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     {EM_ARM,         EM_ARM,     ELFCLASS32, ELFDATA2LSB, (char*)"ARM 32"}
2056   };
2057 
2058   #if  (defined IA32)
2059     static  Elf32_Half running_arch_code=EM_386;
2060   #elif   (defined AMD64)
2061     static  Elf32_Half running_arch_code=EM_X86_64;
2062   #elif  (defined IA64)
2063     static  Elf32_Half running_arch_code=EM_IA_64;
2064   #elif  (defined __sparc) && (defined _LP64)
2065     static  Elf32_Half running_arch_code=EM_SPARCV9;
2066   #elif  (defined __sparc) && (!defined _LP64)
2067     static  Elf32_Half running_arch_code=EM_SPARC;
2068   #elif  (defined __powerpc64__)
2069     static  Elf32_Half running_arch_code=EM_PPC64;
2070   #elif  (defined __powerpc__)
2071     static  Elf32_Half running_arch_code=EM_PPC;
2072   #elif (defined ARM)
2073     static  Elf32_Half running_arch_code=EM_ARM;
2074   #else
2075     #error Method os::dll_load requires that one of following is defined:\
2076          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, ARM
2077   #endif
2078 
2079   // Identify compatability class for VM's architecture and library's architecture
2080   // Obtain string descriptions for architectures
2081 
2082   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2083   int running_arch_index=-1;
2084 
2085   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2086     if (running_arch_code == arch_array[i].code) {
2087       running_arch_index    = i;
2088     }
2089     if (lib_arch.code == arch_array[i].code) {
2090       lib_arch.compat_class = arch_array[i].compat_class;
2091       lib_arch.name         = arch_array[i].name;
2092     }
2093   }
2094 
2095   assert(running_arch_index != -1,
2096     "Didn't find running architecture code (running_arch_code) in arch_array");
2097   if (running_arch_index == -1) {
2098     // Even though running architecture detection failed
2099     // we may still continue with reporting dlerror() message
2100     return NULL;
2101   }
2102 
2103   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2104     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2105     return NULL;
2106   }
2107 
2108   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2109     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2110     return NULL;
2111   }
2112 
2113   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2114     if ( lib_arch.name!=NULL ) {
2115       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2116         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2117         lib_arch.name, arch_array[running_arch_index].name);
2118     } else {
2119       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2120       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2121         lib_arch.code,
2122         arch_array[running_arch_index].name);
2123     }
2124   }
2125 
2126   return NULL;
2127 }
2128 
2129 void* os::dll_lookup(void* handle, const char* name) {
2130   return dlsym(handle, name);
2131 }
2132 
2133 
2134 bool _print_ascii_file(const char* filename, outputStream* st) {
2135   int fd = open(filename, O_RDONLY);
2136   if (fd == -1) {
2137      return false;
2138   }
2139 
2140   char buf[32];
2141   int bytes;
2142   while ((bytes = read(fd, buf, sizeof(buf))) > 0) {
2143     st->print_raw(buf, bytes);
2144   }
2145 
2146   close(fd);
2147 
2148   return true;
2149 }
2150 
2151 void os::print_os_info(outputStream* st) {
2152   st->print("OS:");
2153 
2154   if (!_print_ascii_file("/etc/release", st)) {
2155     st->print("Solaris");
2156   }
2157   st->cr();
2158 
2159   // kernel
2160   st->print("uname:");
2161   struct utsname name;
2162   uname(&name);
2163   st->print(name.sysname); st->print(" ");
2164   st->print(name.release); st->print(" ");
2165   st->print(name.version); st->print(" ");
2166   st->print(name.machine);
2167 
2168   // libthread
2169   if (os::Solaris::T2_libthread()) st->print("  (T2 libthread)");
2170   else st->print("  (T1 libthread)");
2171   st->cr();
2172 
2173   // rlimit
2174   st->print("rlimit:");
2175   struct rlimit rlim;
2176 
2177   st->print(" STACK ");
2178   getrlimit(RLIMIT_STACK, &rlim);
2179   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2180   else st->print("%uk", rlim.rlim_cur >> 10);
2181 
2182   st->print(", CORE ");
2183   getrlimit(RLIMIT_CORE, &rlim);
2184   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2185   else st->print("%uk", rlim.rlim_cur >> 10);
2186 
2187   st->print(", NOFILE ");
2188   getrlimit(RLIMIT_NOFILE, &rlim);
2189   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2190   else st->print("%d", rlim.rlim_cur);
2191 
2192   st->print(", AS ");
2193   getrlimit(RLIMIT_AS, &rlim);
2194   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2195   else st->print("%uk", rlim.rlim_cur >> 10);
2196   st->cr();
2197 
2198   // load average
2199   st->print("load average:");
2200   double loadavg[3];
2201   os::loadavg(loadavg, 3);
2202   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2203   st->cr();
2204 }
2205 
2206 
2207 static bool check_addr0(outputStream* st) {
2208   jboolean status = false;
2209   int fd = open("/proc/self/map",O_RDONLY);
2210   if (fd >= 0) {
2211     prmap_t p;
2212     while(read(fd, &p, sizeof(p)) > 0) {
2213       if (p.pr_vaddr == 0x0) {
2214         st->print("Warning: Address: 0x%x, Size: %dK, ",p.pr_vaddr, p.pr_size/1024, p.pr_mapname);
2215         st->print("Mapped file: %s, ", p.pr_mapname[0] == '\0' ? "None" : p.pr_mapname);
2216         st->print("Access:");
2217         st->print("%s",(p.pr_mflags & MA_READ)  ? "r" : "-");
2218         st->print("%s",(p.pr_mflags & MA_WRITE) ? "w" : "-");
2219         st->print("%s",(p.pr_mflags & MA_EXEC)  ? "x" : "-");
2220         st->cr();
2221         status = true;
2222       }
2223       close(fd);
2224     }
2225   }
2226   return status;
2227 }
2228 
2229 void os::print_memory_info(outputStream* st) {
2230   st->print("Memory:");
2231   st->print(" %dk page", os::vm_page_size()>>10);
2232   st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
2233   st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
2234   st->cr();
2235   (void) check_addr0(st);
2236 }
2237 
2238 // Taken from /usr/include/sys/machsig.h  Supposed to be architecture specific
2239 // but they're the same for all the solaris architectures that we support.
2240 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2241                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2242                           "ILL_COPROC", "ILL_BADSTK" };
2243 
2244 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2245                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2246                           "FPE_FLTINV", "FPE_FLTSUB" };
2247 
2248 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2249 
2250 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2251 
2252 void os::print_siginfo(outputStream* st, void* siginfo) {
2253   st->print("siginfo:");
2254 
2255   const int buflen = 100;
2256   char buf[buflen];
2257   siginfo_t *si = (siginfo_t*)siginfo;
2258   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2259   char *err = strerror(si->si_errno);
2260   if (si->si_errno != 0 && err != NULL) {
2261     st->print("si_errno=%s", err);
2262   } else {
2263     st->print("si_errno=%d", si->si_errno);
2264   }
2265   const int c = si->si_code;
2266   assert(c > 0, "unexpected si_code");
2267   switch (si->si_signo) {
2268   case SIGILL:
2269     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2270     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2271     break;
2272   case SIGFPE:
2273     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2274     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2275     break;
2276   case SIGSEGV:
2277     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2278     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2279     break;
2280   case SIGBUS:
2281     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2282     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2283     break;
2284   default:
2285     st->print(", si_code=%d", si->si_code);
2286     // no si_addr
2287   }
2288 
2289   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2290       UseSharedSpaces) {
2291     FileMapInfo* mapinfo = FileMapInfo::current_info();
2292     if (mapinfo->is_in_shared_space(si->si_addr)) {
2293       st->print("\n\nError accessing class data sharing archive."   \
2294                 " Mapped file inaccessible during execution, "      \
2295                 " possible disk/network problem.");
2296     }
2297   }
2298   st->cr();
2299 }
2300 
2301 // Moved from whole group, because we need them here for diagnostic
2302 // prints.
2303 #define OLDMAXSIGNUM 32
2304 static int Maxsignum = 0;
2305 static int *ourSigFlags = NULL;
2306 
2307 extern "C" void sigINTRHandler(int, siginfo_t*, void*);
2308 
2309 int os::Solaris::get_our_sigflags(int sig) {
2310   assert(ourSigFlags!=NULL, "signal data structure not initialized");
2311   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2312   return ourSigFlags[sig];
2313 }
2314 
2315 void os::Solaris::set_our_sigflags(int sig, int flags) {
2316   assert(ourSigFlags!=NULL, "signal data structure not initialized");
2317   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2318   ourSigFlags[sig] = flags;
2319 }
2320 
2321 
2322 static const char* get_signal_handler_name(address handler,
2323                                            char* buf, int buflen) {
2324   int offset;
2325   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
2326   if (found) {
2327     // skip directory names
2328     const char *p1, *p2;
2329     p1 = buf;
2330     size_t len = strlen(os::file_separator());
2331     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
2332     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
2333   } else {
2334     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
2335   }
2336   return buf;
2337 }
2338 
2339 static void print_signal_handler(outputStream* st, int sig,
2340                                   char* buf, size_t buflen) {
2341   struct sigaction sa;
2342 
2343   sigaction(sig, NULL, &sa);
2344 
2345   st->print("%s: ", os::exception_name(sig, buf, buflen));
2346 
2347   address handler = (sa.sa_flags & SA_SIGINFO)
2348                   ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
2349                   : CAST_FROM_FN_PTR(address, sa.sa_handler);
2350 
2351   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
2352     st->print("SIG_DFL");
2353   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
2354     st->print("SIG_IGN");
2355   } else {
2356     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
2357   }
2358 
2359   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
2360 
2361   address rh = VMError::get_resetted_sighandler(sig);
2362   // May be, handler was resetted by VMError?
2363   if(rh != NULL) {
2364     handler = rh;
2365     sa.sa_flags = VMError::get_resetted_sigflags(sig);
2366   }
2367 
2368   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
2369 
2370   // Check: is it our handler?
2371   if(handler == CAST_FROM_FN_PTR(address, signalHandler) ||
2372      handler == CAST_FROM_FN_PTR(address, sigINTRHandler)) {
2373     // It is our signal handler
2374     // check for flags
2375     if(sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
2376       st->print(
2377         ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
2378         os::Solaris::get_our_sigflags(sig));
2379     }
2380   }
2381   st->cr();
2382 }
2383 
2384 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2385   st->print_cr("Signal Handlers:");
2386   print_signal_handler(st, SIGSEGV, buf, buflen);
2387   print_signal_handler(st, SIGBUS , buf, buflen);
2388   print_signal_handler(st, SIGFPE , buf, buflen);
2389   print_signal_handler(st, SIGPIPE, buf, buflen);
2390   print_signal_handler(st, SIGXFSZ, buf, buflen);
2391   print_signal_handler(st, SIGILL , buf, buflen);
2392   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2393   print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
2394   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2395   print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
2396   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2397   print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
2398   print_signal_handler(st, os::Solaris::SIGinterrupt(), buf, buflen);
2399   print_signal_handler(st, os::Solaris::SIGasync(), buf, buflen);
2400 }
2401 
2402 static char saved_jvm_path[MAXPATHLEN] = { 0 };
2403 
2404 // Find the full path to the current module, libjvm.so or libjvm_g.so
2405 void os::jvm_path(char *buf, jint buflen) {
2406   // Error checking.
2407   if (buflen < MAXPATHLEN) {
2408     assert(false, "must use a large-enough buffer");
2409     buf[0] = '\0';
2410     return;
2411   }
2412   // Lazy resolve the path to current module.
2413   if (saved_jvm_path[0] != 0) {
2414     strcpy(buf, saved_jvm_path);
2415     return;
2416   }
2417 
2418   Dl_info dlinfo;
2419   int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
2420   assert(ret != 0, "cannot locate libjvm");
2421   realpath((char *)dlinfo.dli_fname, buf);
2422 
2423   if (strcmp(Arguments::sun_java_launcher(), "gamma") == 0) {
2424     // Support for the gamma launcher.  Typical value for buf is
2425     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2426     // the right place in the string, then assume we are installed in a JDK and
2427     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2428     // up the path so it looks like libjvm.so is installed there (append a
2429     // fake suffix hotspot/libjvm.so).
2430     const char *p = buf + strlen(buf) - 1;
2431     for (int count = 0; p > buf && count < 5; ++count) {
2432       for (--p; p > buf && *p != '/'; --p)
2433         /* empty */ ;
2434     }
2435 
2436     if (strncmp(p, "/jre/lib/", 9) != 0) {
2437       // Look for JAVA_HOME in the environment.
2438       char* java_home_var = ::getenv("JAVA_HOME");
2439       if (java_home_var != NULL && java_home_var[0] != 0) {
2440         char cpu_arch[12];
2441         char* jrelib_p;
2442         int   len;
2443         sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
2444 #ifdef _LP64
2445         // If we are on sparc running a 64-bit vm, look in jre/lib/sparcv9.
2446         if (strcmp(cpu_arch, "sparc") == 0) {
2447           strcat(cpu_arch, "v9");
2448         } else if (strcmp(cpu_arch, "i386") == 0) {
2449           strcpy(cpu_arch, "amd64");
2450         }
2451 #endif
2452         // Check the current module name "libjvm.so" or "libjvm_g.so".
2453         p = strrchr(buf, '/');
2454         assert(strstr(p, "/libjvm") == p, "invalid library name");
2455         p = strstr(p, "_g") ? "_g" : "";
2456 
2457         realpath(java_home_var, buf);
2458         // determine if this is a legacy image or modules image
2459         // modules image doesn't have "jre" subdirectory
2460         len = strlen(buf);
2461         jrelib_p = buf + len;
2462         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2463         if (0 != access(buf, F_OK)) {
2464           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2465         }
2466 
2467         if (0 == access(buf, F_OK)) {
2468           // Use current module name "libjvm[_g].so" instead of
2469           // "libjvm"debug_only("_g")".so" since for fastdebug version
2470           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2471           // It is used when we are choosing the HPI library's name
2472           // "libhpi[_g].so" in hpi::initialize_get_interface().
2473           len = strlen(buf);
2474           snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2475         } else {
2476           // Go back to path of .so
2477           realpath((char *)dlinfo.dli_fname, buf);
2478         }
2479       }
2480     }
2481   }
2482 
2483   strcpy(saved_jvm_path, buf);
2484 }
2485 
2486 
2487 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2488   // no prefix required, not even "_"
2489 }
2490 
2491 
2492 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2493   // no suffix required
2494 }
2495 
2496 
2497 // sun.misc.Signal
2498 
2499 extern "C" {
2500   static void UserHandler(int sig, void *siginfo, void *context) {
2501     // Ctrl-C is pressed during error reporting, likely because the error
2502     // handler fails to abort. Let VM die immediately.
2503     if (sig == SIGINT && is_error_reported()) {
2504        os::die();
2505     }
2506 
2507     os::signal_notify(sig);
2508     // We do not need to reinstate the signal handler each time...
2509   }
2510 }
2511 
2512 void* os::user_handler() {
2513   return CAST_FROM_FN_PTR(void*, UserHandler);
2514 }
2515 
2516 extern "C" {
2517   typedef void (*sa_handler_t)(int);
2518   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2519 }
2520 
2521 void* os::signal(int signal_number, void* handler) {
2522   struct sigaction sigAct, oldSigAct;
2523   sigfillset(&(sigAct.sa_mask));
2524   sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2525   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2526 
2527   if (sigaction(signal_number, &sigAct, &oldSigAct))
2528     // -1 means registration failed
2529     return (void *)-1;
2530 
2531   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2532 }
2533 
2534 void os::signal_raise(int signal_number) {
2535   raise(signal_number);
2536 }
2537 
2538 /*
2539  * The following code is moved from os.cpp for making this
2540  * code platform specific, which it is by its very nature.
2541  */
2542 
2543 // a counter for each possible signal value
2544 static int Sigexit = 0;
2545 static int Maxlibjsigsigs;
2546 static jint *pending_signals = NULL;
2547 static int *preinstalled_sigs = NULL;
2548 static struct sigaction *chainedsigactions = NULL;
2549 static sema_t sig_sem;
2550 typedef int (*version_getting_t)();
2551 version_getting_t os::Solaris::get_libjsig_version = NULL;
2552 static int libjsigversion = NULL;
2553 
2554 int os::sigexitnum_pd() {
2555   assert(Sigexit > 0, "signal memory not yet initialized");
2556   return Sigexit;
2557 }
2558 
2559 void os::Solaris::init_signal_mem() {
2560   // Initialize signal structures
2561   Maxsignum = SIGRTMAX;
2562   Sigexit = Maxsignum+1;
2563   assert(Maxsignum >0, "Unable to obtain max signal number");
2564 
2565   Maxlibjsigsigs = Maxsignum;
2566 
2567   // pending_signals has one int per signal
2568   // The additional signal is for SIGEXIT - exit signal to signal_thread
2569   pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1));
2570   memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2571 
2572   if (UseSignalChaining) {
2573      chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2574        * (Maxsignum + 1));
2575      memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2576      preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1));
2577      memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2578   }
2579   ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1 ));
2580   memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2581 }
2582 
2583 void os::signal_init_pd() {
2584   int ret;
2585 
2586   ret = ::sema_init(&sig_sem, 0, NULL, NULL);
2587   assert(ret == 0, "sema_init() failed");
2588 }
2589 
2590 void os::signal_notify(int signal_number) {
2591   int ret;
2592 
2593   Atomic::inc(&pending_signals[signal_number]);
2594   ret = ::sema_post(&sig_sem);
2595   assert(ret == 0, "sema_post() failed");
2596 }
2597 
2598 static int check_pending_signals(bool wait_for_signal) {
2599   int ret;
2600   while (true) {
2601     for (int i = 0; i < Sigexit + 1; i++) {
2602       jint n = pending_signals[i];
2603       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2604         return i;
2605       }
2606     }
2607     if (!wait_for_signal) {
2608       return -1;
2609     }
2610     JavaThread *thread = JavaThread::current();
2611     ThreadBlockInVM tbivm(thread);
2612 
2613     bool threadIsSuspended;
2614     do {
2615       thread->set_suspend_equivalent();
2616       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2617       while((ret = ::sema_wait(&sig_sem)) == EINTR)
2618           ;
2619       assert(ret == 0, "sema_wait() failed");
2620 
2621       // were we externally suspended while we were waiting?
2622       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2623       if (threadIsSuspended) {
2624         //
2625         // The semaphore has been incremented, but while we were waiting
2626         // another thread suspended us. We don't want to continue running
2627         // while suspended because that would surprise the thread that
2628         // suspended us.
2629         //
2630         ret = ::sema_post(&sig_sem);
2631         assert(ret == 0, "sema_post() failed");
2632 
2633         thread->java_suspend_self();
2634       }
2635     } while (threadIsSuspended);
2636   }
2637 }
2638 
2639 int os::signal_lookup() {
2640   return check_pending_signals(false);
2641 }
2642 
2643 int os::signal_wait() {
2644   return check_pending_signals(true);
2645 }
2646 
2647 ////////////////////////////////////////////////////////////////////////////////
2648 // Virtual Memory
2649 
2650 static int page_size = -1;
2651 
2652 // The mmap MAP_ALIGN flag is supported on Solaris 9 and later.  init_2() will
2653 // clear this var if support is not available.
2654 static bool has_map_align = true;
2655 
2656 int os::vm_page_size() {
2657   assert(page_size != -1, "must call os::init");
2658   return page_size;
2659 }
2660 
2661 // Solaris allocates memory by pages.
2662 int os::vm_allocation_granularity() {
2663   assert(page_size != -1, "must call os::init");
2664   return page_size;
2665 }
2666 
2667 bool os::commit_memory(char* addr, size_t bytes, bool exec) {
2668   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2669   size_t size = bytes;
2670   return
2671      NULL != Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2672 }
2673 
2674 bool os::commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2675                        bool exec) {
2676   if (commit_memory(addr, bytes, exec)) {
2677     if (UseMPSS && alignment_hint > (size_t)vm_page_size()) {
2678       // If the large page size has been set and the VM
2679       // is using large pages, use the large page size
2680       // if it is smaller than the alignment hint. This is
2681       // a case where the VM wants to use a larger alignment size
2682       // for its own reasons but still want to use large pages
2683       // (which is what matters to setting the mpss range.
2684       size_t page_size = 0;
2685       if (large_page_size() < alignment_hint) {
2686         assert(UseLargePages, "Expected to be here for large page use only");
2687         page_size = large_page_size();
2688       } else {
2689         // If the alignment hint is less than the large page
2690         // size, the VM wants a particular alignment (thus the hint)
2691         // for internal reasons.  Try to set the mpss range using
2692         // the alignment_hint.
2693         page_size = alignment_hint;
2694       }
2695       // Since this is a hint, ignore any failures.
2696       (void)Solaris::set_mpss_range(addr, bytes, page_size);
2697     }
2698     return true;
2699   }
2700   return false;
2701 }
2702 
2703 // Uncommit the pages in a specified region.
2704 void os::free_memory(char* addr, size_t bytes) {
2705   if (madvise(addr, bytes, MADV_FREE) < 0) {
2706     debug_only(warning("MADV_FREE failed."));
2707     return;
2708   }
2709 }
2710 
2711 bool os::create_stack_guard_pages(char* addr, size_t size) {
2712   return os::commit_memory(addr, size);
2713 }
2714 
2715 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2716   return os::uncommit_memory(addr, size);
2717 }
2718 
2719 // Change the page size in a given range.
2720 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2721   assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2722   assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2723   Solaris::set_mpss_range(addr, bytes, alignment_hint);
2724 }
2725 
2726 // Tell the OS to make the range local to the first-touching LWP
2727 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2728   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2729   if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2730     debug_only(warning("MADV_ACCESS_LWP failed."));
2731   }
2732 }
2733 
2734 // Tell the OS that this range would be accessed from different LWPs.
2735 void os::numa_make_global(char *addr, size_t bytes) {
2736   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2737   if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2738     debug_only(warning("MADV_ACCESS_MANY failed."));
2739   }
2740 }
2741 
2742 // Get the number of the locality groups.
2743 size_t os::numa_get_groups_num() {
2744   size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2745   return n != -1 ? n : 1;
2746 }
2747 
2748 // Get a list of leaf locality groups. A leaf lgroup is group that
2749 // doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2750 // board. An LWP is assigned to one of these groups upon creation.
2751 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2752    if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2753      ids[0] = 0;
2754      return 1;
2755    }
2756    int result_size = 0, top = 1, bottom = 0, cur = 0;
2757    for (int k = 0; k < size; k++) {
2758      int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2759                                     (Solaris::lgrp_id_t*)&ids[top], size - top);
2760      if (r == -1) {
2761        ids[0] = 0;
2762        return 1;
2763      }
2764      if (!r) {
2765        // That's a leaf node.
2766        assert (bottom <= cur, "Sanity check");
2767        // Check if the node has memory
2768        if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2769                                    NULL, 0, LGRP_RSRC_MEM) > 0) {
2770          ids[bottom++] = ids[cur];
2771        }
2772      }
2773      top += r;
2774      cur++;
2775    }
2776    if (bottom == 0) {
2777      // Handle a situation, when the OS reports no memory available.
2778      // Assume UMA architecture.
2779      ids[0] = 0;
2780      return 1;
2781    }
2782    return bottom;
2783 }
2784 
2785 // Detect the topology change. Typically happens during CPU plugging-unplugging.
2786 bool os::numa_topology_changed() {
2787   int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2788   if (is_stale != -1 && is_stale) {
2789     Solaris::lgrp_fini(Solaris::lgrp_cookie());
2790     Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2791     assert(c != 0, "Failure to initialize LGRP API");
2792     Solaris::set_lgrp_cookie(c);
2793     return true;
2794   }
2795   return false;
2796 }
2797 
2798 // Get the group id of the current LWP.
2799 int os::numa_get_group_id() {
2800   int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2801   if (lgrp_id == -1) {
2802     return 0;
2803   }
2804   const int size = os::numa_get_groups_num();
2805   int *ids = (int*)alloca(size * sizeof(int));
2806 
2807   // Get the ids of all lgroups with memory; r is the count.
2808   int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2809                                   (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2810   if (r <= 0) {
2811     return 0;
2812   }
2813   return ids[os::random() % r];
2814 }
2815 
2816 // Request information about the page.
2817 bool os::get_page_info(char *start, page_info* info) {
2818   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2819   uint64_t addr = (uintptr_t)start;
2820   uint64_t outdata[2];
2821   uint_t validity = 0;
2822 
2823   if (os::Solaris::meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2824     return false;
2825   }
2826 
2827   info->size = 0;
2828   info->lgrp_id = -1;
2829 
2830   if ((validity & 1) != 0) {
2831     if ((validity & 2) != 0) {
2832       info->lgrp_id = outdata[0];
2833     }
2834     if ((validity & 4) != 0) {
2835       info->size = outdata[1];
2836     }
2837     return true;
2838   }
2839   return false;
2840 }
2841 
2842 // Scan the pages from start to end until a page different than
2843 // the one described in the info parameter is encountered.
2844 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2845   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2846   const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2847   uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT];
2848   uint_t validity[MAX_MEMINFO_CNT];
2849 
2850   size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2851   uint64_t p = (uint64_t)start;
2852   while (p < (uint64_t)end) {
2853     addrs[0] = p;
2854     size_t addrs_count = 1;
2855     while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] < (uint64_t)end) {
2856       addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2857       addrs_count++;
2858     }
2859 
2860     if (os::Solaris::meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2861       return NULL;
2862     }
2863 
2864     size_t i = 0;
2865     for (; i < addrs_count; i++) {
2866       if ((validity[i] & 1) != 0) {
2867         if ((validity[i] & 4) != 0) {
2868           if (outdata[types * i + 1] != page_expected->size) {
2869             break;
2870           }
2871         } else
2872           if (page_expected->size != 0) {
2873             break;
2874           }
2875 
2876         if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
2877           if (outdata[types * i] != page_expected->lgrp_id) {
2878             break;
2879           }
2880         }
2881       } else {
2882         return NULL;
2883       }
2884     }
2885 
2886     if (i != addrs_count) {
2887       if ((validity[i] & 2) != 0) {
2888         page_found->lgrp_id = outdata[types * i];
2889       } else {
2890         page_found->lgrp_id = -1;
2891       }
2892       if ((validity[i] & 4) != 0) {
2893         page_found->size = outdata[types * i + 1];
2894       } else {
2895         page_found->size = 0;
2896       }
2897       return (char*)addrs[i];
2898     }
2899 
2900     p = addrs[addrs_count - 1] + page_size;
2901   }
2902   return end;
2903 }
2904 
2905 bool os::uncommit_memory(char* addr, size_t bytes) {
2906   size_t size = bytes;
2907   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2908   // uncommitted page. Otherwise, the read/write might succeed if we
2909   // have enough swap space to back the physical page.
2910   return
2911     NULL != Solaris::mmap_chunk(addr, size,
2912                                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
2913                                 PROT_NONE);
2914 }
2915 
2916 char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
2917   char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
2918 
2919   if (b == MAP_FAILED) {
2920     return NULL;
2921   }
2922   return b;
2923 }
2924 
2925 char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes, size_t alignment_hint, bool fixed) {
2926   char* addr = requested_addr;
2927   int flags = MAP_PRIVATE | MAP_NORESERVE;
2928 
2929   assert(!(fixed && (alignment_hint > 0)), "alignment hint meaningless with fixed mmap");
2930 
2931   if (fixed) {
2932     flags |= MAP_FIXED;
2933   } else if (has_map_align && (alignment_hint > (size_t) vm_page_size())) {
2934     flags |= MAP_ALIGN;
2935     addr = (char*) alignment_hint;
2936   }
2937 
2938   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2939   // uncommitted page. Otherwise, the read/write might succeed if we
2940   // have enough swap space to back the physical page.
2941   return mmap_chunk(addr, bytes, flags, PROT_NONE);
2942 }
2943 
2944 char* os::reserve_memory(size_t bytes, char* requested_addr, size_t alignment_hint) {
2945   char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint, (requested_addr != NULL));
2946 
2947   guarantee(requested_addr == NULL || requested_addr == addr,
2948             "OS failed to return requested mmap address.");
2949   return addr;
2950 }
2951 
2952 // Reserve memory at an arbitrary address, only if that area is
2953 // available (and not reserved for something else).
2954 
2955 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2956   const int max_tries = 10;
2957   char* base[max_tries];
2958   size_t size[max_tries];
2959 
2960   // Solaris adds a gap between mmap'ed regions.  The size of the gap
2961   // is dependent on the requested size and the MMU.  Our initial gap
2962   // value here is just a guess and will be corrected later.
2963   bool had_top_overlap = false;
2964   bool have_adjusted_gap = false;
2965   size_t gap = 0x400000;
2966 
2967   // Assert only that the size is a multiple of the page size, since
2968   // that's all that mmap requires, and since that's all we really know
2969   // about at this low abstraction level.  If we need higher alignment,
2970   // we can either pass an alignment to this method or verify alignment
2971   // in one of the methods further up the call chain.  See bug 5044738.
2972   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2973 
2974   // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
2975   // Give it a try, if the kernel honors the hint we can return immediately.
2976   char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
2977   volatile int err = errno;
2978   if (addr == requested_addr) {
2979     return addr;
2980   } else if (addr != NULL) {
2981     unmap_memory(addr, bytes);
2982   }
2983 
2984   if (PrintMiscellaneous && Verbose) {
2985     char buf[256];
2986     buf[0] = '\0';
2987     if (addr == NULL) {
2988       jio_snprintf(buf, sizeof(buf), ": %s", strerror(err));
2989     }
2990     warning("attempt_reserve_memory_at: couldn't reserve %d bytes at "
2991             PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
2992             "%s", bytes, requested_addr, addr, buf);
2993   }
2994 
2995   // Address hint method didn't work.  Fall back to the old method.
2996   // In theory, once SNV becomes our oldest supported platform, this
2997   // code will no longer be needed.
2998   //
2999   // Repeatedly allocate blocks until the block is allocated at the
3000   // right spot. Give up after max_tries.
3001   int i;
3002   for (i = 0; i < max_tries; ++i) {
3003     base[i] = reserve_memory(bytes);
3004 
3005     if (base[i] != NULL) {
3006       // Is this the block we wanted?
3007       if (base[i] == requested_addr) {
3008         size[i] = bytes;
3009         break;
3010       }
3011 
3012       // check that the gap value is right
3013       if (had_top_overlap && !have_adjusted_gap) {
3014         size_t actual_gap = base[i-1] - base[i] - bytes;
3015         if (gap != actual_gap) {
3016           // adjust the gap value and retry the last 2 allocations
3017           assert(i > 0, "gap adjustment code problem");
3018           have_adjusted_gap = true;  // adjust the gap only once, just in case
3019           gap = actual_gap;
3020           if (PrintMiscellaneous && Verbose) {
3021             warning("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
3022           }
3023           unmap_memory(base[i], bytes);
3024           unmap_memory(base[i-1], size[i-1]);
3025           i-=2;
3026           continue;
3027         }
3028       }
3029 
3030       // Does this overlap the block we wanted? Give back the overlapped
3031       // parts and try again.
3032       //
3033       // There is still a bug in this code: if top_overlap == bytes,
3034       // the overlap is offset from requested region by the value of gap.
3035       // In this case giving back the overlapped part will not work,
3036       // because we'll give back the entire block at base[i] and
3037       // therefore the subsequent allocation will not generate a new gap.
3038       // This could be fixed with a new algorithm that used larger
3039       // or variable size chunks to find the requested region -
3040       // but such a change would introduce additional complications.
3041       // It's rare enough that the planets align for this bug,
3042       // so we'll just wait for a fix for 6204603/5003415 which
3043       // will provide a mmap flag to allow us to avoid this business.
3044 
3045       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3046       if (top_overlap >= 0 && top_overlap < bytes) {
3047         had_top_overlap = true;
3048         unmap_memory(base[i], top_overlap);
3049         base[i] += top_overlap;
3050         size[i] = bytes - top_overlap;
3051       } else {
3052         size_t bottom_overlap = base[i] + bytes - requested_addr;
3053         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3054           if (PrintMiscellaneous && Verbose && bottom_overlap == 0) {
3055             warning("attempt_reserve_memory_at: possible alignment bug");
3056           }
3057           unmap_memory(requested_addr, bottom_overlap);
3058           size[i] = bytes - bottom_overlap;
3059         } else {
3060           size[i] = bytes;
3061         }
3062       }
3063     }
3064   }
3065 
3066   // Give back the unused reserved pieces.
3067 
3068   for (int j = 0; j < i; ++j) {
3069     if (base[j] != NULL) {
3070       unmap_memory(base[j], size[j]);
3071     }
3072   }
3073 
3074   return (i < max_tries) ? requested_addr : NULL;
3075 }
3076 
3077 bool os::release_memory(char* addr, size_t bytes) {
3078   size_t size = bytes;
3079   return munmap(addr, size) == 0;
3080 }
3081 
3082 static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
3083   assert(addr == (char*)align_size_down((uintptr_t)addr, os::vm_page_size()),
3084          "addr must be page aligned");
3085   int retVal = mprotect(addr, bytes, prot);
3086   return retVal == 0;
3087 }
3088 
3089 // Protect memory (Used to pass readonly pages through
3090 // JNI GetArray<type>Elements with empty arrays.)
3091 // Also, used for serialization page and for compressed oops null pointer
3092 // checking.
3093 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3094                         bool is_committed) {
3095   unsigned int p = 0;
3096   switch (prot) {
3097   case MEM_PROT_NONE: p = PROT_NONE; break;
3098   case MEM_PROT_READ: p = PROT_READ; break;
3099   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3100   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3101   default:
3102     ShouldNotReachHere();
3103   }
3104   // is_committed is unused.
3105   return solaris_mprotect(addr, bytes, p);
3106 }
3107 
3108 // guard_memory and unguard_memory only happens within stack guard pages.
3109 // Since ISM pertains only to the heap, guard and unguard memory should not
3110 /// happen with an ISM region.
3111 bool os::guard_memory(char* addr, size_t bytes) {
3112   return solaris_mprotect(addr, bytes, PROT_NONE);
3113 }
3114 
3115 bool os::unguard_memory(char* addr, size_t bytes) {
3116   return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
3117 }
3118 
3119 // Large page support
3120 
3121 // UseLargePages is the master flag to enable/disable large page memory.
3122 // UseMPSS and UseISM are supported for compatibility reasons. Their combined
3123 // effects can be described in the following table:
3124 //
3125 // UseLargePages UseMPSS UseISM
3126 //    false         *       *   => UseLargePages is the master switch, turning
3127 //                                 it off will turn off both UseMPSS and
3128 //                                 UseISM. VM will not use large page memory
3129 //                                 regardless the settings of UseMPSS/UseISM.
3130 //     true      false    false => Unless future Solaris provides other
3131 //                                 mechanism to use large page memory, this
3132 //                                 combination is equivalent to -UseLargePages,
3133 //                                 VM will not use large page memory
3134 //     true      true     false => JVM will use MPSS for large page memory.
3135 //                                 This is the default behavior.
3136 //     true      false    true  => JVM will use ISM for large page memory.
3137 //     true      true     true  => JVM will use ISM if it is available.
3138 //                                 Otherwise, JVM will fall back to MPSS.
3139 //                                 Becaues ISM is now available on all
3140 //                                 supported Solaris versions, this combination
3141 //                                 is equivalent to +UseISM -UseMPSS.
3142 
3143 typedef int (*getpagesizes_func_type) (size_t[], int);
3144 static size_t _large_page_size = 0;
3145 
3146 bool os::Solaris::ism_sanity_check(bool warn, size_t * page_size) {
3147   // x86 uses either 2M or 4M page, depending on whether PAE (Physical Address
3148   // Extensions) mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. Sparc
3149   // can support multiple page sizes.
3150 
3151   // Don't bother to probe page size because getpagesizes() comes with MPSS.
3152   // ISM is only recommended on old Solaris where there is no MPSS support.
3153   // Simply choose a conservative value as default.
3154   *page_size = LargePageSizeInBytes ? LargePageSizeInBytes :
3155                SPARC_ONLY(4 * M) IA32_ONLY(4 * M) AMD64_ONLY(2 * M)
3156                ARM_ONLY(2 * M);
3157 
3158   // ISM is available on all supported Solaris versions
3159   return true;
3160 }
3161 
3162 // Insertion sort for small arrays (descending order).
3163 static void insertion_sort_descending(size_t* array, int len) {
3164   for (int i = 0; i < len; i++) {
3165     size_t val = array[i];
3166     for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
3167       size_t tmp = array[key];
3168       array[key] = array[key - 1];
3169       array[key - 1] = tmp;
3170     }
3171   }
3172 }
3173 
3174 bool os::Solaris::mpss_sanity_check(bool warn, size_t * page_size) {
3175   getpagesizes_func_type getpagesizes_func =
3176     CAST_TO_FN_PTR(getpagesizes_func_type, dlsym(RTLD_DEFAULT, "getpagesizes"));
3177   if (getpagesizes_func == NULL) {
3178     if (warn) {
3179       warning("MPSS is not supported by the operating system.");
3180     }
3181     return false;
3182   }
3183 
3184   const unsigned int usable_count = VM_Version::page_size_count();
3185   if (usable_count == 1) {
3186     return false;
3187   }
3188 
3189   // Fill the array of page sizes.
3190   int n = getpagesizes_func(_page_sizes, page_sizes_max);
3191   assert(n > 0, "Solaris bug?");
3192   if (n == page_sizes_max) {
3193     // Add a sentinel value (necessary only if the array was completely filled
3194     // since it is static (zeroed at initialization)).
3195     _page_sizes[--n] = 0;
3196     DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
3197   }
3198   assert(_page_sizes[n] == 0, "missing sentinel");
3199 
3200   if (n == 1) return false;     // Only one page size available.
3201 
3202   // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
3203   // select up to usable_count elements.  First sort the array, find the first
3204   // acceptable value, then copy the usable sizes to the top of the array and
3205   // trim the rest.  Make sure to include the default page size :-).
3206   //
3207   // A better policy could get rid of the 4M limit by taking the sizes of the
3208   // important VM memory regions (java heap and possibly the code cache) into
3209   // account.
3210   insertion_sort_descending(_page_sizes, n);
3211   const size_t size_limit =
3212     FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
3213   int beg;
3214   for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */ ;
3215   const int end = MIN2((int)usable_count, n) - 1;
3216   for (int cur = 0; cur < end; ++cur, ++beg) {
3217     _page_sizes[cur] = _page_sizes[beg];
3218   }
3219   _page_sizes[end] = vm_page_size();
3220   _page_sizes[end + 1] = 0;
3221 
3222   if (_page_sizes[end] > _page_sizes[end - 1]) {
3223     // Default page size is not the smallest; sort again.
3224     insertion_sort_descending(_page_sizes, end + 1);
3225   }
3226   *page_size = _page_sizes[0];
3227 
3228   return true;
3229 }
3230 
3231 bool os::large_page_init() {
3232   if (!UseLargePages) {
3233     UseISM = false;
3234     UseMPSS = false;
3235     return false;
3236   }
3237 
3238   // print a warning if any large page related flag is specified on command line
3239   bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
3240                          !FLAG_IS_DEFAULT(UseISM)               ||
3241                          !FLAG_IS_DEFAULT(UseMPSS)              ||
3242                          !FLAG_IS_DEFAULT(LargePageSizeInBytes);
3243   UseISM = UseISM &&
3244            Solaris::ism_sanity_check(warn_on_failure, &_large_page_size);
3245   if (UseISM) {
3246     // ISM disables MPSS to be compatible with old JDK behavior
3247     UseMPSS = false;
3248     _page_sizes[0] = _large_page_size;
3249     _page_sizes[1] = vm_page_size();
3250   }
3251 
3252   UseMPSS = UseMPSS &&
3253             Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
3254 
3255   UseLargePages = UseISM || UseMPSS;
3256   return UseLargePages;
3257 }
3258 
3259 bool os::Solaris::set_mpss_range(caddr_t start, size_t bytes, size_t align) {
3260   // Signal to OS that we want large pages for addresses
3261   // from addr, addr + bytes
3262   struct memcntl_mha mpss_struct;
3263   mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
3264   mpss_struct.mha_pagesize = align;
3265   mpss_struct.mha_flags = 0;
3266   if (memcntl(start, bytes, MC_HAT_ADVISE,
3267               (caddr_t) &mpss_struct, 0, 0) < 0) {
3268     debug_only(warning("Attempt to use MPSS failed."));
3269     return false;
3270   }
3271   return true;
3272 }
3273 
3274 char* os::reserve_memory_special(size_t bytes, char* addr, bool exec) {
3275   // "exec" is passed in but not used.  Creating the shared image for
3276   // the code cache doesn't have an SHM_X executable permission to check.
3277   assert(UseLargePages && UseISM, "only for ISM large pages");
3278 
3279   size_t size = bytes;
3280   char* retAddr = NULL;
3281   int shmid;
3282   key_t ismKey;
3283 
3284   bool warn_on_failure = UseISM &&
3285                         (!FLAG_IS_DEFAULT(UseLargePages)         ||
3286                          !FLAG_IS_DEFAULT(UseISM)                ||
3287                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3288                         );
3289   char msg[128];
3290 
3291   ismKey = IPC_PRIVATE;
3292 
3293   // Create a large shared memory region to attach to based on size.
3294   // Currently, size is the total size of the heap
3295   shmid = shmget(ismKey, size, SHM_R | SHM_W | IPC_CREAT);
3296   if (shmid == -1){
3297      if (warn_on_failure) {
3298        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3299        warning(msg);
3300      }
3301      return NULL;
3302   }
3303 
3304   // Attach to the region
3305   retAddr = (char *) shmat(shmid, 0, SHM_SHARE_MMU | SHM_R | SHM_W);
3306   int err = errno;
3307 
3308   // Remove shmid. If shmat() is successful, the actual shared memory segment
3309   // will be deleted when it's detached by shmdt() or when the process
3310   // terminates. If shmat() is not successful this will remove the shared
3311   // segment immediately.
3312   shmctl(shmid, IPC_RMID, NULL);
3313 
3314   if (retAddr == (char *) -1) {
3315     if (warn_on_failure) {
3316       jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3317       warning(msg);
3318     }
3319     return NULL;
3320   }
3321 
3322   return retAddr;
3323 }
3324 
3325 bool os::release_memory_special(char* base, size_t bytes) {
3326   // detaching the SHM segment will also delete it, see reserve_memory_special()
3327   int rslt = shmdt(base);
3328   return rslt == 0;
3329 }
3330 
3331 size_t os::large_page_size() {
3332   return _large_page_size;
3333 }
3334 
3335 // MPSS allows application to commit large page memory on demand; with ISM
3336 // the entire memory region must be allocated as shared memory.
3337 bool os::can_commit_large_page_memory() {
3338   return UseISM ? false : true;
3339 }
3340 
3341 bool os::can_execute_large_page_memory() {
3342   return UseISM ? false : true;
3343 }
3344 
3345 static int os_sleep(jlong millis, bool interruptible) {
3346   const jlong limit = INT_MAX;
3347   jlong prevtime;
3348   int res;
3349 
3350   while (millis > limit) {
3351     if ((res = os_sleep(limit, interruptible)) != OS_OK)
3352       return res;
3353     millis -= limit;
3354   }
3355 
3356   // Restart interrupted polls with new parameters until the proper delay
3357   // has been completed.
3358 
3359   prevtime = getTimeMillis();
3360 
3361   while (millis > 0) {
3362     jlong newtime;
3363 
3364     if (!interruptible) {
3365       // Following assert fails for os::yield_all:
3366       // assert(!thread->is_Java_thread(), "must not be java thread");
3367       res = poll(NULL, 0, millis);
3368     } else {
3369       JavaThread *jt = JavaThread::current();
3370 
3371       INTERRUPTIBLE_NORESTART_VM_ALWAYS(poll(NULL, 0, millis), res, jt,
3372         os::Solaris::clear_interrupted);
3373     }
3374 
3375     // INTERRUPTIBLE_NORESTART_VM_ALWAYS returns res == OS_INTRPT for
3376     // thread.Interrupt.
3377 
3378     if((res == OS_ERR) && (errno == EINTR)) {
3379       newtime = getTimeMillis();
3380       assert(newtime >= prevtime, "time moving backwards");
3381     /* Doing prevtime and newtime in microseconds doesn't help precision,
3382        and trying to round up to avoid lost milliseconds can result in a
3383        too-short delay. */
3384       millis -= newtime - prevtime;
3385       if(millis <= 0)
3386         return OS_OK;
3387       prevtime = newtime;
3388     } else
3389       return res;
3390   }
3391 
3392   return OS_OK;
3393 }
3394 
3395 // Read calls from inside the vm need to perform state transitions
3396 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3397   INTERRUPTIBLE_RETURN_INT_VM(::read(fd, buf, nBytes), os::Solaris::clear_interrupted);
3398 }
3399 
3400 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3401   assert(thread == Thread::current(),  "thread consistency check");
3402 
3403   // TODO-FIXME: this should be removed.
3404   // On Solaris machines (especially 2.5.1) we found that sometimes the VM gets into a live lock
3405   // situation with a JavaThread being starved out of a lwp. The kernel doesn't seem to generate
3406   // a SIGWAITING signal which would enable the threads library to create a new lwp for the starving
3407   // thread. We suspect that because the Watcher thread keeps waking up at periodic intervals the kernel
3408   // is fooled into believing that the system is making progress. In the code below we block the
3409   // the watcher thread while safepoint is in progress so that it would not appear as though the
3410   // system is making progress.
3411   if (!Solaris::T2_libthread() &&
3412       thread->is_Watcher_thread() && SafepointSynchronize::is_synchronizing() && !Arguments::has_profile()) {
3413     // We now try to acquire the threads lock. Since this lock is held by the VM thread during
3414     // the entire safepoint, the watcher thread will  line up here during the safepoint.
3415     Threads_lock->lock_without_safepoint_check();
3416     Threads_lock->unlock();
3417   }
3418 
3419   if (thread->is_Java_thread()) {
3420     // This is a JavaThread so we honor the _thread_blocked protocol
3421     // even for sleeps of 0 milliseconds. This was originally done
3422     // as a workaround for bug 4338139. However, now we also do it
3423     // to honor the suspend-equivalent protocol.
3424 
3425     JavaThread *jt = (JavaThread *) thread;
3426     ThreadBlockInVM tbivm(jt);
3427 
3428     jt->set_suspend_equivalent();
3429     // cleared by handle_special_suspend_equivalent_condition() or
3430     // java_suspend_self() via check_and_wait_while_suspended()
3431 
3432     int ret_code;
3433     if (millis <= 0) {
3434       thr_yield();
3435       ret_code = 0;
3436     } else {
3437       // The original sleep() implementation did not create an
3438       // OSThreadWaitState helper for sleeps of 0 milliseconds.
3439       // I'm preserving that decision for now.
3440       OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3441 
3442       ret_code = os_sleep(millis, interruptible);
3443     }
3444 
3445     // were we externally suspended while we were waiting?
3446     jt->check_and_wait_while_suspended();
3447 
3448     return ret_code;
3449   }
3450 
3451   // non-JavaThread from this point on:
3452 
3453   if (millis <= 0) {
3454     thr_yield();
3455     return 0;
3456   }
3457 
3458   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3459 
3460   return os_sleep(millis, interruptible);
3461 }
3462 
3463 int os::naked_sleep() {
3464   // %% make the sleep time an integer flag. for now use 1 millisec.
3465   return os_sleep(1, false);
3466 }
3467 
3468 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3469 void os::infinite_sleep() {
3470   while (true) {    // sleep forever ...
3471     ::sleep(100);   // ... 100 seconds at a time
3472   }
3473 }
3474 
3475 // Used to convert frequent JVM_Yield() to nops
3476 bool os::dont_yield() {
3477   if (DontYieldALot) {
3478     static hrtime_t last_time = 0;
3479     hrtime_t diff = getTimeNanos() - last_time;
3480 
3481     if (diff < DontYieldALotInterval * 1000000)
3482       return true;
3483 
3484     last_time += diff;
3485 
3486     return false;
3487   }
3488   else {
3489     return false;
3490   }
3491 }
3492 
3493 // Caveat: Solaris os::yield() causes a thread-state transition whereas
3494 // the linux and win32 implementations do not.  This should be checked.
3495 
3496 void os::yield() {
3497   // Yields to all threads with same or greater priority
3498   os::sleep(Thread::current(), 0, false);
3499 }
3500 
3501 // Note that yield semantics are defined by the scheduling class to which
3502 // the thread currently belongs.  Typically, yield will _not yield to
3503 // other equal or higher priority threads that reside on the dispatch queues
3504 // of other CPUs.
3505 
3506 os::YieldResult os::NakedYield() { thr_yield(); return os::YIELD_UNKNOWN; }
3507 
3508 
3509 // On Solaris we found that yield_all doesn't always yield to all other threads.
3510 // There have been cases where there is a thread ready to execute but it doesn't
3511 // get an lwp as the VM thread continues to spin with sleeps of 1 millisecond.
3512 // The 1 millisecond wait doesn't seem long enough for the kernel to issue a
3513 // SIGWAITING signal which will cause a new lwp to be created. So we count the
3514 // number of times yield_all is called in the one loop and increase the sleep
3515 // time after 8 attempts. If this fails too we increase the concurrency level
3516 // so that the starving thread would get an lwp
3517 
3518 void os::yield_all(int attempts) {
3519   // Yields to all threads, including threads with lower priorities
3520   if (attempts == 0) {
3521     os::sleep(Thread::current(), 1, false);
3522   } else {
3523     int iterations = attempts % 30;
3524     if (iterations == 0 && !os::Solaris::T2_libthread()) {
3525       // thr_setconcurrency and _getconcurrency make sense only under T1.
3526       int noofLWPS = thr_getconcurrency();
3527       if (noofLWPS < (Threads::number_of_threads() + 2)) {
3528         thr_setconcurrency(thr_getconcurrency() + 1);
3529       }
3530     } else if (iterations < 25) {
3531       os::sleep(Thread::current(), 1, false);
3532     } else {
3533       os::sleep(Thread::current(), 10, false);
3534     }
3535   }
3536 }
3537 
3538 // Called from the tight loops to possibly influence time-sharing heuristics
3539 void os::loop_breaker(int attempts) {
3540   os::yield_all(attempts);
3541 }
3542 
3543 
3544 // Interface for setting lwp priorities.  If we are using T2 libthread,
3545 // which forces the use of BoundThreads or we manually set UseBoundThreads,
3546 // all of our threads will be assigned to real lwp's.  Using the thr_setprio
3547 // function is meaningless in this mode so we must adjust the real lwp's priority
3548 // The routines below implement the getting and setting of lwp priorities.
3549 //
3550 // Note: There are three priority scales used on Solaris.  Java priotities
3551 //       which range from 1 to 10, libthread "thr_setprio" scale which range
3552 //       from 0 to 127, and the current scheduling class of the process we
3553 //       are running in.  This is typically from -60 to +60.
3554 //       The setting of the lwp priorities in done after a call to thr_setprio
3555 //       so Java priorities are mapped to libthread priorities and we map from
3556 //       the latter to lwp priorities.  We don't keep priorities stored in
3557 //       Java priorities since some of our worker threads want to set priorities
3558 //       higher than all Java threads.
3559 //
3560 // For related information:
3561 // (1)  man -s 2 priocntl
3562 // (2)  man -s 4 priocntl
3563 // (3)  man dispadmin
3564 // =    librt.so
3565 // =    libthread/common/rtsched.c - thrp_setlwpprio().
3566 // =    ps -cL <pid> ... to validate priority.
3567 // =    sched_get_priority_min and _max
3568 //              pthread_create
3569 //              sched_setparam
3570 //              pthread_setschedparam
3571 //
3572 // Assumptions:
3573 // +    We assume that all threads in the process belong to the same
3574 //              scheduling class.   IE. an homogenous process.
3575 // +    Must be root or in IA group to change change "interactive" attribute.
3576 //              Priocntl() will fail silently.  The only indication of failure is when
3577 //              we read-back the value and notice that it hasn't changed.
3578 // +    Interactive threads enter the runq at the head, non-interactive at the tail.
3579 // +    For RT, change timeslice as well.  Invariant:
3580 //              constant "priority integral"
3581 //              Konst == TimeSlice * (60-Priority)
3582 //              Given a priority, compute appropriate timeslice.
3583 // +    Higher numerical values have higher priority.
3584 
3585 // sched class attributes
3586 typedef struct {
3587         int   schedPolicy;              // classID
3588         int   maxPrio;
3589         int   minPrio;
3590 } SchedInfo;
3591 
3592 
3593 static SchedInfo tsLimits, iaLimits, rtLimits;
3594 
3595 #ifdef ASSERT
3596 static int  ReadBackValidate = 1;
3597 #endif
3598 static int  myClass     = 0;
3599 static int  myMin       = 0;
3600 static int  myMax       = 0;
3601 static int  myCur       = 0;
3602 static bool priocntl_enable = false;
3603 
3604 
3605 // Call the version of priocntl suitable for all supported versions
3606 // of Solaris. We need to call through this wrapper so that we can
3607 // build on Solaris 9 and run on Solaris 8, 9 and 10.
3608 //
3609 // This code should be removed if we ever stop supporting Solaris 8
3610 // and earlier releases.
3611 
3612 static long priocntl_stub(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg);
3613 typedef long (*priocntl_type)(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg);
3614 static priocntl_type priocntl_ptr = priocntl_stub;
3615 
3616 // Stub to set the value of the real pointer, and then call the real
3617 // function.
3618 
3619 static long priocntl_stub(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg) {
3620   // Try Solaris 8- name only.
3621   priocntl_type tmp = (priocntl_type)dlsym(RTLD_DEFAULT, "__priocntl");
3622   guarantee(tmp != NULL, "priocntl function not found.");
3623   priocntl_ptr = tmp;
3624   return (*priocntl_ptr)(PC_VERSION, idtype, id, cmd, arg);
3625 }
3626 
3627 
3628 // lwp_priocntl_init
3629 //
3630 // Try to determine the priority scale for our process.
3631 //
3632 // Return errno or 0 if OK.
3633 //
3634 static
3635 int     lwp_priocntl_init ()
3636 {
3637   int rslt;
3638   pcinfo_t ClassInfo;
3639   pcparms_t ParmInfo;
3640   int i;
3641 
3642   if (!UseThreadPriorities) return 0;
3643 
3644   // We are using Bound threads, we need to determine our priority ranges
3645   if (os::Solaris::T2_libthread() || UseBoundThreads) {
3646     // If ThreadPriorityPolicy is 1, switch tables
3647     if (ThreadPriorityPolicy == 1) {
3648       for (i = 0 ; i < MaxPriority+1; i++)
3649         os::java_to_os_priority[i] = prio_policy1[i];
3650     }
3651   }
3652   // Not using Bound Threads, set to ThreadPolicy 1
3653   else {
3654     for ( i = 0 ; i < MaxPriority+1; i++ ) {
3655       os::java_to_os_priority[i] = prio_policy1[i];
3656     }
3657     return 0;
3658   }
3659 
3660 
3661   // Get IDs for a set of well-known scheduling classes.
3662   // TODO-FIXME: GETCLINFO returns the current # of classes in the
3663   // the system.  We should have a loop that iterates over the
3664   // classID values, which are known to be "small" integers.
3665 
3666   strcpy(ClassInfo.pc_clname, "TS");
3667   ClassInfo.pc_cid = -1;
3668   rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3669   if (rslt < 0) return errno;
3670   assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3671   tsLimits.schedPolicy = ClassInfo.pc_cid;
3672   tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3673   tsLimits.minPrio = -tsLimits.maxPrio;
3674 
3675   strcpy(ClassInfo.pc_clname, "IA");
3676   ClassInfo.pc_cid = -1;
3677   rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3678   if (rslt < 0) return errno;
3679   assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3680   iaLimits.schedPolicy = ClassInfo.pc_cid;
3681   iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3682   iaLimits.minPrio = -iaLimits.maxPrio;
3683 
3684   strcpy(ClassInfo.pc_clname, "RT");
3685   ClassInfo.pc_cid = -1;
3686   rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3687   if (rslt < 0) return errno;
3688   assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3689   rtLimits.schedPolicy = ClassInfo.pc_cid;
3690   rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3691   rtLimits.minPrio = 0;
3692 
3693 
3694   // Query our "current" scheduling class.
3695   // This will normally be IA,TS or, rarely, RT.
3696   memset (&ParmInfo, 0, sizeof(ParmInfo));
3697   ParmInfo.pc_cid = PC_CLNULL;
3698   rslt = (*priocntl_ptr) (PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo );
3699   if ( rslt < 0 ) return errno;
3700   myClass = ParmInfo.pc_cid;
3701 
3702   // We now know our scheduling classId, get specific information
3703   // the class.
3704   ClassInfo.pc_cid = myClass;
3705   ClassInfo.pc_clname[0] = 0;
3706   rslt = (*priocntl_ptr) (PC_VERSION, (idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo );
3707   if ( rslt < 0 ) return errno;
3708 
3709   if (ThreadPriorityVerbose)
3710     tty->print_cr ("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3711 
3712   memset(&ParmInfo, 0, sizeof(pcparms_t));
3713   ParmInfo.pc_cid = PC_CLNULL;
3714   rslt = (*priocntl_ptr)(PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3715   if (rslt < 0) return errno;
3716 
3717   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3718     myMin = rtLimits.minPrio;
3719     myMax = rtLimits.maxPrio;
3720   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3721     iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3722     myMin = iaLimits.minPrio;
3723     myMax = iaLimits.maxPrio;
3724     myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3725   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3726     tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3727     myMin = tsLimits.minPrio;
3728     myMax = tsLimits.maxPrio;
3729     myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3730   } else {
3731     // No clue - punt
3732     if (ThreadPriorityVerbose)
3733       tty->print_cr ("Unknown scheduling class: %s ... \n", ClassInfo.pc_clname);
3734     return EINVAL;      // no clue, punt
3735   }
3736 
3737   if (ThreadPriorityVerbose)
3738         tty->print_cr ("Thread priority Range: [%d..%d]\n", myMin, myMax);
3739 
3740   priocntl_enable = true;  // Enable changing priorities
3741   return 0;
3742 }
3743 
3744 #define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3745 #define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3746 #define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3747 
3748 
3749 // scale_to_lwp_priority
3750 //
3751 // Convert from the libthread "thr_setprio" scale to our current
3752 // lwp scheduling class scale.
3753 //
3754 static
3755 int     scale_to_lwp_priority (int rMin, int rMax, int x)
3756 {
3757   int v;
3758 
3759   if (x == 127) return rMax;            // avoid round-down
3760     v = (((x*(rMax-rMin)))/128)+rMin;
3761   return v;
3762 }
3763 
3764 
3765 // set_lwp_priority
3766 //
3767 // Set the priority of the lwp.  This call should only be made
3768 // when using bound threads (T2 threads are bound by default).
3769 //
3770 int     set_lwp_priority (int ThreadID, int lwpid, int newPrio )
3771 {
3772   int rslt;
3773   int Actual, Expected, prv;
3774   pcparms_t ParmInfo;                   // for GET-SET
3775 #ifdef ASSERT
3776   pcparms_t ReadBack;                   // for readback
3777 #endif
3778 
3779   // Set priority via PC_GETPARMS, update, PC_SETPARMS
3780   // Query current values.
3781   // TODO: accelerate this by eliminating the PC_GETPARMS call.
3782   // Cache "pcparms_t" in global ParmCache.
3783   // TODO: elide set-to-same-value
3784 
3785   // If something went wrong on init, don't change priorities.
3786   if ( !priocntl_enable ) {
3787     if (ThreadPriorityVerbose)
3788       tty->print_cr("Trying to set priority but init failed, ignoring");
3789     return EINVAL;
3790   }
3791 
3792 
3793   // If lwp hasn't started yet, just return
3794   // the _start routine will call us again.
3795   if ( lwpid <= 0 ) {
3796     if (ThreadPriorityVerbose) {
3797       tty->print_cr ("deferring the set_lwp_priority of thread " INTPTR_FORMAT " to %d, lwpid not set",
3798                      ThreadID, newPrio);
3799     }
3800     return 0;
3801   }
3802 
3803   if (ThreadPriorityVerbose) {
3804     tty->print_cr ("set_lwp_priority(" INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3805                    ThreadID, lwpid, newPrio);
3806   }
3807 
3808   memset(&ParmInfo, 0, sizeof(pcparms_t));
3809   ParmInfo.pc_cid = PC_CLNULL;
3810   rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3811   if (rslt < 0) return errno;
3812 
3813   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3814     rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3815     rtInfo->rt_pri     = scale_to_lwp_priority (rtLimits.minPrio, rtLimits.maxPrio, newPrio);
3816     rtInfo->rt_tqsecs  = RT_NOCHANGE;
3817     rtInfo->rt_tqnsecs = RT_NOCHANGE;
3818     if (ThreadPriorityVerbose) {
3819       tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3820     }
3821   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3822     iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3823     int maxClamped     = MIN2(iaLimits.maxPrio, (int)iaInfo->ia_uprilim);
3824     iaInfo->ia_upri    = scale_to_lwp_priority(iaLimits.minPrio, maxClamped, newPrio);
3825     iaInfo->ia_uprilim = IA_NOCHANGE;
3826     iaInfo->ia_mode    = IA_NOCHANGE;
3827     if (ThreadPriorityVerbose) {
3828       tty->print_cr ("IA: [%d...%d] %d->%d\n",
3829                iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3830     }
3831   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3832     tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3833     int maxClamped     = MIN2(tsLimits.maxPrio, (int)tsInfo->ts_uprilim);
3834     prv                = tsInfo->ts_upri;
3835     tsInfo->ts_upri    = scale_to_lwp_priority(tsLimits.minPrio, maxClamped, newPrio);
3836     tsInfo->ts_uprilim = IA_NOCHANGE;
3837     if (ThreadPriorityVerbose) {
3838       tty->print_cr ("TS: %d [%d...%d] %d->%d\n",
3839                prv, tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
3840     }
3841     if (prv == tsInfo->ts_upri) return 0;
3842   } else {
3843     if ( ThreadPriorityVerbose ) {
3844       tty->print_cr ("Unknown scheduling class\n");
3845     }
3846       return EINVAL;    // no clue, punt
3847   }
3848 
3849   rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
3850   if (ThreadPriorityVerbose && rslt) {
3851     tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
3852   }
3853   if (rslt < 0) return errno;
3854 
3855 #ifdef ASSERT
3856   // Sanity check: read back what we just attempted to set.
3857   // In theory it could have changed in the interim ...
3858   //
3859   // The priocntl system call is tricky.
3860   // Sometimes it'll validate the priority value argument and
3861   // return EINVAL if unhappy.  At other times it fails silently.
3862   // Readbacks are prudent.
3863 
3864   if (!ReadBackValidate) return 0;
3865 
3866   memset(&ReadBack, 0, sizeof(pcparms_t));
3867   ReadBack.pc_cid = PC_CLNULL;
3868   rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
3869   assert(rslt >= 0, "priocntl failed");
3870   Actual = Expected = 0xBAD;
3871   assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
3872   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3873     Actual   = RTPRI(ReadBack)->rt_pri;
3874     Expected = RTPRI(ParmInfo)->rt_pri;
3875   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3876     Actual   = IAPRI(ReadBack)->ia_upri;
3877     Expected = IAPRI(ParmInfo)->ia_upri;
3878   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3879     Actual   = TSPRI(ReadBack)->ts_upri;
3880     Expected = TSPRI(ParmInfo)->ts_upri;
3881   } else {
3882     if ( ThreadPriorityVerbose ) {
3883       tty->print_cr("set_lwp_priority: unexpected class in readback: %d\n", ParmInfo.pc_cid);
3884     }
3885   }
3886 
3887   if (Actual != Expected) {
3888     if ( ThreadPriorityVerbose ) {
3889       tty->print_cr ("set_lwp_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
3890              lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
3891     }
3892   }
3893 #endif
3894 
3895   return 0;
3896 }
3897 
3898 
3899 
3900 // Solaris only gives access to 128 real priorities at a time,
3901 // so we expand Java's ten to fill this range.  This would be better
3902 // if we dynamically adjusted relative priorities.
3903 //
3904 // The ThreadPriorityPolicy option allows us to select 2 different
3905 // priority scales.
3906 //
3907 // ThreadPriorityPolicy=0
3908 // Since the Solaris' default priority is MaximumPriority, we do not
3909 // set a priority lower than Max unless a priority lower than
3910 // NormPriority is requested.
3911 //
3912 // ThreadPriorityPolicy=1
3913 // This mode causes the priority table to get filled with
3914 // linear values.  NormPriority get's mapped to 50% of the
3915 // Maximum priority an so on.  This will cause VM threads
3916 // to get unfair treatment against other Solaris processes
3917 // which do not explicitly alter their thread priorities.
3918 //
3919 
3920 
3921 int os::java_to_os_priority[MaxPriority + 1] = {
3922   -99999,         // 0 Entry should never be used
3923 
3924   0,              // 1 MinPriority
3925   32,             // 2
3926   64,             // 3
3927 
3928   96,             // 4
3929   127,            // 5 NormPriority
3930   127,            // 6
3931 
3932   127,            // 7
3933   127,            // 8
3934   127,            // 9 NearMaxPriority
3935 
3936   127             // 10 MaxPriority
3937 };
3938 
3939 
3940 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3941   assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
3942   if ( !UseThreadPriorities ) return OS_OK;
3943   int status = thr_setprio(thread->osthread()->thread_id(), newpri);
3944   if ( os::Solaris::T2_libthread() || (UseBoundThreads && thread->osthread()->is_vm_created()) )
3945     status |= (set_lwp_priority (thread->osthread()->thread_id(),
3946                     thread->osthread()->lwp_id(), newpri ));
3947   return (status == 0) ? OS_OK : OS_ERR;
3948 }
3949 
3950 
3951 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3952   int p;
3953   if ( !UseThreadPriorities ) {
3954     *priority_ptr = NormalPriority;
3955     return OS_OK;
3956   }
3957   int status = thr_getprio(thread->osthread()->thread_id(), &p);
3958   if (status != 0) {
3959     return OS_ERR;
3960   }
3961   *priority_ptr = p;
3962   return OS_OK;
3963 }
3964 
3965 
3966 // Hint to the underlying OS that a task switch would not be good.
3967 // Void return because it's a hint and can fail.
3968 void os::hint_no_preempt() {
3969   schedctl_start(schedctl_init());
3970 }
3971 
3972 void os::interrupt(Thread* thread) {
3973   assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
3974 
3975   OSThread* osthread = thread->osthread();
3976 
3977   int isInterrupted = osthread->interrupted();
3978   if (!isInterrupted) {
3979       osthread->set_interrupted(true);
3980       OrderAccess::fence();
3981       // os::sleep() is implemented with either poll (NULL,0,timeout) or
3982       // by parking on _SleepEvent.  If the former, thr_kill will unwedge
3983       // the sleeper by SIGINTR, otherwise the unpark() will wake the sleeper.
3984       ParkEvent * const slp = thread->_SleepEvent ;
3985       if (slp != NULL) slp->unpark() ;
3986   }
3987 
3988   // For JSR166:  unpark after setting status but before thr_kill -dl
3989   if (thread->is_Java_thread()) {
3990     ((JavaThread*)thread)->parker()->unpark();
3991   }
3992 
3993   // Handle interruptible wait() ...
3994   ParkEvent * const ev = thread->_ParkEvent ;
3995   if (ev != NULL) ev->unpark() ;
3996 
3997   // When events are used everywhere for os::sleep, then this thr_kill
3998   // will only be needed if UseVMInterruptibleIO is true.
3999 
4000   if (!isInterrupted) {
4001     int status = thr_kill(osthread->thread_id(), os::Solaris::SIGinterrupt());
4002     assert_status(status == 0, status, "thr_kill");
4003 
4004     // Bump thread interruption counter
4005     RuntimeService::record_thread_interrupt_signaled_count();
4006   }
4007 }
4008 
4009 
4010 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4011   assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
4012 
4013   OSThread* osthread = thread->osthread();
4014 
4015   bool res = osthread->interrupted();
4016 
4017   // NOTE that since there is no "lock" around these two operations,
4018   // there is the possibility that the interrupted flag will be
4019   // "false" but that the interrupt event will be set. This is
4020   // intentional. The effect of this is that Object.wait() will appear
4021   // to have a spurious wakeup, which is not harmful, and the
4022   // possibility is so rare that it is not worth the added complexity
4023   // to add yet another lock. It has also been recommended not to put
4024   // the interrupted flag into the os::Solaris::Event structure,
4025   // because it hides the issue.
4026   if (res && clear_interrupted) {
4027     osthread->set_interrupted(false);
4028   }
4029   return res;
4030 }
4031 
4032 
4033 void os::print_statistics() {
4034 }
4035 
4036 int os::message_box(const char* title, const char* message) {
4037   int i;
4038   fdStream err(defaultStream::error_fd());
4039   for (i = 0; i < 78; i++) err.print_raw("=");
4040   err.cr();
4041   err.print_raw_cr(title);
4042   for (i = 0; i < 78; i++) err.print_raw("-");
4043   err.cr();
4044   err.print_raw_cr(message);
4045   for (i = 0; i < 78; i++) err.print_raw("=");
4046   err.cr();
4047 
4048   char buf[16];
4049   // Prevent process from exiting upon "read error" without consuming all CPU
4050   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4051 
4052   return buf[0] == 'y' || buf[0] == 'Y';
4053 }
4054 
4055 // A lightweight implementation that does not suspend the target thread and
4056 // thus returns only a hint. Used for profiling only!
4057 ExtendedPC os::get_thread_pc(Thread* thread) {
4058   // Make sure that it is called by the watcher and the Threads lock is owned.
4059   assert(Thread::current()->is_Watcher_thread(), "Must be watcher and own Threads_lock");
4060   // For now, is only used to profile the VM Thread
4061   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4062   ExtendedPC epc;
4063 
4064   GetThreadPC_Callback  cb(ProfileVM_lock);
4065   OSThread *osthread = thread->osthread();
4066   const int time_to_wait = 400; // 400ms wait for initial response
4067   int status = cb.interrupt(thread, time_to_wait);
4068 
4069   if (cb.is_done() ) {
4070     epc = cb.addr();
4071   } else {
4072     DEBUG_ONLY(tty->print_cr("Failed to get pc for thread: %d got %d status",
4073                               osthread->thread_id(), status););
4074     // epc is already NULL
4075   }
4076   return epc;
4077 }
4078 
4079 
4080 // This does not do anything on Solaris. This is basically a hook for being
4081 // able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
4082 void os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, JavaCallArguments* args, Thread* thread) {
4083   f(value, method, args, thread);
4084 }
4085 
4086 // This routine may be used by user applications as a "hook" to catch signals.
4087 // The user-defined signal handler must pass unrecognized signals to this
4088 // routine, and if it returns true (non-zero), then the signal handler must
4089 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4090 // routine will never retun false (zero), but instead will execute a VM panic
4091 // routine kill the process.
4092 //
4093 // If this routine returns false, it is OK to call it again.  This allows
4094 // the user-defined signal handler to perform checks either before or after
4095 // the VM performs its own checks.  Naturally, the user code would be making
4096 // a serious error if it tried to handle an exception (such as a null check
4097 // or breakpoint) that the VM was generating for its own correct operation.
4098 //
4099 // This routine may recognize any of the following kinds of signals:
4100 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
4101 // os::Solaris::SIGasync
4102 // It should be consulted by handlers for any of those signals.
4103 // It explicitly does not recognize os::Solaris::SIGinterrupt
4104 //
4105 // The caller of this routine must pass in the three arguments supplied
4106 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4107 // field of the structure passed to sigaction().  This routine assumes that
4108 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4109 //
4110 // Note that the VM will print warnings if it detects conflicting signal
4111 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4112 //
4113 extern "C" int JVM_handle_solaris_signal(int signo, siginfo_t* siginfo, void* ucontext, int abort_if_unrecognized);
4114 
4115 
4116 void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
4117   JVM_handle_solaris_signal(sig, info, ucVoid, true);
4118 }
4119 
4120 /* Do not delete - if guarantee is ever removed,  a signal handler (even empty)
4121    is needed to provoke threads blocked on IO to return an EINTR
4122    Note: this explicitly does NOT call JVM_handle_solaris_signal and
4123    does NOT participate in signal chaining due to requirement for
4124    NOT setting SA_RESTART to make EINTR work. */
4125 extern "C" void sigINTRHandler(int sig, siginfo_t* info, void* ucVoid) {
4126    if (UseSignalChaining) {
4127       struct sigaction *actp = os::Solaris::get_chained_signal_action(sig);
4128       if (actp && actp->sa_handler) {
4129         vm_exit_during_initialization("Signal chaining detected for VM interrupt signal, try -XX:+UseAltSigs");
4130       }
4131    }
4132 }
4133 
4134 // This boolean allows users to forward their own non-matching signals
4135 // to JVM_handle_solaris_signal, harmlessly.
4136 bool os::Solaris::signal_handlers_are_installed = false;
4137 
4138 // For signal-chaining
4139 bool os::Solaris::libjsig_is_loaded = false;
4140 typedef struct sigaction *(*get_signal_t)(int);
4141 get_signal_t os::Solaris::get_signal_action = NULL;
4142 
4143 struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
4144   struct sigaction *actp = NULL;
4145 
4146   if ((libjsig_is_loaded)  && (sig <= Maxlibjsigsigs)) {
4147     // Retrieve the old signal handler from libjsig
4148     actp = (*get_signal_action)(sig);
4149   }
4150   if (actp == NULL) {
4151     // Retrieve the preinstalled signal handler from jvm
4152     actp = get_preinstalled_handler(sig);
4153   }
4154 
4155   return actp;
4156 }
4157 
4158 static bool call_chained_handler(struct sigaction *actp, int sig,
4159                                  siginfo_t *siginfo, void *context) {
4160   // Call the old signal handler
4161   if (actp->sa_handler == SIG_DFL) {
4162     // It's more reasonable to let jvm treat it as an unexpected exception
4163     // instead of taking the default action.
4164     return false;
4165   } else if (actp->sa_handler != SIG_IGN) {
4166     if ((actp->sa_flags & SA_NODEFER) == 0) {
4167       // automaticlly block the signal
4168       sigaddset(&(actp->sa_mask), sig);
4169     }
4170 
4171     sa_handler_t hand;
4172     sa_sigaction_t sa;
4173     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4174     // retrieve the chained handler
4175     if (siginfo_flag_set) {
4176       sa = actp->sa_sigaction;
4177     } else {
4178       hand = actp->sa_handler;
4179     }
4180 
4181     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4182       actp->sa_handler = SIG_DFL;
4183     }
4184 
4185     // try to honor the signal mask
4186     sigset_t oset;
4187     thr_sigsetmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4188 
4189     // call into the chained handler
4190     if (siginfo_flag_set) {
4191       (*sa)(sig, siginfo, context);
4192     } else {
4193       (*hand)(sig);
4194     }
4195 
4196     // restore the signal mask
4197     thr_sigsetmask(SIG_SETMASK, &oset, 0);
4198   }
4199   // Tell jvm's signal handler the signal is taken care of.
4200   return true;
4201 }
4202 
4203 bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4204   bool chained = false;
4205   // signal-chaining
4206   if (UseSignalChaining) {
4207     struct sigaction *actp = get_chained_signal_action(sig);
4208     if (actp != NULL) {
4209       chained = call_chained_handler(actp, sig, siginfo, context);
4210     }
4211   }
4212   return chained;
4213 }
4214 
4215 struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
4216   assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4217   if (preinstalled_sigs[sig] != 0) {
4218     return &chainedsigactions[sig];
4219   }
4220   return NULL;
4221 }
4222 
4223 void os::Solaris::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4224 
4225   assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
4226   assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4227   chainedsigactions[sig] = oldAct;
4228   preinstalled_sigs[sig] = 1;
4229 }
4230 
4231 void os::Solaris::set_signal_handler(int sig, bool set_installed, bool oktochain) {
4232   // Check for overwrite.
4233   struct sigaction oldAct;
4234   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4235   void* oldhand = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4236                                       : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4237   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4238       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4239       oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
4240     if (AllowUserSignalHandlers || !set_installed) {
4241       // Do not overwrite; user takes responsibility to forward to us.
4242       return;
4243     } else if (UseSignalChaining) {
4244       if (oktochain) {
4245         // save the old handler in jvm
4246         save_preinstalled_handler(sig, oldAct);
4247       } else {
4248         vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal, try -XX:+UseAltSigs.");
4249       }
4250       // libjsig also interposes the sigaction() call below and saves the
4251       // old sigaction on it own.
4252     } else {
4253       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4254                     "%#lx for signal %d.", (long)oldhand, sig));
4255     }
4256   }
4257 
4258   struct sigaction sigAct;
4259   sigfillset(&(sigAct.sa_mask));
4260   sigAct.sa_handler = SIG_DFL;
4261 
4262   sigAct.sa_sigaction = signalHandler;
4263   // Handle SIGSEGV on alternate signal stack if
4264   // not using stack banging
4265   if (!UseStackBanging && sig == SIGSEGV) {
4266     sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
4267   // Interruptible i/o requires SA_RESTART cleared so EINTR
4268   // is returned instead of restarting system calls
4269   } else if (sig == os::Solaris::SIGinterrupt()) {
4270     sigemptyset(&sigAct.sa_mask);
4271     sigAct.sa_handler = NULL;
4272     sigAct.sa_flags = SA_SIGINFO;
4273     sigAct.sa_sigaction = sigINTRHandler;
4274   } else {
4275     sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
4276   }
4277   os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
4278 
4279   sigaction(sig, &sigAct, &oldAct);
4280 
4281   void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4282                                        : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4283   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4284 }
4285 
4286 
4287 #define DO_SIGNAL_CHECK(sig) \
4288   if (!sigismember(&check_signal_done, sig)) \
4289     os::Solaris::check_signal_handler(sig)
4290 
4291 // This method is a periodic task to check for misbehaving JNI applications
4292 // under CheckJNI, we can add any periodic checks here
4293 
4294 void os::run_periodic_checks() {
4295   // A big source of grief is hijacking virt. addr 0x0 on Solaris,
4296   // thereby preventing a NULL checks.
4297   if(!check_addr0_done) check_addr0_done = check_addr0(tty);
4298 
4299   if (check_signals == false) return;
4300 
4301   // SEGV and BUS if overridden could potentially prevent
4302   // generation of hs*.log in the event of a crash, debugging
4303   // such a case can be very challenging, so we absolutely
4304   // check for the following for a good measure:
4305   DO_SIGNAL_CHECK(SIGSEGV);
4306   DO_SIGNAL_CHECK(SIGILL);
4307   DO_SIGNAL_CHECK(SIGFPE);
4308   DO_SIGNAL_CHECK(SIGBUS);
4309   DO_SIGNAL_CHECK(SIGPIPE);
4310   DO_SIGNAL_CHECK(SIGXFSZ);
4311 
4312   // ReduceSignalUsage allows the user to override these handlers
4313   // see comments at the very top and jvm_solaris.h
4314   if (!ReduceSignalUsage) {
4315     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4316     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4317     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4318     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4319   }
4320 
4321   // See comments above for using JVM1/JVM2 and UseAltSigs
4322   DO_SIGNAL_CHECK(os::Solaris::SIGinterrupt());
4323   DO_SIGNAL_CHECK(os::Solaris::SIGasync());
4324 
4325 }
4326 
4327 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4328 
4329 static os_sigaction_t os_sigaction = NULL;
4330 
4331 void os::Solaris::check_signal_handler(int sig) {
4332   char buf[O_BUFLEN];
4333   address jvmHandler = NULL;
4334 
4335   struct sigaction act;
4336   if (os_sigaction == NULL) {
4337     // only trust the default sigaction, in case it has been interposed
4338     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4339     if (os_sigaction == NULL) return;
4340   }
4341 
4342   os_sigaction(sig, (struct sigaction*)NULL, &act);
4343 
4344   address thisHandler = (act.sa_flags & SA_SIGINFO)
4345     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4346     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4347 
4348 
4349   switch(sig) {
4350     case SIGSEGV:
4351     case SIGBUS:
4352     case SIGFPE:
4353     case SIGPIPE:
4354     case SIGXFSZ:
4355     case SIGILL:
4356       jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4357       break;
4358 
4359     case SHUTDOWN1_SIGNAL:
4360     case SHUTDOWN2_SIGNAL:
4361     case SHUTDOWN3_SIGNAL:
4362     case BREAK_SIGNAL:
4363       jvmHandler = (address)user_handler();
4364       break;
4365 
4366     default:
4367       int intrsig = os::Solaris::SIGinterrupt();
4368       int asynsig = os::Solaris::SIGasync();
4369 
4370       if (sig == intrsig) {
4371         jvmHandler = CAST_FROM_FN_PTR(address, sigINTRHandler);
4372       } else if (sig == asynsig) {
4373         jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4374       } else {
4375         return;
4376       }
4377       break;
4378   }
4379 
4380 
4381   if (thisHandler != jvmHandler) {
4382     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4383     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4384     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4385     // No need to check this sig any longer
4386     sigaddset(&check_signal_done, sig);
4387   } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
4388     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4389     tty->print("expected:" PTR32_FORMAT, os::Solaris::get_our_sigflags(sig));
4390     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4391     // No need to check this sig any longer
4392     sigaddset(&check_signal_done, sig);
4393   }
4394 
4395   // Print all the signal handler state
4396   if (sigismember(&check_signal_done, sig)) {
4397     print_signal_handlers(tty, buf, O_BUFLEN);
4398   }
4399 
4400 }
4401 
4402 void os::Solaris::install_signal_handlers() {
4403   bool libjsigdone = false;
4404   signal_handlers_are_installed = true;
4405 
4406   // signal-chaining
4407   typedef void (*signal_setting_t)();
4408   signal_setting_t begin_signal_setting = NULL;
4409   signal_setting_t end_signal_setting = NULL;
4410   begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4411                                         dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4412   if (begin_signal_setting != NULL) {
4413     end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4414                                         dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4415     get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4416                                        dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4417     get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
4418                                          dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
4419     libjsig_is_loaded = true;
4420     if (os::Solaris::get_libjsig_version != NULL) {
4421       libjsigversion =  (*os::Solaris::get_libjsig_version)();
4422     }
4423     assert(UseSignalChaining, "should enable signal-chaining");
4424   }
4425   if (libjsig_is_loaded) {
4426     // Tell libjsig jvm is setting signal handlers
4427     (*begin_signal_setting)();
4428   }
4429 
4430   set_signal_handler(SIGSEGV, true, true);
4431   set_signal_handler(SIGPIPE, true, true);
4432   set_signal_handler(SIGXFSZ, true, true);
4433   set_signal_handler(SIGBUS, true, true);
4434   set_signal_handler(SIGILL, true, true);
4435   set_signal_handler(SIGFPE, true, true);
4436 
4437 
4438   if (os::Solaris::SIGinterrupt() > OLDMAXSIGNUM || os::Solaris::SIGasync() > OLDMAXSIGNUM) {
4439 
4440     // Pre-1.4.1 Libjsig limited to signal chaining signals <= 32 so
4441     // can not register overridable signals which might be > 32
4442     if (libjsig_is_loaded && libjsigversion <= JSIG_VERSION_1_4_1) {
4443     // Tell libjsig jvm has finished setting signal handlers
4444       (*end_signal_setting)();
4445       libjsigdone = true;
4446     }
4447   }
4448 
4449   // Never ok to chain our SIGinterrupt
4450   set_signal_handler(os::Solaris::SIGinterrupt(), true, false);
4451   set_signal_handler(os::Solaris::SIGasync(), true, true);
4452 
4453   if (libjsig_is_loaded && !libjsigdone) {
4454     // Tell libjsig jvm finishes setting signal handlers
4455     (*end_signal_setting)();
4456   }
4457 
4458   // We don't activate signal checker if libjsig is in place, we trust ourselves
4459   // and if UserSignalHandler is installed all bets are off
4460   if (CheckJNICalls) {
4461     if (libjsig_is_loaded) {
4462       tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4463       check_signals = false;
4464     }
4465     if (AllowUserSignalHandlers) {
4466       tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4467       check_signals = false;
4468     }
4469   }
4470 }
4471 
4472 
4473 void report_error(const char* file_name, int line_no, const char* title, const char* format, ...);
4474 
4475 const char * signames[] = {
4476   "SIG0",
4477   "SIGHUP", "SIGINT", "SIGQUIT", "SIGILL", "SIGTRAP",
4478   "SIGABRT", "SIGEMT", "SIGFPE", "SIGKILL", "SIGBUS",
4479   "SIGSEGV", "SIGSYS", "SIGPIPE", "SIGALRM", "SIGTERM",
4480   "SIGUSR1", "SIGUSR2", "SIGCLD", "SIGPWR", "SIGWINCH",
4481   "SIGURG", "SIGPOLL", "SIGSTOP", "SIGTSTP", "SIGCONT",
4482   "SIGTTIN", "SIGTTOU", "SIGVTALRM", "SIGPROF", "SIGXCPU",
4483   "SIGXFSZ", "SIGWAITING", "SIGLWP", "SIGFREEZE", "SIGTHAW",
4484   "SIGCANCEL", "SIGLOST"
4485 };
4486 
4487 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4488   if (0 < exception_code && exception_code <= SIGRTMAX) {
4489     // signal
4490     if (exception_code < sizeof(signames)/sizeof(const char*)) {
4491        jio_snprintf(buf, size, "%s", signames[exception_code]);
4492     } else {
4493        jio_snprintf(buf, size, "SIG%d", exception_code);
4494     }
4495     return buf;
4496   } else {
4497     return NULL;
4498   }
4499 }
4500 
4501 // (Static) wrappers for the new libthread API
4502 int_fnP_thread_t_iP_uP_stack_tP_gregset_t os::Solaris::_thr_getstate;
4503 int_fnP_thread_t_i_gregset_t os::Solaris::_thr_setstate;
4504 int_fnP_thread_t_i os::Solaris::_thr_setmutator;
4505 int_fnP_thread_t os::Solaris::_thr_suspend_mutator;
4506 int_fnP_thread_t os::Solaris::_thr_continue_mutator;
4507 
4508 // (Static) wrapper for getisax(2) call.
4509 os::Solaris::getisax_func_t os::Solaris::_getisax = 0;
4510 
4511 // (Static) wrappers for the liblgrp API
4512 os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4513 os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4514 os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4515 os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4516 os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4517 os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4518 os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4519 os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4520 os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4521 
4522 // (Static) wrapper for meminfo() call.
4523 os::Solaris::meminfo_func_t os::Solaris::_meminfo = 0;
4524 
4525 static address resolve_symbol_lazy(const char* name) {
4526   address addr = (address) dlsym(RTLD_DEFAULT, name);
4527   if(addr == NULL) {
4528     // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4529     addr = (address) dlsym(RTLD_NEXT, name);
4530   }
4531   return addr;
4532 }
4533 
4534 static address resolve_symbol(const char* name) {
4535   address addr = resolve_symbol_lazy(name);
4536   if(addr == NULL) {
4537     fatal(dlerror());
4538   }
4539   return addr;
4540 }
4541 
4542 
4543 
4544 // isT2_libthread()
4545 //
4546 // Routine to determine if we are currently using the new T2 libthread.
4547 //
4548 // We determine if we are using T2 by reading /proc/self/lstatus and
4549 // looking for a thread with the ASLWP bit set.  If we find this status
4550 // bit set, we must assume that we are NOT using T2.  The T2 team
4551 // has approved this algorithm.
4552 //
4553 // We need to determine if we are running with the new T2 libthread
4554 // since setting native thread priorities is handled differently
4555 // when using this library.  All threads created using T2 are bound
4556 // threads. Calling thr_setprio is meaningless in this case.
4557 //
4558 bool isT2_libthread() {
4559   static prheader_t * lwpArray = NULL;
4560   static int lwpSize = 0;
4561   static int lwpFile = -1;
4562   lwpstatus_t * that;
4563   char lwpName [128];
4564   bool isT2 = false;
4565 
4566 #define ADR(x)  ((uintptr_t)(x))
4567 #define LWPINDEX(ary,ix)   ((lwpstatus_t *)(((ary)->pr_entsize * (ix)) + (ADR((ary) + 1))))
4568 
4569   lwpFile = open("/proc/self/lstatus", O_RDONLY, 0);
4570   if (lwpFile < 0) {
4571       if (ThreadPriorityVerbose) warning ("Couldn't open /proc/self/lstatus\n");
4572       return false;
4573   }
4574   lwpSize = 16*1024;
4575   for (;;) {
4576     lseek (lwpFile, 0, SEEK_SET);
4577     lwpArray = (prheader_t *)NEW_C_HEAP_ARRAY(char, lwpSize);
4578     if (read(lwpFile, lwpArray, lwpSize) < 0) {
4579       if (ThreadPriorityVerbose) warning("Error reading /proc/self/lstatus\n");
4580       break;
4581     }
4582     if ((lwpArray->pr_nent * lwpArray->pr_entsize) <= lwpSize) {
4583        // We got a good snapshot - now iterate over the list.
4584       int aslwpcount = 0;
4585       for (int i = 0; i < lwpArray->pr_nent; i++ ) {
4586         that = LWPINDEX(lwpArray,i);
4587         if (that->pr_flags & PR_ASLWP) {
4588           aslwpcount++;
4589         }
4590       }
4591       if (aslwpcount == 0) isT2 = true;
4592       break;
4593     }
4594     lwpSize = lwpArray->pr_nent * lwpArray->pr_entsize;
4595     FREE_C_HEAP_ARRAY(char, lwpArray);  // retry.
4596   }
4597 
4598   FREE_C_HEAP_ARRAY(char, lwpArray);
4599   close (lwpFile);
4600   if (ThreadPriorityVerbose) {
4601     if (isT2) tty->print_cr("We are running with a T2 libthread\n");
4602     else tty->print_cr("We are not running with a T2 libthread\n");
4603   }
4604   return isT2;
4605 }
4606 
4607 
4608 void os::Solaris::libthread_init() {
4609   address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4610 
4611   // Determine if we are running with the new T2 libthread
4612   os::Solaris::set_T2_libthread(isT2_libthread());
4613 
4614   lwp_priocntl_init();
4615 
4616   // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4617   if(func == NULL) {
4618     func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4619     // Guarantee that this VM is running on an new enough OS (5.6 or
4620     // later) that it will have a new enough libthread.so.
4621     guarantee(func != NULL, "libthread.so is too old.");
4622   }
4623 
4624   // Initialize the new libthread getstate API wrappers
4625   func = resolve_symbol("thr_getstate");
4626   os::Solaris::set_thr_getstate(CAST_TO_FN_PTR(int_fnP_thread_t_iP_uP_stack_tP_gregset_t, func));
4627 
4628   func = resolve_symbol("thr_setstate");
4629   os::Solaris::set_thr_setstate(CAST_TO_FN_PTR(int_fnP_thread_t_i_gregset_t, func));
4630 
4631   func = resolve_symbol("thr_setmutator");
4632   os::Solaris::set_thr_setmutator(CAST_TO_FN_PTR(int_fnP_thread_t_i, func));
4633 
4634   func = resolve_symbol("thr_suspend_mutator");
4635   os::Solaris::set_thr_suspend_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4636 
4637   func = resolve_symbol("thr_continue_mutator");
4638   os::Solaris::set_thr_continue_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4639 
4640   int size;
4641   void (*handler_info_func)(address *, int *);
4642   handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4643   handler_info_func(&handler_start, &size);
4644   handler_end = handler_start + size;
4645 }
4646 
4647 
4648 int_fnP_mutex_tP os::Solaris::_mutex_lock;
4649 int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4650 int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4651 int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4652 int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4653 int os::Solaris::_mutex_scope = USYNC_THREAD;
4654 
4655 int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4656 int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4657 int_fnP_cond_tP os::Solaris::_cond_signal;
4658 int_fnP_cond_tP os::Solaris::_cond_broadcast;
4659 int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4660 int_fnP_cond_tP os::Solaris::_cond_destroy;
4661 int os::Solaris::_cond_scope = USYNC_THREAD;
4662 
4663 void os::Solaris::synchronization_init() {
4664   if(UseLWPSynchronization) {
4665     os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4666     os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4667     os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4668     os::Solaris::set_mutex_init(lwp_mutex_init);
4669     os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4670     os::Solaris::set_mutex_scope(USYNC_THREAD);
4671 
4672     os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4673     os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4674     os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4675     os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4676     os::Solaris::set_cond_init(lwp_cond_init);
4677     os::Solaris::set_cond_destroy(lwp_cond_destroy);
4678     os::Solaris::set_cond_scope(USYNC_THREAD);
4679   }
4680   else {
4681     os::Solaris::set_mutex_scope(USYNC_THREAD);
4682     os::Solaris::set_cond_scope(USYNC_THREAD);
4683 
4684     if(UsePthreads) {
4685       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4686       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4687       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4688       os::Solaris::set_mutex_init(pthread_mutex_default_init);
4689       os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4690 
4691       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4692       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4693       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4694       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4695       os::Solaris::set_cond_init(pthread_cond_default_init);
4696       os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4697     }
4698     else {
4699       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4700       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4701       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4702       os::Solaris::set_mutex_init(::mutex_init);
4703       os::Solaris::set_mutex_destroy(::mutex_destroy);
4704 
4705       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4706       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4707       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4708       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4709       os::Solaris::set_cond_init(::cond_init);
4710       os::Solaris::set_cond_destroy(::cond_destroy);
4711     }
4712   }
4713 }
4714 
4715 bool os::Solaris::liblgrp_init() {
4716   void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4717   if (handle != NULL) {
4718     os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4719     os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4720     os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4721     os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4722     os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4723     os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4724     os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4725     os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4726                                        dlsym(handle, "lgrp_cookie_stale")));
4727 
4728     lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4729     set_lgrp_cookie(c);
4730     return true;
4731   }
4732   return false;
4733 }
4734 
4735 void os::Solaris::misc_sym_init() {
4736   address func;
4737 
4738   // getisax
4739   func = resolve_symbol_lazy("getisax");
4740   if (func != NULL) {
4741     os::Solaris::_getisax = CAST_TO_FN_PTR(getisax_func_t, func);
4742   }
4743 
4744   // meminfo
4745   func = resolve_symbol_lazy("meminfo");
4746   if (func != NULL) {
4747     os::Solaris::set_meminfo(CAST_TO_FN_PTR(meminfo_func_t, func));
4748   }
4749 }
4750 
4751 uint_t os::Solaris::getisax(uint32_t* array, uint_t n) {
4752   assert(_getisax != NULL, "_getisax not set");
4753   return _getisax(array, n);
4754 }
4755 
4756 // Symbol doesn't exist in Solaris 8 pset.h
4757 #ifndef PS_MYID
4758 #define PS_MYID -3
4759 #endif
4760 
4761 // int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
4762 typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
4763 static pset_getloadavg_type pset_getloadavg_ptr = NULL;
4764 
4765 void init_pset_getloadavg_ptr(void) {
4766   pset_getloadavg_ptr =
4767     (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
4768   if (PrintMiscellaneous && Verbose && pset_getloadavg_ptr == NULL) {
4769     warning("pset_getloadavg function not found");
4770   }
4771 }
4772 
4773 int os::Solaris::_dev_zero_fd = -1;
4774 
4775 // this is called _before_ the global arguments have been parsed
4776 void os::init(void) {
4777   _initial_pid = getpid();
4778 
4779   max_hrtime = first_hrtime = gethrtime();
4780 
4781   init_random(1234567);
4782 
4783   page_size = sysconf(_SC_PAGESIZE);
4784   if (page_size == -1)
4785     fatal(err_msg("os_solaris.cpp: os::init: sysconf failed (%s)",
4786                   strerror(errno)));
4787   init_page_sizes((size_t) page_size);
4788 
4789   Solaris::initialize_system_info();
4790 
4791   // Initialize misc. symbols as soon as possible, so we can use them
4792   // if we need them.
4793   Solaris::misc_sym_init();
4794 
4795   int fd = open("/dev/zero", O_RDWR);
4796   if (fd < 0) {
4797     fatal(err_msg("os::init: cannot open /dev/zero (%s)", strerror(errno)));
4798   } else {
4799     Solaris::set_dev_zero_fd(fd);
4800 
4801     // Close on exec, child won't inherit.
4802     fcntl(fd, F_SETFD, FD_CLOEXEC);
4803   }
4804 
4805   clock_tics_per_sec = CLK_TCK;
4806 
4807   // check if dladdr1() exists; dladdr1 can provide more information than
4808   // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
4809   // and is available on linker patches for 5.7 and 5.8.
4810   // libdl.so must have been loaded, this call is just an entry lookup
4811   void * hdl = dlopen("libdl.so", RTLD_NOW);
4812   if (hdl)
4813     dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
4814 
4815   // (Solaris only) this switches to calls that actually do locking.
4816   ThreadCritical::initialize();
4817 
4818   main_thread = thr_self();
4819 
4820   // Constant minimum stack size allowed. It must be at least
4821   // the minimum of what the OS supports (thr_min_stack()), and
4822   // enough to allow the thread to get to user bytecode execution.
4823   Solaris::min_stack_allowed = MAX2(thr_min_stack(), Solaris::min_stack_allowed);
4824   // If the pagesize of the VM is greater than 8K determine the appropriate
4825   // number of initial guard pages.  The user can change this with the
4826   // command line arguments, if needed.
4827   if (vm_page_size() > 8*K) {
4828     StackYellowPages = 1;
4829     StackRedPages = 1;
4830     StackShadowPages = round_to((StackShadowPages*8*K), vm_page_size()) / vm_page_size();
4831   }
4832 }
4833 
4834 // To install functions for atexit system call
4835 extern "C" {
4836   static void perfMemory_exit_helper() {
4837     perfMemory_exit();
4838   }
4839 }
4840 
4841 // this is called _after_ the global arguments have been parsed
4842 jint os::init_2(void) {
4843   // try to enable extended file IO ASAP, see 6431278
4844   os::Solaris::try_enable_extended_io();
4845 
4846   // Allocate a single page and mark it as readable for safepoint polling.  Also
4847   // use this first mmap call to check support for MAP_ALIGN.
4848   address polling_page = (address)Solaris::mmap_chunk((char*)page_size,
4849                                                       page_size,
4850                                                       MAP_PRIVATE | MAP_ALIGN,
4851                                                       PROT_READ);
4852   if (polling_page == NULL) {
4853     has_map_align = false;
4854     polling_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE,
4855                                                 PROT_READ);
4856   }
4857 
4858   os::set_polling_page(polling_page);
4859 
4860 #ifndef PRODUCT
4861   if( Verbose && PrintMiscellaneous )
4862     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4863 #endif
4864 
4865   if (!UseMembar) {
4866     address mem_serialize_page = (address)Solaris::mmap_chunk( NULL, page_size, MAP_PRIVATE, PROT_READ | PROT_WRITE );
4867     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4868     os::set_memory_serialize_page( mem_serialize_page );
4869 
4870 #ifndef PRODUCT
4871     if(Verbose && PrintMiscellaneous)
4872       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4873 #endif
4874 }
4875 
4876   FLAG_SET_DEFAULT(UseLargePages, os::large_page_init());
4877 
4878   // Check minimum allowable stack size for thread creation and to initialize
4879   // the java system classes, including StackOverflowError - depends on page
4880   // size.  Add a page for compiler2 recursion in main thread.
4881   // Add in BytesPerWord times page size to account for VM stack during
4882   // class initialization depending on 32 or 64 bit VM.
4883   guarantee((Solaris::min_stack_allowed >=
4884     (StackYellowPages+StackRedPages+StackShadowPages+BytesPerWord
4885      COMPILER2_PRESENT(+1)) * page_size),
4886     "need to increase Solaris::min_stack_allowed on this platform");
4887 
4888   size_t threadStackSizeInBytes = ThreadStackSize * K;
4889   if (threadStackSizeInBytes != 0 &&
4890     threadStackSizeInBytes < Solaris::min_stack_allowed) {
4891     tty->print_cr("\nThe stack size specified is too small, Specify at least %dk",
4892                   Solaris::min_stack_allowed/K);
4893     return JNI_ERR;
4894   }
4895 
4896   // For 64kbps there will be a 64kb page size, which makes
4897   // the usable default stack size quite a bit less.  Increase the
4898   // stack for 64kb (or any > than 8kb) pages, this increases
4899   // virtual memory fragmentation (since we're not creating the
4900   // stack on a power of 2 boundary.  The real fix for this
4901   // should be to fix the guard page mechanism.
4902 
4903   if (vm_page_size() > 8*K) {
4904       threadStackSizeInBytes = (threadStackSizeInBytes != 0)
4905          ? threadStackSizeInBytes +
4906            ((StackYellowPages + StackRedPages) * vm_page_size())
4907          : 0;
4908       ThreadStackSize = threadStackSizeInBytes/K;
4909   }
4910 
4911   // Make the stack size a multiple of the page size so that
4912   // the yellow/red zones can be guarded.
4913   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4914         vm_page_size()));
4915 
4916   Solaris::libthread_init();
4917 
4918   if (UseNUMA) {
4919     if (!Solaris::liblgrp_init()) {
4920       UseNUMA = false;
4921     } else {
4922       size_t lgrp_limit = os::numa_get_groups_num();
4923       int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit);
4924       size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
4925       FREE_C_HEAP_ARRAY(int, lgrp_ids);
4926       if (lgrp_num < 2) {
4927         // There's only one locality group, disable NUMA.
4928         UseNUMA = false;
4929       }
4930     }
4931     if (!UseNUMA && ForceNUMA) {
4932       UseNUMA = true;
4933     }
4934   }
4935 
4936   Solaris::signal_sets_init();
4937   Solaris::init_signal_mem();
4938   Solaris::install_signal_handlers();
4939 
4940   if (libjsigversion < JSIG_VERSION_1_4_1) {
4941     Maxlibjsigsigs = OLDMAXSIGNUM;
4942   }
4943 
4944   // initialize synchronization primitives to use either thread or
4945   // lwp synchronization (controlled by UseLWPSynchronization)
4946   Solaris::synchronization_init();
4947 
4948   if (MaxFDLimit) {
4949     // set the number of file descriptors to max. print out error
4950     // if getrlimit/setrlimit fails but continue regardless.
4951     struct rlimit nbr_files;
4952     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4953     if (status != 0) {
4954       if (PrintMiscellaneous && (Verbose || WizardMode))
4955         perror("os::init_2 getrlimit failed");
4956     } else {
4957       nbr_files.rlim_cur = nbr_files.rlim_max;
4958       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4959       if (status != 0) {
4960         if (PrintMiscellaneous && (Verbose || WizardMode))
4961           perror("os::init_2 setrlimit failed");
4962       }
4963     }
4964   }
4965 
4966   // Initialize HPI.
4967   jint hpi_result = hpi::initialize();
4968   if (hpi_result != JNI_OK) {
4969     tty->print_cr("There was an error trying to initialize the HPI library.");
4970     return hpi_result;
4971   }
4972 
4973   // Calculate theoretical max. size of Threads to guard gainst
4974   // artifical out-of-memory situations, where all available address-
4975   // space has been reserved by thread stacks. Default stack size is 1Mb.
4976   size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
4977     JavaThread::stack_size_at_create() : (1*K*K);
4978   assert(pre_thread_stack_size != 0, "Must have a stack");
4979   // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
4980   // we should start doing Virtual Memory banging. Currently when the threads will
4981   // have used all but 200Mb of space.
4982   size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
4983   Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
4984 
4985   // at-exit methods are called in the reverse order of their registration.
4986   // In Solaris 7 and earlier, atexit functions are called on return from
4987   // main or as a result of a call to exit(3C). There can be only 32 of
4988   // these functions registered and atexit() does not set errno. In Solaris
4989   // 8 and later, there is no limit to the number of functions registered
4990   // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
4991   // functions are called upon dlclose(3DL) in addition to return from main
4992   // and exit(3C).
4993 
4994   if (PerfAllowAtExitRegistration) {
4995     // only register atexit functions if PerfAllowAtExitRegistration is set.
4996     // atexit functions can be delayed until process exit time, which
4997     // can be problematic for embedded VM situations. Embedded VMs should
4998     // call DestroyJavaVM() to assure that VM resources are released.
4999 
5000     // note: perfMemory_exit_helper atexit function may be removed in
5001     // the future if the appropriate cleanup code can be added to the
5002     // VM_Exit VMOperation's doit method.
5003     if (atexit(perfMemory_exit_helper) != 0) {
5004       warning("os::init2 atexit(perfMemory_exit_helper) failed");
5005     }
5006   }
5007 
5008   // Init pset_loadavg function pointer
5009   init_pset_getloadavg_ptr();
5010 
5011   return JNI_OK;
5012 }
5013 
5014 void os::init_3(void) {
5015   return;
5016 }
5017 
5018 // Mark the polling page as unreadable
5019 void os::make_polling_page_unreadable(void) {
5020   if( mprotect((char *)_polling_page, page_size, PROT_NONE) != 0 )
5021     fatal("Could not disable polling page");
5022 };
5023 
5024 // Mark the polling page as readable
5025 void os::make_polling_page_readable(void) {
5026   if( mprotect((char *)_polling_page, page_size, PROT_READ) != 0 )
5027     fatal("Could not enable polling page");
5028 };
5029 
5030 // OS interface.
5031 
5032 int os::stat(const char *path, struct stat *sbuf) {
5033   char pathbuf[MAX_PATH];
5034   if (strlen(path) > MAX_PATH - 1) {
5035     errno = ENAMETOOLONG;
5036     return -1;
5037   }
5038   hpi::native_path(strcpy(pathbuf, path));
5039   return ::stat(pathbuf, sbuf);
5040 }
5041 
5042 
5043 bool os::check_heap(bool force) { return true; }
5044 
5045 typedef int (*vsnprintf_t)(char* buf, size_t count, const char* fmt, va_list argptr);
5046 static vsnprintf_t sol_vsnprintf = NULL;
5047 
5048 int local_vsnprintf(char* buf, size_t count, const char* fmt, va_list argptr) {
5049   if (!sol_vsnprintf) {
5050     //search  for the named symbol in the objects that were loaded after libjvm
5051     void* where = RTLD_NEXT;
5052     if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5053         sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5054     if (!sol_vsnprintf){
5055       //search  for the named symbol in the objects that were loaded before libjvm
5056       where = RTLD_DEFAULT;
5057       if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5058         sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5059       assert(sol_vsnprintf != NULL, "vsnprintf not found");
5060     }
5061   }
5062   return (*sol_vsnprintf)(buf, count, fmt, argptr);
5063 }
5064 
5065 
5066 // Is a (classpath) directory empty?
5067 bool os::dir_is_empty(const char* path) {
5068   DIR *dir = NULL;
5069   struct dirent *ptr;
5070 
5071   dir = opendir(path);
5072   if (dir == NULL) return true;
5073 
5074   /* Scan the directory */
5075   bool result = true;
5076   char buf[sizeof(struct dirent) + MAX_PATH];
5077   struct dirent *dbuf = (struct dirent *) buf;
5078   while (result && (ptr = readdir(dir, dbuf)) != NULL) {
5079     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5080       result = false;
5081     }
5082   }
5083   closedir(dir);
5084   return result;
5085 }
5086 
5087 // create binary file, rewriting existing file if required
5088 int os::create_binary_file(const char* path, bool rewrite_existing) {
5089   int oflags = O_WRONLY | O_CREAT;
5090   if (!rewrite_existing) {
5091     oflags |= O_EXCL;
5092   }
5093   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5094 }
5095 
5096 // return current position of file pointer
5097 jlong os::current_file_offset(int fd) {
5098   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5099 }
5100 
5101 // move file pointer to the specified offset
5102 jlong os::seek_to_file_offset(int fd, jlong offset) {
5103   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5104 }
5105 
5106 // Map a block of memory.
5107 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
5108                      char *addr, size_t bytes, bool read_only,
5109                      bool allow_exec) {
5110   int prot;
5111   int flags;
5112 
5113   if (read_only) {
5114     prot = PROT_READ;
5115     flags = MAP_SHARED;
5116   } else {
5117     prot = PROT_READ | PROT_WRITE;
5118     flags = MAP_PRIVATE;
5119   }
5120 
5121   if (allow_exec) {
5122     prot |= PROT_EXEC;
5123   }
5124 
5125   if (addr != NULL) {
5126     flags |= MAP_FIXED;
5127   }
5128 
5129   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5130                                      fd, file_offset);
5131   if (mapped_address == MAP_FAILED) {
5132     return NULL;
5133   }
5134   return mapped_address;
5135 }
5136 
5137 
5138 // Remap a block of memory.
5139 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
5140                        char *addr, size_t bytes, bool read_only,
5141                        bool allow_exec) {
5142   // same as map_memory() on this OS
5143   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5144                         allow_exec);
5145 }
5146 
5147 
5148 // Unmap a block of memory.
5149 bool os::unmap_memory(char* addr, size_t bytes) {
5150   return munmap(addr, bytes) == 0;
5151 }
5152 
5153 void os::pause() {
5154   char filename[MAX_PATH];
5155   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5156     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5157   } else {
5158     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5159   }
5160 
5161   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5162   if (fd != -1) {
5163     struct stat buf;
5164     close(fd);
5165     while (::stat(filename, &buf) == 0) {
5166       (void)::poll(NULL, 0, 100);
5167     }
5168   } else {
5169     jio_fprintf(stderr,
5170       "Could not open pause file '%s', continuing immediately.\n", filename);
5171   }
5172 }
5173 
5174 #ifndef PRODUCT
5175 #ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5176 // Turn this on if you need to trace synch operations.
5177 // Set RECORD_SYNCH_LIMIT to a large-enough value,
5178 // and call record_synch_enable and record_synch_disable
5179 // around the computation of interest.
5180 
5181 void record_synch(char* name, bool returning);  // defined below
5182 
5183 class RecordSynch {
5184   char* _name;
5185  public:
5186   RecordSynch(char* name) :_name(name)
5187                  { record_synch(_name, false); }
5188   ~RecordSynch() { record_synch(_name,   true);  }
5189 };
5190 
5191 #define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
5192 extern "C" ret name params {                                    \
5193   typedef ret name##_t params;                                  \
5194   static name##_t* implem = NULL;                               \
5195   static int callcount = 0;                                     \
5196   if (implem == NULL) {                                         \
5197     implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
5198     if (implem == NULL)  fatal(dlerror());                      \
5199   }                                                             \
5200   ++callcount;                                                  \
5201   RecordSynch _rs(#name);                                       \
5202   inner;                                                        \
5203   return implem args;                                           \
5204 }
5205 // in dbx, examine callcounts this way:
5206 // for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
5207 
5208 #define CHECK_POINTER_OK(p) \
5209   (Universe::perm_gen() == NULL || !Universe::is_reserved_heap((oop)(p)))
5210 #define CHECK_MU \
5211   if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
5212 #define CHECK_CV \
5213   if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
5214 #define CHECK_P(p) \
5215   if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
5216 
5217 #define CHECK_MUTEX(mutex_op) \
5218 CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
5219 
5220 CHECK_MUTEX(   mutex_lock)
5221 CHECK_MUTEX(  _mutex_lock)
5222 CHECK_MUTEX( mutex_unlock)
5223 CHECK_MUTEX(_mutex_unlock)
5224 CHECK_MUTEX( mutex_trylock)
5225 CHECK_MUTEX(_mutex_trylock)
5226 
5227 #define CHECK_COND(cond_op) \
5228 CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU;CHECK_CV);
5229 
5230 CHECK_COND( cond_wait);
5231 CHECK_COND(_cond_wait);
5232 CHECK_COND(_cond_wait_cancel);
5233 
5234 #define CHECK_COND2(cond_op) \
5235 CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU;CHECK_CV);
5236 
5237 CHECK_COND2( cond_timedwait);
5238 CHECK_COND2(_cond_timedwait);
5239 CHECK_COND2(_cond_timedwait_cancel);
5240 
5241 // do the _lwp_* versions too
5242 #define mutex_t lwp_mutex_t
5243 #define cond_t  lwp_cond_t
5244 CHECK_MUTEX(  _lwp_mutex_lock)
5245 CHECK_MUTEX(  _lwp_mutex_unlock)
5246 CHECK_MUTEX(  _lwp_mutex_trylock)
5247 CHECK_MUTEX( __lwp_mutex_lock)
5248 CHECK_MUTEX( __lwp_mutex_unlock)
5249 CHECK_MUTEX( __lwp_mutex_trylock)
5250 CHECK_MUTEX(___lwp_mutex_lock)
5251 CHECK_MUTEX(___lwp_mutex_unlock)
5252 
5253 CHECK_COND(  _lwp_cond_wait);
5254 CHECK_COND( __lwp_cond_wait);
5255 CHECK_COND(___lwp_cond_wait);
5256 
5257 CHECK_COND2(  _lwp_cond_timedwait);
5258 CHECK_COND2( __lwp_cond_timedwait);
5259 #undef mutex_t
5260 #undef cond_t
5261 
5262 CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5263 CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5264 CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5265 CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5266 CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5267 CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5268 CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5269 CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5270 
5271 
5272 // recording machinery:
5273 
5274 enum { RECORD_SYNCH_LIMIT = 200 };
5275 char* record_synch_name[RECORD_SYNCH_LIMIT];
5276 void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
5277 bool record_synch_returning[RECORD_SYNCH_LIMIT];
5278 thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
5279 int record_synch_count = 0;
5280 bool record_synch_enabled = false;
5281 
5282 // in dbx, examine recorded data this way:
5283 // for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
5284 
5285 void record_synch(char* name, bool returning) {
5286   if (record_synch_enabled) {
5287     if (record_synch_count < RECORD_SYNCH_LIMIT) {
5288       record_synch_name[record_synch_count] = name;
5289       record_synch_returning[record_synch_count] = returning;
5290       record_synch_thread[record_synch_count] = thr_self();
5291       record_synch_arg0ptr[record_synch_count] = &name;
5292       record_synch_count++;
5293     }
5294     // put more checking code here:
5295     // ...
5296   }
5297 }
5298 
5299 void record_synch_enable() {
5300   // start collecting trace data, if not already doing so
5301   if (!record_synch_enabled)  record_synch_count = 0;
5302   record_synch_enabled = true;
5303 }
5304 
5305 void record_synch_disable() {
5306   // stop collecting trace data
5307   record_synch_enabled = false;
5308 }
5309 
5310 #endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5311 #endif // PRODUCT
5312 
5313 const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5314 const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
5315                                (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5316 
5317 
5318 // JVMTI & JVM monitoring and management support
5319 // The thread_cpu_time() and current_thread_cpu_time() are only
5320 // supported if is_thread_cpu_time_supported() returns true.
5321 // They are not supported on Solaris T1.
5322 
5323 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5324 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5325 // of a thread.
5326 //
5327 // current_thread_cpu_time() and thread_cpu_time(Thread *)
5328 // returns the fast estimate available on the platform.
5329 
5330 // hrtime_t gethrvtime() return value includes
5331 // user time but does not include system time
5332 jlong os::current_thread_cpu_time() {
5333   return (jlong) gethrvtime();
5334 }
5335 
5336 jlong os::thread_cpu_time(Thread *thread) {
5337   // return user level CPU time only to be consistent with
5338   // what current_thread_cpu_time returns.
5339   // thread_cpu_time_info() must be changed if this changes
5340   return os::thread_cpu_time(thread, false /* user time only */);
5341 }
5342 
5343 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5344   if (user_sys_cpu_time) {
5345     return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5346   } else {
5347     return os::current_thread_cpu_time();
5348   }
5349 }
5350 
5351 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5352   char proc_name[64];
5353   int count;
5354   prusage_t prusage;
5355   jlong lwp_time;
5356   int fd;
5357 
5358   sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
5359                      getpid(),
5360                      thread->osthread()->lwp_id());
5361   fd = open(proc_name, O_RDONLY);
5362   if ( fd == -1 ) return -1;
5363 
5364   do {
5365     count = pread(fd,
5366                   (void *)&prusage.pr_utime,
5367                   thr_time_size,
5368                   thr_time_off);
5369   } while (count < 0 && errno == EINTR);
5370   close(fd);
5371   if ( count < 0 ) return -1;
5372 
5373   if (user_sys_cpu_time) {
5374     // user + system CPU time
5375     lwp_time = (((jlong)prusage.pr_stime.tv_sec +
5376                  (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
5377                  (jlong)prusage.pr_stime.tv_nsec +
5378                  (jlong)prusage.pr_utime.tv_nsec;
5379   } else {
5380     // user level CPU time only
5381     lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
5382                 (jlong)prusage.pr_utime.tv_nsec;
5383   }
5384 
5385   return(lwp_time);
5386 }
5387 
5388 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5389   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5390   info_ptr->may_skip_backward = false;    // elapsed time not wall time
5391   info_ptr->may_skip_forward = false;     // elapsed time not wall time
5392   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5393 }
5394 
5395 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5396   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5397   info_ptr->may_skip_backward = false;    // elapsed time not wall time
5398   info_ptr->may_skip_forward = false;     // elapsed time not wall time
5399   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5400 }
5401 
5402 bool os::is_thread_cpu_time_supported() {
5403   if ( os::Solaris::T2_libthread() || UseBoundThreads ) {
5404     return true;
5405   } else {
5406     return false;
5407   }
5408 }
5409 
5410 // System loadavg support.  Returns -1 if load average cannot be obtained.
5411 // Return the load average for our processor set if the primitive exists
5412 // (Solaris 9 and later).  Otherwise just return system wide loadavg.
5413 int os::loadavg(double loadavg[], int nelem) {
5414   if (pset_getloadavg_ptr != NULL) {
5415     return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
5416   } else {
5417     return ::getloadavg(loadavg, nelem);
5418   }
5419 }
5420 
5421 //---------------------------------------------------------------------------------
5422 
5423 static address same_page(address x, address y) {
5424   intptr_t page_bits = -os::vm_page_size();
5425   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
5426     return x;
5427   else if (x > y)
5428     return (address)(intptr_t(y) | ~page_bits) + 1;
5429   else
5430     return (address)(intptr_t(y) & page_bits);
5431 }
5432 
5433 bool os::find(address addr, outputStream* st) {
5434   Dl_info dlinfo;
5435   memset(&dlinfo, 0, sizeof(dlinfo));
5436   if (dladdr(addr, &dlinfo)) {
5437 #ifdef _LP64
5438     st->print("0x%016lx: ", addr);
5439 #else
5440     st->print("0x%08x: ", addr);
5441 #endif
5442     if (dlinfo.dli_sname != NULL)
5443       st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
5444     else if (dlinfo.dli_fname)
5445       st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
5446     else
5447       st->print("<absolute address>");
5448     if (dlinfo.dli_fname)  st->print(" in %s", dlinfo.dli_fname);
5449 #ifdef _LP64
5450     if (dlinfo.dli_fbase)  st->print(" at 0x%016lx", dlinfo.dli_fbase);
5451 #else
5452     if (dlinfo.dli_fbase)  st->print(" at 0x%08x", dlinfo.dli_fbase);
5453 #endif
5454     st->cr();
5455 
5456     if (Verbose) {
5457       // decode some bytes around the PC
5458       address begin = same_page(addr-40, addr);
5459       address end   = same_page(addr+40, addr);
5460       address       lowest = (address) dlinfo.dli_sname;
5461       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5462       if (begin < lowest)  begin = lowest;
5463       Dl_info dlinfo2;
5464       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
5465           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5466         end = (address) dlinfo2.dli_saddr;
5467       Disassembler::decode(begin, end, st);
5468     }
5469     return true;
5470   }
5471   return false;
5472 }
5473 
5474 // Following function has been added to support HotSparc's libjvm.so running
5475 // under Solaris production JDK 1.2.2 / 1.3.0.  These came from
5476 // src/solaris/hpi/native_threads in the EVM codebase.
5477 //
5478 // NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
5479 // libraries and should thus be removed. We will leave it behind for a while
5480 // until we no longer want to able to run on top of 1.3.0 Solaris production
5481 // JDK. See 4341971.
5482 
5483 #define STACK_SLACK 0x800
5484 
5485 extern "C" {
5486   intptr_t sysThreadAvailableStackWithSlack() {
5487     stack_t st;
5488     intptr_t retval, stack_top;
5489     retval = thr_stksegment(&st);
5490     assert(retval == 0, "incorrect return value from thr_stksegment");
5491     assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
5492     assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
5493     stack_top=(intptr_t)st.ss_sp-st.ss_size;
5494     return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
5495   }
5496 }
5497 
5498 // Just to get the Kernel build to link on solaris for testing.
5499 
5500 extern "C" {
5501 class ASGCT_CallTrace;
5502 void AsyncGetCallTrace(ASGCT_CallTrace *trace, jint depth, void* ucontext)
5503   KERNEL_RETURN;
5504 }
5505 
5506 
5507 // ObjectMonitor park-unpark infrastructure ...
5508 //
5509 // We implement Solaris and Linux PlatformEvents with the
5510 // obvious condvar-mutex-flag triple.
5511 // Another alternative that works quite well is pipes:
5512 // Each PlatformEvent consists of a pipe-pair.
5513 // The thread associated with the PlatformEvent
5514 // calls park(), which reads from the input end of the pipe.
5515 // Unpark() writes into the other end of the pipe.
5516 // The write-side of the pipe must be set NDELAY.
5517 // Unfortunately pipes consume a large # of handles.
5518 // Native solaris lwp_park() and lwp_unpark() work nicely, too.
5519 // Using pipes for the 1st few threads might be workable, however.
5520 //
5521 // park() is permitted to return spuriously.
5522 // Callers of park() should wrap the call to park() in
5523 // an appropriate loop.  A litmus test for the correct
5524 // usage of park is the following: if park() were modified
5525 // to immediately return 0 your code should still work,
5526 // albeit degenerating to a spin loop.
5527 //
5528 // An interesting optimization for park() is to use a trylock()
5529 // to attempt to acquire the mutex.  If the trylock() fails
5530 // then we know that a concurrent unpark() operation is in-progress.
5531 // in that case the park() code could simply set _count to 0
5532 // and return immediately.  The subsequent park() operation *might*
5533 // return immediately.  That's harmless as the caller of park() is
5534 // expected to loop.  By using trylock() we will have avoided a
5535 // avoided a context switch caused by contention on the per-thread mutex.
5536 //
5537 // TODO-FIXME:
5538 // 1.  Reconcile Doug's JSR166 j.u.c park-unpark with the
5539 //     objectmonitor implementation.
5540 // 2.  Collapse the JSR166 parker event, and the
5541 //     objectmonitor ParkEvent into a single "Event" construct.
5542 // 3.  In park() and unpark() add:
5543 //     assert (Thread::current() == AssociatedWith).
5544 // 4.  add spurious wakeup injection on a -XX:EarlyParkReturn=N switch.
5545 //     1-out-of-N park() operations will return immediately.
5546 //
5547 // _Event transitions in park()
5548 //   -1 => -1 : illegal
5549 //    1 =>  0 : pass - return immediately
5550 //    0 => -1 : block
5551 //
5552 // _Event serves as a restricted-range semaphore.
5553 //
5554 // Another possible encoding of _Event would be with
5555 // explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
5556 //
5557 // TODO-FIXME: add DTRACE probes for:
5558 // 1.   Tx parks
5559 // 2.   Ty unparks Tx
5560 // 3.   Tx resumes from park
5561 
5562 
5563 // value determined through experimentation
5564 #define ROUNDINGFIX 11
5565 
5566 // utility to compute the abstime argument to timedwait.
5567 // TODO-FIXME: switch from compute_abstime() to unpackTime().
5568 
5569 static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
5570   // millis is the relative timeout time
5571   // abstime will be the absolute timeout time
5572   if (millis < 0)  millis = 0;
5573   struct timeval now;
5574   int status = gettimeofday(&now, NULL);
5575   assert(status == 0, "gettimeofday");
5576   jlong seconds = millis / 1000;
5577   jlong max_wait_period;
5578 
5579   if (UseLWPSynchronization) {
5580     // forward port of fix for 4275818 (not sleeping long enough)
5581     // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
5582     // _lwp_cond_timedwait() used a round_down algorithm rather
5583     // than a round_up. For millis less than our roundfactor
5584     // it rounded down to 0 which doesn't meet the spec.
5585     // For millis > roundfactor we may return a bit sooner, but
5586     // since we can not accurately identify the patch level and
5587     // this has already been fixed in Solaris 9 and 8 we will
5588     // leave it alone rather than always rounding down.
5589 
5590     if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
5591        // It appears that when we go directly through Solaris _lwp_cond_timedwait()
5592            // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
5593            max_wait_period = 21000000;
5594   } else {
5595     max_wait_period = 50000000;
5596   }
5597   millis %= 1000;
5598   if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
5599      seconds = max_wait_period;
5600   }
5601   abstime->tv_sec = now.tv_sec  + seconds;
5602   long       usec = now.tv_usec + millis * 1000;
5603   if (usec >= 1000000) {
5604     abstime->tv_sec += 1;
5605     usec -= 1000000;
5606   }
5607   abstime->tv_nsec = usec * 1000;
5608   return abstime;
5609 }
5610 
5611 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5612 // Conceptually TryPark() should be equivalent to park(0).
5613 
5614 int os::PlatformEvent::TryPark() {
5615   for (;;) {
5616     const int v = _Event ;
5617     guarantee ((v == 0) || (v == 1), "invariant") ;
5618     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
5619   }
5620 }
5621 
5622 void os::PlatformEvent::park() {           // AKA: down()
5623   // Invariant: Only the thread associated with the Event/PlatformEvent
5624   // may call park().
5625   int v ;
5626   for (;;) {
5627       v = _Event ;
5628       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5629   }
5630   guarantee (v >= 0, "invariant") ;
5631   if (v == 0) {
5632      // Do this the hard way by blocking ...
5633      // See http://monaco.sfbay/detail.jsf?cr=5094058.
5634      // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
5635      // Only for SPARC >= V8PlusA
5636 #if defined(__sparc) && defined(COMPILER2)
5637      if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
5638 #endif
5639      int status = os::Solaris::mutex_lock(_mutex);
5640      assert_status(status == 0, status,  "mutex_lock");
5641      guarantee (_nParked == 0, "invariant") ;
5642      ++ _nParked ;
5643      while (_Event < 0) {
5644         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5645         // Treat this the same as if the wait was interrupted
5646         // With usr/lib/lwp going to kernel, always handle ETIME
5647         status = os::Solaris::cond_wait(_cond, _mutex);
5648         if (status == ETIME) status = EINTR ;
5649         assert_status(status == 0 || status == EINTR, status, "cond_wait");
5650      }
5651      -- _nParked ;
5652      _Event = 0 ;
5653      status = os::Solaris::mutex_unlock(_mutex);
5654      assert_status(status == 0, status, "mutex_unlock");
5655   }
5656 }
5657 
5658 int os::PlatformEvent::park(jlong millis) {
5659   guarantee (_nParked == 0, "invariant") ;
5660   int v ;
5661   for (;;) {
5662       v = _Event ;
5663       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5664   }
5665   guarantee (v >= 0, "invariant") ;
5666   if (v != 0) return OS_OK ;
5667 
5668   int ret = OS_TIMEOUT;
5669   timestruc_t abst;
5670   compute_abstime (&abst, millis);
5671 
5672   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5673   // For Solaris SPARC set fprs.FEF=0 prior to parking.
5674   // Only for SPARC >= V8PlusA
5675 #if defined(__sparc) && defined(COMPILER2)
5676  if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
5677 #endif
5678   int status = os::Solaris::mutex_lock(_mutex);
5679   assert_status(status == 0, status, "mutex_lock");
5680   guarantee (_nParked == 0, "invariant") ;
5681   ++ _nParked ;
5682   while (_Event < 0) {
5683      int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
5684      assert_status(status == 0 || status == EINTR ||
5685                    status == ETIME || status == ETIMEDOUT,
5686                    status, "cond_timedwait");
5687      if (!FilterSpuriousWakeups) break ;                // previous semantics
5688      if (status == ETIME || status == ETIMEDOUT) break ;
5689      // We consume and ignore EINTR and spurious wakeups.
5690   }
5691   -- _nParked ;
5692   if (_Event >= 0) ret = OS_OK ;
5693   _Event = 0 ;
5694   status = os::Solaris::mutex_unlock(_mutex);
5695   assert_status(status == 0, status, "mutex_unlock");
5696   return ret;
5697 }
5698 
5699 void os::PlatformEvent::unpark() {
5700   int v, AnyWaiters;
5701 
5702   // Increment _Event.
5703   // Another acceptable implementation would be to simply swap 1
5704   // into _Event:
5705   //   if (Swap (&_Event, 1) < 0) {
5706   //      mutex_lock (_mutex) ; AnyWaiters = nParked; mutex_unlock (_mutex) ;
5707   //      if (AnyWaiters) cond_signal (_cond) ;
5708   //   }
5709 
5710   for (;;) {
5711     v = _Event ;
5712     if (v > 0) {
5713        // The LD of _Event could have reordered or be satisfied
5714        // by a read-aside from this processor's write buffer.
5715        // To avoid problems execute a barrier and then
5716        // ratify the value.  A degenerate CAS() would also work.
5717        // Viz., CAS (v+0, &_Event, v) == v).
5718        OrderAccess::fence() ;
5719        if (_Event == v) return ;
5720        continue ;
5721     }
5722     if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5723   }
5724 
5725   // If the thread associated with the event was parked, wake it.
5726   if (v < 0) {
5727      int status ;
5728      // Wait for the thread assoc with the PlatformEvent to vacate.
5729      status = os::Solaris::mutex_lock(_mutex);
5730      assert_status(status == 0, status, "mutex_lock");
5731      AnyWaiters = _nParked ;
5732      status = os::Solaris::mutex_unlock(_mutex);
5733      assert_status(status == 0, status, "mutex_unlock");
5734      guarantee (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5735      if (AnyWaiters != 0) {
5736        // We intentional signal *after* dropping the lock
5737        // to avoid a common class of futile wakeups.
5738        status = os::Solaris::cond_signal(_cond);
5739        assert_status(status == 0, status, "cond_signal");
5740      }
5741   }
5742 }
5743 
5744 // JSR166
5745 // -------------------------------------------------------
5746 
5747 /*
5748  * The solaris and linux implementations of park/unpark are fairly
5749  * conservative for now, but can be improved. They currently use a
5750  * mutex/condvar pair, plus _counter.
5751  * Park decrements _counter if > 0, else does a condvar wait.  Unpark
5752  * sets count to 1 and signals condvar.  Only one thread ever waits
5753  * on the condvar. Contention seen when trying to park implies that someone
5754  * is unparking you, so don't wait. And spurious returns are fine, so there
5755  * is no need to track notifications.
5756  */
5757 
5758 #define NANOSECS_PER_SEC 1000000000
5759 #define NANOSECS_PER_MILLISEC 1000000
5760 #define MAX_SECS 100000000
5761 
5762 /*
5763  * This code is common to linux and solaris and will be moved to a
5764  * common place in dolphin.
5765  *
5766  * The passed in time value is either a relative time in nanoseconds
5767  * or an absolute time in milliseconds. Either way it has to be unpacked
5768  * into suitable seconds and nanoseconds components and stored in the
5769  * given timespec structure.
5770  * Given time is a 64-bit value and the time_t used in the timespec is only
5771  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5772  * overflow if times way in the future are given. Further on Solaris versions
5773  * prior to 10 there is a restriction (see cond_timedwait) that the specified
5774  * number of seconds, in abstime, is less than current_time  + 100,000,000.
5775  * As it will be 28 years before "now + 100000000" will overflow we can
5776  * ignore overflow and just impose a hard-limit on seconds using the value
5777  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5778  * years from "now".
5779  */
5780 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5781   assert (time > 0, "convertTime");
5782 
5783   struct timeval now;
5784   int status = gettimeofday(&now, NULL);
5785   assert(status == 0, "gettimeofday");
5786 
5787   time_t max_secs = now.tv_sec + MAX_SECS;
5788 
5789   if (isAbsolute) {
5790     jlong secs = time / 1000;
5791     if (secs > max_secs) {
5792       absTime->tv_sec = max_secs;
5793     }
5794     else {
5795       absTime->tv_sec = secs;
5796     }
5797     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5798   }
5799   else {
5800     jlong secs = time / NANOSECS_PER_SEC;
5801     if (secs >= MAX_SECS) {
5802       absTime->tv_sec = max_secs;
5803       absTime->tv_nsec = 0;
5804     }
5805     else {
5806       absTime->tv_sec = now.tv_sec + secs;
5807       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5808       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5809         absTime->tv_nsec -= NANOSECS_PER_SEC;
5810         ++absTime->tv_sec; // note: this must be <= max_secs
5811       }
5812     }
5813   }
5814   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5815   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5816   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5817   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5818 }
5819 
5820 void Parker::park(bool isAbsolute, jlong time) {
5821 
5822   // Optional fast-path check:
5823   // Return immediately if a permit is available.
5824   if (_counter > 0) {
5825       _counter = 0 ;
5826       OrderAccess::fence();
5827       return ;
5828   }
5829 
5830   // Optional fast-exit: Check interrupt before trying to wait
5831   Thread* thread = Thread::current();
5832   assert(thread->is_Java_thread(), "Must be JavaThread");
5833   JavaThread *jt = (JavaThread *)thread;
5834   if (Thread::is_interrupted(thread, false)) {
5835     return;
5836   }
5837 
5838   // First, demultiplex/decode time arguments
5839   timespec absTime;
5840   if (time < 0) { // don't wait at all
5841     return;
5842   }
5843   if (time > 0) {
5844     // Warning: this code might be exposed to the old Solaris time
5845     // round-down bugs.  Grep "roundingFix" for details.
5846     unpackTime(&absTime, isAbsolute, time);
5847   }
5848 
5849   // Enter safepoint region
5850   // Beware of deadlocks such as 6317397.
5851   // The per-thread Parker:: _mutex is a classic leaf-lock.
5852   // In particular a thread must never block on the Threads_lock while
5853   // holding the Parker:: mutex.  If safepoints are pending both the
5854   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5855   ThreadBlockInVM tbivm(jt);
5856 
5857   // Don't wait if cannot get lock since interference arises from
5858   // unblocking.  Also. check interrupt before trying wait
5859   if (Thread::is_interrupted(thread, false) ||
5860       os::Solaris::mutex_trylock(_mutex) != 0) {
5861     return;
5862   }
5863 
5864   int status ;
5865 
5866   if (_counter > 0)  { // no wait needed
5867     _counter = 0;
5868     status = os::Solaris::mutex_unlock(_mutex);
5869     assert (status == 0, "invariant") ;
5870     OrderAccess::fence();
5871     return;
5872   }
5873 
5874 #ifdef ASSERT
5875   // Don't catch signals while blocked; let the running threads have the signals.
5876   // (This allows a debugger to break into the running thread.)
5877   sigset_t oldsigs;
5878   sigset_t* allowdebug_blocked = os::Solaris::allowdebug_blocked_signals();
5879   thr_sigsetmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5880 #endif
5881 
5882   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5883   jt->set_suspend_equivalent();
5884   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5885 
5886   // Do this the hard way by blocking ...
5887   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5888   // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
5889   // Only for SPARC >= V8PlusA
5890 #if defined(__sparc) && defined(COMPILER2)
5891   if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
5892 #endif
5893 
5894   if (time == 0) {
5895     status = os::Solaris::cond_wait (_cond, _mutex) ;
5896   } else {
5897     status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
5898   }
5899   // Note that an untimed cond_wait() can sometimes return ETIME on older
5900   // versions of the Solaris.
5901   assert_status(status == 0 || status == EINTR ||
5902                 status == ETIME || status == ETIMEDOUT,
5903                 status, "cond_timedwait");
5904 
5905 #ifdef ASSERT
5906   thr_sigsetmask(SIG_SETMASK, &oldsigs, NULL);
5907 #endif
5908   _counter = 0 ;
5909   status = os::Solaris::mutex_unlock(_mutex);
5910   assert_status(status == 0, status, "mutex_unlock") ;
5911 
5912   // If externally suspended while waiting, re-suspend
5913   if (jt->handle_special_suspend_equivalent_condition()) {
5914     jt->java_suspend_self();
5915   }
5916   OrderAccess::fence();
5917 }
5918 
5919 void Parker::unpark() {
5920   int s, status ;
5921   status = os::Solaris::mutex_lock (_mutex) ;
5922   assert (status == 0, "invariant") ;
5923   s = _counter;
5924   _counter = 1;
5925   status = os::Solaris::mutex_unlock (_mutex) ;
5926   assert (status == 0, "invariant") ;
5927 
5928   if (s < 1) {
5929     status = os::Solaris::cond_signal (_cond) ;
5930     assert (status == 0, "invariant") ;
5931   }
5932 }
5933 
5934 extern char** environ;
5935 
5936 // Run the specified command in a separate process. Return its exit value,
5937 // or -1 on failure (e.g. can't fork a new process).
5938 // Unlike system(), this function can be called from signal handler. It
5939 // doesn't block SIGINT et al.
5940 int os::fork_and_exec(char* cmd) {
5941   char * argv[4];
5942   argv[0] = (char *)"sh";
5943   argv[1] = (char *)"-c";
5944   argv[2] = cmd;
5945   argv[3] = NULL;
5946 
5947   // fork is async-safe, fork1 is not so can't use in signal handler
5948   pid_t pid;
5949   Thread* t = ThreadLocalStorage::get_thread_slow();
5950   if (t != NULL && t->is_inside_signal_handler()) {
5951     pid = fork();
5952   } else {
5953     pid = fork1();
5954   }
5955 
5956   if (pid < 0) {
5957     // fork failed
5958     warning("fork failed: %s", strerror(errno));
5959     return -1;
5960 
5961   } else if (pid == 0) {
5962     // child process
5963 
5964     // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
5965     execve("/usr/bin/sh", argv, environ);
5966 
5967     // execve failed
5968     _exit(-1);
5969 
5970   } else  {
5971     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5972     // care about the actual exit code, for now.
5973 
5974     int status;
5975 
5976     // Wait for the child process to exit.  This returns immediately if
5977     // the child has already exited. */
5978     while (waitpid(pid, &status, 0) < 0) {
5979         switch (errno) {
5980         case ECHILD: return 0;
5981         case EINTR: break;
5982         default: return -1;
5983         }
5984     }
5985 
5986     if (WIFEXITED(status)) {
5987        // The child exited normally; get its exit code.
5988        return WEXITSTATUS(status);
5989     } else if (WIFSIGNALED(status)) {
5990        // The child exited because of a signal
5991        // The best value to return is 0x80 + signal number,
5992        // because that is what all Unix shells do, and because
5993        // it allows callers to distinguish between process exit and
5994        // process death by signal.
5995        return 0x80 + WTERMSIG(status);
5996     } else {
5997        // Unknown exit code; pass it through
5998        return status;
5999     }
6000   }
6001 }
6002 
6003 // is_headless_jre()
6004 //
6005 // Test for the existence of libmawt in motif21 or xawt directories
6006 // in order to report if we are running in a headless jre
6007 //
6008 bool os::is_headless_jre() {
6009     struct stat statbuf;
6010     char buf[MAXPATHLEN];
6011     char libmawtpath[MAXPATHLEN];
6012     const char *xawtstr  = "/xawt/libmawt.so";
6013     const char *motifstr = "/motif21/libmawt.so";
6014     char *p;
6015 
6016     // Get path to libjvm.so
6017     os::jvm_path(buf, sizeof(buf));
6018 
6019     // Get rid of libjvm.so
6020     p = strrchr(buf, '/');
6021     if (p == NULL) return false;
6022     else *p = '\0';
6023 
6024     // Get rid of client or server
6025     p = strrchr(buf, '/');
6026     if (p == NULL) return false;
6027     else *p = '\0';
6028 
6029     // check xawt/libmawt.so
6030     strcpy(libmawtpath, buf);
6031     strcat(libmawtpath, xawtstr);
6032     if (::stat(libmawtpath, &statbuf) == 0) return false;
6033 
6034     // check motif21/libmawt.so
6035     strcpy(libmawtpath, buf);
6036     strcat(libmawtpath, motifstr);
6037     if (::stat(libmawtpath, &statbuf) == 0) return false;
6038 
6039     return true;
6040 }
6041 
6042