os::numa_make_global(char *addr, size_t bytes) {
2722   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2723   if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2724     debug_only(warning("MADV_ACCESS_MANY failed."));
2725   }
2726 }
2727 
2728 // Get the number of the locality groups.
2729 size_t os::numa_get_groups_num() {
2730   size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2731   return n != -1 ? n : 1;
2732 }
2733 
2734 // Get a list of leaf locality groups. A leaf lgroup is group that
2735 // doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2736 // board. An LWP is assigned to one of these groups upon creation.
2737 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2738    if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2739      ids[0] = 0;
2740      return 1;
2741    }
2742    int result_size = 0, top = 1, bottom = 0, cur = 0;
2743    for (int k = 0; k < size; k++) {
2744      int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2745                                     (Solaris::lgrp_id_t*)&ids[top], size - top);
2746      if (r == -1) {
2747        ids[0] = 0;
2748        return 1;
2749      }
2750      if (!r) {
2751        // That's a leaf node.
2752        assert (bottom <= cur, "Sanity check");
2753        // Check if the node has memory
2754        if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2755                                    NULL, 0, LGRP_RSRC_MEM) > 0) {
2756          ids[bottom++] = ids[cur];
2757        }
2758      }
2759      top += r;
2760      cur++;
2761    }
2762    if (bottom == 0) {
2763      // Handle a situation, when the OS reports no memory available.
2764      // Assume UMA architecture.
2765      ids[0] = 0;
2766      return 1;
2767    }
2768    return bottom;
2769 }
2770 
2771 // Detect the topology change. Typically happens during CPU plugging-unplugging.
2772 bool os::numa_topology_changed() {
2773   int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2774   if (is_stale != -1 && is_stale) {
2775     Solaris::lgrp_fini(Solaris::lgrp_cookie());
2776     Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2777     assert(c != 0, "Failure to initialize LGRP API");
2778     Solaris::set_lgrp_cookie(c);
2779     return true;
2780   }
2781   return false;
2782 }
2783 
2784 // Get the group id of the current LWP.
2785 int os::numa_get_group_id() {
2786   int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2787   if (lgrp_id == -1) {
2788     return 0;
2789   }
2790   const int size = os::numa_get_groups_num();
2791   int *ids = (int*)alloca(size * sizeof(int));
2792 
2793   // Get the ids of all lgroups with memory; r is the count.
2794   int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2795                                   (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2796   if (r <= 0) {
2797     return 0;
2798   }
2799   return ids[os::random() % r];
2800 }
2801 
2802 // Request information about the page.
2803 bool os::get_page_info(char *start, page_info* info) {
2804   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2805   uint64_t addr = (uintptr_t)start;
2806   uint64_t outdata[2];
2807   uint_t validity = 0;
2808 
2809   if (os::Solaris::meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2810     return false;
2811   }
2812 
2813   info->size = 0;
2814   info->lgrp_id = -1;
2815 
2816   if ((validity & 1) != 0) {
2817     if ((validity & 2) != 0) {
2818       info->lgrp_id = outdata[0];
2819     }
2820     if ((validity & 4) != 0) {
2821       info->size = outdata[1];
2822     }
2823     return true;
2824   }
2825   return false;
2826 }
2827 
2828 // Scan the pages from start to end until a page different than
2829 // the one described in the info parameter is encountered.
2830 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2831   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2832   const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2833   uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT];
2834   uint_t validity[MAX_MEMINFO_CNT];
2835 
2836   size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2837   uint64_t p = (uint64_t)start;
2838   while (p < (uint64_t)end) {
2839     addrs[0] = p;
2840     size_t addrs_count = 1;
2841     while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] < (uint64_t)end) {
2842       addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2843       addrs_count++;
2844     }
2845 
2846     if (os::Solaris::meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2847       return NULL;
2848     }
2849 
2850     size_t i = 0;
2851     for (; i < addrs_count; i++) {
2852       if ((validity[i] & 1) != 0) {
2853         if ((validity[i] & 4) != 0) {
2854           if (outdata[types * i + 1] != page_expected->size) {
2855             break;
2856           }
2857         } else
2858           if (page_expected->size != 0) {
2859             break;
2860           }
2861 
2862         if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
2863           if (outdata[types * i] != page_expected->lgrp_id) {
2864             break;
2865           }
2866         }
2867       } else {
2868         return NULL;
2869       }
2870     }
2871 
2872     if (i != addrs_count) {
2873       if ((validity[i] & 2) != 0) {
2874         page_found->lgrp_id = outdata[types * i];
2875       } else {
2876         page_found->lgrp_id = -1;
2877       }
2878       if ((validity[i] & 4) != 0) {
2879         page_found->size = outdata[types * i + 1];
2880       } else {
2881         page_found->size = 0;
2882       }
2883       return (char*)addrs[i];
2884     }
2885 
2886     p = addrs[addrs_count - 1] + page_size;
2887   }
2888   return end;
2889 }
2890 
2891 bool os::uncommit_memory(char* addr, size_t bytes) {
2892   size_t size = bytes;
2893   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2894   // uncommitted page. Otherwise, the read/write might succeed if we
2895   // have enough swap space to back the physical page.
2896   return
2897     NULL != Solaris::mmap_chunk(addr, size,
2898                                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
2899                                 PROT_NONE);
2900 }
2901 
2902 char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
2903   char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
2904 
2905   if (b == MAP_FAILED) {
2906     return NULL;
2907   }
2908   return b;
2909 }
2910 
2911 char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes, size_t alignment_hint, bool fixed) {
2912   char* addr = requested_addr;
2913   int flags = MAP_PRIVATE | MAP_NORESERVE;
2914 
2915   assert(!(fixed && (alignment_hint > 0)), "alignment hint meaningless with fixed mmap");
2916 
2917   if (fixed) {
2918     flags |= MAP_FIXED;
2919   } else if (has_map_align && (alignment_hint > (size_t) vm_page_size())) {
2920     flags |= MAP_ALIGN;
2921     addr = (char*) alignment_hint;
2922   }
2923 
2924   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2925   // uncommitted page. Otherwise, the read/write might succeed if we
2926   // have enough swap space to back the physical page.
2927   return mmap_chunk(addr, bytes, flags, PROT_NONE);
2928 }
2929 
2930 char* os::reserve_memory(size_t bytes, char* requested_addr, size_t alignment_hint) {
2931   char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint, (requested_addr != NULL));
2932 
2933   guarantee(requested_addr == NULL || requested_addr == addr,
2934             "OS failed to return requested mmap address.");
2935   return addr;
2936 }
2937 
2938 // Reserve memory at an arbitrary address, only if that area is
2939 // available (and not reserved for something else).
2940 
2941 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2942   const int max_tries = 10;
2943   char* base[max_tries];
2944   size_t size[max_tries];
2945 
2946   // Solaris adds a gap between mmap'ed regions.  The size of the gap
2947   // is dependent on the requested size and the MMU.  Our initial gap
2948   // value here is just a guess and will be corrected later.
2949   bool had_top_overlap = false;
2950   bool have_adjusted_gap = false;
2951   size_t gap = 0x400000;
2952 
2953   // Assert only that the size is a multiple of the page size, since
2954   // that's all that mmap requires, and since that's all we really know
2955   // about at this low abstraction level.  If we need higher alignment,
2956   // we can either pass an alignment to this method or verify alignment
2957   // in one of the methods further up the call chain.  See bug 5044738.
2958   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2959 
2960   // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
2961   // Give it a try, if the kernel honors the hint we can return immediately.
2962   char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
2963   volatile int err = errno;
2964   if (addr == requested_addr) {
2965     return addr;
2966   } else if (addr != NULL) {
2967     unmap_memory(addr, bytes);
2968   }
2969 
2970   if (PrintMiscellaneous && Verbose) {
2971     char buf[256];
2972     buf[0] = '\0';
2973     if (addr == NULL) {
2974       jio_snprintf(buf, sizeof(buf), ": %s", strerror(err));
2975     }
2976     warning("attempt_reserve_memory_at: couldn't reserve %d bytes at "
2977             PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
2978             "%s", bytes, requested_addr, addr, buf);
2979   }
2980 
2981   // Address hint method didn't work.  Fall back to the old method.
2982   // In theory, once SNV becomes our oldest supported platform, this
2983   // code will no longer be needed.
2984   //
2985   // Repeatedly allocate blocks until the block is allocated at the
2986   // right spot. Give up after max_tries.
2987   int i;
2988   for (i = 0; i < max_tries; ++i) {
2989     base[i] = reserve_memory(bytes);
2990 
2991     if (base[i] != NULL) {
2992       // Is this the block we wanted?
2993       if (base[i] == requested_addr) {
2994         size[i] = bytes;
2995         break;
2996       }
2997 
2998       // check that the gap value is right
2999       if (had_top_overlap && !have_adjusted_gap) {
3000         size_t actual_gap = base[i-1] - base[i] - bytes;
3001         if (gap != actual_gap) {
3002           // adjust the gap value and retry the last 2 allocations
3003           assert(i > 0, "gap adjustment code problem");
3004           have_adjusted_gap = true;  // adjust the gap only once, just in case
3005           gap = actual_gap;
3006           if (PrintMiscellaneous && Verbose) {
3007             warning("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
3008           }
3009           unmap_memory(base[i], bytes);
3010           unmap_memory(base[i-1], size[i-1]);
3011           i-=2;
3012           continue;
3013         }
3014       }
3015 
3016       // Does this overlap the block we wanted? Give back the overlapped
3017       // parts and try again.
3018       //
3019       // There is still a bug in this code: if top_overlap == bytes,
3020       // the overlap is offset from requested region by the value of gap.
3021       // In this case giving back the overlapped part will not work,
3022       // because we'll give back the entire block at base[i] and
3023       // therefore the subsequent allocation will not generate a new gap.
3024       // This could be fixed with a new algorithm that used larger
3025       // or variable size chunks to find the requested region -
3026       // but such a change would introduce additional complications.
3027       // It's rare enough that the planets align for this bug,
3028       // so we'll just wait for a fix for 6204603/5003415 which
3029       // will provide a mmap flag to allow us to avoid this business.
3030 
3031       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3032       if (top_overlap >= 0 && top_overlap < bytes) {
3033         had_top_overlap = true;
3034         unmap_memory(base[i], top_overlap);
3035         base[i] += top_overlap;
3036         size[i] = bytes - top_overlap;
3037       } else {
3038         size_t bottom_overlap = base[i] + bytes - requested_addr;
3039         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3040           if (PrintMiscellaneous && Verbose && bottom_overlap == 0) {
3041             warning("attempt_reserve_memory_at: possible alignment bug");
3042           }
3043           unmap_memory(requested_addr, bottom_overlap);
3044           size[i] = bytes - bottom_overlap;
3045         } else {
3046           size[i] = bytes;
3047         }
3048       }
3049     }
3050   }
3051 
3052   // Give back the unused reserved pieces.
3053 
3054   for (int j = 0; j < i; ++j) {
3055     if (base[j] != NULL) {
3056       unmap_memory(base[j], size[j]);
3057     }
3058   }
3059 
3060   return (i < max_tries) ? requested_addr : NULL;
3061 }
3062 
3063 bool os::release_memory(char* addr, size_t bytes) {
3064   size_t size = bytes;
3065   return munmap(addr, size) == 0;
3066 }
3067 
3068 static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
3069   assert(addr == (char*)align_size_down((uintptr_t)addr, os::vm_page_size()),
3070          "addr must be page aligned");
3071   int retVal = mprotect(addr, bytes, prot);
3072   return retVal == 0;
3073 }
3074 
3075 // Protect memory (Used to pass readonly pages through
3076 // JNI GetArray<type>Elements with empty arrays.)
3077 // Also, used for serialization page and for compressed oops null pointer
3078 // checking.
3079 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3080                         bool is_committed) {
3081   unsigned int p = 0;
3082   switch (prot) {
3083   case MEM_PROT_NONE: p = PROT_NONE; break;
3084   case MEM_PROT_READ: p = PROT_READ; break;
3085   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3086   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3087   default:
3088     ShouldNotReachHere();
3089   }
3090   // is_committed is unused.
3091   return solaris_mprotect(addr, bytes, p);
3092 }
3093 
3094 // guard_memory and unguard_memory only happens within stack guard pages.
3095 // Since ISM pertains only to the heap, guard and unguard memory should not
3096 /// happen with an ISM region.
3097 bool os::guard_memory(char* addr, size_t bytes) {
3098   return solaris_mprotect(addr, bytes, PROT_NONE);
3099 }
3100 
3101 bool os::unguard_memory(char* addr, size_t bytes) {
3102   return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
3103 }
3104 
3105 // Large page support
3106 
3107 // UseLargePages is the master flag to enable/disable large page memory.
3108 // UseMPSS and UseISM are supported for compatibility reasons. Their combined
3109 // effects can be described in the following table:
3110 //
3111 // UseLargePages UseMPSS UseISM
3112 //    false         *       *   => UseLargePages is the master switch, turning
3113 //                                 it off will turn off both UseMPSS and
3114 //                                 UseISM. VM will not use large page memory
3115 //                                 regardless the settings of UseMPSS/UseISM.
3116 //     true      false    false => Unless future Solaris provides other
3117 //                                 mechanism to use large page memory, this
3118 //                                 combination is equivalent to -UseLargePages,
3119 //                                 VM will not use large page memory
3120 //     true      true     false => JVM will use MPSS for large page memory.
3121 //                                 This is the default behavior.
3122 //     true      false    true  => JVM will use ISM for large page memory.
3123 //     true      true     true  => JVM will use ISM if it is available.
3124 //                                 Otherwise, JVM will fall back to MPSS.
3125 //                                 Becaues ISM is now available on all
3126 //                                 supported Solaris versions, this combination
3127 //                                 is equivalent to +UseISM -UseMPSS.
3128 
3129 typedef int (*getpagesizes_func_type) (size_t[], int);
3130 static size_t _large_page_size = 0;
3131 
3132 bool os::Solaris::ism_sanity_check(bool warn, size_t * page_size) {
3133   // x86 uses either 2M or 4M page, depending on whether PAE (Physical Address
3134   // Extensions) mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. Sparc
3135   // can support multiple page sizes.
3136 
3137   // Don't bother to probe page size because getpagesizes() comes with MPSS.
3138   // ISM is only recommended on old Solaris where there is no MPSS support.
3139   // Simply choose a conservative value as default.
3140   *page_size = LargePageSizeInBytes ? LargePageSizeInBytes :
3141                SPARC_ONLY(4 * M) IA32_ONLY(4 * M) AMD64_ONLY(2 * M);
3142 
3143   // ISM is available on all supported Solaris versions
3144   return true;
3145 }
3146 
3147 // Insertion sort for small arrays (descending order).
3148 static void insertion_sort_descending(size_t* array, int len) {
3149   for (int i = 0; i < len; i++) {
3150     size_t val = array[i];
3151     for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
3152       size_t tmp = array[key];
3153       array[key] = array[key - 1];
3154       array[key - 1] = tmp;
3155     }
3156   }
3157 }
3158 
3159 bool os::Solaris::mpss_sanity_check(bool warn, size_t * page_size) {
3160   getpagesizes_func_type getpagesizes_func =
3161     CAST_TO_FN_PTR(getpagesizes_func_type, dlsym(RTLD_DEFAULT, "getpagesizes"));
3162   if (getpagesizes_func == NULL) {
3163     if (warn) {
3164       warning("MPSS is not supported by the operating system.");
3165     }
3166     return false;
3167   }
3168 
3169   const unsigned int usable_count = VM_Version::page_size_count();
3170   if (usable_count == 1) {
3171     return false;
3172   }
3173 
3174   // Fill the array of page sizes.
3175   int n = getpagesizes_func(_page_sizes, page_sizes_max);
3176   assert(n > 0, "Solaris bug?");
3177   if (n == page_sizes_max) {
3178     // Add a sentinel value (necessary only if the array was completely filled
3179     // since it is static (zeroed at initialization)).
3180     _page_sizes[--n] = 0;
3181     DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
3182   }
3183   assert(_page_sizes[n] == 0, "missing sentinel");
3184 
3185   if (n == 1) return false;     // Only one page size available.
3186 
3187   // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
3188   // select up to usable_count elements.  First sort the array, find the first
3189   // acceptable value, then copy the usable sizes to the top of the array and
3190   // trim the rest.  Make sure to include the default page size :-).
3191   //
3192   // A better policy could get rid of the 4M limit by taking the sizes of the
3193   // important VM memory regions (java heap and possibly the code cache) into
3194   // account.
3195   insertion_sort_descending(_page_sizes, n);
3196   const size_t size_limit =
3197     FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
3198   int beg;
3199   for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */ ;
3200   const int end = MIN2((int)usable_count, n) - 1;
3201   for (int cur = 0; cur < end; ++cur, ++beg) {
3202     _page_sizes[cur] = _page_sizes[beg];
3203   }
3204   _page_sizes[end] = vm_page_size();
3205   _page_sizes[end + 1] = 0;
3206 
3207   if (_page_sizes[end] > _page_sizes[end - 1]) {
3208     // Default page size is not the smallest; sort again.
3209     insertion_sort_descending(_page_sizes, end + 1);
3210   }
3211   *page_size = _page_sizes[0];
3212 
3213   return true;
3214 }
3215 
3216 bool os::large_page_init() {
3217   if (!UseLargePages) {
3218     UseISM = false;
3219     UseMPSS = false;
3220     return false;
3221   }
3222 
3223   // print a warning if any large page related flag is specified on command line
3224   bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
3225                          !FLAG_IS_DEFAULT(UseISM)               ||
3226                          !FLAG_IS_DEFAULT(UseMPSS)              ||
3227                          !FLAG_IS_DEFAULT(LargePageSizeInBytes);
3228   UseISM = UseISM &&
3229            Solaris::ism_sanity_check(warn_on_failure, &_large_page_size);
3230   if (UseISM) {
3231     // ISM disables MPSS to be compatible with old JDK behavior
3232     UseMPSS = false;
3233     _page_sizes[0] = _large_page_size;
3234     _page_sizes[1] = vm_page_size();
3235   }
3236 
3237   UseMPSS = UseMPSS &&
3238             Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
3239 
3240   UseLargePages = UseISM || UseMPSS;
3241   return UseLargePages;
3242 }
3243 
3244 bool os::Solaris::set_mpss_range(caddr_t start, size_t bytes, size_t align) {
3245   // Signal to OS that we want large pages for addresses
3246   // from addr, addr + bytes
3247   struct memcntl_mha mpss_struct;
3248   mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
3249   mpss_struct.mha_pagesize = align;
3250   mpss_struct.mha_flags = 0;
3251   if (memcntl(start, bytes, MC_HAT_ADVISE,
3252               (caddr_t) &mpss_struct, 0, 0) < 0) {
3253     debug_only(warning("Attempt to use MPSS failed."));
3254     return false;
3255   }
3256   return true;
3257 }
3258 
3259 char* os::reserve_memory_special(size_t bytes, char* addr, bool exec) {
3260   // "exec" is passed in but not used.  Creating the shared image for
3261   // the code cache doesn't have an SHM_X executable permission to check.
3262   assert(UseLargePages && UseISM, "only for ISM large pages");
3263 
3264   size_t size = bytes;
3265   char* retAddr = NULL;
3266   int shmid;
3267   key_t ismKey;
3268 
3269   bool warn_on_failure = UseISM &&
3270                         (!FLAG_IS_DEFAULT(UseLargePages)         ||
3271                          !FLAG_IS_DEFAULT(UseISM)                ||
3272                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3273                         );
3274   char msg[128];
3275 
3276   ismKey = IPC_PRIVATE;
3277 
3278   // Create a large shared memory region to attach to based on size.
3279   // Currently, size is the total size of the heap
3280   shmid = shmget(ismKey, size, SHM_R | SHM_W | IPC_CREAT);
3281   if (shmid == -1){
3282      if (warn_on_failure) {
3283        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3284        warning(msg);
3285      }
3286      return NULL;
3287   }
3288 
3289   // Attach to the region
3290   retAddr = (char *) shmat(shmid, 0, SHM_SHARE_MMU | SHM_R | SHM_W);
3291   int err = errno;
3292 
3293   // Remove shmid. If shmat() is successful, the actual shared memory segment
3294   // will be deleted when it's detached by shmdt() or when the process
3295   // terminates. If shmat() is not successful this will remove the shared
3296   // segment immediately.
3297   shmctl(shmid, IPC_RMID, NULL);
3298 
3299   if (retAddr == (char *) -1) {
3300     if (warn_on_failure) {
3301       jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3302       warning(msg);
3303     }
3304     return NULL;
3305   }
3306 
3307   return retAddr;
3308 }
3309 
3310 bool os::release_memory_special(char* base, size_t bytes) {
3311   // detaching the SHM segment will also delete it, see reserve_memory_special()
3312   int rslt = shmdt(base);
3313   return rslt == 0;
3314 }
3315 
3316 size_t os::large_page_size() {
3317   return _large_page_size;
3318 }
3319 
3320 // MPSS allows application to commit large page memory on demand; with ISM
3321 // the entire memory region must be allocated as shared memory.
3322 bool os::can_commit_large_page_memory() {
3323   return UseISM ? false : true;
3324 }
3325 
3326 bool os::can_execute_large_page_memory() {
3327   return UseISM ? false : true;
3328 }
3329 
3330 static int os_sleep(jlong millis, bool interruptible) {
3331   const jlong limit = INT_MAX;
3332   jlong prevtime;
3333   int res;
3334 
3335   while (millis > limit) {
3336     if ((res = os_sleep(limit, interruptible)) != OS_OK)
3337       return res;
3338     millis -= limit;
3339   }
3340 
3341   // Restart interrupted polls with new parameters until the proper delay
3342   // has been completed.
3343 
3344   prevtime = getTimeMillis();
3345 
3346   while (millis > 0) {
3347     jlong newtime;
3348 
3349     if (!interruptible) {
3350       // Following assert fails for os::yield_all:
3351       // assert(!thread->is_Java_thread(), "must not be java thread");
3352       res = poll(NULL, 0, millis);
3353     } else {
3354       JavaThread *jt = JavaThread::current();
3355 
3356       INTERRUPTIBLE_NORESTART_VM_ALWAYS(poll(NULL, 0, millis), res, jt,
3357         os::Solaris::clear_interrupted);
3358     }
3359 
3360     // INTERRUPTIBLE_NORESTART_VM_ALWAYS returns res == OS_INTRPT for
3361     // thread.Interrupt.
3362 
3363     if((res == OS_ERR) && (errno == EINTR)) {
3364       newtime = getTimeMillis();
3365       assert(newtime >= prevtime, "time moving backwards");
3366     /* Doing prevtime and newtime in microseconds doesn't help precision,
3367        and trying to round up to avoid lost milliseconds can result in a
3368        too-short delay. */
3369       millis -= newtime - prevtime;
3370       if(millis <= 0)
3371         return OS_OK;
3372       prevtime = newtime;
3373     } else
3374       return res;
3375   }
3376 
3377   return OS_OK;
3378 }
3379 
3380 // Read calls from inside the vm need to perform state transitions
3381 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3382   INTERRUPTIBLE_RETURN_INT_VM(::read(fd, buf, nBytes), os::Solaris::clear_interrupted);
3383 }
3384 
3385 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3386   assert(thread == Thread::current(),  "thread consistency check");
3387 
3388   // TODO-FIXME: this should be removed.
3389   // On Solaris machines (especially 2.5.1) we found that sometimes the VM gets into a live lock
3390   // situation with a JavaThread being starved out of a lwp. The kernel doesn't seem to generate
3391   // a SIGWAITING signal which would enable the threads library to create a new lwp for the starving
3392   // thread. We suspect that because the Watcher thread keeps waking up at periodic intervals the kernel
3393   // is fooled into believing that the system is making progress. In the code below we block the
3394   // the watcher thread while safepoint is in progress so that it would not appear as though the
3395   // system is making progress.
3396   if (!Solaris::T2_libthread() &&
3397       thread->is_Watcher_thread() && SafepointSynchronize::is_synchronizing() && !Arguments::has_profile()) {
3398     // We now try to acquire the threads lock. Since this lock is held by the VM thread during
3399     // the entire safepoint, the watcher thread will  line up here during the safepoint.
3400     Threads_lock->lock_without_safepoint_check();
3401     Threads_lock->unlock();
3402   }
3403 
3404   if (thread->is_Java_thread()) {
3405     // This is a JavaThread so we honor the _thread_blocked protocol
3406     // even for sleeps of 0 milliseconds. This was originally done
3407     // as a workaround for bug 4338139. However, now we also do it
3408     // to honor the suspend-equivalent protocol.
3409 
3410     JavaThread *jt = (JavaThread *) thread;
3411     ThreadBlockInVM tbivm(jt);
3412 
3413     jt->set_suspend_equivalent();
3414     // cleared by handle_special_suspend_equivalent_condition() or
3415     // java_suspend_self() via check_and_wait_while_suspended()
3416 
3417     int ret_code;
3418     if (millis <= 0) {
3419       thr_yield();
3420       ret_code = 0;
3421     } else {
3422       // The original sleep() implementation did not create an
3423       // OSThreadWaitState helper for sleeps of 0 milliseconds.
3424       // I'm preserving that decision for now.
3425       OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3426 
3427       ret_code = os_sleep(millis, interruptible);
3428     }
3429 
3430     // were we externally suspended while we were waiting?
3431     jt->check_and_wait_while_suspended();
3432 
3433     return ret_code;
3434   }
3435 
3436   // non-JavaThread from this point on:
3437 
3438   if (millis <= 0) {
3439     thr_yield();
3440     return 0;
3441   }
3442 
3443   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3444 
3445   return os_sleep(millis, interruptible);
3446 }
3447 
3448 int os::naked_sleep() {
3449   // %% make the sleep time an integer flag. for now use 1 millisec.
3450   return os_sleep(1, false);
3451 }
3452 
3453 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3454 void os::infinite_sleep() {
3455   while (true) {    // sleep forever ...
3456     ::sleep(100);   // ... 100 seconds at a time
3457   }
3458 }
3459 
3460 // Used to convert frequent JVM_Yield() to nops
3461 bool os::dont_yield() {
3462   if (DontYieldALot) {
3463     static hrtime_t last_time = 0;
3464     hrtime_t diff = getTimeNanos() - last_time;
3465 
3466     if (diff < DontYieldALotInterval * 1000000)
3467       return true;
3468 
3469     last_time += diff;
3470 
3471     return false;
3472   }
3473   else {
3474     return false;
3475   }
3476 }
3477 
3478 // Caveat: Solaris os::yield() causes a thread-state transition whereas
3479 // the linux and win32 implementations do not.  This should be checked.
3480 
3481 void os::yield() {
3482   // Yields to all threads with same or greater priority
3483   os::sleep(Thread::current(), 0, false);
3484 }
3485 
3486 // Note that yield semantics are defined by the scheduling class to which
3487 // the thread currently belongs.  Typically, yield will _not yield to
3488 // other equal or higher priority threads that reside on the dispatch queues
3489 // of other CPUs.
3490 
3491 os::YieldResult os::NakedYield() { thr_yield(); return os::YIELD_UNKNOWN; }
3492 
3493 
3494 // On Solaris we found that yield_all doesn't always yield to all other threads.
3495 // There have been cases where there is a thread ready to execute but it doesn't
3496 // get an lwp as the VM thread continues to spin with sleeps of 1 millisecond.
3497 // The 1 millisecond wait doesn't seem long enough for the kernel to issue a
3498 // SIGWAITING signal which will cause a new lwp to be created. So we count the
3499 // number of times yield_all is called in the one loop and increase the sleep
3500 // time after 8 attempts. If this fails too we increase the concurrency level
3501 // so that the starving thread would get an lwp
3502 
3503 void os::yield_all(int attempts) {
3504   // Yields to all threads, including threads with lower priorities
3505   if (attempts == 0) {
3506     os::sleep(Thread::current(), 1, false);
3507   } else {
3508     int iterations = attempts % 30;
3509     if (iterations == 0 && !os::Solaris::T2_libthread()) {
3510       // thr_setconcurrency and _getconcurrency make sense only under T1.
3511       int noofLWPS = thr_getconcurrency();
3512       if (noofLWPS < (Threads::number_of_threads() + 2)) {
3513         thr_setconcurrency(thr_getconcurrency() + 1);
3514       }
3515     } else if (iterations < 25) {
3516       os::sleep(Thread::current(), 1, false);
3517     } else {
3518       os::sleep(Thread::current(), 10, false);
3519     }
3520   }
3521 }
3522 
3523 // Called from the tight loops to possibly influence time-sharing heuristics
3524 void os::loop_breaker(int attempts) {
3525   os::yield_all(attempts);
3526 }
3527 
3528 
3529 // Interface for setting lwp priorities.  If we are using T2 libthread,
3530 // which forces the use of BoundThreads or we manually set UseBoundThreads,
3531 // all of our threads will be assigned to real lwp's.  Using the thr_setprio
3532 // function is meaningless in this mode so we must adjust the real lwp's priority
3533 // The routines below implement the getting and setting of lwp priorities.
3534 //
3535 // Note: There are three priority scales used on Solaris.  Java priotities
3536 //       which range from 1 to 10, libthread "thr_setprio" scale which range
3537 //       from 0 to 127, and the current scheduling class of the process we
3538 //       are running in.  This is typically from -60 to +60.
3539 //       The setting of the lwp priorities in done after a call to thr_setprio
3540 //       so Java priorities are mapped to libthread priorities and we map from
3541 //       the latter to lwp priorities.  We don't keep priorities stored in
3542 //       Java priorities since some of our worker threads want to set priorities
3543 //       higher than all Java threads.
3544 //
3545 // For related information:
3546 // (1)  man -s 2 priocntl
3547 // (2)  man -s 4 priocntl
3548 // (3)  man dispadmin
3549 // =    librt.so
3550 // =    libthread/common/rtsched.c - thrp_setlwpprio().
3551 // =    ps -cL <pid> ... to validate priority.
3552 // =    sched_get_priority_min and _max
3553 //              pthread_create
3554 //              sched_setparam
3555 //              pthread_setschedparam
3556 //
3557 // Assumptions:
3558 // +    We assume that all threads in the process belong to the same
3559 //              scheduling class.   IE. an homogenous process.
3560 // +    Must be root or in IA group to change change "interactive" attribute.
3561 //              Priocntl() will fail silently.  The only indication of failure is when
3562 //              we read-back the value and notice that it hasn't changed.
3563 // +    Interactive threads enter the runq at the head, non-interactive at the tail.
3564 // +    For RT, change timeslice as well.  Invariant:
3565 //              constant "priority integral"
3566 //              Konst == TimeSlice * (60-Priority)
3567 //              Given a priority, compute appropriate timeslice.
3568 // +    Higher numerical values have higher priority.
3569 
3570 // sched class attributes
3571 typedef struct {
3572         int   schedPolicy;              // classID
3573         int   maxPrio;
3574         int   minPrio;
3575 } SchedInfo;
3576 
3577 
3578 static SchedInfo tsLimits, iaLimits, rtLimits;
3579 
3580 #ifdef ASSERT
3581 static int  ReadBackValidate = 1;
3582 #endif
3583 static int  myClass     = 0;
3584 static int  myMin       = 0;
3585 static int  myMax       = 0;
3586 static int  myCur       = 0;
3587 static bool priocntl_enable = false;
3588 
3589 
3590 // Call the version of priocntl suitable for all supported versions
3591 // of Solaris. We need to call through this wrapper so that we can
3592 // build on Solaris 9 and run on Solaris 8, 9 and 10.
3593 //
3594 // This code should be removed if we ever stop supporting Solaris 8
3595 // and earlier releases.
3596 
3597 static long priocntl_stub(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg);
3598 typedef long (*priocntl_type)(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg);
3599 static priocntl_type priocntl_ptr = priocntl_stub;
3600 
3601 // Stub to set the value of the real pointer, and then call the real
3602 // function.
3603 
3604 static long priocntl_stub(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg) {
3605   // Try Solaris 8- name only.
3606   priocntl_type tmp = (priocntl_type)dlsym(RTLD_DEFAULT, "__priocntl");
3607   guarantee(tmp != NULL, "priocntl function not found.");
3608   priocntl_ptr = tmp;
3609   return (*priocntl_ptr)(PC_VERSION, idtype, id, cmd, arg);
3610 }
3611 
3612 
3613 // lwp_priocntl_init
3614 //
3615 // Try to determine the priority scale for our process.
3616 //
3617 // Return errno or 0 if OK.
3618 //
3619 static
3620 int     lwp_priocntl_init ()
3621 {
3622   int rslt;
3623   pcinfo_t ClassInfo;
3624   pcparms_t ParmInfo;
3625   int i;
3626 
3627   if (!UseThreadPriorities) return 0;
3628 
3629   // We are using Bound threads, we need to determine our priority ranges
3630   if (os::Solaris::T2_libthread() || UseBoundThreads) {
3631     // If ThreadPriorityPolicy is 1, switch tables
3632     if (ThreadPriorityPolicy == 1) {
3633       for (i = 0 ; i < MaxPriority+1; i++)
3634         os::java_to_os_priority[i] = prio_policy1[i];
3635     }
3636   }
3637   // Not using Bound Threads, set to ThreadPolicy 1
3638   else {
3639     for ( i = 0 ; i < MaxPriority+1; i++ ) {
3640       os::java_to_os_priority[i] = prio_policy1[i];
3641     }
3642     return 0;
3643   }
3644 
3645 
3646   // Get IDs for a set of well-known scheduling classes.
3647   // TODO-FIXME: GETCLINFO returns the current # of classes in the
3648   // the system.  We should have a loop that iterates over the
3649   // classID values, which are known to be "small" integers.
3650 
3651   strcpy(ClassInfo.pc_clname, "TS");
3652   ClassInfo.pc_cid = -1;
3653   rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3654   if (rslt < 0) return errno;
3655   assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3656   tsLimits.schedPolicy = ClassInfo.pc_cid;
3657   tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3658   tsLimits.minPrio = -tsLimits.maxPrio;
3659 
3660   strcpy(ClassInfo.pc_clname, "IA");
3661   ClassInfo.pc_cid = -1;
3662   rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3663   if (rslt < 0) return errno;
3664   assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3665   iaLimits.schedPolicy = ClassInfo.pc_cid;
3666   iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3667   iaLimits.minPrio = -iaLimits.maxPrio;
3668 
3669   strcpy(ClassInfo.pc_clname, "RT");
3670   ClassInfo.pc_cid = -1;
3671   rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3672   if (rslt < 0) return errno;
3673   assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3674   rtLimits.schedPolicy = ClassInfo.pc_cid;
3675   rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3676   rtLimits.minPrio = 0;
3677 
3678 
3679   // Query our "current" scheduling class.
3680   // This will normally be IA,TS or, rarely, RT.
3681   memset (&ParmInfo, 0, sizeof(ParmInfo));
3682   ParmInfo.pc_cid = PC_CLNULL;
3683   rslt = (*priocntl_ptr) (PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo );
3684   if ( rslt < 0 ) return errno;
3685   myClass = ParmInfo.pc_cid;
3686 
3687   // We now know our scheduling classId, get specific information
3688   // the class.
3689   ClassInfo.pc_cid = myClass;
3690   ClassInfo.pc_clname[0] = 0;
3691   rslt = (*priocntl_ptr) (PC_VERSION, (idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo );
3692   if ( rslt < 0 ) return errno;
3693 
3694   if (ThreadPriorityVerbose)
3695     tty->print_cr ("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3696 
3697   memset(&ParmInfo, 0, sizeof(pcparms_t));
3698   ParmInfo.pc_cid = PC_CLNULL;
3699   rslt = (*priocntl_ptr)(PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3700   if (rslt < 0) return errno;
3701 
3702   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3703     myMin = rtLimits.minPrio;
3704     myMax = rtLimits.maxPrio;
3705   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3706     iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3707     myMin = iaLimits.minPrio;
3708     myMax = iaLimits.maxPrio;
3709     myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3710   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3711     tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3712     myMin = tsLimits.minPrio;
3713     myMax = tsLimits.maxPrio;
3714     myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3715   } else {
3716     // No clue - punt
3717     if (ThreadPriorityVerbose)
3718       tty->print_cr ("Unknown scheduling class: %s ... \n", ClassInfo.pc_clname);
3719     return EINVAL;      // no clue, punt
3720   }
3721 
3722   if (ThreadPriorityVerbose)
3723         tty->print_cr ("Thread priority Range: [%d..%d]\n", myMin, myMax);
3724 
3725   priocntl_enable = true;  // Enable changing priorities
3726   return 0;
3727 }
3728 
3729 #define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3730 #define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3731 #define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3732 
3733 
3734 // scale_to_lwp_priority
3735 //
3736 // Convert from the libthread "thr_setprio" scale to our current
3737 // lwp scheduling class scale.
3738 //
3739 static
3740 int     scale_to_lwp_priority (int rMin, int rMax, int x)
3741 {
3742   int v;
3743 
3744   if (x == 127) return rMax;            // avoid round-down
3745     v = (((x*(rMax-rMin)))/128)+rMin;
3746   return v;
3747 }
3748 
3749 
3750 // set_lwp_priority
3751 //
3752 // Set the priority of the lwp.  This call should only be made
3753 // when using bound threads (T2 threads are bound by default).
3754 //
3755 int     set_lwp_priority (int ThreadID, int lwpid, int newPrio )
3756 {
3757   int rslt;
3758   int Actual, Expected, prv;
3759   pcparms_t ParmInfo;                   // for GET-SET
3760 #ifdef ASSERT
3761   pcparms_t ReadBack;                   // for readback
3762 #endif
3763 
3764   // Set priority via PC_GETPARMS, update, PC_SETPARMS
3765   // Query current values.
3766   // TODO: accelerate this by eliminating the PC_GETPARMS call.
3767   // Cache "pcparms_t" in global ParmCache.
3768   // TODO: elide set-to-same-value
3769 
3770   // If something went wrong on init, don't change priorities.
3771   if ( !priocntl_enable ) {
3772     if (ThreadPriorityVerbose)
3773       tty->print_cr("Trying to set priority but init failed, ignoring");
3774     return EINVAL;
3775   }
3776 
3777 
3778   // If lwp hasn't started yet, just return
3779   // the _start routine will call us again.
3780   if ( lwpid <= 0 ) {
3781     if (ThreadPriorityVerbose) {
3782       tty->print_cr ("deferring the set_lwp_priority of thread " INTPTR_FORMAT " to %d, lwpid not set",
3783                      ThreadID, newPrio);
3784     }
3785     return 0;
3786   }
3787 
3788   if (ThreadPriorityVerbose) {
3789     tty->print_cr ("set_lwp_priority(" INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3790                    ThreadID, lwpid, newPrio);
3791   }
3792 
3793   memset(&ParmInfo, 0, sizeof(pcparms_t));
3794   ParmInfo.pc_cid = PC_CLNULL;
3795   rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3796   if (rslt < 0) return errno;
3797 
3798   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3799     rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3800     rtInfo->rt_pri     = scale_to_lwp_priority (rtLimits.minPrio, rtLimits.maxPrio, newPrio);
3801     rtInfo->rt_tqsecs  = RT_NOCHANGE;
3802     rtInfo->rt_tqnsecs = RT_NOCHANGE;
3803     if (ThreadPriorityVerbose) {
3804       tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3805     }
3806   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3807     iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3808     int maxClamped     = MIN2(iaLimits.maxPrio, (int)iaInfo->ia_uprilim);
3809     iaInfo->ia_upri    = scale_to_lwp_priority(iaLimits.minPrio, maxClamped, newPrio);
3810     iaInfo->ia_uprilim = IA_NOCHANGE;
3811     iaInfo->ia_mode    = IA_NOCHANGE;
3812     if (ThreadPriorityVerbose) {
3813       tty->print_cr ("IA: [%d...%d] %d->%d\n",
3814                iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3815     }
3816   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3817     tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3818     int maxClamped     = MIN2(tsLimits.maxPrio, (int)tsInfo->ts_uprilim);
3819     prv                = tsInfo->ts_upri;
3820     tsInfo->ts_upri    = scale_to_lwp_priority(tsLimits.minPrio, maxClamped, newPrio);
3821     tsInfo->ts_uprilim = IA_NOCHANGE;
3822     if (ThreadPriorityVerbose) {
3823       tty->print_cr ("TS: %d [%d...%d] %d->%d\n",
3824                prv, tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
3825     }
3826     if (prv == tsInfo->ts_upri) return 0;
3827   } else {
3828     if ( ThreadPriorityVerbose ) {
3829       tty->print_cr ("Unknown scheduling class\n");
3830     }
3831       return EINVAL;    // no clue, punt
3832   }
3833 
3834   rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
3835   if (ThreadPriorityVerbose && rslt) {
3836     tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
3837   }
3838   if (rslt < 0) return errno;
3839 
3840 #ifdef ASSERT
3841   // Sanity check: read back what we just attempted to set.
3842   // In theory it could have changed in the interim ...
3843   //
3844   // The priocntl system call is tricky.
3845   // Sometimes it'll validate the priority value argument and
3846   // return EINVAL if unhappy.  At other times it fails silently.
3847   // Readbacks are prudent.
3848 
3849   if (!ReadBackValidate) return 0;
3850 
3851   memset(&ReadBack, 0, sizeof(pcparms_t));
3852   ReadBack.pc_cid = PC_CLNULL;
3853   rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
3854   assert(rslt >= 0, "priocntl failed");
3855   Actual = Expected = 0xBAD;
3856   assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
3857   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3858     Actual   = RTPRI(ReadBack)->rt_pri;
3859     Expected = RTPRI(ParmInfo)->rt_pri;
3860   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3861     Actual   = IAPRI(ReadBack)->ia_upri;
3862     Expected = IAPRI(ParmInfo)->ia_upri;
3863   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3864     Actual   = TSPRI(ReadBack)->ts_upri;
3865     Expected = TSPRI(ParmInfo)->ts_upri;
3866   } else {
3867     if ( ThreadPriorityVerbose ) {
3868       tty->print_cr("set_lwp_priority: unexpected class in readback: %d\n", ParmInfo.pc_cid);
3869     }
3870   }
3871 
3872   if (Actual != Expected) {
3873     if ( ThreadPriorityVerbose ) {
3874       tty->print_cr ("set_lwp_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
3875              lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
3876     }
3877   }
3878 #endif
3879 
3880   return 0;
3881 }
3882 
3883 
3884 
3885 // Solaris only gives access to 128 real priorities at a time,
3886 // so we expand Java's ten to fill this range.  This would be better
3887 // if we dynamically adjusted relative priorities.
3888 //
3889 // The ThreadPriorityPolicy option allows us to select 2 different
3890 // priority scales.
3891 //
3892 // ThreadPriorityPolicy=0
3893 // Since the Solaris' default priority is MaximumPriority, we do not
3894 // set a priority lower than Max unless a priority lower than
3895 // NormPriority is requested.
3896 //
3897 // ThreadPriorityPolicy=1
3898 // This mode causes the priority table to get filled with
3899 // linear values.  NormPriority get's mapped to 50% of the
3900 // Maximum priority an so on.  This will cause VM threads
3901 // to get unfair treatment against other Solaris processes
3902 // which do not explicitly alter their thread priorities.
3903 //
3904 
3905 
3906 int os::java_to_os_priority[MaxPriority + 1] = {
3907   -99999,         // 0 Entry should never be used
3908 
3909   0,              // 1 MinPriority
3910   32,             // 2
3911   64,             // 3
3912 
3913   96,             // 4
3914   127,            // 5 NormPriority
3915   127,            // 6
3916 
3917   127,            // 7
3918   127,            // 8
3919   127,            // 9 NearMaxPriority
3920 
3921   127             // 10 MaxPriority
3922 };
3923 
3924 
3925 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3926   assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
3927   if ( !UseThreadPriorities ) return OS_OK;
3928   int status = thr_setprio(thread->osthread()->thread_id(), newpri);
3929   if ( os::Solaris::T2_libthread() || (UseBoundThreads && thread->osthread()->is_vm_created()) )
3930     status |= (set_lwp_priority (thread->osthread()->thread_id(),
3931                     thread->osthread()->lwp_id(), newpri ));
3932   return (status == 0) ? OS_OK : OS_ERR;
3933 }
3934 
3935 
3936 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3937   int p;
3938   if ( !UseThreadPriorities ) {
3939     *priority_ptr = NormalPriority;
3940     return OS_OK;
3941   }
3942   int status = thr_getprio(thread->osthread()->thread_id(), &p);
3943   if (status != 0) {
3944     return OS_ERR;
3945   }
3946   *priority_ptr = p;
3947   return OS_OK;
3948 }
3949 
3950 
3951 // Hint to the underlying OS that a task switch would not be good.
3952 // Void return because it's a hint and can fail.
3953 void os::hint_no_preempt() {
3954   schedctl_start(schedctl_init());
3955 }
3956 
3957 void os::interrupt(Thread* thread) {
3958   assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
3959 
3960   OSThread* osthread = thread->osthread();
3961 
3962   int isInterrupted = osthread->interrupted();
3963   if (!isInterrupted) {
3964       osthread->set_interrupted(true);
3965       OrderAccess::fence();
3966       // os::sleep() is implemented with either poll (NULL,0,timeout) or
3967       // by parking on _SleepEvent.  If the former, thr_kill will unwedge
3968       // the sleeper by SIGINTR, otherwise the unpark() will wake the sleeper.
3969       ParkEvent * const slp = thread->_SleepEvent ;
3970       if (slp != NULL) slp->unpark() ;
3971   }
3972 
3973   // For JSR166:  unpark after setting status but before thr_kill -dl
3974   if (thread->is_Java_thread()) {
3975     ((JavaThread*)thread)->parker()->unpark();
3976   }
3977 
3978   // Handle interruptible wait() ...
3979   ParkEvent * const ev = thread->_ParkEvent ;
3980   if (ev != NULL) ev->unpark() ;
3981 
3982   // When events are used everywhere for os::sleep, then this thr_kill
3983   // will only be needed if UseVMInterruptibleIO is true.
3984 
3985   if (!isInterrupted) {
3986     int status = thr_kill(osthread->thread_id(), os::Solaris::SIGinterrupt());
3987     assert_status(status == 0, status, "thr_kill");
3988 
3989     // Bump thread interruption counter
3990     RuntimeService::record_thread_interrupt_signaled_count();
3991   }
3992 }
3993 
3994 
3995 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3996   assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
3997 
3998   OSThread* osthread = thread->osthread();
3999 
4000   bool res = osthread->interrupted();
4001 
4002   // NOTE that since there is no "lock" around these two operations,
4003   // there is the possibility that the interrupted flag will be
4004   // "false" but that the interrupt event will be set. This is
4005   // intentional. The effect of this is that Object.wait() will appear
4006   // to have a spurious wakeup, which is not harmful, and the
4007   // possibility is so rare that it is not worth the added complexity
4008   // to add yet another lock. It has also been recommended not to put
4009   // the interrupted flag into the os::Solaris::Event structure,
4010   // because it hides the issue.
4011   if (res && clear_interrupted) {
4012     osthread->set_interrupted(false);
4013   }
4014   return res;
4015 }
4016 
4017 
4018 void os::print_statistics() {
4019 }
4020 
4021 int os::message_box(const char* title, const char* message) {
4022   int i;
4023   fdStream err(defaultStream::error_fd());
4024   for (i = 0; i < 78; i++) err.print_raw("=");
4025   err.cr();
4026   err.print_raw_cr(title);
4027   for (i = 0; i < 78; i++) err.print_raw("-");
4028   err.cr();
4029   err.print_raw_cr(message);
4030   for (i = 0; i < 78; i++) err.print_raw("=");
4031   err.cr();
4032 
4033   char buf[16];
4034   // Prevent process from exiting upon "read error" without consuming all CPU
4035   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4036 
4037   return buf[0] == 'y' || buf[0] == 'Y';
4038 }
4039 
4040 // A lightweight implementation that does not suspend the target thread and
4041 // thus returns only a hint. Used for profiling only!
4042 ExtendedPC os::get_thread_pc(Thread* thread) {
4043   // Make sure that it is called by the watcher and the Threads lock is owned.
4044   assert(Thread::current()->is_Watcher_thread(), "Must be watcher and own Threads_lock");
4045   // For now, is only used to profile the VM Thread
4046   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4047   ExtendedPC epc;
4048 
4049   GetThreadPC_Callback  cb(ProfileVM_lock);
4050   OSThread *osthread = thread->osthread();
4051   const int time_to_wait = 400; // 400ms wait for initial response
4052   int status = cb.interrupt(thread, time_to_wait);
4053 
4054   if (cb.is_done() ) {
4055     epc = cb.addr();
4056   } else {
4057     DEBUG_ONLY(tty->print_cr("Failed to get pc for thread: %d got %d status",
4058                               osthread->thread_id(), status););
4059     // epc is already NULL
4060   }
4061   return epc;
4062 }
4063 
4064 
4065 // This does not do anything on Solaris. This is basically a hook for being
4066 // able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
4067 void os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, JavaCallArguments* args, Thread* thread) {
4068   f(value, method, args, thread);
4069 }
4070 
4071 // This routine may be used by user applications as a "hook" to catch signals.
4072 // The user-defined signal handler must pass unrecognized signals to this
4073 // routine, and if it returns true (non-zero), then the signal handler must
4074 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4075 // routine will never retun false (zero), but instead will execute a VM panic
4076 // routine kill the process.
4077 //
4078 // If this routine returns false, it is OK to call it again.  This allows
4079 // the user-defined signal handler to perform checks either before or after
4080 // the VM performs its own checks.  Naturally, the user code would be making
4081 // a serious error if it tried to handle an exception (such as a null check
4082 // or breakpoint) that the VM was generating for its own correct operation.
4083 //
4084 // This routine may recognize any of the following kinds of signals:
4085 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
4086 // os::Solaris::SIGasync
4087 // It should be consulted by handlers for any of those signals.
4088 // It explicitly does not recognize os::Solaris::SIGinterrupt
4089 //
4090 // The caller of this routine must pass in the three arguments supplied
4091 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4092 // field of the structure passed to sigaction().  This routine assumes that
4093 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4094 //
4095 // Note that the VM will print warnings if it detects conflicting signal
4096 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4097 //
4098 extern "C" int JVM_handle_solaris_signal(int signo, siginfo_t* siginfo, void* ucontext, int abort_if_unrecognized);
4099 
4100 
4101 void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
4102   JVM_handle_solaris_signal(sig, info, ucVoid, true);
4103 }
4104 
4105 /* Do not delete - if guarantee is ever removed,  a signal handler (even empty)
4106    is needed to provoke threads blocked on IO to return an EINTR
4107    Note: this explicitly does NOT call JVM_handle_solaris_signal and
4108    does NOT participate in signal chaining due to requirement for
4109    NOT setting SA_RESTART to make EINTR work. */
4110 extern "C" void sigINTRHandler(int sig, siginfo_t* info, void* ucVoid) {
4111    if (UseSignalChaining) {
4112       struct sigaction *actp = os::Solaris::get_chained_signal_action(sig);
4113       if (actp && actp->sa_handler) {
4114         vm_exit_during_initialization("Signal chaining detected for VM interrupt signal, try -XX:+UseAltSigs");
4115       }
4116    }
4117 }
4118 
4119 // This boolean allows users to forward their own non-matching signals
4120 // to JVM_handle_solaris_signal, harmlessly.
4121 bool os::Solaris::signal_handlers_are_installed = false;
4122 
4123 // For signal-chaining
4124 bool os::Solaris::libjsig_is_loaded = false;
4125 typedef struct sigaction *(*get_signal_t)(int);
4126 get_signal_t os::Solaris::get_signal_action = NULL;
4127 
4128 struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
4129   struct sigaction *actp = NULL;
4130 
4131   if ((libjsig_is_loaded)  && (sig <= Maxlibjsigsigs)) {
4132     // Retrieve the old signal handler from libjsig
4133     actp = (*get_signal_action)(sig);
4134   }
4135   if (actp == NULL) {
4136     // Retrieve the preinstalled signal handler from jvm
4137     actp = get_preinstalled_handler(sig);
4138   }
4139 
4140   return actp;
4141 }
4142 
4143 static bool call_chained_handler(struct sigaction *actp, int sig,
4144                                  siginfo_t *siginfo, void *context) {
4145   // Call the old signal handler
4146   if (actp->sa_handler == SIG_DFL) {
4147     // It's more reasonable to let jvm treat it as an unexpected exception
4148     // instead of taking the default action.
4149     return false;
4150   } else if (actp->sa_handler != SIG_IGN) {
4151     if ((actp->sa_flags & SA_NODEFER) == 0) {
4152       // automaticlly block the signal
4153       sigaddset(&(actp->sa_mask), sig);
4154     }
4155 
4156     sa_handler_t hand;
4157     sa_sigaction_t sa;
4158     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4159     // retrieve the chained handler
4160     if (siginfo_flag_set) {
4161       sa = actp->sa_sigaction;
4162     } else {
4163       hand = actp->sa_handler;
4164     }
4165 
4166     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4167       actp->sa_handler = SIG_DFL;
4168     }
4169 
4170     // try to honor the signal mask
4171     sigset_t oset;
4172     thr_sigsetmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4173 
4174     // call into the chained handler
4175     if (siginfo_flag_set) {
4176       (*sa)(sig, siginfo, context);
4177     } else {
4178       (*hand)(sig);
4179     }
4180 
4181     // restore the signal mask
4182     thr_sigsetmask(SIG_SETMASK, &oset, 0);
4183   }
4184   // Tell jvm's signal handler the signal is taken care of.
4185   return true;
4186 }
4187 
4188 bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4189   bool chained = false;
4190   // signal-chaining
4191   if (UseSignalChaining) {
4192     struct sigaction *actp = get_chained_signal_action(sig);
4193     if (actp != NULL) {
4194       chained = call_chained_handler(actp, sig, siginfo, context);
4195     }
4196   }
4197   return chained;
4198 }
4199 
4200 struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
4201   assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4202   if (preinstalled_sigs[sig] != 0) {
4203     return &chainedsigactions[sig];
4204   }
4205   return NULL;
4206 }
4207 
4208 void os::Solaris::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4209 
4210   assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
4211   assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4212   chainedsigactions[sig] = oldAct;
4213   preinstalled_sigs[sig] = 1;
4214 }
4215 
4216 void os::Solaris::set_signal_handler(int sig, bool set_installed, bool oktochain) {
4217   // Check for overwrite.
4218   struct sigaction oldAct;
4219   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4220   void* oldhand = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4221                                       : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4222   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4223       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4224       oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
4225     if (AllowUserSignalHandlers || !set_installed) {
4226       // Do not overwrite; user takes responsibility to forward to us.
4227       return;
4228     } else if (UseSignalChaining) {
4229       if (oktochain) {
4230         // save the old handler in jvm
4231         save_preinstalled_handler(sig, oldAct);
4232       } else {
4233         vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal, try -XX:+UseAltSigs.");
4234       }
4235       // libjsig also interposes the sigaction() call below and saves the
4236       // old sigaction on it own.
4237     } else {
4238       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4239                     "%#lx for signal %d.", (long)oldhand, sig));
4240     }
4241   }
4242 
4243   struct sigaction sigAct;
4244   sigfillset(&(sigAct.sa_mask));
4245   sigAct.sa_handler = SIG_DFL;
4246 
4247   sigAct.sa_sigaction = signalHandler;
4248   // Handle SIGSEGV on alternate signal stack if
4249   // not using stack banging
4250   if (!UseStackBanging && sig == SIGSEGV) {
4251     sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
4252   // Interruptible i/o requires SA_RESTART cleared so EINTR
4253   // is returned instead of restarting system calls
4254   } else if (sig == os::Solaris::SIGinterrupt()) {
4255     sigemptyset(&sigAct.sa_mask);
4256     sigAct.sa_handler = NULL;
4257     sigAct.sa_flags = SA_SIGINFO;
4258     sigAct.sa_sigaction = sigINTRHandler;
4259   } else {
4260     sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
4261   }
4262   os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
4263 
4264   sigaction(sig, &sigAct, &oldAct);
4265 
4266   void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4267                                        : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4268   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4269 }
4270 
4271 
4272 #define DO_SIGNAL_CHECK(sig) \
4273   if (!sigismember(&check_signal_done, sig)) \
4274     os::Solaris::check_signal_handler(sig)
4275 
4276 // This method is a periodic task to check for misbehaving JNI applications
4277 // under CheckJNI, we can add any periodic checks here
4278 
4279 void os::run_periodic_checks() {
4280   // A big source of grief is hijacking virt. addr 0x0 on Solaris,
4281   // thereby preventing a NULL checks.
4282   if(!check_addr0_done) check_addr0_done = check_addr0(tty);
4283 
4284   if (check_signals == false) return;
4285 
4286   // SEGV and BUS if overridden could potentially prevent
4287   // generation of hs*.log in the event of a crash, debugging
4288   // such a case can be very challenging, so we absolutely
4289   // check for the following for a good measure:
4290   DO_SIGNAL_CHECK(SIGSEGV);
4291   DO_SIGNAL_CHECK(SIGILL);
4292   DO_SIGNAL_CHECK(SIGFPE);
4293   DO_SIGNAL_CHECK(SIGBUS);
4294   DO_SIGNAL_CHECK(SIGPIPE);
4295   DO_SIGNAL_CHECK(SIGXFSZ);
4296 
4297   // ReduceSignalUsage allows the user to override these handlers
4298   // see comments at the very top and jvm_solaris.h
4299   if (!ReduceSignalUsage) {
4300     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4301     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4302     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4303     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4304   }
4305 
4306   // See comments above for using JVM1/JVM2 and UseAltSigs
4307   DO_SIGNAL_CHECK(os::Solaris::SIGinterrupt());
4308   DO_SIGNAL_CHECK(os::Solaris::SIGasync());
4309 
4310 }
4311 
4312 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4313 
4314 static os_sigaction_t os_sigaction = NULL;
4315 
4316 void os::Solaris::check_signal_handler(int sig) {
4317   char buf[O_BUFLEN];
4318   address jvmHandler = NULL;
4319 
4320   struct sigaction act;
4321   if (os_sigaction == NULL) {
4322     // only trust the default sigaction, in case it has been interposed
4323     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4324     if (os_sigaction == NULL) return;
4325   }
4326 
4327   os_sigaction(sig, (struct sigaction*)NULL, &act);
4328 
4329   address thisHandler = (act.sa_flags & SA_SIGINFO)
4330     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4331     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4332 
4333 
4334   switch(sig) {
4335     case SIGSEGV:
4336     case SIGBUS:
4337     case SIGFPE:
4338     case SIGPIPE:
4339     case SIGXFSZ:
4340     case SIGILL:
4341       jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4342       break;
4343 
4344     case SHUTDOWN1_SIGNAL:
4345     case SHUTDOWN2_SIGNAL:
4346     case SHUTDOWN3_SIGNAL:
4347     case BREAK_SIGNAL:
4348       jvmHandler = (address)user_handler();
4349       break;
4350 
4351     default:
4352       int intrsig = os::Solaris::SIGinterrupt();
4353       int asynsig = os::Solaris::SIGasync();
4354 
4355       if (sig == intrsig) {
4356         jvmHandler = CAST_FROM_FN_PTR(address, sigINTRHandler);
4357       } else if (sig == asynsig) {
4358         jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4359       } else {
4360         return;
4361       }
4362       break;
4363   }
4364 
4365 
4366   if (thisHandler != jvmHandler) {
4367     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4368     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4369     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4370     // No need to check this sig any longer
4371     sigaddset(&check_signal_done, sig);
4372   } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
4373     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4374     tty->print("expected:" PTR32_FORMAT, os::Solaris::get_our_sigflags(sig));
4375     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4376     // No need to check this sig any longer
4377     sigaddset(&check_signal_done, sig);
4378   }
4379 
4380   // Print all the signal handler state
4381   if (sigismember(&check_signal_done, sig)) {
4382     print_signal_handlers(tty, buf, O_BUFLEN);
4383   }
4384 
4385 }
4386 
4387 void os::Solaris::install_signal_handlers() {
4388   bool libjsigdone = false;
4389   signal_handlers_are_installed = true;
4390 
4391   // signal-chaining
4392   typedef void (*signal_setting_t)();
4393   signal_setting_t begin_signal_setting = NULL;
4394   signal_setting_t end_signal_setting = NULL;
4395   begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4396                                         dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4397   if (begin_signal_setting != NULL) {
4398     end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4399                                         dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4400     get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4401                                        dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4402     get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
4403                                          dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
4404     libjsig_is_loaded = true;
4405     if (os::Solaris::get_libjsig_version != NULL) {
4406       libjsigversion =  (*os::Solaris::get_libjsig_version)();
4407     }
4408     assert(UseSignalChaining, "should enable signal-chaining");
4409   }
4410   if (libjsig_is_loaded) {
4411     // Tell libjsig jvm is setting signal handlers
4412     (*begin_signal_setting)();
4413   }
4414 
4415   set_signal_handler(SIGSEGV, true, true);
4416   set_signal_handler(SIGPIPE, true, true);
4417   set_signal_handler(SIGXFSZ, true, true);
4418   set_signal_handler(SIGBUS, true, true);
4419   set_signal_handler(SIGILL, true, true);
4420   set_signal_handler(SIGFPE, true, true);
4421 
4422 
4423   if (os::Solaris::SIGinterrupt() > OLDMAXSIGNUM || os::Solaris::SIGasync() > OLDMAXSIGNUM) {
4424 
4425     // Pre-1.4.1 Libjsig limited to signal chaining signals <= 32 so
4426     // can not register overridable signals which might be > 32
4427     if (libjsig_is_loaded && libjsigversion <= JSIG_VERSION_1_4_1) {
4428     // Tell libjsig jvm has finished setting signal handlers
4429       (*end_signal_setting)();
4430       libjsigdone = true;
4431     }
4432   }
4433 
4434   // Never ok to chain our SIGinterrupt
4435   set_signal_handler(os::Solaris::SIGinterrupt(), true, false);
4436   set_signal_handler(os::Solaris::SIGasync(), true, true);
4437 
4438   if (libjsig_is_loaded && !libjsigdone) {
4439     // Tell libjsig jvm finishes setting signal handlers
4440     (*end_signal_setting)();
4441   }
4442 
4443   // We don't activate signal checker if libjsig is in place, we trust ourselves
4444   // and if UserSignalHandler is installed all bets are off
4445   if (CheckJNICalls) {
4446     if (libjsig_is_loaded) {
4447       tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4448       check_signals = false;
4449     }
4450     if (AllowUserSignalHandlers) {
4451       tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4452       check_signals = false;
4453     }
4454   }
4455 }
4456 
4457 
4458 void report_error(const char* file_name, int line_no, const char* title, const char* format, ...);
4459 
4460 const char * signames[] = {
4461   "SIG0",
4462   "SIGHUP", "SIGINT", "SIGQUIT", "SIGILL", "SIGTRAP",
4463   "SIGABRT", "SIGEMT", "SIGFPE", "SIGKILL", "SIGBUS",
4464   "SIGSEGV", "SIGSYS", "SIGPIPE", "SIGALRM", "SIGTERM",
4465   "SIGUSR1", "SIGUSR2", "SIGCLD", "SIGPWR", "SIGWINCH",
4466   "SIGURG", "SIGPOLL", "SIGSTOP", "SIGTSTP", "SIGCONT",
4467   "SIGTTIN", "SIGTTOU", "SIGVTALRM", "SIGPROF", "SIGXCPU",
4468   "SIGXFSZ", "SIGWAITING", "SIGLWP", "SIGFREEZE", "SIGTHAW",
4469   "SIGCANCEL", "SIGLOST"
4470 };
4471 
4472 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4473   if (0 < exception_code && exception_code <= SIGRTMAX) {
4474     // signal
4475     if (exception_code < sizeof(signames)/sizeof(const char*)) {
4476        jio_snprintf(buf, size, "%s", signames[exception_code]);
4477     } else {
4478        jio_snprintf(buf, size, "SIG%d", exception_code);
4479     }
4480     return buf;
4481   } else {
4482     return NULL;
4483   }
4484 }
4485 
4486 // (Static) wrappers for the new libthread API
4487 int_fnP_thread_t_iP_uP_stack_tP_gregset_t os::Solaris::_thr_getstate;
4488 int_fnP_thread_t_i_gregset_t os::Solaris::_thr_setstate;
4489 int_fnP_thread_t_i os::Solaris::_thr_setmutator;
4490 int_fnP_thread_t os::Solaris::_thr_suspend_mutator;
4491 int_fnP_thread_t os::Solaris::_thr_continue_mutator;
4492 
4493 // (Static) wrapper for getisax(2) call.
4494 os::Solaris::getisax_func_t os::Solaris::_getisax = 0;
4495 
4496 // (Static) wrappers for the liblgrp API
4497 os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4498 os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4499 os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4500 os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4501 os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4502 os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4503 os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4504 os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4505 os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4506 
4507 // (Static) wrapper for meminfo() call.
4508 os::Solaris::meminfo_func_t os::Solaris::_meminfo = 0;
4509 
4510 static address resolve_symbol_lazy(const char* name) {
4511   address addr = (address) dlsym(RTLD_DEFAULT, name);
4512   if(addr == NULL) {
4513     // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4514     addr = (address) dlsym(RTLD_NEXT, name);
4515   }
4516   return addr;
4517 }
4518 
4519 static address resolve_symbol(const char* name) {
4520   address addr = resolve_symbol_lazy(name);
4521   if(addr == NULL) {
4522     fatal(dlerror());
4523   }
4524   return addr;
4525 }
4526 
4527 
4528 
4529 // isT2_libthread()
4530 //
4531 // Routine to determine if we are currently using the new T2 libthread.
4532 //
4533 // We determine if we are using T2 by reading /proc/self/lstatus and
4534 // looking for a thread with the ASLWP bit set.  If we find this status
4535 // bit set, we must assume that we are NOT using T2.  The T2 team
4536 // has approved this algorithm.
4537 //
4538 // We need to determine if we are running with the new T2 libthread
4539 // since setting native thread priorities is handled differently
4540 // when using this library.  All threads created using T2 are bound
4541 // threads. Calling thr_setprio is meaningless in this case.
4542 //
4543 bool isT2_libthread() {
4544   static prheader_t * lwpArray = NULL;
4545   static int lwpSize = 0;
4546   static int lwpFile = -1;
4547   lwpstatus_t * that;
4548   char lwpName [128];
4549   bool isT2 = false;
4550 
4551 #define ADR(x)  ((uintptr_t)(x))
4552 #define LWPINDEX(ary,ix)   ((lwpstatus_t *)(((ary)->pr_entsize * (ix)) + (ADR((ary) + 1))))
4553 
4554   lwpFile = open("/proc/self/lstatus", O_RDONLY, 0);
4555   if (lwpFile < 0) {
4556       if (ThreadPriorityVerbose) warning ("Couldn't open /proc/self/lstatus\n");
4557       return false;
4558   }
4559   lwpSize = 16*1024;
4560   for (;;) {
4561     lseek (lwpFile, 0, SEEK_SET);
4562     lwpArray = (prheader_t *)NEW_C_HEAP_ARRAY(char, lwpSize);
4563     if (read(lwpFile, lwpArray, lwpSize) < 0) {
4564       if (ThreadPriorityVerbose) warning("Error reading /proc/self/lstatus\n");
4565       break;
4566     }
4567     if ((lwpArray->pr_nent * lwpArray->pr_entsize) <= lwpSize) {
4568        // We got a good snapshot - now iterate over the list.
4569       int aslwpcount = 0;
4570       for (int i = 0; i < lwpArray->pr_nent; i++ ) {
4571         that = LWPINDEX(lwpArray,i);
4572         if (that->pr_flags & PR_ASLWP) {
4573           aslwpcount++;
4574         }
4575       }
4576       if (aslwpcount == 0) isT2 = true;
4577       break;
4578     }
4579     lwpSize = lwpArray->pr_nent * lwpArray->pr_entsize;
4580     FREE_C_HEAP_ARRAY(char, lwpArray);  // retry.
4581   }
4582 
4583   FREE_C_HEAP_ARRAY(char, lwpArray);
4584   close (lwpFile);
4585   if (ThreadPriorityVerbose) {
4586     if (isT2) tty->print_cr("We are running with a T2 libthread\n");
4587     else tty->print_cr("We are not running with a T2 libthread\n");
4588   }
4589   return isT2;
4590 }
4591 
4592 
4593 void os::Solaris::libthread_init() {
4594   address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4595 
4596   // Determine if we are running with the new T2 libthread
4597   os::Solaris::set_T2_libthread(isT2_libthread());
4598 
4599   lwp_priocntl_init();
4600 
4601   // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4602   if(func == NULL) {
4603     func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4604     // Guarantee that this VM is running on an new enough OS (5.6 or
4605     // later) that it will have a new enough libthread.so.
4606     guarantee(func != NULL, "libthread.so is too old.");
4607   }
4608 
4609   // Initialize the new libthread getstate API wrappers
4610   func = resolve_symbol("thr_getstate");
4611   os::Solaris::set_thr_getstate(CAST_TO_FN_PTR(int_fnP_thread_t_iP_uP_stack_tP_gregset_t, func));
4612 
4613   func = resolve_symbol("thr_setstate");
4614   os::Solaris::set_thr_setstate(CAST_TO_FN_PTR(int_fnP_thread_t_i_gregset_t, func));
4615 
4616   func = resolve_symbol("thr_setmutator");
4617   os::Solaris::set_thr_setmutator(CAST_TO_FN_PTR(int_fnP_thread_t_i, func));
4618 
4619   func = resolve_symbol("thr_suspend_mutator");
4620   os::Solaris::set_thr_suspend_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4621 
4622   func = resolve_symbol("thr_continue_mutator");
4623   os::Solaris::set_thr_continue_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4624 
4625   int size;
4626   void (*handler_info_func)(address *, int *);
4627   handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4628   handler_info_func(&handler_start, &size);
4629   handler_end = handler_start + size;
4630 }
4631 
4632 
4633 int_fnP_mutex_tP os::Solaris::_mutex_lock;
4634 int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4635 int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4636 int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4637 int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4638 int os::Solaris::_mutex_scope = USYNC_THREAD;
4639 
4640 int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4641 int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4642 int_fnP_cond_tP os::Solaris::_cond_signal;
4643 int_fnP_cond_tP os::Solaris::_cond_broadcast;
4644 int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4645 int_fnP_cond_tP os::Solaris::_cond_destroy;
4646 int os::Solaris::_cond_scope = USYNC_THREAD;
4647 
4648 void os::Solaris::synchronization_init() {
4649   if(UseLWPSynchronization) {
4650     os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4651     os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4652     os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4653     os::Solaris::set_mutex_init(lwp_mutex_init);
4654     os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4655     os::Solaris::set_mutex_scope(USYNC_THREAD);
4656 
4657     os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4658     os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4659     os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4660     os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4661     os::Solaris::set_cond_init(lwp_cond_init);
4662     os::Solaris::set_cond_destroy(lwp_cond_destroy);
4663     os::Solaris::set_cond_scope(USYNC_THREAD);
4664   }
4665   else {
4666     os::Solaris::set_mutex_scope(USYNC_THREAD);
4667     os::Solaris::set_cond_scope(USYNC_THREAD);
4668 
4669     if(UsePthreads) {
4670       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4671       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4672       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4673       os::Solaris::set_mutex_init(pthread_mutex_default_init);
4674       os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4675 
4676       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4677       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4678       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4679       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4680       os::Solaris::set_cond_init(pthread_cond_default_init);
4681       os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4682     }
4683     else {
4684       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4685       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4686       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4687       os::Solaris::set_mutex_init(::mutex_init);
4688       os::Solaris::set_mutex_destroy(::mutex_destroy);
4689 
4690       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4691       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4692       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4693       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4694       os::Solaris::set_cond_init(::cond_init);
4695       os::Solaris::set_cond_destroy(::cond_destroy);
4696     }
4697   }
4698 }
4699 
4700 bool os::Solaris::liblgrp_init() {
4701   void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4702   if (handle != NULL) {
4703     os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4704     os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4705     os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4706     os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4707     os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4708     os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4709     os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4710     os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4711                                        dlsym(handle, "lgrp_cookie_stale")));
4712 
4713     lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4714     set_lgrp_cookie(c);
4715     return true;
4716   }
4717   return false;
4718 }
4719 
4720 void os::Solaris::misc_sym_init() {
4721   address func;
4722 
4723   // getisax
4724   func = resolve_symbol_lazy("getisax");
4725   if (func != NULL) {
4726     os::Solaris::_getisax = CAST_TO_FN_PTR(getisax_func_t, func);
4727   }
4728 
4729   // meminfo
4730   func = resolve_symbol_lazy("meminfo");
4731   if (func != NULL) {
4732     os::Solaris::set_meminfo(CAST_TO_FN_PTR(meminfo_func_t, func));
4733   }
4734 }
4735 
4736 uint_t os::Solaris::getisax(uint32_t* array, uint_t n) {
4737   assert(_getisax != NULL, "_getisax not set");
4738   return _getisax(array, n);
4739 }
4740 
4741 // Symbol doesn't exist in Solaris 8 pset.h
4742 #ifndef PS_MYID
4743 #define PS_MYID -3
4744 #endif
4745 
4746 // int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
4747 typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
4748 static pset_getloadavg_type pset_getloadavg_ptr = NULL;
4749 
4750 void init_pset_getloadavg_ptr(void) {
4751   pset_getloadavg_ptr =
4752     (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
4753   if (PrintMiscellaneous && Verbose && pset_getloadavg_ptr == NULL) {
4754     warning("pset_getloadavg function not found");
4755   }
4756 }
4757 
4758 int os::Solaris::_dev_zero_fd = -1;
4759 
4760 // this is called _before_ the global arguments have been parsed
4761 void os::init(void) {
4762   _initial_pid = getpid();
4763 
4764   max_hrtime = first_hrtime = gethrtime();
4765 
4766   init_random(1234567);
4767 
4768   page_size = sysconf(_SC_PAGESIZE);
4769   if (page_size == -1)
4770     fatal(err_msg("os_solaris.cpp: os::init: sysconf failed (%s)",
4771                   strerror(errno)));
4772   init_page_sizes((size_t) page_size);
4773 
4774   Solaris::initialize_system_info();
4775 
4776   // Initialize misc. symbols as soon as possible, so we can use them
4777   // if we need them.
4778   Solaris::misc_sym_init();
4779 
4780   int fd = open("/dev/zero", O_RDWR);
4781   if (fd < 0) {
4782     fatal(err_msg("os::init: cannot open /dev/zero (%s)", strerror(errno)));
4783   } else {
4784     Solaris::set_dev_zero_fd(fd);
4785 
4786     // Close on exec, child won't inherit.
4787     fcntl(fd, F_SETFD, FD_CLOEXEC);
4788   }
4789 
4790   clock_tics_per_sec = CLK_TCK;
4791 
4792   // check if dladdr1() exists; dladdr1 can provide more information than
4793   // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
4794   // and is available on linker patches for 5.7 and 5.8.
4795   // libdl.so must have been loaded, this call is just an entry lookup
4796   void * hdl = dlopen("libdl.so", RTLD_NOW);
4797   if (hdl)
4798     dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
4799 
4800   // (Solaris only) this switches to calls that actually do locking.
4801   ThreadCritical::initialize();
4802 
4803   main_thread = thr_self();
4804 
4805   // Constant minimum stack size allowed. It must be at least
4806   // the minimum of what the OS supports (thr_min_stack()), and
4807   // enough to allow the thread to get to user bytecode execution.
4808   Solaris::min_stack_allowed = MAX2(thr_min_stack(), Solaris::min_stack_allowed);
4809   // If the pagesize of the VM is greater than 8K determine the appropriate
4810   // number of initial guard pages.  The user can change this with the
4811   // command line arguments, if needed.
4812   if (vm_page_size() > 8*K) {
4813     StackYellowPages = 1;
4814     StackRedPages = 1;
4815     StackShadowPages = round_to((StackShadowPages*8*K), vm_page_size()) / vm_page_size();
4816   }
4817 }
4818 
4819 // To install functions for atexit system call
4820 extern "C" {
4821   static void perfMemory_exit_helper() {
4822     perfMemory_exit();
4823   }
4824 }
4825 
4826 // this is called _after_ the global arguments have been parsed
4827 jint os::init_2(void) {
4828   // try to enable extended file IO ASAP, see 6431278
4829   os::Solaris::try_enable_extended_io();
4830 
4831   // Allocate a single page and mark it as readable for safepoint polling.  Also
4832   // use this first mmap call to check support for MAP_ALIGN.
4833   address polling_page = (address)Solaris::mmap_chunk((char*)page_size,
4834                                                       page_size,
4835                                                       MAP_PRIVATE | MAP_ALIGN,
4836                                                       PROT_READ);
4837   if (polling_page == NULL) {
4838     has_map_align = false;
4839     polling_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE,
4840                                                 PROT_READ);
4841   }
4842 
4843   os::set_polling_page(polling_page);
4844 
4845 #ifndef PRODUCT
4846   if( Verbose && PrintMiscellaneous )
4847     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4848 #endif
4849 
4850   if (!UseMembar) {
4851     address mem_serialize_page = (address)Solaris::mmap_chunk( NULL, page_size, MAP_PRIVATE, PROT_READ | PROT_WRITE );
4852     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4853     os::set_memory_serialize_page( mem_serialize_page );
4854 
4855 #ifndef PRODUCT
4856     if(Verbose && PrintMiscellaneous)
4857       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4858 #endif
4859 }
4860 
4861   FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
4862 
4863   // Check minimum allowable stack size for thread creation and to initialize
4864   // the java system classes, including StackOverflowError - depends on page
4865   // size.  Add a page for compiler2 recursion in main thread.
4866   // Add in BytesPerWord times page size to account for VM stack during
4867   // class initialization depending on 32 or 64 bit VM.
4868   guarantee((Solaris::min_stack_allowed >=
4869     (StackYellowPages+StackRedPages+StackShadowPages+BytesPerWord
4870      COMPILER2_PRESENT(+1)) * page_size),
4871     "need to increase Solaris::min_stack_allowed on this platform");
4872 
4873   size_t threadStackSizeInBytes = ThreadStackSize * K;
4874   if (threadStackSizeInBytes != 0 &&
4875     threadStackSizeInBytes < Solaris::min_stack_allowed) {
4876     tty->print_cr("\nThe stack size specified is too small, Specify at least %dk",
4877                   Solaris::min_stack_allowed/K);
4878     return JNI_ERR;
4879   }
4880 
4881   // For 64kbps there will be a 64kb page size, which makes
4882   // the usable default stack size quite a bit less.  Increase the
4883   // stack for 64kb (or any > than 8kb) pages, this increases
4884   // virtual memory fragmentation (since we're not creating the
4885   // stack on a power of 2 boundary.  The real fix for this
4886   // should be to fix the guard page mechanism.
4887 
4888   if (vm_page_size() > 8*K) {
4889       threadStackSizeInBytes = (threadStackSizeInBytes != 0)
4890          ? threadStackSizeInBytes +
4891            ((StackYellowPages + StackRedPages) * vm_page_size())
4892          : 0;
4893       ThreadStackSize = threadStackSizeInBytes/K;
4894   }
4895 
4896   // Make the stack size a multiple of the page size so that
4897   // the yellow/red zones can be guarded.
4898   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4899         vm_page_size()));
4900 
4901   Solaris::libthread_init();
4902 
4903   if (UseNUMA) {
4904     if (!Solaris::liblgrp_init()) {
4905       UseNUMA = false;
4906     } else {
4907       size_t lgrp_limit = os::numa_get_groups_num();
4908       int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit);
4909       size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
4910       FREE_C_HEAP_ARRAY(int, lgrp_ids);
4911       if (lgrp_num < 2) {
4912         // There's only one locality group, disable NUMA.
4913         UseNUMA = false;
4914       }
4915     }
4916     if (!UseNUMA && ForceNUMA) {
4917       UseNUMA = true;
4918     }
4919   }
4920 
4921   Solaris::signal_sets_init();
4922   Solaris::init_signal_mem();
4923   Solaris::install_signal_handlers();
4924 
4925   if (libjsigversion < JSIG_VERSION_1_4_1) {
4926     Maxlibjsigsigs = OLDMAXSIGNUM;
4927   }
4928 
4929   // initialize synchronization primitives to use either thread or
4930   // lwp synchronization (controlled by UseLWPSynchronization)
4931   Solaris::synchronization_init();
4932 
4933   if (MaxFDLimit) {
4934     // set the number of file descriptors to max. print out error
4935     // if getrlimit/setrlimit fails but continue regardless.
4936     struct rlimit nbr_files;
4937     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4938     if (status != 0) {
4939       if (PrintMiscellaneous && (Verbose || WizardMode))
4940         perror("os::init_2 getrlimit failed");
4941     } else {
4942       nbr_files.rlim_cur = nbr_files.rlim_max;
4943       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4944       if (status != 0) {
4945         if (PrintMiscellaneous && (Verbose || WizardMode))
4946           perror("os::init_2 setrlimit failed");
4947       }
4948     }
4949   }
4950 
4951   // Initialize HPI.
4952   jint hpi_result = hpi::initialize();
4953   if (hpi_result != JNI_OK) {
4954     tty->print_cr("There was an error trying to initialize the HPI library.");
4955     return hpi_result;
4956   }
4957 
4958   // Calculate theoretical max. size of Threads to guard gainst
4959   // artifical out-of-memory situations, where all available address-
4960   // space has been reserved by thread stacks. Default stack size is 1Mb.
4961   size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
4962     JavaThread::stack_size_at_create() : (1*K*K);
4963   assert(pre_thread_stack_size != 0, "Must have a stack");
4964   // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
4965   // we should start doing Virtual Memory banging. Currently when the threads will
4966   // have used all but 200Mb of space.
4967   size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
4968   Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
4969 
4970   // at-exit methods are called in the reverse order of their registration.
4971   // In Solaris 7 and earlier, atexit functions are called on return from
4972   // main or as a result of a call to exit(3C). There can be only 32 of
4973   // these functions registered and atexit() does not set errno. In Solaris
4974   // 8 and later, there is no limit to the number of functions registered
4975   // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
4976   // functions are called upon dlclose(3DL) in addition to return from main
4977   // and exit(3C).
4978 
4979   if (PerfAllowAtExitRegistration) {
4980     // only register atexit functions if PerfAllowAtExitRegistration is set.
4981     // atexit functions can be delayed until process exit time, which
4982     // can be problematic for embedded VM situations. Embedded VMs should
4983     // call DestroyJavaVM() to assure that VM resources are released.
4984 
4985     // note: perfMemory_exit_helper atexit function may be removed in
4986     // the future if the appropriate cleanup code can be added to the
4987     // VM_Exit VMOperation's doit method.
4988     if (atexit(perfMemory_exit_helper) != 0) {
4989       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4990     }
4991   }
4992 
4993   // Init pset_loadavg function pointer
4994   init_pset_getloadavg_ptr();
4995 
4996   return JNI_OK;
4997 }
4998 
4999 
5000 // Mark the polling page as unreadable
5001 void os::make_polling_page_unreadable(void) {
5002   if( mprotect((char *)_polling_page, page_size, PROT_NONE) != 0 )
5003     fatal("Could not disable polling page");
5004 };
5005 
5006 // Mark the polling page as readable
5007 void os::make_polling_page_readable(void) {
5008   if( mprotect((char *)_polling_page, page_size, PROT_READ) != 0 )
5009     fatal("Could not enable polling page");
5010 };
5011 
5012 // OS interface.
5013 
5014 int os::stat(const char *path, struct stat *sbuf) {
5015   char pathbuf[MAX_PATH];
5016   if (strlen(path) > MAX_PATH - 1) {
5017     errno = ENAMETOOLONG;
5018     return -1;
5019   }
5020   hpi::native_path(strcpy(pathbuf, path));
5021   return ::stat(pathbuf, sbuf);
5022 }
5023 
5024 
5025 bool os::check_heap(bool force) { return true; }
5026 
5027 typedef int (*vsnprintf_t)(char* buf, size_t count, const char* fmt, va_list argptr);
5028 static vsnprintf_t sol_vsnprintf = NULL;
5029 
5030 int local_vsnprintf(char* buf, size_t count, const char* fmt, va_list argptr) {
5031   if (!sol_vsnprintf) {
5032     //search  for the named symbol in the objects that were loaded after libjvm
5033     void* where = RTLD_NEXT;
5034     if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5035         sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5036     if (!sol_vsnprintf){
5037       //search  for the named symbol in the objects that were loaded before libjvm
5038       where = RTLD_DEFAULT;
5039       if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5040         sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5041       assert(sol_vsnprintf != NULL, "vsnprintf not found");
5042     }
5043   }
5044   return (*sol_vsnprintf)(buf, count, fmt, argptr);
5045 }
5046 
5047 
5048 // Is a (classpath) directory empty?
5049 bool os::dir_is_empty(const char* path) {
5050   DIR *dir = NULL;
5051   struct dirent *ptr;
5052 
5053   dir = opendir(path);
5054   if (dir == NULL) return true;
5055 
5056   /* Scan the directory */
5057   bool result = true;
5058   char buf[sizeof(struct dirent) + MAX_PATH];
5059   struct dirent *dbuf = (struct dirent *) buf;
5060   while (result && (ptr = readdir(dir, dbuf)) != NULL) {
5061     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5062       result = false;
5063     }
5064   }
5065   closedir(dir);
5066   return result;
5067 }
5068 
5069 // create binary file, rewriting existing file if required
5070 int os::create_binary_file(const char* path, bool rewrite_existing) {
5071   int oflags = O_WRONLY | O_CREAT;
5072   if (!rewrite_existing) {
5073     oflags |= O_EXCL;
5074   }
5075   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5076 }
5077 
5078 // return current position of file pointer
5079 jlong os::current_file_offset(int fd) {
5080   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5081 }
5082 
5083 // move file pointer to the specified offset
5084 jlong os::seek_to_file_offset(int fd, jlong offset) {
5085   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5086 }
5087 
5088 // Map a block of memory.
5089 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
5090                      char *addr, size_t bytes, bool read_only,
5091                      bool allow_exec) {
5092   int prot;
5093   int flags;
5094 
5095   if (read_only) {
5096     prot = PROT_READ;
5097     flags = MAP_SHARED;
5098   } else {
5099     prot = PROT_READ | PROT_WRITE;
5100     flags = MAP_PRIVATE;
5101   }
5102 
5103   if (allow_exec) {
5104     prot |= PROT_EXEC;
5105   }
5106 
5107   if (addr != NULL) {
5108     flags |= MAP_FIXED;
5109   }
5110 
5111   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5112                                      fd, file_offset);
5113   if (mapped_address == MAP_FAILED) {
5114     return NULL;
5115   }
5116   return mapped_address;
5117 }
5118 
5119 
5120 // Remap a block of memory.
5121 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
5122                        char *addr, size_t bytes, bool read_only,
5123                        bool allow_exec) {
5124   // same as map_memory() on this OS
5125   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5126                         allow_exec);
5127 }
5128 
5129 
5130 // Unmap a block of memory.
5131 bool os::unmap_memory(char* addr, size_t bytes) {
5132   return munmap(addr, bytes) == 0;
5133 }
5134 
5135 void os::pause() {
5136   char filename[MAX_PATH];
5137   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5138     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5139   } else {
5140     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5141   }
5142 
5143   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5144   if (fd != -1) {
5145     struct stat buf;
5146     close(fd);
5147     while (::stat(filename, &buf) == 0) {
5148       (void)::poll(NULL, 0, 100);
5149     }
5150   } else {
5151     jio_fprintf(stderr,
5152       "Could not open pause file '%s', continuing immediately.\n", filename);
5153   }
5154 }
5155 
5156 #ifndef PRODUCT
5157 #ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5158 // Turn this on if you need to trace synch operations.
5159 // Set RECORD_SYNCH_LIMIT to a large-enough value,
5160 // and call record_synch_enable and record_synch_disable
5161 // around the computation of interest.
5162 
5163 void record_synch(char* name, bool returning);  // defined below
5164 
5165 class RecordSynch {
5166   char* _name;
5167  public:
5168   RecordSynch(char* name) :_name(name)
5169                  { record_synch(_name, false); }
5170   ~RecordSynch() { record_synch(_name,   true);  }
5171 };
5172 
5173 #define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
5174 extern "C" ret name params {                                    \
5175   typedef ret name##_t params;                                  \
5176   static name##_t* implem = NULL;                               \
5177   static int callcount = 0;                                     \
5178   if (implem == NULL) {                                         \
5179     implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
5180     if (implem == NULL)  fatal(dlerror());                      \
5181   }                                                             \
5182   ++callcount;                                                  \
5183   RecordSynch _rs(#name);                                       \
5184   inner;                                                        \
5185   return implem args;                                           \
5186 }
5187 // in dbx, examine callcounts this way:
5188 // for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
5189 
5190 #define CHECK_POINTER_OK(p) \
5191   (Universe::perm_gen() == NULL || !Universe::is_reserved_heap((oop)(p)))
5192 #define CHECK_MU \
5193   if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
5194 #define CHECK_CV \
5195   if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
5196 #define CHECK_P(p) \
5197   if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
5198 
5199 #define CHECK_MUTEX(mutex_op) \
5200 CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
5201 
5202 CHECK_MUTEX(   mutex_lock)
5203 CHECK_MUTEX(  _mutex_lock)
5204 CHECK_MUTEX( mutex_unlock)
5205 CHECK_MUTEX(_mutex_unlock)
5206 CHECK_MUTEX( mutex_trylock)
5207 CHECK_MUTEX(_mutex_trylock)
5208 
5209 #define CHECK_COND(cond_op) \
5210 CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU;CHECK_CV);
5211 
5212 CHECK_COND( cond_wait);
5213 CHECK_COND(_cond_wait);
5214 CHECK_COND(_cond_wait_cancel);
5215 
5216 #define CHECK_COND2(cond_op) \
5217 CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU;CHECK_CV);
5218 
5219 CHECK_COND2( cond_timedwait);
5220 CHECK_COND2(_cond_timedwait);
5221 CHECK_COND2(_cond_timedwait_cancel);
5222 
5223 // do the _lwp_* versions too
5224 #define mutex_t lwp_mutex_t
5225 #define cond_t  lwp_cond_t
5226 CHECK_MUTEX(  _lwp_mutex_lock)
5227 CHECK_MUTEX(  _lwp_mutex_unlock)
5228 CHECK_MUTEX(  _lwp_mutex_trylock)
5229 CHECK_MUTEX( __lwp_mutex_lock)
5230 CHECK_MUTEX( __lwp_mutex_unlock)
5231 CHECK_MUTEX( __lwp_mutex_trylock)
5232 CHECK_MUTEX(___lwp_mutex_lock)
5233 CHECK_MUTEX(___lwp_mutex_unlock)
5234 
5235 CHECK_COND(  _lwp_cond_wait);
5236 CHECK_COND( __lwp_cond_wait);
5237 CHECK_COND(___lwp_cond_wait);
5238 
5239 CHECK_COND2(  _lwp_cond_timedwait);
5240 CHECK_COND2( __lwp_cond_timedwait);
5241 #undef mutex_t
5242 #undef cond_t
5243 
5244 CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5245 CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5246 CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5247 CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5248 CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5249 CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5250 CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5251 CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5252 
5253 
5254 // recording machinery:
5255 
5256 enum { RECORD_SYNCH_LIMIT = 200 };
5257 char* record_synch_name[RECORD_SYNCH_LIMIT];
5258 void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
5259 bool record_synch_returning[RECORD_SYNCH_LIMIT];
5260 thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
5261 int record_synch_count = 0;
5262 bool record_synch_enabled = false;
5263 
5264 // in dbx, examine recorded data this way:
5265 // for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
5266 
5267 void record_synch(char* name, bool returning) {
5268   if (record_synch_enabled) {
5269     if (record_synch_count < RECORD_SYNCH_LIMIT) {
5270       record_synch_name[record_synch_count] = name;
5271       record_synch_returning[record_synch_count] = returning;
5272       record_synch_thread[record_synch_count] = thr_self();
5273       record_synch_arg0ptr[record_synch_count] = &name;
5274       record_synch_count++;
5275     }
5276     // put more checking code here:
5277     // ...
5278   }
5279 }
5280 
5281 void record_synch_enable() {
5282   // start collecting trace data, if not already doing so
5283   if (!record_synch_enabled)  record_synch_count = 0;
5284   record_synch_enabled = true;
5285 }
5286 
5287 void record_synch_disable() {
5288   // stop collecting trace data
5289   record_synch_enabled = false;
5290 }
5291 
5292 #endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5293 #endif // PRODUCT
5294 
5295 const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5296 const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
5297                                (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5298 
5299 
5300 // JVMTI & JVM monitoring and management support
5301 // The thread_cpu_time() and current_thread_cpu_time() are only
5302 // supported if is_thread_cpu_time_supported() returns true.
5303 // They are not supported on Solaris T1.
5304 
5305 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5306 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5307 // of a thread.
5308 //
5309 // current_thread_cpu_time() and thread_cpu_time(Thread *)
5310 // returns the fast estimate available on the platform.
5311 
5312 // hrtime_t gethrvtime() return value includes
5313 // user time but does not include system time
5314 jlong os::current_thread_cpu_time() {
5315   return (jlong) gethrvtime();
5316 }
5317 
5318 jlong os::thread_cpu_time(Thread *thread) {
5319   // return user level CPU time only to be consistent with
5320   // what current_thread_cpu_time returns.
5321   // thread_cpu_time_info() must be changed if this changes
5322   return os::thread_cpu_time(thread, false /* user time only */);
5323 }
5324 
5325 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5326   if (user_sys_cpu_time) {
5327     return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5328   } else {
5329     return os::current_thread_cpu_time();
5330   }
5331 }
5332 
5333 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5334   char proc_name[64];
5335   int count;
5336   prusage_t prusage;
5337   jlong lwp_time;
5338   int fd;
5339 
5340   sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
5341                      getpid(),
5342                      thread->osthread()->lwp_id());
5343   fd = open(proc_name, O_RDONLY);
5344   if ( fd == -1 ) return -1;
5345 
5346   do {
5347     count = pread(fd,
5348                   (void *)&prusage.pr_utime,
5349                   thr_time_size,
5350                   thr_time_off);
5351   } while (count < 0 && errno == EINTR);
5352   close(fd);
5353   if ( count < 0 ) return -1;
5354 
5355   if (user_sys_cpu_time) {
5356     // user + system CPU time
5357     lwp_time = (((jlong)prusage.pr_stime.tv_sec +
5358                  (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
5359                  (jlong)prusage.pr_stime.tv_nsec +
5360                  (jlong)prusage.pr_utime.tv_nsec;
5361   } else {
5362     // user level CPU time only
5363     lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
5364                 (jlong)prusage.pr_utime.tv_nsec;
5365   }
5366 
5367   return(lwp_time);
5368 }
5369 
5370 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5371   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5372   info_ptr->may_skip_backward = false;    // elapsed time not wall time
5373   info_ptr->may_skip_forward = false;     // elapsed time not wall time
5374   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5375 }
5376 
5377 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5378   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5379   info_ptr->may_skip_backward = false;    // elapsed time not wall time
5380   info_ptr->may_skip_forward = false;     // elapsed time not wall time
5381   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5382 }
5383 
5384 bool os::is_thread_cpu_time_supported() {
5385   if ( os::Solaris::T2_libthread() || UseBoundThreads ) {
5386     return true;
5387   } else {
5388     return false;
5389   }
5390 }
5391 
5392 // System loadavg support.  Returns -1 if load average cannot be obtained.
5393 // Return the load average for our processor set if the primitive exists
5394 // (Solaris 9 and later).  Otherwise just return system wide loadavg.
5395 int os::loadavg(double loadavg[], int nelem) {
5396   if (pset_getloadavg_ptr != NULL) {
5397     return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
5398   } else {
5399     return ::getloadavg(loadavg, nelem);
5400   }
5401 }
5402 
5403 //---------------------------------------------------------------------------------
5404 #ifndef PRODUCT
5405 
5406 static address same_page(address x, address y) {
5407   intptr_t page_bits = -os::vm_page_size();
5408   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
5409     return x;
5410   else if (x > y)
5411     return (address)(intptr_t(y) | ~page_bits) + 1;
5412   else
5413     return (address)(intptr_t(y) & page_bits);
5414 }
5415 
5416 bool os::find(address addr) {
5417   Dl_info dlinfo;
5418   memset(&dlinfo, 0, sizeof(dlinfo));
5419   if (dladdr(addr, &dlinfo)) {
5420 #ifdef _LP64
5421     tty->print("0x%016lx: ", addr);
5422 #else
5423     tty->print("0x%08x: ", addr);
5424 #endif
5425     if (dlinfo.dli_sname != NULL)
5426       tty->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
5427     else if (dlinfo.dli_fname)
5428       tty->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
5429     else
5430       tty->print("<absolute address>");
5431     if (dlinfo.dli_fname)  tty->print(" in %s", dlinfo.dli_fname);
5432 #ifdef _LP64
5433     if (dlinfo.dli_fbase)  tty->print(" at 0x%016lx", dlinfo.dli_fbase);
5434 #else
5435     if (dlinfo.dli_fbase)  tty->print(" at 0x%08x", dlinfo.dli_fbase);
5436 #endif
5437     tty->cr();
5438 
5439     if (Verbose) {
5440       // decode some bytes around the PC
5441       address begin = same_page(addr-40, addr);
5442       address end   = same_page(addr+40, addr);
5443       address       lowest = (address) dlinfo.dli_sname;
5444       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5445       if (begin < lowest)  begin = lowest;
5446       Dl_info dlinfo2;
5447       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
5448           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5449         end = (address) dlinfo2.dli_saddr;
5450       Disassembler::decode(begin, end);
5451     }
5452     return true;
5453   }
5454   return false;
5455 }
5456 
5457 #endif
5458 
5459 
5460 // Following function has been added to support HotSparc's libjvm.so running
5461 // under Solaris production JDK 1.2.2 / 1.3.0.  These came from
5462 // src/solaris/hpi/native_threads in the EVM codebase.
5463 //
5464 // NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
5465 // libraries and should thus be removed. We will leave it behind for a while
5466 // until we no longer want to able to run on top of 1.3.0 Solaris production
5467 // JDK. See 4341971.
5468 
5469 #define STACK_SLACK 0x800
5470 
5471 extern "C" {
5472   intptr_t sysThreadAvailableStackWithSlack() {
5473     stack_t st;
5474     intptr_t retval, stack_top;
5475     retval = thr_stksegment(&st);
5476     assert(retval == 0, "incorrect return value from thr_stksegment");
5477     assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
5478     assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
5479     stack_top=(intptr_t)st.ss_sp-st.ss_size;
5480     return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
5481   }
5482 }
5483 
5484 // Just to get the Kernel build to link on solaris for testing.
5485 
5486 extern "C" {
5487 class ASGCT_CallTrace;
5488 void AsyncGetCallTrace(ASGCT_CallTrace *trace, jint depth, void* ucontext)
5489   KERNEL_RETURN;
5490 }
5491 
5492 
5493 // ObjectMonitor park-unpark infrastructure ...
5494 //
5495 // We implement Solaris and Linux PlatformEvents with the
5496 // obvious condvar-mutex-flag triple.
5497 // Another alternative that works quite well is pipes:
5498 // Each PlatformEvent consists of a pipe-pair.
5499 // The thread associated with the PlatformEvent
5500 // calls park(), which reads from the input end of the pipe.
5501 // Unpark() writes into the other end of the pipe.
5502 // The write-side of the pipe must be set NDELAY.
5503 // Unfortunately pipes consume a large # of handles.
5504 // Native solaris lwp_park() and lwp_unpark() work nicely, too.
5505 // Using pipes for the 1st few threads might be workable, however.
5506 //
5507 // park() is permitted to return spuriously.
5508 // Callers of park() should wrap the call to park() in
5509 // an appropriate loop.  A litmus test for the correct
5510 // usage of park is the following: if park() were modified
5511 // to immediately return 0 your code should still work,
5512 // albeit degenerating to a spin loop.
5513 //
5514 // An interesting optimization for park() is to use a trylock()
5515 // to attempt to acquire the mutex.  If the trylock() fails
5516 // then we know that a concurrent unpark() operation is in-progress.
5517 // in that case the park() code could simply set _count to 0
5518 // and return immediately.  The subsequent park() operation *might*
5519 // return immediately.  That's harmless as the caller of park() is
5520 // expected to loop.  By using trylock() we will have avoided a
5521 // avoided a context switch caused by contention on the per-thread mutex.
5522 //
5523 // TODO-FIXME:
5524 // 1.  Reconcile Doug's JSR166 j.u.c park-unpark with the
5525 //     objectmonitor implementation.
5526 // 2.  Collapse the JSR166 parker event, and the
5527 //     objectmonitor ParkEvent into a single "Event" construct.
5528 // 3.  In park() and unpark() add:
5529 //     assert (Thread::current() == AssociatedWith).
5530 // 4.  add spurious wakeup injection on a -XX:EarlyParkReturn=N switch.
5531 //     1-out-of-N park() operations will return immediately.
5532 //
5533 // _Event transitions in park()
5534 //   -1 => -1 : illegal
5535 //    1 =>  0 : pass - return immediately
5536 //    0 => -1 : block
5537 //
5538 // _Event serves as a restricted-range semaphore.
5539 //
5540 // Another possible encoding of _Event would be with
5541 // explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
5542 //
5543 // TODO-FIXME: add DTRACE probes for:
5544 // 1.   Tx parks
5545 // 2.   Ty unparks Tx
5546 // 3.   Tx resumes from park
5547 
5548 
5549 // value determined through experimentation
5550 #define ROUNDINGFIX 11
5551 
5552 // utility to compute the abstime argument to timedwait.
5553 // TODO-FIXME: switch from compute_abstime() to unpackTime().
5554 
5555 static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
5556   // millis is the relative timeout time
5557   // abstime will be the absolute timeout time
5558   if (millis < 0)  millis = 0;
5559   struct timeval now;
5560   int status = gettimeofday(&now, NULL);
5561   assert(status == 0, "gettimeofday");
5562   jlong seconds = millis / 1000;
5563   jlong max_wait_period;
5564 
5565   if (UseLWPSynchronization) {
5566     // forward port of fix for 4275818 (not sleeping long enough)
5567     // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
5568     // _lwp_cond_timedwait() used a round_down algorithm rather
5569     // than a round_up. For millis less than our roundfactor
5570     // it rounded down to 0 which doesn't meet the spec.
5571     // For millis > roundfactor we may return a bit sooner, but
5572     // since we can not accurately identify the patch level and
5573     // this has already been fixed in Solaris 9 and 8 we will
5574     // leave it alone rather than always rounding down.
5575 
5576     if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
5577        // It appears that when we go directly through Solaris _lwp_cond_timedwait()
5578            // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
5579            max_wait_period = 21000000;
5580   } else {
5581     max_wait_period = 50000000;
5582   }
5583   millis %= 1000;
5584   if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
5585      seconds = max_wait_period;
5586   }
5587   abstime->tv_sec = now.tv_sec  + seconds;
5588   long       usec = now.tv_usec + millis * 1000;
5589   if (usec >= 1000000) {
5590     abstime->tv_sec += 1;
5591     usec -= 1000000;
5592   }
5593   abstime->tv_nsec = usec * 1000;
5594   return abstime;
5595 }
5596 
5597 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5598 // Conceptually TryPark() should be equivalent to park(0).
5599 
5600 int os::PlatformEvent::TryPark() {
5601   for (;;) {
5602     const int v = _Event ;
5603     guarantee ((v == 0) || (v == 1), "invariant") ;
5604     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5605   }
5606 }
5607 
5608 void os::PlatformEvent::park() {           // AKA: down()
5609   // Invariant: Only the thread associated with the Event/PlatformEvent
5610   // may call park().
5611   int v ;
5612   for (;;) {
5613       v = _Event ;
5614       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5615   }
5616   guarantee (v >= 0, "invariant") ;
5617   if (v == 0) {
5618      // Do this the hard way by blocking ...
5619      // See http://monaco.sfbay/detail.jsf?cr=5094058.
5620      // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
5621      // Only for SPARC >= V8PlusA
5622 #if defined(__sparc) && defined(COMPILER2)
5623      if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
5624 #endif
5625      int status = os::Solaris::mutex_lock(_mutex);
5626      assert_status(status == 0, status,  "mutex_lock");
5627      guarantee (_nParked == 0, "invariant") ;
5628      ++ _nParked ;
5629      while (_Event < 0) {
5630         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5631         // Treat this the same as if the wait was interrupted
5632         // With usr/lib/lwp going to kernel, always handle ETIME
5633         status = os::Solaris::cond_wait(_cond, _mutex);
5634         if (status == ETIME) status = EINTR ;
5635         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5636      }
5637      -- _nParked ;
5638      _Event = 0 ;
5639      status = os::Solaris::mutex_unlock(_mutex);
5640      assert_status(status == 0, status, "mutex_unlock");
5641   }
5642 }
5643 
5644 int os::PlatformEvent::park(jlong millis) {
5645   guarantee (_nParked == 0, "invariant") ;
5646   int v ;
5647   for (;;) {
5648       v = _Event ;
5649       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5650   }
5651   guarantee (v >= 0, "invariant") ;
5652   if (v != 0) return OS_OK ;
5653 
5654   int ret = OS_TIMEOUT;
5655   timestruc_t abst;
5656   compute_abstime (&abst, millis);
5657 
5658   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5659   // For Solaris SPARC set fprs.FEF=0 prior to parking.
5660   // Only for SPARC >= V8PlusA
5661 #if defined(__sparc) && defined(COMPILER2)
5662  if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
5663 #endif
5664   int status = os::Solaris::mutex_lock(_mutex);
5665   assert_status(status == 0, status, "mutex_lock");
5666   guarantee (_nParked == 0, "invariant") ;
5667   ++ _nParked ;
5668   while (_Event < 0) {
5669      int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
5670      assert_status(status == 0 || status == EINTR ||
5671                    status == ETIME || status == ETIMEDOUT,
5672                    status, "cond_timedwait");
5673      if (!FilterSpuriousWakeups) break ;                // previous semantics
5674      if (status == ETIME || status == ETIMEDOUT) break ;
5675      // We consume and ignore EINTR and spurious wakeups.
5676   }
5677   -- _nParked ;
5678   if (_Event >= 0) ret = OS_OK ;
5679   _Event = 0 ;
5680   status = os::Solaris::mutex_unlock(_mutex);
5681   assert_status(status == 0, status, "mutex_unlock");
5682   return ret;
5683 }
5684 
5685 void os::PlatformEvent::unpark() {
5686   int v, AnyWaiters;
5687 
5688   // Increment _Event.
5689   // Another acceptable implementation would be to simply swap 1
5690   // into _Event:
5691   //   if (Swap (&_Event, 1) < 0) {
5692   //      mutex_lock (_mutex) ; AnyWaiters = nParked; mutex_unlock (_mutex) ;
5693   //      if (AnyWaiters) cond_signal (_cond) ;
5694   //   }
5695 
5696   for (;;) {
5697     v = _Event ;
5698     if (v > 0) {
5699        // The LD of _Event could have reordered or be satisfied
5700        // by a read-aside from this processor's write buffer.
5701        // To avoid problems execute a barrier and then
5702        // ratify the value.  A degenerate CAS() would also work.
5703        // Viz., CAS (v+0, &_Event, v) == v).
5704        OrderAccess::fence() ;
5705        if (_Event == v) return ;
5706        continue ;
5707     }
5708     if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5709   }
5710 
5711   // If the thread associated with the event was parked, wake it.
5712   if (v < 0) {
5713      int status ;
5714      // Wait for the thread assoc with the PlatformEvent to vacate.
5715      status = os::Solaris::mutex_lock(_mutex);
5716      assert_status(status == 0, status, "mutex_lock");
5717      AnyWaiters = _nParked ;
5718      status = os::Solaris::mutex_unlock(_mutex);
5719      assert_status(status == 0, status, "mutex_unlock");
5720      guarantee (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5721      if (AnyWaiters != 0) {
5722        // We intentional signal *after* dropping the lock
5723        // to avoid a common class of futile wakeups.
5724        status = os::Solaris::cond_signal(_cond);
5725        assert_status(status == 0, status, "cond_signal");
5726      }
5727   }
5728 }
5729 
5730 // JSR166
5731 // -------------------------------------------------------
5732 
5733 /*
5734  * The solaris and linux implementations of park/unpark are fairly
5735  * conservative for now, but can be improved. They currently use a
5736  * mutex/condvar pair, plus _counter.
5737  * Park decrements _counter if > 0, else does a condvar wait.  Unpark
5738  * sets count to 1 and signals condvar.  Only one thread ever waits
5739  * on the condvar. Contention seen when trying to park implies that someone
5740  * is unparking you, so don't wait. And spurious returns are fine, so there
5741  * is no need to track notifications.
5742  */
5743 
5744 #define NANOSECS_PER_SEC 1000000000
5745 #define NANOSECS_PER_MILLISEC 1000000
5746 #define MAX_SECS 100000000
5747 
5748 /*
5749  * This code is common to linux and solaris and will be moved to a
5750  * common place in dolphin.
5751  *
5752  * The passed in time value is either a relative time in nanoseconds
5753  * or an absolute time in milliseconds. Either way it has to be unpacked
5754  * into suitable seconds and nanoseconds components and stored in the
5755  * given timespec structure.
5756  * Given time is a 64-bit value and the time_t used in the timespec is only
5757  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5758  * overflow if times way in the future are given. Further on Solaris versions
5759  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5760  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5761  * As it will be 28 years before "now + 100000000" will overflow we can
5762  * ignore overflow and just impose a hard-limit on seconds using the value
5763  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5764  * years from "now".
5765  */
5766 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5767   assert (time > 0, "convertTime");
5768 
5769   struct timeval now;
5770   int status = gettimeofday(&now, NULL);
5771   assert(status == 0, "gettimeofday");
5772 
5773   time_t max_secs = now.tv_sec + MAX_SECS;
5774 
5775   if (isAbsolute) {
5776     jlong secs = time / 1000;
5777     if (secs > max_secs) {
5778       absTime->tv_sec = max_secs;
5779     }
5780     else {
5781       absTime->tv_sec = secs;
5782     }
5783     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5784   }
5785   else {
5786     jlong secs = time / NANOSECS_PER_SEC;
5787     if (secs >= MAX_SECS) {
5788       absTime->tv_sec = max_secs;
5789       absTime->tv_nsec = 0;
5790     }
5791     else {
5792       absTime->tv_sec = now.tv_sec + secs;
5793       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5794       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5795         absTime->tv_nsec -= NANOSECS_PER_SEC;
5796         ++absTime->tv_sec; // note: this must be <= max_secs
5797       }
5798     }
5799   }
5800   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5801   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5802   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5803   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5804 }
5805 
5806 void Parker::park(bool isAbsolute, jlong time) {
5807 
5808   // Optional fast-path check:
5809   // Return immediately if a permit is available.
5810   if (_counter > 0) {
5811       _counter = 0 ;
5812       OrderAccess::fence();
5813       return ;
5814   }
5815 
5816   // Optional fast-exit: Check interrupt before trying to wait
5817   Thread* thread = Thread::current();
5818   assert(thread->is_Java_thread(), "Must be JavaThread");
5819   JavaThread *jt = (JavaThread *)thread;
5820   if (Thread::is_interrupted(thread, false)) {
5821     return;
5822   }
5823 
5824   // First, demultiplex/decode time arguments
5825   timespec absTime;
5826   if (time < 0) { // don't wait at all
5827     return;
5828   }
5829   if (time > 0) {
5830     // Warning: this code might be exposed to the old Solaris time
5831     // round-down bugs.  Grep "roundingFix" for details.
5832     unpackTime(&absTime, isAbsolute, time);
5833   }
5834 
5835   // Enter safepoint region
5836   // Beware of deadlocks such as 6317397.
5837   // The per-thread Parker:: _mutex is a classic leaf-lock.
5838   // In particular a thread must never block on the Threads_lock while
5839   // holding the Parker:: mutex.  If safepoints are pending both the
5840   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5841   ThreadBlockInVM tbivm(jt);
5842 
5843   // Don't wait if cannot get lock since interference arises from
5844   // unblocking.  Also. check interrupt before trying wait
5845   if (Thread::is_interrupted(thread, false) ||
5846       os::Solaris::mutex_trylock(_mutex) != 0) {
5847     return;
5848   }
5849 
5850   int status ;
5851 
5852   if (_counter > 0)  { // no wait needed
5853     _counter = 0;
5854     status = os::Solaris::mutex_unlock(_mutex);
5855     assert (status == 0, "invariant") ;
5856     OrderAccess::fence();
5857     return;
5858   }
5859 
5860 #ifdef ASSERT
5861   // Don't catch signals while blocked; let the running threads have the signals.
5862   // (This allows a debugger to break into the running thread.)
5863   sigset_t oldsigs;
5864   sigset_t* allowdebug_blocked = os::Solaris::allowdebug_blocked_signals();
5865   thr_sigsetmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5866 #endif
5867 
5868   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5869   jt->set_suspend_equivalent();
5870   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5871 
5872   // Do this the hard way by blocking ...
5873   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5874   // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
5875   // Only for SPARC >= V8PlusA
5876 #if defined(__sparc) && defined(COMPILER2)
5877   if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
5878 #endif
5879 
5880   if (time == 0) {
5881     status = os::Solaris::cond_wait (_cond, _mutex) ;
5882   } else {
5883     status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
5884   }
5885   // Note that an untimed cond_wait() can sometimes return ETIME on older
5886   // versions of the Solaris.
5887   assert_status(status == 0 || status == EINTR ||
5888                 status == ETIME || status == ETIMEDOUT,
5889                 status, "cond_timedwait");
5890 
5891 #ifdef ASSERT
5892   thr_sigsetmask(SIG_SETMASK, &oldsigs, NULL);
5893 #endif
5894   _counter = 0 ;
5895   status = os::Solaris::mutex_unlock(_mutex);
5896   assert_status(status == 0, status, "mutex_unlock") ;
5897 
5898   // If externally suspended while waiting, re-suspend
5899   if (jt->handle_special_suspend_equivalent_condition()) {
5900     jt->java_suspend_self();
5901   }
5902 
5903   OrderAccess::fence();
5904 }
5905 
5906 void Parker::unpark() {
5907   int s, status ;
5908   status = os::Solaris::mutex_lock (_mutex) ;
5909   assert (status == 0, "invariant") ;
5910   s = _counter;
5911   _counter = 1;
5912   status = os::Solaris::mutex_unlock (_mutex) ;
5913   assert (status == 0, "invariant") ;
5914 
5915   if (s < 1) {
5916     status = os::Solaris::cond_signal (_cond) ;
5917     assert (status == 0, "invariant") ;
5918   }
5919 }
5920 
5921 extern char** environ;
5922 
5923 // Run the specified command in a separate process. Return its exit value,
5924 // or -1 on failure (e.g. can't fork a new process).
5925 // Unlike system(), this function can be called from signal handler. It
5926 // doesn't block SIGINT et al.
5927 int os::fork_and_exec(char* cmd) {
5928   char * argv[4];
5929   argv[0] = (char *)"sh";
5930   argv[1] = (char *)"-c";
5931   argv[2] = cmd;
5932   argv[3] = NULL;
5933 
5934   // fork is async-safe, fork1 is not so can't use in signal handler
5935   pid_t pid;
5936   Thread* t = ThreadLocalStorage::get_thread_slow();
5937   if (t != NULL && t->is_inside_signal_handler()) {
5938     pid = fork();
5939   } else {
5940     pid = fork1();
5941   }
5942 
5943   if (pid < 0) {
5944     // fork failed
5945     warning("fork failed: %s", strerror(errno));
5946     return -1;
5947 
5948   } else if (pid == 0) {
5949     // child process
5950 
5951     // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
5952     execve("/usr/bin/sh", argv, environ);
5953 
5954     // execve failed
5955     _exit(-1);
5956 
5957   } else  {
5958     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5959     // care about the actual exit code, for now.
5960 
5961     int status;
5962 
5963     // Wait for the child process to exit.  This returns immediately if
5964     // the child has already exited. */
5965     while (waitpid(pid, &status, 0) < 0) {
5966         switch (errno) {
5967         case ECHILD: return 0;
5968         case EINTR: break;
5969         default: return -1;
5970         }
5971     }
5972 
5973     if (WIFEXITED(status)) {
5974        // The child exited normally; get its exit code.
5975        return WEXITSTATUS(status);
5976     } else if (WIFSIGNALED(status)) {
5977        // The child exited because of a signal
5978        // The best value to return is 0x80 + signal number,
5979        // because that is what all Unix shells do, and because
5980        // it allows callers to distinguish between process exit and
5981        // process death by signal.
5982        return 0x80 + WTERMSIG(status);
5983     } else {
5984        // Unknown exit code; pass it through
5985        return status;
5986     }
5987   }
5988 }