1 /*
   2  * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "interpreter/interpreter.hpp"
  33 #include "jvm_solaris.h"
  34 #include "memory/allocation.inline.hpp"
  35 #include "memory/filemap.hpp"
  36 #include "mutex_solaris.inline.hpp"
  37 #include "oops/oop.inline.hpp"
  38 #include "os_share_solaris.hpp"
  39 #include "prims/jniFastGetField.hpp"
  40 #include "prims/jvm.h"
  41 #include "prims/jvm_misc.hpp"
  42 #include "runtime/arguments.hpp"
  43 #include "runtime/extendedPC.hpp"
  44 #include "runtime/globals.hpp"
  45 #include "runtime/interfaceSupport.hpp"
  46 #include "runtime/java.hpp"
  47 #include "runtime/javaCalls.hpp"
  48 #include "runtime/mutexLocker.hpp"
  49 #include "runtime/objectMonitor.hpp"
  50 #include "runtime/osThread.hpp"
  51 #include "runtime/perfMemory.hpp"
  52 #include "runtime/sharedRuntime.hpp"
  53 #include "runtime/statSampler.hpp"
  54 #include "runtime/stubRoutines.hpp"
  55 #include "runtime/threadCritical.hpp"
  56 #include "runtime/timer.hpp"
  57 #include "services/attachListener.hpp"
  58 #include "services/runtimeService.hpp"
  59 #include "thread_solaris.inline.hpp"
  60 #include "utilities/decoder.hpp"
  61 #include "utilities/defaultStream.hpp"
  62 #include "utilities/events.hpp"
  63 #include "utilities/growableArray.hpp"
  64 #include "utilities/vmError.hpp"
  65 #ifdef TARGET_ARCH_x86
  66 # include "assembler_x86.inline.hpp"
  67 # include "nativeInst_x86.hpp"
  68 #endif
  69 #ifdef TARGET_ARCH_sparc
  70 # include "assembler_sparc.inline.hpp"
  71 # include "nativeInst_sparc.hpp"
  72 #endif
  73 #ifdef COMPILER1
  74 #include "c1/c1_Runtime1.hpp"
  75 #endif
  76 #ifdef COMPILER2
  77 #include "opto/runtime.hpp"
  78 #endif
  79 
  80 // put OS-includes here
  81 # include <dlfcn.h>
  82 # include <errno.h>
  83 # include <exception>
  84 # include <link.h>
  85 # include <poll.h>
  86 # include <pthread.h>
  87 # include <pwd.h>
  88 # include <schedctl.h>
  89 # include <setjmp.h>
  90 # include <signal.h>
  91 # include <stdio.h>
  92 # include <alloca.h>
  93 # include <sys/filio.h>
  94 # include <sys/ipc.h>
  95 # include <sys/lwp.h>
  96 # include <sys/machelf.h>     // for elf Sym structure used by dladdr1
  97 # include <sys/mman.h>
  98 # include <sys/processor.h>
  99 # include <sys/procset.h>
 100 # include <sys/pset.h>
 101 # include <sys/resource.h>
 102 # include <sys/shm.h>
 103 # include <sys/socket.h>
 104 # include <sys/stat.h>
 105 # include <sys/systeminfo.h>
 106 # include <sys/time.h>
 107 # include <sys/times.h>
 108 # include <sys/types.h>
 109 # include <sys/wait.h>
 110 # include <sys/utsname.h>
 111 # include <thread.h>
 112 # include <unistd.h>
 113 # include <sys/priocntl.h>
 114 # include <sys/rtpriocntl.h>
 115 # include <sys/tspriocntl.h>
 116 # include <sys/iapriocntl.h>
 117 # include <sys/fxpriocntl.h>
 118 # include <sys/loadavg.h>
 119 # include <string.h>
 120 # include <stdio.h>
 121 
 122 # define _STRUCTURED_PROC 1  //  this gets us the new structured proc interfaces of 5.6 & later
 123 # include <sys/procfs.h>     //  see comment in <sys/procfs.h>
 124 
 125 #define MAX_PATH (2 * K)
 126 
 127 // for timer info max values which include all bits
 128 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 129 
 130 #ifdef _GNU_SOURCE
 131 // See bug #6514594
 132 extern "C" int madvise(caddr_t, size_t, int);
 133 extern "C" int memcntl(caddr_t addr, size_t len, int cmd, caddr_t arg,
 134                        int attr, int mask);
 135 #endif //_GNU_SOURCE
 136 
 137 /*
 138   MPSS Changes Start.
 139   The JVM binary needs to be built and run on pre-Solaris 9
 140   systems, but the constants needed by MPSS are only in Solaris 9
 141   header files.  They are textually replicated here to allow
 142   building on earlier systems.  Once building on Solaris 8 is
 143   no longer a requirement, these #defines can be replaced by ordinary
 144   system .h inclusion.
 145 
 146   In earlier versions of the  JDK and Solaris, we used ISM for large pages.
 147   But ISM requires shared memory to achieve this and thus has many caveats.
 148   MPSS is a fully transparent and is a cleaner way to get large pages.
 149   Although we still require keeping ISM for backward compatiblitiy as well as
 150   giving the opportunity to use large pages on older systems it is
 151   recommended that MPSS be used for Solaris 9 and above.
 152 
 153 */
 154 
 155 #ifndef MC_HAT_ADVISE
 156 
 157 struct memcntl_mha {
 158   uint_t          mha_cmd;        /* command(s) */
 159   uint_t          mha_flags;
 160   size_t          mha_pagesize;
 161 };
 162 #define MC_HAT_ADVISE   7       /* advise hat map size */
 163 #define MHA_MAPSIZE_VA  0x1     /* set preferred page size */
 164 #define MAP_ALIGN       0x200   /* addr specifies alignment */
 165 
 166 #endif
 167 // MPSS Changes End.
 168 
 169 
 170 // Here are some liblgrp types from sys/lgrp_user.h to be able to
 171 // compile on older systems without this header file.
 172 
 173 #ifndef MADV_ACCESS_LWP
 174 # define  MADV_ACCESS_LWP         7       /* next LWP to access heavily */
 175 #endif
 176 #ifndef MADV_ACCESS_MANY
 177 # define  MADV_ACCESS_MANY        8       /* many processes to access heavily */
 178 #endif
 179 
 180 #ifndef LGRP_RSRC_CPU
 181 # define LGRP_RSRC_CPU           0       /* CPU resources */
 182 #endif
 183 #ifndef LGRP_RSRC_MEM
 184 # define LGRP_RSRC_MEM           1       /* memory resources */
 185 #endif
 186 
 187 // Some more macros from sys/mman.h that are not present in Solaris 8.
 188 
 189 #ifndef MAX_MEMINFO_CNT
 190 /*
 191  * info_req request type definitions for meminfo
 192  * request types starting with MEMINFO_V are used for Virtual addresses
 193  * and should not be mixed with MEMINFO_PLGRP which is targeted for Physical
 194  * addresses
 195  */
 196 # define MEMINFO_SHIFT           16
 197 # define MEMINFO_MASK            (0xFF << MEMINFO_SHIFT)
 198 # define MEMINFO_VPHYSICAL       (0x01 << MEMINFO_SHIFT) /* get physical addr */
 199 # define MEMINFO_VLGRP           (0x02 << MEMINFO_SHIFT) /* get lgroup */
 200 # define MEMINFO_VPAGESIZE       (0x03 << MEMINFO_SHIFT) /* size of phys page */
 201 # define MEMINFO_VREPLCNT        (0x04 << MEMINFO_SHIFT) /* no. of replica */
 202 # define MEMINFO_VREPL           (0x05 << MEMINFO_SHIFT) /* physical replica */
 203 # define MEMINFO_VREPL_LGRP      (0x06 << MEMINFO_SHIFT) /* lgrp of replica */
 204 # define MEMINFO_PLGRP           (0x07 << MEMINFO_SHIFT) /* lgroup for paddr */
 205 
 206 /* maximum number of addresses meminfo() can process at a time */
 207 # define MAX_MEMINFO_CNT 256
 208 
 209 /* maximum number of request types */
 210 # define MAX_MEMINFO_REQ 31
 211 #endif
 212 
 213 // see thr_setprio(3T) for the basis of these numbers
 214 #define MinimumPriority 0
 215 #define NormalPriority  64
 216 #define MaximumPriority 127
 217 
 218 // Values for ThreadPriorityPolicy == 1
 219 int prio_policy1[CriticalPriority+1] = {
 220   -99999,  0, 16,  32,  48,  64,
 221           80, 96, 112, 124, 127, 127 };
 222 
 223 // System parameters used internally
 224 static clock_t clock_tics_per_sec = 100;
 225 
 226 // Track if we have called enable_extended_FILE_stdio (on Solaris 10u4+)
 227 static bool enabled_extended_FILE_stdio = false;
 228 
 229 // For diagnostics to print a message once. see run_periodic_checks
 230 static bool check_addr0_done = false;
 231 static sigset_t check_signal_done;
 232 static bool check_signals = true;
 233 
 234 address os::Solaris::handler_start;  // start pc of thr_sighndlrinfo
 235 address os::Solaris::handler_end;    // end pc of thr_sighndlrinfo
 236 
 237 address os::Solaris::_main_stack_base = NULL;  // 4352906 workaround
 238 
 239 
 240 // "default" initializers for missing libc APIs
 241 extern "C" {
 242   static int lwp_mutex_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
 243   static int lwp_mutex_destroy(mutex_t *mx)                 { return 0; }
 244 
 245   static int lwp_cond_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
 246   static int lwp_cond_destroy(cond_t *cv)                   { return 0; }
 247 }
 248 
 249 // "default" initializers for pthread-based synchronization
 250 extern "C" {
 251   static int pthread_mutex_default_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
 252   static int pthread_cond_default_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
 253 }
 254 
 255 // Thread Local Storage
 256 // This is common to all Solaris platforms so it is defined here,
 257 // in this common file.
 258 // The declarations are in the os_cpu threadLS*.hpp files.
 259 //
 260 // Static member initialization for TLS
 261 Thread* ThreadLocalStorage::_get_thread_cache[ThreadLocalStorage::_pd_cache_size] = {NULL};
 262 
 263 #ifndef PRODUCT
 264 #define _PCT(n,d)       ((100.0*(double)(n))/(double)(d))
 265 
 266 int ThreadLocalStorage::_tcacheHit = 0;
 267 int ThreadLocalStorage::_tcacheMiss = 0;
 268 
 269 void ThreadLocalStorage::print_statistics() {
 270   int total = _tcacheMiss+_tcacheHit;
 271   tty->print_cr("Thread cache hits %d misses %d total %d percent %f\n",
 272                 _tcacheHit, _tcacheMiss, total, _PCT(_tcacheHit, total));
 273 }
 274 #undef _PCT
 275 #endif // PRODUCT
 276 
 277 Thread* ThreadLocalStorage::get_thread_via_cache_slowly(uintptr_t raw_id,
 278                                                         int index) {
 279   Thread *thread = get_thread_slow();
 280   if (thread != NULL) {
 281     address sp = os::current_stack_pointer();
 282     guarantee(thread->_stack_base == NULL ||
 283               (sp <= thread->_stack_base &&
 284                  sp >= thread->_stack_base - thread->_stack_size) ||
 285                is_error_reported(),
 286               "sp must be inside of selected thread stack");
 287 
 288     thread->set_self_raw_id(raw_id);  // mark for quick retrieval
 289     _get_thread_cache[ index ] = thread;
 290   }
 291   return thread;
 292 }
 293 
 294 
 295 static const double all_zero[ sizeof(Thread) / sizeof(double) + 1 ] = {0};
 296 #define NO_CACHED_THREAD ((Thread*)all_zero)
 297 
 298 void ThreadLocalStorage::pd_set_thread(Thread* thread) {
 299 
 300   // Store the new value before updating the cache to prevent a race
 301   // between get_thread_via_cache_slowly() and this store operation.
 302   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
 303 
 304   // Update thread cache with new thread if setting on thread create,
 305   // or NO_CACHED_THREAD (zeroed) thread if resetting thread on exit.
 306   uintptr_t raw = pd_raw_thread_id();
 307   int ix = pd_cache_index(raw);
 308   _get_thread_cache[ix] = thread == NULL ? NO_CACHED_THREAD : thread;
 309 }
 310 
 311 void ThreadLocalStorage::pd_init() {
 312   for (int i = 0; i < _pd_cache_size; i++) {
 313     _get_thread_cache[i] = NO_CACHED_THREAD;
 314   }
 315 }
 316 
 317 // Invalidate all the caches (happens to be the same as pd_init).
 318 void ThreadLocalStorage::pd_invalidate_all() { pd_init(); }
 319 
 320 #undef NO_CACHED_THREAD
 321 
 322 // END Thread Local Storage
 323 
 324 static inline size_t adjust_stack_size(address base, size_t size) {
 325   if ((ssize_t)size < 0) {
 326     // 4759953: Compensate for ridiculous stack size.
 327     size = max_intx;
 328   }
 329   if (size > (size_t)base) {
 330     // 4812466: Make sure size doesn't allow the stack to wrap the address space.
 331     size = (size_t)base;
 332   }
 333   return size;
 334 }
 335 
 336 static inline stack_t get_stack_info() {
 337   stack_t st;
 338   int retval = thr_stksegment(&st);
 339   st.ss_size = adjust_stack_size((address)st.ss_sp, st.ss_size);
 340   assert(retval == 0, "incorrect return value from thr_stksegment");
 341   assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
 342   assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
 343   return st;
 344 }
 345 
 346 address os::current_stack_base() {
 347   int r = thr_main() ;
 348   guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
 349   bool is_primordial_thread = r;
 350 
 351   // Workaround 4352906, avoid calls to thr_stksegment by
 352   // thr_main after the first one (it looks like we trash
 353   // some data, causing the value for ss_sp to be incorrect).
 354   if (!is_primordial_thread || os::Solaris::_main_stack_base == NULL) {
 355     stack_t st = get_stack_info();
 356     if (is_primordial_thread) {
 357       // cache initial value of stack base
 358       os::Solaris::_main_stack_base = (address)st.ss_sp;
 359     }
 360     return (address)st.ss_sp;
 361   } else {
 362     guarantee(os::Solaris::_main_stack_base != NULL, "Attempt to use null cached stack base");
 363     return os::Solaris::_main_stack_base;
 364   }
 365 }
 366 
 367 size_t os::current_stack_size() {
 368   size_t size;
 369 
 370   int r = thr_main() ;
 371   guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
 372   if(!r) {
 373     size = get_stack_info().ss_size;
 374   } else {
 375     struct rlimit limits;
 376     getrlimit(RLIMIT_STACK, &limits);
 377     size = adjust_stack_size(os::Solaris::_main_stack_base, (size_t)limits.rlim_cur);
 378   }
 379   // base may not be page aligned
 380   address base = current_stack_base();
 381   address bottom = (address)align_size_up((intptr_t)(base - size), os::vm_page_size());;
 382   return (size_t)(base - bottom);
 383 }
 384 
 385 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
 386   return localtime_r(clock, res);
 387 }
 388 
 389 // interruptible infrastructure
 390 
 391 // setup_interruptible saves the thread state before going into an
 392 // interruptible system call.
 393 // The saved state is used to restore the thread to
 394 // its former state whether or not an interrupt is received.
 395 // Used by classloader os::read
 396 // os::restartable_read calls skip this layer and stay in _thread_in_native
 397 
 398 void os::Solaris::setup_interruptible(JavaThread* thread) {
 399 
 400   JavaThreadState thread_state = thread->thread_state();
 401 
 402   assert(thread_state != _thread_blocked, "Coming from the wrong thread");
 403   assert(thread_state != _thread_in_native, "Native threads skip setup_interruptible");
 404   OSThread* osthread = thread->osthread();
 405   osthread->set_saved_interrupt_thread_state(thread_state);
 406   thread->frame_anchor()->make_walkable(thread);
 407   ThreadStateTransition::transition(thread, thread_state, _thread_blocked);
 408 }
 409 
 410 // Version of setup_interruptible() for threads that are already in
 411 // _thread_blocked. Used by os_sleep().
 412 void os::Solaris::setup_interruptible_already_blocked(JavaThread* thread) {
 413   thread->frame_anchor()->make_walkable(thread);
 414 }
 415 
 416 JavaThread* os::Solaris::setup_interruptible() {
 417   JavaThread* thread = (JavaThread*)ThreadLocalStorage::thread();
 418   setup_interruptible(thread);
 419   return thread;
 420 }
 421 
 422 void os::Solaris::try_enable_extended_io() {
 423   typedef int (*enable_extended_FILE_stdio_t)(int, int);
 424 
 425   if (!UseExtendedFileIO) {
 426     return;
 427   }
 428 
 429   enable_extended_FILE_stdio_t enabler =
 430     (enable_extended_FILE_stdio_t) dlsym(RTLD_DEFAULT,
 431                                          "enable_extended_FILE_stdio");
 432   if (enabler) {
 433     enabler(-1, -1);
 434   }
 435 }
 436 
 437 
 438 #ifdef ASSERT
 439 
 440 JavaThread* os::Solaris::setup_interruptible_native() {
 441   JavaThread* thread = (JavaThread*)ThreadLocalStorage::thread();
 442   JavaThreadState thread_state = thread->thread_state();
 443   assert(thread_state == _thread_in_native, "Assumed thread_in_native");
 444   return thread;
 445 }
 446 
 447 void os::Solaris::cleanup_interruptible_native(JavaThread* thread) {
 448   JavaThreadState thread_state = thread->thread_state();
 449   assert(thread_state == _thread_in_native, "Assumed thread_in_native");
 450 }
 451 #endif
 452 
 453 // cleanup_interruptible reverses the effects of setup_interruptible
 454 // setup_interruptible_already_blocked() does not need any cleanup.
 455 
 456 void os::Solaris::cleanup_interruptible(JavaThread* thread) {
 457   OSThread* osthread = thread->osthread();
 458 
 459   ThreadStateTransition::transition(thread, _thread_blocked, osthread->saved_interrupt_thread_state());
 460 }
 461 
 462 // I/O interruption related counters called in _INTERRUPTIBLE
 463 
 464 void os::Solaris::bump_interrupted_before_count() {
 465   RuntimeService::record_interrupted_before_count();
 466 }
 467 
 468 void os::Solaris::bump_interrupted_during_count() {
 469   RuntimeService::record_interrupted_during_count();
 470 }
 471 
 472 static int _processors_online = 0;
 473 
 474          jint os::Solaris::_os_thread_limit = 0;
 475 volatile jint os::Solaris::_os_thread_count = 0;
 476 
 477 julong os::available_memory() {
 478   return Solaris::available_memory();
 479 }
 480 
 481 julong os::Solaris::available_memory() {
 482   return (julong)sysconf(_SC_AVPHYS_PAGES) * os::vm_page_size();
 483 }
 484 
 485 julong os::Solaris::_physical_memory = 0;
 486 
 487 julong os::physical_memory() {
 488    return Solaris::physical_memory();
 489 }
 490 
 491 julong os::allocatable_physical_memory(julong size) {
 492 #ifdef _LP64
 493    return size;
 494 #else
 495    julong result = MIN2(size, (julong)3835*M);
 496    if (!is_allocatable(result)) {
 497      // Memory allocations will be aligned but the alignment
 498      // is not known at this point.  Alignments will
 499      // be at most to LargePageSizeInBytes.  Protect
 500      // allocations from alignments up to illegal
 501      // values. If at this point 2G is illegal.
 502      julong reasonable_size = (julong)2*G - 2 * LargePageSizeInBytes;
 503      result =  MIN2(size, reasonable_size);
 504    }
 505    return result;
 506 #endif
 507 }
 508 
 509 static hrtime_t first_hrtime = 0;
 510 static const hrtime_t hrtime_hz = 1000*1000*1000;
 511 const int LOCK_BUSY = 1;
 512 const int LOCK_FREE = 0;
 513 const int LOCK_INVALID = -1;
 514 static volatile hrtime_t max_hrtime = 0;
 515 static volatile int max_hrtime_lock = LOCK_FREE;     // Update counter with LSB as lock-in-progress
 516 
 517 
 518 void os::Solaris::initialize_system_info() {
 519   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 520   _processors_online = sysconf (_SC_NPROCESSORS_ONLN);
 521   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 522 }
 523 
 524 int os::active_processor_count() {
 525   int online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
 526   pid_t pid = getpid();
 527   psetid_t pset = PS_NONE;
 528   // Are we running in a processor set or is there any processor set around?
 529   if (pset_bind(PS_QUERY, P_PID, pid, &pset) == 0) {
 530     uint_t pset_cpus;
 531     // Query the number of cpus available to us.
 532     if (pset_info(pset, NULL, &pset_cpus, NULL) == 0) {
 533       assert(pset_cpus > 0 && pset_cpus <= online_cpus, "sanity check");
 534       _processors_online = pset_cpus;
 535       return pset_cpus;
 536     }
 537   }
 538   // Otherwise return number of online cpus
 539   return online_cpus;
 540 }
 541 
 542 static bool find_processors_in_pset(psetid_t        pset,
 543                                     processorid_t** id_array,
 544                                     uint_t*         id_length) {
 545   bool result = false;
 546   // Find the number of processors in the processor set.
 547   if (pset_info(pset, NULL, id_length, NULL) == 0) {
 548     // Make up an array to hold their ids.
 549     *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length);
 550     // Fill in the array with their processor ids.
 551     if (pset_info(pset, NULL, id_length, *id_array) == 0) {
 552       result = true;
 553     }
 554   }
 555   return result;
 556 }
 557 
 558 // Callers of find_processors_online() must tolerate imprecise results --
 559 // the system configuration can change asynchronously because of DR
 560 // or explicit psradm operations.
 561 //
 562 // We also need to take care that the loop (below) terminates as the
 563 // number of processors online can change between the _SC_NPROCESSORS_ONLN
 564 // request and the loop that builds the list of processor ids.   Unfortunately
 565 // there's no reliable way to determine the maximum valid processor id,
 566 // so we use a manifest constant, MAX_PROCESSOR_ID, instead.  See p_online
 567 // man pages, which claim the processor id set is "sparse, but
 568 // not too sparse".  MAX_PROCESSOR_ID is used to ensure that we eventually
 569 // exit the loop.
 570 //
 571 // In the future we'll be able to use sysconf(_SC_CPUID_MAX), but that's
 572 // not available on S8.0.
 573 
 574 static bool find_processors_online(processorid_t** id_array,
 575                                    uint*           id_length) {
 576   const processorid_t MAX_PROCESSOR_ID = 100000 ;
 577   // Find the number of processors online.
 578   *id_length = sysconf(_SC_NPROCESSORS_ONLN);
 579   // Make up an array to hold their ids.
 580   *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length);
 581   // Processors need not be numbered consecutively.
 582   long found = 0;
 583   processorid_t next = 0;
 584   while (found < *id_length && next < MAX_PROCESSOR_ID) {
 585     processor_info_t info;
 586     if (processor_info(next, &info) == 0) {
 587       // NB, PI_NOINTR processors are effectively online ...
 588       if (info.pi_state == P_ONLINE || info.pi_state == P_NOINTR) {
 589         (*id_array)[found] = next;
 590         found += 1;
 591       }
 592     }
 593     next += 1;
 594   }
 595   if (found < *id_length) {
 596       // The loop above didn't identify the expected number of processors.
 597       // We could always retry the operation, calling sysconf(_SC_NPROCESSORS_ONLN)
 598       // and re-running the loop, above, but there's no guarantee of progress
 599       // if the system configuration is in flux.  Instead, we just return what
 600       // we've got.  Note that in the worst case find_processors_online() could
 601       // return an empty set.  (As a fall-back in the case of the empty set we
 602       // could just return the ID of the current processor).
 603       *id_length = found ;
 604   }
 605 
 606   return true;
 607 }
 608 
 609 static bool assign_distribution(processorid_t* id_array,
 610                                 uint           id_length,
 611                                 uint*          distribution,
 612                                 uint           distribution_length) {
 613   // We assume we can assign processorid_t's to uint's.
 614   assert(sizeof(processorid_t) == sizeof(uint),
 615          "can't convert processorid_t to uint");
 616   // Quick check to see if we won't succeed.
 617   if (id_length < distribution_length) {
 618     return false;
 619   }
 620   // Assign processor ids to the distribution.
 621   // Try to shuffle processors to distribute work across boards,
 622   // assuming 4 processors per board.
 623   const uint processors_per_board = ProcessDistributionStride;
 624   // Find the maximum processor id.
 625   processorid_t max_id = 0;
 626   for (uint m = 0; m < id_length; m += 1) {
 627     max_id = MAX2(max_id, id_array[m]);
 628   }
 629   // The next id, to limit loops.
 630   const processorid_t limit_id = max_id + 1;
 631   // Make up markers for available processors.
 632   bool* available_id = NEW_C_HEAP_ARRAY(bool, limit_id);
 633   for (uint c = 0; c < limit_id; c += 1) {
 634     available_id[c] = false;
 635   }
 636   for (uint a = 0; a < id_length; a += 1) {
 637     available_id[id_array[a]] = true;
 638   }
 639   // Step by "boards", then by "slot", copying to "assigned".
 640   // NEEDS_CLEANUP: The assignment of processors should be stateful,
 641   //                remembering which processors have been assigned by
 642   //                previous calls, etc., so as to distribute several
 643   //                independent calls of this method.  What we'd like is
 644   //                It would be nice to have an API that let us ask
 645   //                how many processes are bound to a processor,
 646   //                but we don't have that, either.
 647   //                In the short term, "board" is static so that
 648   //                subsequent distributions don't all start at board 0.
 649   static uint board = 0;
 650   uint assigned = 0;
 651   // Until we've found enough processors ....
 652   while (assigned < distribution_length) {
 653     // ... find the next available processor in the board.
 654     for (uint slot = 0; slot < processors_per_board; slot += 1) {
 655       uint try_id = board * processors_per_board + slot;
 656       if ((try_id < limit_id) && (available_id[try_id] == true)) {
 657         distribution[assigned] = try_id;
 658         available_id[try_id] = false;
 659         assigned += 1;
 660         break;
 661       }
 662     }
 663     board += 1;
 664     if (board * processors_per_board + 0 >= limit_id) {
 665       board = 0;
 666     }
 667   }
 668   if (available_id != NULL) {
 669     FREE_C_HEAP_ARRAY(bool, available_id);
 670   }
 671   return true;
 672 }
 673 
 674 void os::set_native_thread_name(const char *name) {
 675   // Not yet implemented.
 676   return;
 677 }
 678 
 679 bool os::distribute_processes(uint length, uint* distribution) {
 680   bool result = false;
 681   // Find the processor id's of all the available CPUs.
 682   processorid_t* id_array  = NULL;
 683   uint           id_length = 0;
 684   // There are some races between querying information and using it,
 685   // since processor sets can change dynamically.
 686   psetid_t pset = PS_NONE;
 687   // Are we running in a processor set?
 688   if ((pset_bind(PS_QUERY, P_PID, P_MYID, &pset) == 0) && pset != PS_NONE) {
 689     result = find_processors_in_pset(pset, &id_array, &id_length);
 690   } else {
 691     result = find_processors_online(&id_array, &id_length);
 692   }
 693   if (result == true) {
 694     if (id_length >= length) {
 695       result = assign_distribution(id_array, id_length, distribution, length);
 696     } else {
 697       result = false;
 698     }
 699   }
 700   if (id_array != NULL) {
 701     FREE_C_HEAP_ARRAY(processorid_t, id_array);
 702   }
 703   return result;
 704 }
 705 
 706 bool os::bind_to_processor(uint processor_id) {
 707   // We assume that a processorid_t can be stored in a uint.
 708   assert(sizeof(uint) == sizeof(processorid_t),
 709          "can't convert uint to processorid_t");
 710   int bind_result =
 711     processor_bind(P_LWPID,                       // bind LWP.
 712                    P_MYID,                        // bind current LWP.
 713                    (processorid_t) processor_id,  // id.
 714                    NULL);                         // don't return old binding.
 715   return (bind_result == 0);
 716 }
 717 
 718 bool os::getenv(const char* name, char* buffer, int len) {
 719   char* val = ::getenv( name );
 720   if ( val == NULL
 721   ||   strlen(val) + 1  >  len ) {
 722     if (len > 0)  buffer[0] = 0; // return a null string
 723     return false;
 724   }
 725   strcpy( buffer, val );
 726   return true;
 727 }
 728 
 729 
 730 // Return true if user is running as root.
 731 
 732 bool os::have_special_privileges() {
 733   static bool init = false;
 734   static bool privileges = false;
 735   if (!init) {
 736     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 737     init = true;
 738   }
 739   return privileges;
 740 }
 741 
 742 
 743 void os::init_system_properties_values() {
 744   char arch[12];
 745   sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
 746 
 747   // The next steps are taken in the product version:
 748   //
 749   // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
 750   // This library should be located at:
 751   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
 752   //
 753   // If "/jre/lib/" appears at the right place in the path, then we
 754   // assume libjvm[_g].so is installed in a JDK and we use this path.
 755   //
 756   // Otherwise exit with message: "Could not create the Java virtual machine."
 757   //
 758   // The following extra steps are taken in the debugging version:
 759   //
 760   // If "/jre/lib/" does NOT appear at the right place in the path
 761   // instead of exit check for $JAVA_HOME environment variable.
 762   //
 763   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 764   // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
 765   // it looks like libjvm[_g].so is installed there
 766   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
 767   //
 768   // Otherwise exit.
 769   //
 770   // Important note: if the location of libjvm.so changes this
 771   // code needs to be changed accordingly.
 772 
 773   // The next few definitions allow the code to be verbatim:
 774 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
 775 #define free(p) FREE_C_HEAP_ARRAY(char, p)
 776 #define getenv(n) ::getenv(n)
 777 
 778 #define EXTENSIONS_DIR  "/lib/ext"
 779 #define ENDORSED_DIR    "/lib/endorsed"
 780 #define COMMON_DIR      "/usr/jdk/packages"
 781 
 782   {
 783     /* sysclasspath, java_home, dll_dir */
 784     {
 785         char *home_path;
 786         char *dll_path;
 787         char *pslash;
 788         char buf[MAXPATHLEN];
 789         os::jvm_path(buf, sizeof(buf));
 790 
 791         // Found the full path to libjvm.so.
 792         // Now cut the path to <java_home>/jre if we can.
 793         *(strrchr(buf, '/')) = '\0';  /* get rid of /libjvm.so */
 794         pslash = strrchr(buf, '/');
 795         if (pslash != NULL)
 796             *pslash = '\0';           /* get rid of /{client|server|hotspot} */
 797         dll_path = malloc(strlen(buf) + 1);
 798         if (dll_path == NULL)
 799             return;
 800         strcpy(dll_path, buf);
 801         Arguments::set_dll_dir(dll_path);
 802 
 803         if (pslash != NULL) {
 804             pslash = strrchr(buf, '/');
 805             if (pslash != NULL) {
 806                 *pslash = '\0';       /* get rid of /<arch> */
 807                 pslash = strrchr(buf, '/');
 808                 if (pslash != NULL)
 809                     *pslash = '\0';   /* get rid of /lib */
 810             }
 811         }
 812 
 813         home_path = malloc(strlen(buf) + 1);
 814         if (home_path == NULL)
 815             return;
 816         strcpy(home_path, buf);
 817         Arguments::set_java_home(home_path);
 818 
 819         if (!set_boot_path('/', ':'))
 820             return;
 821     }
 822 
 823     /*
 824      * Where to look for native libraries
 825      */
 826     {
 827       // Use dlinfo() to determine the correct java.library.path.
 828       //
 829       // If we're launched by the Java launcher, and the user
 830       // does not set java.library.path explicitly on the commandline,
 831       // the Java launcher sets LD_LIBRARY_PATH for us and unsets
 832       // LD_LIBRARY_PATH_32 and LD_LIBRARY_PATH_64.  In this case
 833       // dlinfo returns LD_LIBRARY_PATH + crle settings (including
 834       // /usr/lib), which is exactly what we want.
 835       //
 836       // If the user does set java.library.path, it completely
 837       // overwrites this setting, and always has.
 838       //
 839       // If we're not launched by the Java launcher, we may
 840       // get here with any/all of the LD_LIBRARY_PATH[_32|64]
 841       // settings.  Again, dlinfo does exactly what we want.
 842 
 843       Dl_serinfo     _info, *info = &_info;
 844       Dl_serpath     *path;
 845       char*          library_path;
 846       char           *common_path;
 847       int            i;
 848 
 849       // determine search path count and required buffer size
 850       if (dlinfo(RTLD_SELF, RTLD_DI_SERINFOSIZE, (void *)info) == -1) {
 851         vm_exit_during_initialization("dlinfo SERINFOSIZE request", dlerror());
 852       }
 853 
 854       // allocate new buffer and initialize
 855       info = (Dl_serinfo*)malloc(_info.dls_size);
 856       if (info == NULL) {
 857         vm_exit_out_of_memory(_info.dls_size,
 858                               "init_system_properties_values info");
 859       }
 860       info->dls_size = _info.dls_size;
 861       info->dls_cnt = _info.dls_cnt;
 862 
 863       // obtain search path information
 864       if (dlinfo(RTLD_SELF, RTLD_DI_SERINFO, (void *)info) == -1) {
 865         free(info);
 866         vm_exit_during_initialization("dlinfo SERINFO request", dlerror());
 867       }
 868 
 869       path = &info->dls_serpath[0];
 870 
 871       // Note: Due to a legacy implementation, most of the library path
 872       // is set in the launcher.  This was to accomodate linking restrictions
 873       // on legacy Solaris implementations (which are no longer supported).
 874       // Eventually, all the library path setting will be done here.
 875       //
 876       // However, to prevent the proliferation of improperly built native
 877       // libraries, the new path component /usr/jdk/packages is added here.
 878 
 879       // Determine the actual CPU architecture.
 880       char cpu_arch[12];
 881       sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
 882 #ifdef _LP64
 883       // If we are a 64-bit vm, perform the following translations:
 884       //   sparc   -> sparcv9
 885       //   i386    -> amd64
 886       if (strcmp(cpu_arch, "sparc") == 0)
 887         strcat(cpu_arch, "v9");
 888       else if (strcmp(cpu_arch, "i386") == 0)
 889         strcpy(cpu_arch, "amd64");
 890 #endif
 891 
 892       // Construct the invariant part of ld_library_path. Note that the
 893       // space for the colon and the trailing null are provided by the
 894       // nulls included by the sizeof operator.
 895       size_t bufsize = sizeof(COMMON_DIR) + sizeof("/lib/") + strlen(cpu_arch);
 896       common_path = malloc(bufsize);
 897       if (common_path == NULL) {
 898         free(info);
 899         vm_exit_out_of_memory(bufsize,
 900                               "init_system_properties_values common_path");
 901       }
 902       sprintf(common_path, COMMON_DIR "/lib/%s", cpu_arch);
 903 
 904       // struct size is more than sufficient for the path components obtained
 905       // through the dlinfo() call, so only add additional space for the path
 906       // components explicitly added here.
 907       bufsize = info->dls_size + strlen(common_path);
 908       library_path = malloc(bufsize);
 909       if (library_path == NULL) {
 910         free(info);
 911         free(common_path);
 912         vm_exit_out_of_memory(bufsize,
 913                               "init_system_properties_values library_path");
 914       }
 915       library_path[0] = '\0';
 916 
 917       // Construct the desired Java library path from the linker's library
 918       // search path.
 919       //
 920       // For compatibility, it is optimal that we insert the additional path
 921       // components specific to the Java VM after those components specified
 922       // in LD_LIBRARY_PATH (if any) but before those added by the ld.so
 923       // infrastructure.
 924       if (info->dls_cnt == 0) { // Not sure this can happen, but allow for it
 925         strcpy(library_path, common_path);
 926       } else {
 927         int inserted = 0;
 928         for (i = 0; i < info->dls_cnt; i++, path++) {
 929           uint_t flags = path->dls_flags & LA_SER_MASK;
 930           if (((flags & LA_SER_LIBPATH) == 0) && !inserted) {
 931             strcat(library_path, common_path);
 932             strcat(library_path, os::path_separator());
 933             inserted = 1;
 934           }
 935           strcat(library_path, path->dls_name);
 936           strcat(library_path, os::path_separator());
 937         }
 938         // eliminate trailing path separator
 939         library_path[strlen(library_path)-1] = '\0';
 940       }
 941 
 942       // happens before argument parsing - can't use a trace flag
 943       // tty->print_raw("init_system_properties_values: native lib path: ");
 944       // tty->print_raw_cr(library_path);
 945 
 946       // callee copies into its own buffer
 947       Arguments::set_library_path(library_path);
 948 
 949       free(common_path);
 950       free(library_path);
 951       free(info);
 952     }
 953 
 954     /*
 955      * Extensions directories.
 956      *
 957      * Note that the space for the colon and the trailing null are provided
 958      * by the nulls included by the sizeof operator (so actually one byte more
 959      * than necessary is allocated).
 960      */
 961     {
 962         char *buf = (char *) malloc(strlen(Arguments::get_java_home()) +
 963             sizeof(EXTENSIONS_DIR) + sizeof(COMMON_DIR) +
 964             sizeof(EXTENSIONS_DIR));
 965         sprintf(buf, "%s" EXTENSIONS_DIR ":" COMMON_DIR EXTENSIONS_DIR,
 966             Arguments::get_java_home());
 967         Arguments::set_ext_dirs(buf);
 968     }
 969 
 970     /* Endorsed standards default directory. */
 971     {
 972         char * buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
 973         sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
 974         Arguments::set_endorsed_dirs(buf);
 975     }
 976   }
 977 
 978 #undef malloc
 979 #undef free
 980 #undef getenv
 981 #undef EXTENSIONS_DIR
 982 #undef ENDORSED_DIR
 983 #undef COMMON_DIR
 984 
 985 }
 986 
 987 void os::breakpoint() {
 988   BREAKPOINT;
 989 }
 990 
 991 bool os::obsolete_option(const JavaVMOption *option)
 992 {
 993   if (!strncmp(option->optionString, "-Xt", 3)) {
 994     return true;
 995   } else if (!strncmp(option->optionString, "-Xtm", 4)) {
 996     return true;
 997   } else if (!strncmp(option->optionString, "-Xverifyheap", 12)) {
 998     return true;
 999   } else if (!strncmp(option->optionString, "-Xmaxjitcodesize", 16)) {
1000     return true;
1001   }
1002   return false;
1003 }
1004 
1005 bool os::Solaris::valid_stack_address(Thread* thread, address sp) {
1006   address  stackStart  = (address)thread->stack_base();
1007   address  stackEnd    = (address)(stackStart - (address)thread->stack_size());
1008   if (sp < stackStart && sp >= stackEnd ) return true;
1009   return false;
1010 }
1011 
1012 extern "C" void breakpoint() {
1013   // use debugger to set breakpoint here
1014 }
1015 
1016 static thread_t main_thread;
1017 
1018 // Thread start routine for all new Java threads
1019 extern "C" void* java_start(void* thread_addr) {
1020   // Try to randomize the cache line index of hot stack frames.
1021   // This helps when threads of the same stack traces evict each other's
1022   // cache lines. The threads can be either from the same JVM instance, or
1023   // from different JVM instances. The benefit is especially true for
1024   // processors with hyperthreading technology.
1025   static int counter = 0;
1026   int pid = os::current_process_id();
1027   alloca(((pid ^ counter++) & 7) * 128);
1028 
1029   int prio;
1030   Thread* thread = (Thread*)thread_addr;
1031   OSThread* osthr = thread->osthread();
1032 
1033   osthr->set_lwp_id( _lwp_self() );  // Store lwp in case we are bound
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 the creator called set priority before we started,
1044   // we need to call set_native_priority now that we have an lwp.
1045   // We used to get the priority from thr_getprio (we called
1046   // thr_setprio way back in create_thread) and pass it to
1047   // set_native_priority, but Solaris scales the priority
1048   // in java_to_os_priority, so when we read it back here,
1049   // we pass trash to set_native_priority instead of what's
1050   // in java_to_os_priority. So we save the native priority
1051   // in the osThread and recall it here.
1052 
1053   if ( osthr->thread_id() != -1 ) {
1054     if ( UseThreadPriorities ) {
1055       int prio = osthr->native_priority();
1056       if (ThreadPriorityVerbose) {
1057         tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is "
1058                       INTPTR_FORMAT ", setting priority: %d\n",
1059                       osthr->thread_id(), osthr->lwp_id(), prio);
1060       }
1061       os::set_native_priority(thread, prio);
1062     }
1063   } else if (ThreadPriorityVerbose) {
1064     warning("Can't set priority in _start routine, thread id hasn't been set\n");
1065   }
1066 
1067   assert(osthr->get_state() == RUNNABLE, "invalid os thread state");
1068 
1069   // initialize signal mask for this thread
1070   os::Solaris::hotspot_sigmask(thread);
1071 
1072   thread->run();
1073 
1074   // One less thread is executing
1075   // When the VMThread gets here, the main thread may have already exited
1076   // which frees the CodeHeap containing the Atomic::dec code
1077   if (thread != VMThread::vm_thread() && VMThread::vm_thread() != NULL) {
1078     Atomic::dec(&os::Solaris::_os_thread_count);
1079   }
1080 
1081   if (UseDetachedThreads) {
1082     thr_exit(NULL);
1083     ShouldNotReachHere();
1084   }
1085   return NULL;
1086 }
1087 
1088 static OSThread* create_os_thread(Thread* thread, thread_t thread_id) {
1089   // Allocate the OSThread object
1090   OSThread* osthread = new OSThread(NULL, NULL);
1091   if (osthread == NULL) return NULL;
1092 
1093   // Store info on the Solaris thread into the OSThread
1094   osthread->set_thread_id(thread_id);
1095   osthread->set_lwp_id(_lwp_self());
1096   thread->_schedctl = (void *) schedctl_init () ;
1097 
1098   if (UseNUMA) {
1099     int lgrp_id = os::numa_get_group_id();
1100     if (lgrp_id != -1) {
1101       thread->set_lgrp_id(lgrp_id);
1102     }
1103   }
1104 
1105   if ( ThreadPriorityVerbose ) {
1106     tty->print_cr("In create_os_thread, Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT "\n",
1107                   osthread->thread_id(), osthread->lwp_id() );
1108   }
1109 
1110   // Initial thread state is INITIALIZED, not SUSPENDED
1111   osthread->set_state(INITIALIZED);
1112 
1113   return osthread;
1114 }
1115 
1116 void os::Solaris::hotspot_sigmask(Thread* thread) {
1117 
1118   //Save caller's signal mask
1119   sigset_t sigmask;
1120   thr_sigsetmask(SIG_SETMASK, NULL, &sigmask);
1121   OSThread *osthread = thread->osthread();
1122   osthread->set_caller_sigmask(sigmask);
1123 
1124   thr_sigsetmask(SIG_UNBLOCK, os::Solaris::unblocked_signals(), NULL);
1125   if (!ReduceSignalUsage) {
1126     if (thread->is_VM_thread()) {
1127       // Only the VM thread handles BREAK_SIGNAL ...
1128       thr_sigsetmask(SIG_UNBLOCK, vm_signals(), NULL);
1129     } else {
1130       // ... all other threads block BREAK_SIGNAL
1131       assert(!sigismember(vm_signals(), SIGINT), "SIGINT should not be blocked");
1132       thr_sigsetmask(SIG_BLOCK, vm_signals(), NULL);
1133     }
1134   }
1135 }
1136 
1137 bool os::create_attached_thread(JavaThread* thread) {
1138 #ifdef ASSERT
1139   thread->verify_not_published();
1140 #endif
1141   OSThread* osthread = create_os_thread(thread, thr_self());
1142   if (osthread == NULL) {
1143      return false;
1144   }
1145 
1146   // Initial thread state is RUNNABLE
1147   osthread->set_state(RUNNABLE);
1148   thread->set_osthread(osthread);
1149 
1150   // initialize signal mask for this thread
1151   // and save the caller's signal mask
1152   os::Solaris::hotspot_sigmask(thread);
1153 
1154   return true;
1155 }
1156 
1157 bool os::create_main_thread(JavaThread* thread) {
1158 #ifdef ASSERT
1159   thread->verify_not_published();
1160 #endif
1161   if (_starting_thread == NULL) {
1162     _starting_thread = create_os_thread(thread, main_thread);
1163      if (_starting_thread == NULL) {
1164         return false;
1165      }
1166   }
1167 
1168   // The primodial thread is runnable from the start
1169   _starting_thread->set_state(RUNNABLE);
1170 
1171   thread->set_osthread(_starting_thread);
1172 
1173   // initialize signal mask for this thread
1174   // and save the caller's signal mask
1175   os::Solaris::hotspot_sigmask(thread);
1176 
1177   return true;
1178 }
1179 
1180 // _T2_libthread is true if we believe we are running with the newer
1181 // SunSoft lwp/libthread.so (2.8 patch, 2.9 default)
1182 bool os::Solaris::_T2_libthread = false;
1183 
1184 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
1185   // Allocate the OSThread object
1186   OSThread* osthread = new OSThread(NULL, NULL);
1187   if (osthread == NULL) {
1188     return false;
1189   }
1190 
1191   if ( ThreadPriorityVerbose ) {
1192     char *thrtyp;
1193     switch ( thr_type ) {
1194       case vm_thread:
1195         thrtyp = (char *)"vm";
1196         break;
1197       case cgc_thread:
1198         thrtyp = (char *)"cgc";
1199         break;
1200       case pgc_thread:
1201         thrtyp = (char *)"pgc";
1202         break;
1203       case java_thread:
1204         thrtyp = (char *)"java";
1205         break;
1206       case compiler_thread:
1207         thrtyp = (char *)"compiler";
1208         break;
1209       case watcher_thread:
1210         thrtyp = (char *)"watcher";
1211         break;
1212       default:
1213         thrtyp = (char *)"unknown";
1214         break;
1215     }
1216     tty->print_cr("In create_thread, creating a %s thread\n", thrtyp);
1217   }
1218 
1219   // Calculate stack size if it's not specified by caller.
1220   if (stack_size == 0) {
1221     // The default stack size 1M (2M for LP64).
1222     stack_size = (BytesPerWord >> 2) * K * K;
1223 
1224     switch (thr_type) {
1225     case os::java_thread:
1226       // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
1227       if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
1228       break;
1229     case os::compiler_thread:
1230       if (CompilerThreadStackSize > 0) {
1231         stack_size = (size_t)(CompilerThreadStackSize * K);
1232         break;
1233       } // else fall through:
1234         // use VMThreadStackSize if CompilerThreadStackSize is not defined
1235     case os::vm_thread:
1236     case os::pgc_thread:
1237     case os::cgc_thread:
1238     case os::watcher_thread:
1239       if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
1240       break;
1241     }
1242   }
1243   stack_size = MAX2(stack_size, os::Solaris::min_stack_allowed);
1244 
1245   // Initial state is ALLOCATED but not INITIALIZED
1246   osthread->set_state(ALLOCATED);
1247 
1248   if (os::Solaris::_os_thread_count > os::Solaris::_os_thread_limit) {
1249     // We got lots of threads. Check if we still have some address space left.
1250     // Need to be at least 5Mb of unreserved address space. We do check by
1251     // trying to reserve some.
1252     const size_t VirtualMemoryBangSize = 20*K*K;
1253     char* mem = os::reserve_memory(VirtualMemoryBangSize);
1254     if (mem == NULL) {
1255       delete osthread;
1256       return false;
1257     } else {
1258       // Release the memory again
1259       os::release_memory(mem, VirtualMemoryBangSize);
1260     }
1261   }
1262 
1263   // Setup osthread because the child thread may need it.
1264   thread->set_osthread(osthread);
1265 
1266   // Create the Solaris thread
1267   // explicit THR_BOUND for T2_libthread case in case
1268   // that assumption is not accurate, but our alternate signal stack
1269   // handling is based on it which must have bound threads
1270   thread_t tid = 0;
1271   long     flags = (UseDetachedThreads ? THR_DETACHED : 0) | THR_SUSPENDED
1272                    | ((UseBoundThreads || os::Solaris::T2_libthread() ||
1273                        (thr_type == vm_thread) ||
1274                        (thr_type == cgc_thread) ||
1275                        (thr_type == pgc_thread) ||
1276                        (thr_type == compiler_thread && BackgroundCompilation)) ?
1277                       THR_BOUND : 0);
1278   int      status;
1279 
1280   // 4376845 -- libthread/kernel don't provide enough LWPs to utilize all CPUs.
1281   //
1282   // On multiprocessors systems, libthread sometimes under-provisions our
1283   // process with LWPs.  On a 30-way systems, for instance, we could have
1284   // 50 user-level threads in ready state and only 2 or 3 LWPs assigned
1285   // to our process.  This can result in under utilization of PEs.
1286   // I suspect the problem is related to libthread's LWP
1287   // pool management and to the kernel's SIGBLOCKING "last LWP parked"
1288   // upcall policy.
1289   //
1290   // The following code is palliative -- it attempts to ensure that our
1291   // process has sufficient LWPs to take advantage of multiple PEs.
1292   // Proper long-term cures include using user-level threads bound to LWPs
1293   // (THR_BOUND) or using LWP-based synchronization.  Note that there is a
1294   // slight timing window with respect to sampling _os_thread_count, but
1295   // the race is benign.  Also, we should periodically recompute
1296   // _processors_online as the min of SC_NPROCESSORS_ONLN and the
1297   // the number of PEs in our partition.  You might be tempted to use
1298   // THR_NEW_LWP here, but I'd recommend against it as that could
1299   // result in undesirable growth of the libthread's LWP pool.
1300   // The fix below isn't sufficient; for instance, it doesn't take into count
1301   // LWPs parked on IO.  It does, however, help certain CPU-bound benchmarks.
1302   //
1303   // Some pathologies this scheme doesn't handle:
1304   // *  Threads can block, releasing the LWPs.  The LWPs can age out.
1305   //    When a large number of threads become ready again there aren't
1306   //    enough LWPs available to service them.  This can occur when the
1307   //    number of ready threads oscillates.
1308   // *  LWPs/Threads park on IO, thus taking the LWP out of circulation.
1309   //
1310   // Finally, we should call thr_setconcurrency() periodically to refresh
1311   // the LWP pool and thwart the LWP age-out mechanism.
1312   // The "+3" term provides a little slop -- we want to slightly overprovision.
1313 
1314   if (AdjustConcurrency && os::Solaris::_os_thread_count < (_processors_online+3)) {
1315     if (!(flags & THR_BOUND)) {
1316       thr_setconcurrency (os::Solaris::_os_thread_count);       // avoid starvation
1317     }
1318   }
1319   // Although this doesn't hurt, we should warn of undefined behavior
1320   // when using unbound T1 threads with schedctl().  This should never
1321   // happen, as the compiler and VM threads are always created bound
1322   DEBUG_ONLY(
1323       if ((VMThreadHintNoPreempt || CompilerThreadHintNoPreempt) &&
1324           (!os::Solaris::T2_libthread() && (!(flags & THR_BOUND))) &&
1325           ((thr_type == vm_thread) || (thr_type == cgc_thread) ||
1326            (thr_type == pgc_thread) || (thr_type == compiler_thread && BackgroundCompilation))) {
1327          warning("schedctl behavior undefined when Compiler/VM/GC Threads are Unbound");
1328       }
1329   );
1330 
1331 
1332   // Mark that we don't have an lwp or thread id yet.
1333   // In case we attempt to set the priority before the thread starts.
1334   osthread->set_lwp_id(-1);
1335   osthread->set_thread_id(-1);
1336 
1337   status = thr_create(NULL, stack_size, java_start, thread, flags, &tid);
1338   if (status != 0) {
1339     if (PrintMiscellaneous && (Verbose || WizardMode)) {
1340       perror("os::create_thread");
1341     }
1342     thread->set_osthread(NULL);
1343     // Need to clean up stuff we've allocated so far
1344     delete osthread;
1345     return false;
1346   }
1347 
1348   Atomic::inc(&os::Solaris::_os_thread_count);
1349 
1350   // Store info on the Solaris thread into the OSThread
1351   osthread->set_thread_id(tid);
1352 
1353   // Remember that we created this thread so we can set priority on it
1354   osthread->set_vm_created();
1355 
1356   // Set the default thread priority.  If using bound threads, setting
1357   // lwp priority will be delayed until thread start.
1358   set_native_priority(thread,
1359                       DefaultThreadPriority == -1 ?
1360                         java_to_os_priority[NormPriority] :
1361                         DefaultThreadPriority);
1362 
1363   // Initial thread state is INITIALIZED, not SUSPENDED
1364   osthread->set_state(INITIALIZED);
1365 
1366   // The thread is returned suspended (in state INITIALIZED), and is started higher up in the call chain
1367   return true;
1368 }
1369 
1370 /* defined for >= Solaris 10. This allows builds on earlier versions
1371  *  of Solaris to take advantage of the newly reserved Solaris JVM signals
1372  *  With SIGJVM1, SIGJVM2, INTERRUPT_SIGNAL is SIGJVM1, ASYNC_SIGNAL is SIGJVM2
1373  *  and -XX:+UseAltSigs does nothing since these should have no conflict
1374  */
1375 #if !defined(SIGJVM1)
1376 #define SIGJVM1 39
1377 #define SIGJVM2 40
1378 #endif
1379 
1380 debug_only(static bool signal_sets_initialized = false);
1381 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
1382 int os::Solaris::_SIGinterrupt = INTERRUPT_SIGNAL;
1383 int os::Solaris::_SIGasync = ASYNC_SIGNAL;
1384 
1385 bool os::Solaris::is_sig_ignored(int sig) {
1386       struct sigaction oact;
1387       sigaction(sig, (struct sigaction*)NULL, &oact);
1388       void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
1389                                      : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
1390       if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
1391            return true;
1392       else
1393            return false;
1394 }
1395 
1396 // Note: SIGRTMIN is a macro that calls sysconf() so it will
1397 // dynamically detect SIGRTMIN value for the system at runtime, not buildtime
1398 static bool isJVM1available() {
1399   return SIGJVM1 < SIGRTMIN;
1400 }
1401 
1402 void os::Solaris::signal_sets_init() {
1403   // Should also have an assertion stating we are still single-threaded.
1404   assert(!signal_sets_initialized, "Already initialized");
1405   // Fill in signals that are necessarily unblocked for all threads in
1406   // the VM. Currently, we unblock the following signals:
1407   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
1408   //                         by -Xrs (=ReduceSignalUsage));
1409   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
1410   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
1411   // the dispositions or masks wrt these signals.
1412   // Programs embedding the VM that want to use the above signals for their
1413   // own purposes must, at this time, use the "-Xrs" option to prevent
1414   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
1415   // (See bug 4345157, and other related bugs).
1416   // In reality, though, unblocking these signals is really a nop, since
1417   // these signals are not blocked by default.
1418   sigemptyset(&unblocked_sigs);
1419   sigemptyset(&allowdebug_blocked_sigs);
1420   sigaddset(&unblocked_sigs, SIGILL);
1421   sigaddset(&unblocked_sigs, SIGSEGV);
1422   sigaddset(&unblocked_sigs, SIGBUS);
1423   sigaddset(&unblocked_sigs, SIGFPE);
1424 
1425   if (isJVM1available) {
1426     os::Solaris::set_SIGinterrupt(SIGJVM1);
1427     os::Solaris::set_SIGasync(SIGJVM2);
1428   } else if (UseAltSigs) {
1429     os::Solaris::set_SIGinterrupt(ALT_INTERRUPT_SIGNAL);
1430     os::Solaris::set_SIGasync(ALT_ASYNC_SIGNAL);
1431   } else {
1432     os::Solaris::set_SIGinterrupt(INTERRUPT_SIGNAL);
1433     os::Solaris::set_SIGasync(ASYNC_SIGNAL);
1434   }
1435 
1436   sigaddset(&unblocked_sigs, os::Solaris::SIGinterrupt());
1437   sigaddset(&unblocked_sigs, os::Solaris::SIGasync());
1438 
1439   if (!ReduceSignalUsage) {
1440    if (!os::Solaris::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
1441       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
1442       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
1443    }
1444    if (!os::Solaris::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
1445       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
1446       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
1447    }
1448    if (!os::Solaris::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
1449       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
1450       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
1451    }
1452   }
1453   // Fill in signals that are blocked by all but the VM thread.
1454   sigemptyset(&vm_sigs);
1455   if (!ReduceSignalUsage)
1456     sigaddset(&vm_sigs, BREAK_SIGNAL);
1457   debug_only(signal_sets_initialized = true);
1458 
1459   // For diagnostics only used in run_periodic_checks
1460   sigemptyset(&check_signal_done);
1461 }
1462 
1463 // These are signals that are unblocked while a thread is running Java.
1464 // (For some reason, they get blocked by default.)
1465 sigset_t* os::Solaris::unblocked_signals() {
1466   assert(signal_sets_initialized, "Not initialized");
1467   return &unblocked_sigs;
1468 }
1469 
1470 // These are the signals that are blocked while a (non-VM) thread is
1471 // running Java. Only the VM thread handles these signals.
1472 sigset_t* os::Solaris::vm_signals() {
1473   assert(signal_sets_initialized, "Not initialized");
1474   return &vm_sigs;
1475 }
1476 
1477 // These are signals that are blocked during cond_wait to allow debugger in
1478 sigset_t* os::Solaris::allowdebug_blocked_signals() {
1479   assert(signal_sets_initialized, "Not initialized");
1480   return &allowdebug_blocked_sigs;
1481 }
1482 
1483 
1484 void _handle_uncaught_cxx_exception() {
1485   VMError err("An uncaught C++ exception");
1486   err.report_and_die();
1487 }
1488 
1489 
1490 // First crack at OS-specific initialization, from inside the new thread.
1491 void os::initialize_thread() {
1492   int r = thr_main() ;
1493   guarantee (r == 0 || r == 1, "CR6501650 or CR6493689") ;
1494   if (r) {
1495     JavaThread* jt = (JavaThread *)Thread::current();
1496     assert(jt != NULL,"Sanity check");
1497     size_t stack_size;
1498     address base = jt->stack_base();
1499     if (Arguments::created_by_java_launcher()) {
1500       // Use 2MB to allow for Solaris 7 64 bit mode.
1501       stack_size = JavaThread::stack_size_at_create() == 0
1502         ? 2048*K : JavaThread::stack_size_at_create();
1503 
1504       // There are rare cases when we may have already used more than
1505       // the basic stack size allotment before this method is invoked.
1506       // Attempt to allow for a normally sized java_stack.
1507       size_t current_stack_offset = (size_t)(base - (address)&stack_size);
1508       stack_size += ReservedSpace::page_align_size_down(current_stack_offset);
1509     } else {
1510       // 6269555: If we were not created by a Java launcher, i.e. if we are
1511       // running embedded in a native application, treat the primordial thread
1512       // as much like a native attached thread as possible.  This means using
1513       // the current stack size from thr_stksegment(), unless it is too large
1514       // to reliably setup guard pages.  A reasonable max size is 8MB.
1515       size_t current_size = current_stack_size();
1516       // This should never happen, but just in case....
1517       if (current_size == 0) current_size = 2 * K * K;
1518       stack_size = current_size > (8 * K * K) ? (8 * K * K) : current_size;
1519     }
1520     address bottom = (address)align_size_up((intptr_t)(base - stack_size), os::vm_page_size());;
1521     stack_size = (size_t)(base - bottom);
1522 
1523     assert(stack_size > 0, "Stack size calculation problem");
1524 
1525     if (stack_size > jt->stack_size()) {
1526       NOT_PRODUCT(
1527         struct rlimit limits;
1528         getrlimit(RLIMIT_STACK, &limits);
1529         size_t size = adjust_stack_size(base, (size_t)limits.rlim_cur);
1530         assert(size >= jt->stack_size(), "Stack size problem in main thread");
1531       )
1532       tty->print_cr(
1533         "Stack size of %d Kb exceeds current limit of %d Kb.\n"
1534         "(Stack sizes are rounded up to a multiple of the system page size.)\n"
1535         "See limit(1) to increase the stack size limit.",
1536         stack_size / K, jt->stack_size() / K);
1537       vm_exit(1);
1538     }
1539     assert(jt->stack_size() >= stack_size,
1540           "Attempt to map more stack than was allocated");
1541     jt->set_stack_size(stack_size);
1542   }
1543 
1544    // 5/22/01: Right now alternate signal stacks do not handle
1545    // throwing stack overflow exceptions, see bug 4463178
1546    // Until a fix is found for this, T2 will NOT imply alternate signal
1547    // stacks.
1548    // If using T2 libthread threads, install an alternate signal stack.
1549    // Because alternate stacks associate with LWPs on Solaris,
1550    // see sigaltstack(2), if using UNBOUND threads, or if UseBoundThreads
1551    // we prefer to explicitly stack bang.
1552    // If not using T2 libthread, but using UseBoundThreads any threads
1553    // (primordial thread, jni_attachCurrentThread) we do not create,
1554    // probably are not bound, therefore they can not have an alternate
1555    // signal stack. Since our stack banging code is generated and
1556    // is shared across threads, all threads must be bound to allow
1557    // using alternate signal stacks.  The alternative is to interpose
1558    // on _lwp_create to associate an alt sig stack with each LWP,
1559    // and this could be a problem when the JVM is embedded.
1560    // We would prefer to use alternate signal stacks with T2
1561    // Since there is currently no accurate way to detect T2
1562    // we do not. Assuming T2 when running T1 causes sig 11s or assertions
1563    // on installing alternate signal stacks
1564 
1565 
1566    // 05/09/03: removed alternate signal stack support for Solaris
1567    // The alternate signal stack mechanism is no longer needed to
1568    // handle stack overflow. This is now handled by allocating
1569    // guard pages (red zone) and stackbanging.
1570    // Initially the alternate signal stack mechanism was removed because
1571    // it did not work with T1 llibthread. Alternate
1572    // signal stacks MUST have all threads bound to lwps. Applications
1573    // can create their own threads and attach them without their being
1574    // bound under T1. This is frequently the case for the primordial thread.
1575    // If we were ever to reenable this mechanism we would need to
1576    // use the dynamic check for T2 libthread.
1577 
1578   os::Solaris::init_thread_fpu_state();
1579   std::set_terminate(_handle_uncaught_cxx_exception);
1580 }
1581 
1582 
1583 
1584 // Free Solaris resources related to the OSThread
1585 void os::free_thread(OSThread* osthread) {
1586   assert(osthread != NULL, "os::free_thread but osthread not set");
1587 
1588 
1589   // We are told to free resources of the argument thread,
1590   // but we can only really operate on the current thread.
1591   // The main thread must take the VMThread down synchronously
1592   // before the main thread exits and frees up CodeHeap
1593   guarantee((Thread::current()->osthread() == osthread
1594      || (osthread == VMThread::vm_thread()->osthread())), "os::free_thread but not current thread");
1595   if (Thread::current()->osthread() == osthread) {
1596     // Restore caller's signal mask
1597     sigset_t sigmask = osthread->caller_sigmask();
1598     thr_sigsetmask(SIG_SETMASK, &sigmask, NULL);
1599   }
1600   delete osthread;
1601 }
1602 
1603 void os::pd_start_thread(Thread* thread) {
1604   int status = thr_continue(thread->osthread()->thread_id());
1605   assert_status(status == 0, status, "thr_continue failed");
1606 }
1607 
1608 
1609 intx os::current_thread_id() {
1610   return (intx)thr_self();
1611 }
1612 
1613 static pid_t _initial_pid = 0;
1614 
1615 int os::current_process_id() {
1616   return (int)(_initial_pid ? _initial_pid : getpid());
1617 }
1618 
1619 int os::allocate_thread_local_storage() {
1620   // %%%       in Win32 this allocates a memory segment pointed to by a
1621   //           register.  Dan Stein can implement a similar feature in
1622   //           Solaris.  Alternatively, the VM can do the same thing
1623   //           explicitly: malloc some storage and keep the pointer in a
1624   //           register (which is part of the thread's context) (or keep it
1625   //           in TLS).
1626   // %%%       In current versions of Solaris, thr_self and TSD can
1627   //           be accessed via short sequences of displaced indirections.
1628   //           The value of thr_self is available as %g7(36).
1629   //           The value of thr_getspecific(k) is stored in %g7(12)(4)(k*4-4),
1630   //           assuming that the current thread already has a value bound to k.
1631   //           It may be worth experimenting with such access patterns,
1632   //           and later having the parameters formally exported from a Solaris
1633   //           interface.  I think, however, that it will be faster to
1634   //           maintain the invariant that %g2 always contains the
1635   //           JavaThread in Java code, and have stubs simply
1636   //           treat %g2 as a caller-save register, preserving it in a %lN.
1637   thread_key_t tk;
1638   if (thr_keycreate( &tk, NULL ) )
1639     fatal(err_msg("os::allocate_thread_local_storage: thr_keycreate failed "
1640                   "(%s)", strerror(errno)));
1641   return int(tk);
1642 }
1643 
1644 void os::free_thread_local_storage(int index) {
1645   // %%% don't think we need anything here
1646   // if ( pthread_key_delete((pthread_key_t) tk) )
1647   //   fatal("os::free_thread_local_storage: pthread_key_delete failed");
1648 }
1649 
1650 #define SMALLINT 32   // libthread allocate for tsd_common is a version specific
1651                       // small number - point is NO swap space available
1652 void os::thread_local_storage_at_put(int index, void* value) {
1653   // %%% this is used only in threadLocalStorage.cpp
1654   if (thr_setspecific((thread_key_t)index, value)) {
1655     if (errno == ENOMEM) {
1656        vm_exit_out_of_memory(SMALLINT, "thr_setspecific: out of swap space");
1657     } else {
1658       fatal(err_msg("os::thread_local_storage_at_put: thr_setspecific failed "
1659                     "(%s)", strerror(errno)));
1660     }
1661   } else {
1662       ThreadLocalStorage::set_thread_in_slot ((Thread *) value) ;
1663   }
1664 }
1665 
1666 // This function could be called before TLS is initialized, for example, when
1667 // VM receives an async signal or when VM causes a fatal error during
1668 // initialization. Return NULL if thr_getspecific() fails.
1669 void* os::thread_local_storage_at(int index) {
1670   // %%% this is used only in threadLocalStorage.cpp
1671   void* r = NULL;
1672   return thr_getspecific((thread_key_t)index, &r) != 0 ? NULL : r;
1673 }
1674 
1675 
1676 // gethrtime can move backwards if read from one cpu and then a different cpu
1677 // getTimeNanos is guaranteed to not move backward on Solaris
1678 // local spinloop created as faster for a CAS on an int than
1679 // a CAS on a 64bit jlong. Also Atomic::cmpxchg for jlong is not
1680 // supported on sparc v8 or pre supports_cx8 intel boxes.
1681 // oldgetTimeNanos for systems which do not support CAS on 64bit jlong
1682 // i.e. sparc v8 and pre supports_cx8 (i486) intel boxes
1683 inline hrtime_t oldgetTimeNanos() {
1684   int gotlock = LOCK_INVALID;
1685   hrtime_t newtime = gethrtime();
1686 
1687   for (;;) {
1688 // grab lock for max_hrtime
1689     int curlock = max_hrtime_lock;
1690     if (curlock & LOCK_BUSY)  continue;
1691     if (gotlock = Atomic::cmpxchg(LOCK_BUSY, &max_hrtime_lock, LOCK_FREE) != LOCK_FREE) continue;
1692     if (newtime > max_hrtime) {
1693       max_hrtime = newtime;
1694     } else {
1695       newtime = max_hrtime;
1696     }
1697     // release lock
1698     max_hrtime_lock = LOCK_FREE;
1699     return newtime;
1700   }
1701 }
1702 // gethrtime can move backwards if read from one cpu and then a different cpu
1703 // getTimeNanos is guaranteed to not move backward on Solaris
1704 inline hrtime_t getTimeNanos() {
1705   if (VM_Version::supports_cx8()) {
1706     const hrtime_t now = gethrtime();
1707     // Use atomic long load since 32-bit x86 uses 2 registers to keep long.
1708     const hrtime_t prev = Atomic::load((volatile jlong*)&max_hrtime);
1709     if (now <= prev)  return prev;   // same or retrograde time;
1710     const hrtime_t obsv = Atomic::cmpxchg(now, (volatile jlong*)&max_hrtime, prev);
1711     assert(obsv >= prev, "invariant");   // Monotonicity
1712     // If the CAS succeeded then we're done and return "now".
1713     // If the CAS failed and the observed value "obs" is >= now then
1714     // we should return "obs".  If the CAS failed and now > obs > prv then
1715     // some other thread raced this thread and installed a new value, in which case
1716     // we could either (a) retry the entire operation, (b) retry trying to install now
1717     // or (c) just return obs.  We use (c).   No loop is required although in some cases
1718     // we might discard a higher "now" value in deference to a slightly lower but freshly
1719     // installed obs value.   That's entirely benign -- it admits no new orderings compared
1720     // to (a) or (b) -- and greatly reduces coherence traffic.
1721     // We might also condition (c) on the magnitude of the delta between obs and now.
1722     // Avoiding excessive CAS operations to hot RW locations is critical.
1723     // See http://blogs.sun.com/dave/entry/cas_and_cache_trivia_invalidate
1724     return (prev == obsv) ? now : obsv ;
1725   } else {
1726     return oldgetTimeNanos();
1727   }
1728 }
1729 
1730 // Time since start-up in seconds to a fine granularity.
1731 // Used by VMSelfDestructTimer and the MemProfiler.
1732 double os::elapsedTime() {
1733   return (double)(getTimeNanos() - first_hrtime) / (double)hrtime_hz;
1734 }
1735 
1736 jlong os::elapsed_counter() {
1737   return (jlong)(getTimeNanos() - first_hrtime);
1738 }
1739 
1740 jlong os::elapsed_frequency() {
1741    return hrtime_hz;
1742 }
1743 
1744 // Return the real, user, and system times in seconds from an
1745 // arbitrary fixed point in the past.
1746 bool os::getTimesSecs(double* process_real_time,
1747                   double* process_user_time,
1748                   double* process_system_time) {
1749   struct tms ticks;
1750   clock_t real_ticks = times(&ticks);
1751 
1752   if (real_ticks == (clock_t) (-1)) {
1753     return false;
1754   } else {
1755     double ticks_per_second = (double) clock_tics_per_sec;
1756     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1757     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1758     // For consistency return the real time from getTimeNanos()
1759     // converted to seconds.
1760     *process_real_time = ((double) getTimeNanos()) / ((double) NANOUNITS);
1761 
1762     return true;
1763   }
1764 }
1765 
1766 bool os::supports_vtime() { return true; }
1767 
1768 bool os::enable_vtime() {
1769   int fd = ::open("/proc/self/ctl", O_WRONLY);
1770   if (fd == -1)
1771     return false;
1772 
1773   long cmd[] = { PCSET, PR_MSACCT };
1774   int res = ::write(fd, cmd, sizeof(long) * 2);
1775   ::close(fd);
1776   if (res != sizeof(long) * 2)
1777     return false;
1778 
1779   return true;
1780 }
1781 
1782 bool os::vtime_enabled() {
1783   int fd = ::open("/proc/self/status", O_RDONLY);
1784   if (fd == -1)
1785     return false;
1786 
1787   pstatus_t status;
1788   int res = os::read(fd, (void*) &status, sizeof(pstatus_t));
1789   ::close(fd);
1790   if (res != sizeof(pstatus_t))
1791     return false;
1792 
1793   return status.pr_flags & PR_MSACCT;
1794 }
1795 
1796 double os::elapsedVTime() {
1797   return (double)gethrvtime() / (double)hrtime_hz;
1798 }
1799 
1800 // Used internally for comparisons only
1801 // getTimeMillis guaranteed to not move backwards on Solaris
1802 jlong getTimeMillis() {
1803   jlong nanotime = getTimeNanos();
1804   return (jlong)(nanotime / NANOSECS_PER_MILLISEC);
1805 }
1806 
1807 // Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1808 jlong os::javaTimeMillis() {
1809   timeval t;
1810   if (gettimeofday( &t, NULL) == -1)
1811     fatal(err_msg("os::javaTimeMillis: gettimeofday (%s)", strerror(errno)));
1812   return jlong(t.tv_sec) * 1000  +  jlong(t.tv_usec) / 1000;
1813 }
1814 
1815 jlong os::javaTimeNanos() {
1816   return (jlong)getTimeNanos();
1817 }
1818 
1819 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1820   info_ptr->max_value = ALL_64_BITS;      // gethrtime() uses all 64 bits
1821   info_ptr->may_skip_backward = false;    // not subject to resetting or drifting
1822   info_ptr->may_skip_forward = false;     // not subject to resetting or drifting
1823   info_ptr->kind = JVMTI_TIMER_ELAPSED;   // elapsed not CPU time
1824 }
1825 
1826 char * os::local_time_string(char *buf, size_t buflen) {
1827   struct tm t;
1828   time_t long_time;
1829   time(&long_time);
1830   localtime_r(&long_time, &t);
1831   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1832                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1833                t.tm_hour, t.tm_min, t.tm_sec);
1834   return buf;
1835 }
1836 
1837 // Note: os::shutdown() might be called very early during initialization, or
1838 // called from signal handler. Before adding something to os::shutdown(), make
1839 // sure it is async-safe and can handle partially initialized VM.
1840 void os::shutdown() {
1841 
1842   // allow PerfMemory to attempt cleanup of any persistent resources
1843   perfMemory_exit();
1844 
1845   // needs to remove object in file system
1846   AttachListener::abort();
1847 
1848   // flush buffered output, finish log files
1849   ostream_abort();
1850 
1851   // Check for abort hook
1852   abort_hook_t abort_hook = Arguments::abort_hook();
1853   if (abort_hook != NULL) {
1854     abort_hook();
1855   }
1856 }
1857 
1858 // Note: os::abort() might be called very early during initialization, or
1859 // called from signal handler. Before adding something to os::abort(), make
1860 // sure it is async-safe and can handle partially initialized VM.
1861 void os::abort(bool dump_core) {
1862   os::shutdown();
1863   if (dump_core) {
1864 #ifndef PRODUCT
1865     fdStream out(defaultStream::output_fd());
1866     out.print_raw("Current thread is ");
1867     char buf[16];
1868     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1869     out.print_raw_cr(buf);
1870     out.print_raw_cr("Dumping core ...");
1871 #endif
1872     ::abort(); // dump core (for debugging)
1873   }
1874 
1875   ::exit(1);
1876 }
1877 
1878 // Die immediately, no exit hook, no abort hook, no cleanup.
1879 void os::die() {
1880   _exit(-1);
1881 }
1882 
1883 // unused
1884 void os::set_error_file(const char *logfile) {}
1885 
1886 // DLL functions
1887 
1888 const char* os::dll_file_extension() { return ".so"; }
1889 
1890 // This must be hard coded because it's the system's temporary
1891 // directory not the java application's temp directory, ala java.io.tmpdir.
1892 const char* os::get_temp_directory() { return "/tmp"; }
1893 
1894 static bool file_exists(const char* filename) {
1895   struct stat statbuf;
1896   if (filename == NULL || strlen(filename) == 0) {
1897     return false;
1898   }
1899   return os::stat(filename, &statbuf) == 0;
1900 }
1901 
1902 void os::dll_build_name(char* buffer, size_t buflen,
1903                         const char* pname, const char* fname) {
1904   const size_t pnamelen = pname ? strlen(pname) : 0;
1905 
1906   // Quietly truncate on buffer overflow.  Should be an error.
1907   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1908     *buffer = '\0';
1909     return;
1910   }
1911 
1912   if (pnamelen == 0) {
1913     snprintf(buffer, buflen, "lib%s.so", fname);
1914   } else if (strchr(pname, *os::path_separator()) != NULL) {
1915     int n;
1916     char** pelements = split_path(pname, &n);
1917     for (int i = 0 ; i < n ; i++) {
1918       // really shouldn't be NULL but what the heck, check can't hurt
1919       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1920         continue; // skip the empty path values
1921       }
1922       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1923       if (file_exists(buffer)) {
1924         break;
1925       }
1926     }
1927     // release the storage
1928     for (int i = 0 ; i < n ; i++) {
1929       if (pelements[i] != NULL) {
1930         FREE_C_HEAP_ARRAY(char, pelements[i]);
1931       }
1932     }
1933     if (pelements != NULL) {
1934       FREE_C_HEAP_ARRAY(char*, pelements);
1935     }
1936   } else {
1937     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1938   }
1939 }
1940 
1941 const char* os::get_current_directory(char *buf, int buflen) {
1942   return getcwd(buf, buflen);
1943 }
1944 
1945 // check if addr is inside libjvm[_g].so
1946 bool os::address_is_in_vm(address addr) {
1947   static address libjvm_base_addr;
1948   Dl_info dlinfo;
1949 
1950   if (libjvm_base_addr == NULL) {
1951     dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1952     libjvm_base_addr = (address)dlinfo.dli_fbase;
1953     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1954   }
1955 
1956   if (dladdr((void *)addr, &dlinfo)) {
1957     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1958   }
1959 
1960   return false;
1961 }
1962 
1963 typedef int (*dladdr1_func_type) (void *, Dl_info *, void **, int);
1964 static dladdr1_func_type dladdr1_func = NULL;
1965 
1966 bool os::dll_address_to_function_name(address addr, char *buf,
1967                                       int buflen, int * offset) {
1968   Dl_info dlinfo;
1969 
1970   // dladdr1_func was initialized in os::init()
1971   if (dladdr1_func){
1972       // yes, we have dladdr1
1973 
1974       // Support for dladdr1 is checked at runtime; it may be
1975       // available even if the vm is built on a machine that does
1976       // not have dladdr1 support.  Make sure there is a value for
1977       // RTLD_DL_SYMENT.
1978       #ifndef RTLD_DL_SYMENT
1979       #define RTLD_DL_SYMENT 1
1980       #endif
1981 #ifdef _LP64
1982       Elf64_Sym * info;
1983 #else
1984       Elf32_Sym * info;
1985 #endif
1986       if (dladdr1_func((void *)addr, &dlinfo, (void **)&info,
1987                        RTLD_DL_SYMENT)) {
1988         if ((char *)dlinfo.dli_saddr + info->st_size > (char *)addr) {
1989           if (buf != NULL) {
1990             if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen))
1991               jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1992             }
1993             if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1994             return true;
1995         }
1996       }
1997       if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1998         if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1999            buf, buflen, offset, dlinfo.dli_fname)) {
2000           return true;
2001         }
2002       }
2003       if (buf != NULL) buf[0] = '\0';
2004       if (offset != NULL) *offset  = -1;
2005       return false;
2006   } else {
2007       // no, only dladdr is available
2008       if (dladdr((void *)addr, &dlinfo)) {
2009         if (buf != NULL) {
2010           if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen))
2011             jio_snprintf(buf, buflen, dlinfo.dli_sname);
2012         }
2013         if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
2014         return true;
2015       } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
2016         if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
2017           buf, buflen, offset, dlinfo.dli_fname)) {
2018           return true;
2019         }
2020       }
2021       if (buf != NULL) buf[0] = '\0';
2022       if (offset != NULL) *offset  = -1;
2023       return false;
2024   }
2025 }
2026 
2027 bool os::dll_address_to_library_name(address addr, char* buf,
2028                                      int buflen, int* offset) {
2029   Dl_info dlinfo;
2030 
2031   if (dladdr((void*)addr, &dlinfo)){
2032      if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
2033      if (offset) *offset = addr - (address)dlinfo.dli_fbase;
2034      return true;
2035   } else {
2036      if (buf) buf[0] = '\0';
2037      if (offset) *offset = -1;
2038      return false;
2039   }
2040 }
2041 
2042 // Prints the names and full paths of all opened dynamic libraries
2043 // for current process
2044 void os::print_dll_info(outputStream * st) {
2045     Dl_info dli;
2046     void *handle;
2047     Link_map *map;
2048     Link_map *p;
2049 
2050     st->print_cr("Dynamic libraries:"); st->flush();
2051 
2052     if (!dladdr(CAST_FROM_FN_PTR(void *, os::print_dll_info), &dli)) {
2053         st->print_cr("Error: Cannot print dynamic libraries.");
2054         return;
2055     }
2056     handle = dlopen(dli.dli_fname, RTLD_LAZY);
2057     if (handle == NULL) {
2058         st->print_cr("Error: Cannot print dynamic libraries.");
2059         return;
2060     }
2061     dlinfo(handle, RTLD_DI_LINKMAP, &map);
2062     if (map == NULL) {
2063         st->print_cr("Error: Cannot print dynamic libraries.");
2064         return;
2065     }
2066 
2067     while (map->l_prev != NULL)
2068         map = map->l_prev;
2069 
2070     while (map != NULL) {
2071         st->print_cr(PTR_FORMAT " \t%s", map->l_addr, map->l_name);
2072         map = map->l_next;
2073     }
2074 
2075     dlclose(handle);
2076 }
2077 
2078   // Loads .dll/.so and
2079   // in case of error it checks if .dll/.so was built for the
2080   // same architecture as Hotspot is running on
2081 
2082 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
2083 {
2084   void * result= ::dlopen(filename, RTLD_LAZY);
2085   if (result != NULL) {
2086     // Successful loading
2087     return result;
2088   }
2089 
2090   Elf32_Ehdr elf_head;
2091 
2092   // Read system error message into ebuf
2093   // It may or may not be overwritten below
2094   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
2095   ebuf[ebuflen-1]='\0';
2096   int diag_msg_max_length=ebuflen-strlen(ebuf);
2097   char* diag_msg_buf=ebuf+strlen(ebuf);
2098 
2099   if (diag_msg_max_length==0) {
2100     // No more space in ebuf for additional diagnostics message
2101     return NULL;
2102   }
2103 
2104 
2105   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
2106 
2107   if (file_descriptor < 0) {
2108     // Can't open library, report dlerror() message
2109     return NULL;
2110   }
2111 
2112   bool failed_to_read_elf_head=
2113     (sizeof(elf_head)!=
2114         (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
2115 
2116   ::close(file_descriptor);
2117   if (failed_to_read_elf_head) {
2118     // file i/o error - report dlerror() msg
2119     return NULL;
2120   }
2121 
2122   typedef struct {
2123     Elf32_Half  code;         // Actual value as defined in elf.h
2124     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
2125     char        elf_class;    // 32 or 64 bit
2126     char        endianess;    // MSB or LSB
2127     char*       name;         // String representation
2128   } arch_t;
2129 
2130   static const arch_t arch_array[]={
2131     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2132     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
2133     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
2134     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
2135     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2136     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
2137     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
2138     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
2139     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
2140     {EM_ARM,         EM_ARM,     ELFCLASS32, ELFDATA2LSB, (char*)"ARM 32"}
2141   };
2142 
2143   #if  (defined IA32)
2144     static  Elf32_Half running_arch_code=EM_386;
2145   #elif   (defined AMD64)
2146     static  Elf32_Half running_arch_code=EM_X86_64;
2147   #elif  (defined IA64)
2148     static  Elf32_Half running_arch_code=EM_IA_64;
2149   #elif  (defined __sparc) && (defined _LP64)
2150     static  Elf32_Half running_arch_code=EM_SPARCV9;
2151   #elif  (defined __sparc) && (!defined _LP64)
2152     static  Elf32_Half running_arch_code=EM_SPARC;
2153   #elif  (defined __powerpc64__)
2154     static  Elf32_Half running_arch_code=EM_PPC64;
2155   #elif  (defined __powerpc__)
2156     static  Elf32_Half running_arch_code=EM_PPC;
2157   #elif (defined ARM)
2158     static  Elf32_Half running_arch_code=EM_ARM;
2159   #else
2160     #error Method os::dll_load requires that one of following is defined:\
2161          IA32, AMD64, IA64, __sparc, __powerpc__, ARM, ARM
2162   #endif
2163 
2164   // Identify compatability class for VM's architecture and library's architecture
2165   // Obtain string descriptions for architectures
2166 
2167   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2168   int running_arch_index=-1;
2169 
2170   for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2171     if (running_arch_code == arch_array[i].code) {
2172       running_arch_index    = i;
2173     }
2174     if (lib_arch.code == arch_array[i].code) {
2175       lib_arch.compat_class = arch_array[i].compat_class;
2176       lib_arch.name         = arch_array[i].name;
2177     }
2178   }
2179 
2180   assert(running_arch_index != -1,
2181     "Didn't find running architecture code (running_arch_code) in arch_array");
2182   if (running_arch_index == -1) {
2183     // Even though running architecture detection failed
2184     // we may still continue with reporting dlerror() message
2185     return NULL;
2186   }
2187 
2188   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2189     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2190     return NULL;
2191   }
2192 
2193   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2194     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2195     return NULL;
2196   }
2197 
2198   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2199     if ( lib_arch.name!=NULL ) {
2200       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2201         " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2202         lib_arch.name, arch_array[running_arch_index].name);
2203     } else {
2204       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2205       " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2206         lib_arch.code,
2207         arch_array[running_arch_index].name);
2208     }
2209   }
2210 
2211   return NULL;
2212 }
2213 
2214 void* os::dll_lookup(void* handle, const char* name) {
2215   return dlsym(handle, name);
2216 }
2217 
2218 int os::stat(const char *path, struct stat *sbuf) {
2219   char pathbuf[MAX_PATH];
2220   if (strlen(path) > MAX_PATH - 1) {
2221     errno = ENAMETOOLONG;
2222     return -1;
2223   }
2224   os::native_path(strcpy(pathbuf, path));
2225   return ::stat(pathbuf, sbuf);
2226 }
2227 
2228 static bool _print_ascii_file(const char* filename, outputStream* st) {
2229   int fd = ::open(filename, O_RDONLY);
2230   if (fd == -1) {
2231      return false;
2232   }
2233 
2234   char buf[32];
2235   int bytes;
2236   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2237     st->print_raw(buf, bytes);
2238   }
2239 
2240   ::close(fd);
2241 
2242   return true;
2243 }
2244 
2245 void os::print_os_info(outputStream* st) {
2246   st->print("OS:");
2247 
2248   if (!_print_ascii_file("/etc/release", st)) {
2249     st->print("Solaris");
2250   }
2251   st->cr();
2252 
2253   // kernel
2254   st->print("uname:");
2255   struct utsname name;
2256   uname(&name);
2257   st->print(name.sysname); st->print(" ");
2258   st->print(name.release); st->print(" ");
2259   st->print(name.version); st->print(" ");
2260   st->print(name.machine);
2261 
2262   // libthread
2263   if (os::Solaris::T2_libthread()) st->print("  (T2 libthread)");
2264   else st->print("  (T1 libthread)");
2265   st->cr();
2266 
2267   // rlimit
2268   st->print("rlimit:");
2269   struct rlimit rlim;
2270 
2271   st->print(" STACK ");
2272   getrlimit(RLIMIT_STACK, &rlim);
2273   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2274   else st->print("%uk", rlim.rlim_cur >> 10);
2275 
2276   st->print(", CORE ");
2277   getrlimit(RLIMIT_CORE, &rlim);
2278   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2279   else st->print("%uk", rlim.rlim_cur >> 10);
2280 
2281   st->print(", NOFILE ");
2282   getrlimit(RLIMIT_NOFILE, &rlim);
2283   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2284   else st->print("%d", rlim.rlim_cur);
2285 
2286   st->print(", AS ");
2287   getrlimit(RLIMIT_AS, &rlim);
2288   if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2289   else st->print("%uk", rlim.rlim_cur >> 10);
2290   st->cr();
2291 
2292   // load average
2293   st->print("load average:");
2294   double loadavg[3];
2295   os::loadavg(loadavg, 3);
2296   st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2297   st->cr();
2298 }
2299 
2300 
2301 static bool check_addr0(outputStream* st) {
2302   jboolean status = false;
2303   int fd = ::open("/proc/self/map",O_RDONLY);
2304   if (fd >= 0) {
2305     prmap_t p;
2306     while(::read(fd, &p, sizeof(p)) > 0) {
2307       if (p.pr_vaddr == 0x0) {
2308         st->print("Warning: Address: 0x%x, Size: %dK, ",p.pr_vaddr, p.pr_size/1024, p.pr_mapname);
2309         st->print("Mapped file: %s, ", p.pr_mapname[0] == '\0' ? "None" : p.pr_mapname);
2310         st->print("Access:");
2311         st->print("%s",(p.pr_mflags & MA_READ)  ? "r" : "-");
2312         st->print("%s",(p.pr_mflags & MA_WRITE) ? "w" : "-");
2313         st->print("%s",(p.pr_mflags & MA_EXEC)  ? "x" : "-");
2314         st->cr();
2315         status = true;
2316       }
2317       ::close(fd);
2318     }
2319   }
2320   return status;
2321 }
2322 
2323 void os::pd_print_cpu_info(outputStream* st) {
2324   // Nothing to do for now.
2325 }
2326 
2327 void os::print_memory_info(outputStream* st) {
2328   st->print("Memory:");
2329   st->print(" %dk page", os::vm_page_size()>>10);
2330   st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
2331   st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
2332   st->cr();
2333   (void) check_addr0(st);
2334 }
2335 
2336 // Taken from /usr/include/sys/machsig.h  Supposed to be architecture specific
2337 // but they're the same for all the solaris architectures that we support.
2338 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2339                           "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2340                           "ILL_COPROC", "ILL_BADSTK" };
2341 
2342 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2343                           "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2344                           "FPE_FLTINV", "FPE_FLTSUB" };
2345 
2346 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2347 
2348 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2349 
2350 void os::print_siginfo(outputStream* st, void* siginfo) {
2351   st->print("siginfo:");
2352 
2353   const int buflen = 100;
2354   char buf[buflen];
2355   siginfo_t *si = (siginfo_t*)siginfo;
2356   st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2357   char *err = strerror(si->si_errno);
2358   if (si->si_errno != 0 && err != NULL) {
2359     st->print("si_errno=%s", err);
2360   } else {
2361     st->print("si_errno=%d", si->si_errno);
2362   }
2363   const int c = si->si_code;
2364   assert(c > 0, "unexpected si_code");
2365   switch (si->si_signo) {
2366   case SIGILL:
2367     st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2368     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2369     break;
2370   case SIGFPE:
2371     st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2372     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2373     break;
2374   case SIGSEGV:
2375     st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2376     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2377     break;
2378   case SIGBUS:
2379     st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2380     st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2381     break;
2382   default:
2383     st->print(", si_code=%d", si->si_code);
2384     // no si_addr
2385   }
2386 
2387   if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2388       UseSharedSpaces) {
2389     FileMapInfo* mapinfo = FileMapInfo::current_info();
2390     if (mapinfo->is_in_shared_space(si->si_addr)) {
2391       st->print("\n\nError accessing class data sharing archive."   \
2392                 " Mapped file inaccessible during execution, "      \
2393                 " possible disk/network problem.");
2394     }
2395   }
2396   st->cr();
2397 }
2398 
2399 // Moved from whole group, because we need them here for diagnostic
2400 // prints.
2401 #define OLDMAXSIGNUM 32
2402 static int Maxsignum = 0;
2403 static int *ourSigFlags = NULL;
2404 
2405 extern "C" void sigINTRHandler(int, siginfo_t*, void*);
2406 
2407 int os::Solaris::get_our_sigflags(int sig) {
2408   assert(ourSigFlags!=NULL, "signal data structure not initialized");
2409   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2410   return ourSigFlags[sig];
2411 }
2412 
2413 void os::Solaris::set_our_sigflags(int sig, int flags) {
2414   assert(ourSigFlags!=NULL, "signal data structure not initialized");
2415   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2416   ourSigFlags[sig] = flags;
2417 }
2418 
2419 
2420 static const char* get_signal_handler_name(address handler,
2421                                            char* buf, int buflen) {
2422   int offset;
2423   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
2424   if (found) {
2425     // skip directory names
2426     const char *p1, *p2;
2427     p1 = buf;
2428     size_t len = strlen(os::file_separator());
2429     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
2430     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
2431   } else {
2432     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
2433   }
2434   return buf;
2435 }
2436 
2437 static void print_signal_handler(outputStream* st, int sig,
2438                                   char* buf, size_t buflen) {
2439   struct sigaction sa;
2440 
2441   sigaction(sig, NULL, &sa);
2442 
2443   st->print("%s: ", os::exception_name(sig, buf, buflen));
2444 
2445   address handler = (sa.sa_flags & SA_SIGINFO)
2446                   ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
2447                   : CAST_FROM_FN_PTR(address, sa.sa_handler);
2448 
2449   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
2450     st->print("SIG_DFL");
2451   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
2452     st->print("SIG_IGN");
2453   } else {
2454     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
2455   }
2456 
2457   st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
2458 
2459   address rh = VMError::get_resetted_sighandler(sig);
2460   // May be, handler was resetted by VMError?
2461   if(rh != NULL) {
2462     handler = rh;
2463     sa.sa_flags = VMError::get_resetted_sigflags(sig);
2464   }
2465 
2466   st->print(", sa_flags="   PTR32_FORMAT, sa.sa_flags);
2467 
2468   // Check: is it our handler?
2469   if(handler == CAST_FROM_FN_PTR(address, signalHandler) ||
2470      handler == CAST_FROM_FN_PTR(address, sigINTRHandler)) {
2471     // It is our signal handler
2472     // check for flags
2473     if(sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
2474       st->print(
2475         ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
2476         os::Solaris::get_our_sigflags(sig));
2477     }
2478   }
2479   st->cr();
2480 }
2481 
2482 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2483   st->print_cr("Signal Handlers:");
2484   print_signal_handler(st, SIGSEGV, buf, buflen);
2485   print_signal_handler(st, SIGBUS , buf, buflen);
2486   print_signal_handler(st, SIGFPE , buf, buflen);
2487   print_signal_handler(st, SIGPIPE, buf, buflen);
2488   print_signal_handler(st, SIGXFSZ, buf, buflen);
2489   print_signal_handler(st, SIGILL , buf, buflen);
2490   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2491   print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
2492   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2493   print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
2494   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2495   print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
2496   print_signal_handler(st, os::Solaris::SIGinterrupt(), buf, buflen);
2497   print_signal_handler(st, os::Solaris::SIGasync(), buf, buflen);
2498 }
2499 
2500 static char saved_jvm_path[MAXPATHLEN] = { 0 };
2501 
2502 // Find the full path to the current module, libjvm.so or libjvm_g.so
2503 void os::jvm_path(char *buf, jint buflen) {
2504   // Error checking.
2505   if (buflen < MAXPATHLEN) {
2506     assert(false, "must use a large-enough buffer");
2507     buf[0] = '\0';
2508     return;
2509   }
2510   // Lazy resolve the path to current module.
2511   if (saved_jvm_path[0] != 0) {
2512     strcpy(buf, saved_jvm_path);
2513     return;
2514   }
2515 
2516   Dl_info dlinfo;
2517   int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
2518   assert(ret != 0, "cannot locate libjvm");
2519   realpath((char *)dlinfo.dli_fname, buf);
2520 
2521   if (Arguments::created_by_gamma_launcher()) {
2522     // Support for the gamma launcher.  Typical value for buf is
2523     // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
2524     // the right place in the string, then assume we are installed in a JDK and
2525     // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
2526     // up the path so it looks like libjvm.so is installed there (append a
2527     // fake suffix hotspot/libjvm.so).
2528     const char *p = buf + strlen(buf) - 1;
2529     for (int count = 0; p > buf && count < 5; ++count) {
2530       for (--p; p > buf && *p != '/'; --p)
2531         /* empty */ ;
2532     }
2533 
2534     if (strncmp(p, "/jre/lib/", 9) != 0) {
2535       // Look for JAVA_HOME in the environment.
2536       char* java_home_var = ::getenv("JAVA_HOME");
2537       if (java_home_var != NULL && java_home_var[0] != 0) {
2538         char cpu_arch[12];
2539         char* jrelib_p;
2540         int   len;
2541         sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
2542 #ifdef _LP64
2543         // If we are on sparc running a 64-bit vm, look in jre/lib/sparcv9.
2544         if (strcmp(cpu_arch, "sparc") == 0) {
2545           strcat(cpu_arch, "v9");
2546         } else if (strcmp(cpu_arch, "i386") == 0) {
2547           strcpy(cpu_arch, "amd64");
2548         }
2549 #endif
2550         // Check the current module name "libjvm.so" or "libjvm_g.so".
2551         p = strrchr(buf, '/');
2552         assert(strstr(p, "/libjvm") == p, "invalid library name");
2553         p = strstr(p, "_g") ? "_g" : "";
2554 
2555         realpath(java_home_var, buf);
2556         // determine if this is a legacy image or modules image
2557         // modules image doesn't have "jre" subdirectory
2558         len = strlen(buf);
2559         jrelib_p = buf + len;
2560         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2561         if (0 != access(buf, F_OK)) {
2562           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2563         }
2564 
2565         if (0 == access(buf, F_OK)) {
2566           // Use current module name "libjvm[_g].so" instead of
2567           // "libjvm"debug_only("_g")".so" since for fastdebug version
2568           // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2569           len = strlen(buf);
2570           snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2571         } else {
2572           // Go back to path of .so
2573           realpath((char *)dlinfo.dli_fname, buf);
2574         }
2575       }
2576     }
2577   }
2578 
2579   strcpy(saved_jvm_path, buf);
2580 }
2581 
2582 
2583 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2584   // no prefix required, not even "_"
2585 }
2586 
2587 
2588 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2589   // no suffix required
2590 }
2591 
2592 // This method is a copy of JDK's sysGetLastErrorString
2593 // from src/solaris/hpi/src/system_md.c
2594 
2595 size_t os::lasterror(char *buf, size_t len) {
2596 
2597   if (errno == 0)  return 0;
2598 
2599   const char *s = ::strerror(errno);
2600   size_t n = ::strlen(s);
2601   if (n >= len) {
2602     n = len - 1;
2603   }
2604   ::strncpy(buf, s, n);
2605   buf[n] = '\0';
2606   return n;
2607 }
2608 
2609 
2610 // sun.misc.Signal
2611 
2612 extern "C" {
2613   static void UserHandler(int sig, void *siginfo, void *context) {
2614     // Ctrl-C is pressed during error reporting, likely because the error
2615     // handler fails to abort. Let VM die immediately.
2616     if (sig == SIGINT && is_error_reported()) {
2617        os::die();
2618     }
2619 
2620     os::signal_notify(sig);
2621     // We do not need to reinstate the signal handler each time...
2622   }
2623 }
2624 
2625 void* os::user_handler() {
2626   return CAST_FROM_FN_PTR(void*, UserHandler);
2627 }
2628 
2629 extern "C" {
2630   typedef void (*sa_handler_t)(int);
2631   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2632 }
2633 
2634 void* os::signal(int signal_number, void* handler) {
2635   struct sigaction sigAct, oldSigAct;
2636   sigfillset(&(sigAct.sa_mask));
2637   sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2638   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2639 
2640   if (sigaction(signal_number, &sigAct, &oldSigAct))
2641     // -1 means registration failed
2642     return (void *)-1;
2643 
2644   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2645 }
2646 
2647 void os::signal_raise(int signal_number) {
2648   raise(signal_number);
2649 }
2650 
2651 /*
2652  * The following code is moved from os.cpp for making this
2653  * code platform specific, which it is by its very nature.
2654  */
2655 
2656 // a counter for each possible signal value
2657 static int Sigexit = 0;
2658 static int Maxlibjsigsigs;
2659 static jint *pending_signals = NULL;
2660 static int *preinstalled_sigs = NULL;
2661 static struct sigaction *chainedsigactions = NULL;
2662 static sema_t sig_sem;
2663 typedef int (*version_getting_t)();
2664 version_getting_t os::Solaris::get_libjsig_version = NULL;
2665 static int libjsigversion = NULL;
2666 
2667 int os::sigexitnum_pd() {
2668   assert(Sigexit > 0, "signal memory not yet initialized");
2669   return Sigexit;
2670 }
2671 
2672 void os::Solaris::init_signal_mem() {
2673   // Initialize signal structures
2674   Maxsignum = SIGRTMAX;
2675   Sigexit = Maxsignum+1;
2676   assert(Maxsignum >0, "Unable to obtain max signal number");
2677 
2678   Maxlibjsigsigs = Maxsignum;
2679 
2680   // pending_signals has one int per signal
2681   // The additional signal is for SIGEXIT - exit signal to signal_thread
2682   pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1));
2683   memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2684 
2685   if (UseSignalChaining) {
2686      chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2687        * (Maxsignum + 1));
2688      memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2689      preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1));
2690      memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2691   }
2692   ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1 ));
2693   memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2694 }
2695 
2696 void os::signal_init_pd() {
2697   int ret;
2698 
2699   ret = ::sema_init(&sig_sem, 0, NULL, NULL);
2700   assert(ret == 0, "sema_init() failed");
2701 }
2702 
2703 void os::signal_notify(int signal_number) {
2704   int ret;
2705 
2706   Atomic::inc(&pending_signals[signal_number]);
2707   ret = ::sema_post(&sig_sem);
2708   assert(ret == 0, "sema_post() failed");
2709 }
2710 
2711 static int check_pending_signals(bool wait_for_signal) {
2712   int ret;
2713   while (true) {
2714     for (int i = 0; i < Sigexit + 1; i++) {
2715       jint n = pending_signals[i];
2716       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2717         return i;
2718       }
2719     }
2720     if (!wait_for_signal) {
2721       return -1;
2722     }
2723     JavaThread *thread = JavaThread::current();
2724     ThreadBlockInVM tbivm(thread);
2725 
2726     bool threadIsSuspended;
2727     do {
2728       thread->set_suspend_equivalent();
2729       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2730       while((ret = ::sema_wait(&sig_sem)) == EINTR)
2731           ;
2732       assert(ret == 0, "sema_wait() failed");
2733 
2734       // were we externally suspended while we were waiting?
2735       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2736       if (threadIsSuspended) {
2737         //
2738         // The semaphore has been incremented, but while we were waiting
2739         // another thread suspended us. We don't want to continue running
2740         // while suspended because that would surprise the thread that
2741         // suspended us.
2742         //
2743         ret = ::sema_post(&sig_sem);
2744         assert(ret == 0, "sema_post() failed");
2745 
2746         thread->java_suspend_self();
2747       }
2748     } while (threadIsSuspended);
2749   }
2750 }
2751 
2752 int os::signal_lookup() {
2753   return check_pending_signals(false);
2754 }
2755 
2756 int os::signal_wait() {
2757   return check_pending_signals(true);
2758 }
2759 
2760 ////////////////////////////////////////////////////////////////////////////////
2761 // Virtual Memory
2762 
2763 static int page_size = -1;
2764 
2765 // The mmap MAP_ALIGN flag is supported on Solaris 9 and later.  init_2() will
2766 // clear this var if support is not available.
2767 static bool has_map_align = true;
2768 
2769 int os::vm_page_size() {
2770   assert(page_size != -1, "must call os::init");
2771   return page_size;
2772 }
2773 
2774 // Solaris allocates memory by pages.
2775 int os::vm_allocation_granularity() {
2776   assert(page_size != -1, "must call os::init");
2777   return page_size;
2778 }
2779 
2780 bool os::commit_memory(char* addr, size_t bytes, bool exec) {
2781   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2782   size_t size = bytes;
2783   char *res = Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2784   if (res != NULL) {
2785     if (UseNUMAInterleaving) {
2786       numa_make_global(addr, bytes);
2787     }
2788     return true;
2789   }
2790   return false;
2791 }
2792 
2793 bool os::commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2794                        bool exec) {
2795   if (commit_memory(addr, bytes, exec)) {
2796     if (UseMPSS && alignment_hint > (size_t)vm_page_size()) {
2797       // If the large page size has been set and the VM
2798       // is using large pages, use the large page size
2799       // if it is smaller than the alignment hint. This is
2800       // a case where the VM wants to use a larger alignment size
2801       // for its own reasons but still want to use large pages
2802       // (which is what matters to setting the mpss range.
2803       size_t page_size = 0;
2804       if (large_page_size() < alignment_hint) {
2805         assert(UseLargePages, "Expected to be here for large page use only");
2806         page_size = large_page_size();
2807       } else {
2808         // If the alignment hint is less than the large page
2809         // size, the VM wants a particular alignment (thus the hint)
2810         // for internal reasons.  Try to set the mpss range using
2811         // the alignment_hint.
2812         page_size = alignment_hint;
2813       }
2814       // Since this is a hint, ignore any failures.
2815       (void)Solaris::set_mpss_range(addr, bytes, page_size);
2816     }
2817     return true;
2818   }
2819   return false;
2820 }
2821 
2822 // Uncommit the pages in a specified region.
2823 void os::free_memory(char* addr, size_t bytes, size_t alignment_hint) {
2824   if (madvise(addr, bytes, MADV_FREE) < 0) {
2825     debug_only(warning("MADV_FREE failed."));
2826     return;
2827   }
2828 }
2829 
2830 bool os::create_stack_guard_pages(char* addr, size_t size) {
2831   return os::commit_memory(addr, size);
2832 }
2833 
2834 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2835   return os::uncommit_memory(addr, size);
2836 }
2837 
2838 // Change the page size in a given range.
2839 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2840   assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2841   assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2842   if (UseLargePages && UseMPSS) {
2843     Solaris::set_mpss_range(addr, bytes, alignment_hint);
2844   }
2845 }
2846 
2847 // Tell the OS to make the range local to the first-touching LWP
2848 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2849   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2850   if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2851     debug_only(warning("MADV_ACCESS_LWP failed."));
2852   }
2853 }
2854 
2855 // Tell the OS that this range would be accessed from different LWPs.
2856 void os::numa_make_global(char *addr, size_t bytes) {
2857   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2858   if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2859     debug_only(warning("MADV_ACCESS_MANY failed."));
2860   }
2861 }
2862 
2863 // Get the number of the locality groups.
2864 size_t os::numa_get_groups_num() {
2865   size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2866   return n != -1 ? n : 1;
2867 }
2868 
2869 // Get a list of leaf locality groups. A leaf lgroup is group that
2870 // doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2871 // board. An LWP is assigned to one of these groups upon creation.
2872 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2873    if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2874      ids[0] = 0;
2875      return 1;
2876    }
2877    int result_size = 0, top = 1, bottom = 0, cur = 0;
2878    for (int k = 0; k < size; k++) {
2879      int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2880                                     (Solaris::lgrp_id_t*)&ids[top], size - top);
2881      if (r == -1) {
2882        ids[0] = 0;
2883        return 1;
2884      }
2885      if (!r) {
2886        // That's a leaf node.
2887        assert (bottom <= cur, "Sanity check");
2888        // Check if the node has memory
2889        if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2890                                    NULL, 0, LGRP_RSRC_MEM) > 0) {
2891          ids[bottom++] = ids[cur];
2892        }
2893      }
2894      top += r;
2895      cur++;
2896    }
2897    if (bottom == 0) {
2898      // Handle a situation, when the OS reports no memory available.
2899      // Assume UMA architecture.
2900      ids[0] = 0;
2901      return 1;
2902    }
2903    return bottom;
2904 }
2905 
2906 // Detect the topology change. Typically happens during CPU plugging-unplugging.
2907 bool os::numa_topology_changed() {
2908   int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2909   if (is_stale != -1 && is_stale) {
2910     Solaris::lgrp_fini(Solaris::lgrp_cookie());
2911     Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2912     assert(c != 0, "Failure to initialize LGRP API");
2913     Solaris::set_lgrp_cookie(c);
2914     return true;
2915   }
2916   return false;
2917 }
2918 
2919 // Get the group id of the current LWP.
2920 int os::numa_get_group_id() {
2921   int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2922   if (lgrp_id == -1) {
2923     return 0;
2924   }
2925   const int size = os::numa_get_groups_num();
2926   int *ids = (int*)alloca(size * sizeof(int));
2927 
2928   // Get the ids of all lgroups with memory; r is the count.
2929   int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2930                                   (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2931   if (r <= 0) {
2932     return 0;
2933   }
2934   return ids[os::random() % r];
2935 }
2936 
2937 // Request information about the page.
2938 bool os::get_page_info(char *start, page_info* info) {
2939   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2940   uint64_t addr = (uintptr_t)start;
2941   uint64_t outdata[2];
2942   uint_t validity = 0;
2943 
2944   if (os::Solaris::meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2945     return false;
2946   }
2947 
2948   info->size = 0;
2949   info->lgrp_id = -1;
2950 
2951   if ((validity & 1) != 0) {
2952     if ((validity & 2) != 0) {
2953       info->lgrp_id = outdata[0];
2954     }
2955     if ((validity & 4) != 0) {
2956       info->size = outdata[1];
2957     }
2958     return true;
2959   }
2960   return false;
2961 }
2962 
2963 // Scan the pages from start to end until a page different than
2964 // the one described in the info parameter is encountered.
2965 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2966   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2967   const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2968   uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT];
2969   uint_t validity[MAX_MEMINFO_CNT];
2970 
2971   size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2972   uint64_t p = (uint64_t)start;
2973   while (p < (uint64_t)end) {
2974     addrs[0] = p;
2975     size_t addrs_count = 1;
2976     while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] < (uint64_t)end) {
2977       addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2978       addrs_count++;
2979     }
2980 
2981     if (os::Solaris::meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2982       return NULL;
2983     }
2984 
2985     size_t i = 0;
2986     for (; i < addrs_count; i++) {
2987       if ((validity[i] & 1) != 0) {
2988         if ((validity[i] & 4) != 0) {
2989           if (outdata[types * i + 1] != page_expected->size) {
2990             break;
2991           }
2992         } else
2993           if (page_expected->size != 0) {
2994             break;
2995           }
2996 
2997         if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
2998           if (outdata[types * i] != page_expected->lgrp_id) {
2999             break;
3000           }
3001         }
3002       } else {
3003         return NULL;
3004       }
3005     }
3006 
3007     if (i != addrs_count) {
3008       if ((validity[i] & 2) != 0) {
3009         page_found->lgrp_id = outdata[types * i];
3010       } else {
3011         page_found->lgrp_id = -1;
3012       }
3013       if ((validity[i] & 4) != 0) {
3014         page_found->size = outdata[types * i + 1];
3015       } else {
3016         page_found->size = 0;
3017       }
3018       return (char*)addrs[i];
3019     }
3020 
3021     p = addrs[addrs_count - 1] + page_size;
3022   }
3023   return end;
3024 }
3025 
3026 bool os::uncommit_memory(char* addr, size_t bytes) {
3027   size_t size = bytes;
3028   // Map uncommitted pages PROT_NONE so we fail early if we touch an
3029   // uncommitted page. Otherwise, the read/write might succeed if we
3030   // have enough swap space to back the physical page.
3031   return
3032     NULL != Solaris::mmap_chunk(addr, size,
3033                                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
3034                                 PROT_NONE);
3035 }
3036 
3037 char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
3038   char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
3039 
3040   if (b == MAP_FAILED) {
3041     return NULL;
3042   }
3043   return b;
3044 }
3045 
3046 char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes, size_t alignment_hint, bool fixed) {
3047   char* addr = requested_addr;
3048   int flags = MAP_PRIVATE | MAP_NORESERVE;
3049 
3050   assert(!(fixed && (alignment_hint > 0)), "alignment hint meaningless with fixed mmap");
3051 
3052   if (fixed) {
3053     flags |= MAP_FIXED;
3054   } else if (has_map_align && (alignment_hint > (size_t) vm_page_size())) {
3055     flags |= MAP_ALIGN;
3056     addr = (char*) alignment_hint;
3057   }
3058 
3059   // Map uncommitted pages PROT_NONE so we fail early if we touch an
3060   // uncommitted page. Otherwise, the read/write might succeed if we
3061   // have enough swap space to back the physical page.
3062   return mmap_chunk(addr, bytes, flags, PROT_NONE);
3063 }
3064 
3065 char* os::reserve_memory(size_t bytes, char* requested_addr, size_t alignment_hint) {
3066   char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint, (requested_addr != NULL));
3067 
3068   guarantee(requested_addr == NULL || requested_addr == addr,
3069             "OS failed to return requested mmap address.");
3070   return addr;
3071 }
3072 
3073 // Reserve memory at an arbitrary address, only if that area is
3074 // available (and not reserved for something else).
3075 
3076 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3077   const int max_tries = 10;
3078   char* base[max_tries];
3079   size_t size[max_tries];
3080 
3081   // Solaris adds a gap between mmap'ed regions.  The size of the gap
3082   // is dependent on the requested size and the MMU.  Our initial gap
3083   // value here is just a guess and will be corrected later.
3084   bool had_top_overlap = false;
3085   bool have_adjusted_gap = false;
3086   size_t gap = 0x400000;
3087 
3088   // Assert only that the size is a multiple of the page size, since
3089   // that's all that mmap requires, and since that's all we really know
3090   // about at this low abstraction level.  If we need higher alignment,
3091   // we can either pass an alignment to this method or verify alignment
3092   // in one of the methods further up the call chain.  See bug 5044738.
3093   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3094 
3095   // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
3096   // Give it a try, if the kernel honors the hint we can return immediately.
3097   char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
3098   volatile int err = errno;
3099   if (addr == requested_addr) {
3100     return addr;
3101   } else if (addr != NULL) {
3102     unmap_memory(addr, bytes);
3103   }
3104 
3105   if (PrintMiscellaneous && Verbose) {
3106     char buf[256];
3107     buf[0] = '\0';
3108     if (addr == NULL) {
3109       jio_snprintf(buf, sizeof(buf), ": %s", strerror(err));
3110     }
3111     warning("attempt_reserve_memory_at: couldn't reserve " SIZE_FORMAT " bytes at "
3112             PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
3113             "%s", bytes, requested_addr, addr, buf);
3114   }
3115 
3116   // Address hint method didn't work.  Fall back to the old method.
3117   // In theory, once SNV becomes our oldest supported platform, this
3118   // code will no longer be needed.
3119   //
3120   // Repeatedly allocate blocks until the block is allocated at the
3121   // right spot. Give up after max_tries.
3122   int i;
3123   for (i = 0; i < max_tries; ++i) {
3124     base[i] = reserve_memory(bytes);
3125 
3126     if (base[i] != NULL) {
3127       // Is this the block we wanted?
3128       if (base[i] == requested_addr) {
3129         size[i] = bytes;
3130         break;
3131       }
3132 
3133       // check that the gap value is right
3134       if (had_top_overlap && !have_adjusted_gap) {
3135         size_t actual_gap = base[i-1] - base[i] - bytes;
3136         if (gap != actual_gap) {
3137           // adjust the gap value and retry the last 2 allocations
3138           assert(i > 0, "gap adjustment code problem");
3139           have_adjusted_gap = true;  // adjust the gap only once, just in case
3140           gap = actual_gap;
3141           if (PrintMiscellaneous && Verbose) {
3142             warning("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
3143           }
3144           unmap_memory(base[i], bytes);
3145           unmap_memory(base[i-1], size[i-1]);
3146           i-=2;
3147           continue;
3148         }
3149       }
3150 
3151       // Does this overlap the block we wanted? Give back the overlapped
3152       // parts and try again.
3153       //
3154       // There is still a bug in this code: if top_overlap == bytes,
3155       // the overlap is offset from requested region by the value of gap.
3156       // In this case giving back the overlapped part will not work,
3157       // because we'll give back the entire block at base[i] and
3158       // therefore the subsequent allocation will not generate a new gap.
3159       // This could be fixed with a new algorithm that used larger
3160       // or variable size chunks to find the requested region -
3161       // but such a change would introduce additional complications.
3162       // It's rare enough that the planets align for this bug,
3163       // so we'll just wait for a fix for 6204603/5003415 which
3164       // will provide a mmap flag to allow us to avoid this business.
3165 
3166       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3167       if (top_overlap >= 0 && top_overlap < bytes) {
3168         had_top_overlap = true;
3169         unmap_memory(base[i], top_overlap);
3170         base[i] += top_overlap;
3171         size[i] = bytes - top_overlap;
3172       } else {
3173         size_t bottom_overlap = base[i] + bytes - requested_addr;
3174         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3175           if (PrintMiscellaneous && Verbose && bottom_overlap == 0) {
3176             warning("attempt_reserve_memory_at: possible alignment bug");
3177           }
3178           unmap_memory(requested_addr, bottom_overlap);
3179           size[i] = bytes - bottom_overlap;
3180         } else {
3181           size[i] = bytes;
3182         }
3183       }
3184     }
3185   }
3186 
3187   // Give back the unused reserved pieces.
3188 
3189   for (int j = 0; j < i; ++j) {
3190     if (base[j] != NULL) {
3191       unmap_memory(base[j], size[j]);
3192     }
3193   }
3194 
3195   return (i < max_tries) ? requested_addr : NULL;
3196 }
3197 
3198 bool os::release_memory(char* addr, size_t bytes) {
3199   size_t size = bytes;
3200   return munmap(addr, size) == 0;
3201 }
3202 
3203 static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
3204   assert(addr == (char*)align_size_down((uintptr_t)addr, os::vm_page_size()),
3205          "addr must be page aligned");
3206   int retVal = mprotect(addr, bytes, prot);
3207   return retVal == 0;
3208 }
3209 
3210 // Protect memory (Used to pass readonly pages through
3211 // JNI GetArray<type>Elements with empty arrays.)
3212 // Also, used for serialization page and for compressed oops null pointer
3213 // checking.
3214 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3215                         bool is_committed) {
3216   unsigned int p = 0;
3217   switch (prot) {
3218   case MEM_PROT_NONE: p = PROT_NONE; break;
3219   case MEM_PROT_READ: p = PROT_READ; break;
3220   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3221   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3222   default:
3223     ShouldNotReachHere();
3224   }
3225   // is_committed is unused.
3226   return solaris_mprotect(addr, bytes, p);
3227 }
3228 
3229 // guard_memory and unguard_memory only happens within stack guard pages.
3230 // Since ISM pertains only to the heap, guard and unguard memory should not
3231 /// happen with an ISM region.
3232 bool os::guard_memory(char* addr, size_t bytes) {
3233   return solaris_mprotect(addr, bytes, PROT_NONE);
3234 }
3235 
3236 bool os::unguard_memory(char* addr, size_t bytes) {
3237   return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
3238 }
3239 
3240 // Large page support
3241 
3242 // UseLargePages is the master flag to enable/disable large page memory.
3243 // UseMPSS and UseISM are supported for compatibility reasons. Their combined
3244 // effects can be described in the following table:
3245 //
3246 // UseLargePages UseMPSS UseISM
3247 //    false         *       *   => UseLargePages is the master switch, turning
3248 //                                 it off will turn off both UseMPSS and
3249 //                                 UseISM. VM will not use large page memory
3250 //                                 regardless the settings of UseMPSS/UseISM.
3251 //     true      false    false => Unless future Solaris provides other
3252 //                                 mechanism to use large page memory, this
3253 //                                 combination is equivalent to -UseLargePages,
3254 //                                 VM will not use large page memory
3255 //     true      true     false => JVM will use MPSS for large page memory.
3256 //                                 This is the default behavior.
3257 //     true      false    true  => JVM will use ISM for large page memory.
3258 //     true      true     true  => JVM will use ISM if it is available.
3259 //                                 Otherwise, JVM will fall back to MPSS.
3260 //                                 Becaues ISM is now available on all
3261 //                                 supported Solaris versions, this combination
3262 //                                 is equivalent to +UseISM -UseMPSS.
3263 
3264 static size_t _large_page_size = 0;
3265 
3266 bool os::Solaris::ism_sanity_check(bool warn, size_t * page_size) {
3267   // x86 uses either 2M or 4M page, depending on whether PAE (Physical Address
3268   // Extensions) mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. Sparc
3269   // can support multiple page sizes.
3270 
3271   // Don't bother to probe page size because getpagesizes() comes with MPSS.
3272   // ISM is only recommended on old Solaris where there is no MPSS support.
3273   // Simply choose a conservative value as default.
3274   *page_size = LargePageSizeInBytes ? LargePageSizeInBytes :
3275                SPARC_ONLY(4 * M) IA32_ONLY(4 * M) AMD64_ONLY(2 * M)
3276                ARM_ONLY(2 * M);
3277 
3278   // ISM is available on all supported Solaris versions
3279   return true;
3280 }
3281 
3282 // Insertion sort for small arrays (descending order).
3283 static void insertion_sort_descending(size_t* array, int len) {
3284   for (int i = 0; i < len; i++) {
3285     size_t val = array[i];
3286     for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
3287       size_t tmp = array[key];
3288       array[key] = array[key - 1];
3289       array[key - 1] = tmp;
3290     }
3291   }
3292 }
3293 
3294 bool os::Solaris::mpss_sanity_check(bool warn, size_t * page_size) {
3295   const unsigned int usable_count = VM_Version::page_size_count();
3296   if (usable_count == 1) {
3297     return false;
3298   }
3299 
3300   // Find the right getpagesizes interface.  When solaris 11 is the minimum
3301   // build platform, getpagesizes() (without the '2') can be called directly.
3302   typedef int (*gps_t)(size_t[], int);
3303   gps_t gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes2"));
3304   if (gps_func == NULL) {
3305     gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes"));
3306     if (gps_func == NULL) {
3307       if (warn) {
3308         warning("MPSS is not supported by the operating system.");
3309       }
3310       return false;
3311     }
3312   }
3313 
3314   // Fill the array of page sizes.
3315   int n = (*gps_func)(_page_sizes, page_sizes_max);
3316   assert(n > 0, "Solaris bug?");
3317 
3318   if (n == page_sizes_max) {
3319     // Add a sentinel value (necessary only if the array was completely filled
3320     // since it is static (zeroed at initialization)).
3321     _page_sizes[--n] = 0;
3322     DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
3323   }
3324   assert(_page_sizes[n] == 0, "missing sentinel");
3325   trace_page_sizes("available page sizes", _page_sizes, n);
3326 
3327   if (n == 1) return false;     // Only one page size available.
3328 
3329   // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
3330   // select up to usable_count elements.  First sort the array, find the first
3331   // acceptable value, then copy the usable sizes to the top of the array and
3332   // trim the rest.  Make sure to include the default page size :-).
3333   //
3334   // A better policy could get rid of the 4M limit by taking the sizes of the
3335   // important VM memory regions (java heap and possibly the code cache) into
3336   // account.
3337   insertion_sort_descending(_page_sizes, n);
3338   const size_t size_limit =
3339     FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
3340   int beg;
3341   for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */ ;
3342   const int end = MIN2((int)usable_count, n) - 1;
3343   for (int cur = 0; cur < end; ++cur, ++beg) {
3344     _page_sizes[cur] = _page_sizes[beg];
3345   }
3346   _page_sizes[end] = vm_page_size();
3347   _page_sizes[end + 1] = 0;
3348 
3349   if (_page_sizes[end] > _page_sizes[end - 1]) {
3350     // Default page size is not the smallest; sort again.
3351     insertion_sort_descending(_page_sizes, end + 1);
3352   }
3353   *page_size = _page_sizes[0];
3354 
3355   trace_page_sizes("usable page sizes", _page_sizes, end + 1);
3356   return true;
3357 }
3358 
3359 void os::large_page_init() {
3360   if (!UseLargePages) {
3361     UseISM = false;
3362     UseMPSS = false;
3363     return;
3364   }
3365 
3366   // print a warning if any large page related flag is specified on command line
3367   bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
3368                          !FLAG_IS_DEFAULT(UseISM)               ||
3369                          !FLAG_IS_DEFAULT(UseMPSS)              ||
3370                          !FLAG_IS_DEFAULT(LargePageSizeInBytes);
3371   UseISM = UseISM &&
3372            Solaris::ism_sanity_check(warn_on_failure, &_large_page_size);
3373   if (UseISM) {
3374     // ISM disables MPSS to be compatible with old JDK behavior
3375     UseMPSS = false;
3376     _page_sizes[0] = _large_page_size;
3377     _page_sizes[1] = vm_page_size();
3378   }
3379 
3380   UseMPSS = UseMPSS &&
3381             Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
3382 
3383   UseLargePages = UseISM || UseMPSS;
3384 }
3385 
3386 bool os::Solaris::set_mpss_range(caddr_t start, size_t bytes, size_t align) {
3387   // Signal to OS that we want large pages for addresses
3388   // from addr, addr + bytes
3389   struct memcntl_mha mpss_struct;
3390   mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
3391   mpss_struct.mha_pagesize = align;
3392   mpss_struct.mha_flags = 0;
3393   if (memcntl(start, bytes, MC_HAT_ADVISE,
3394               (caddr_t) &mpss_struct, 0, 0) < 0) {
3395     debug_only(warning("Attempt to use MPSS failed."));
3396     return false;
3397   }
3398   return true;
3399 }
3400 
3401 char* os::reserve_memory_special(size_t size, char* addr, bool exec) {
3402   // "exec" is passed in but not used.  Creating the shared image for
3403   // the code cache doesn't have an SHM_X executable permission to check.
3404   assert(UseLargePages && UseISM, "only for ISM large pages");
3405 
3406   char* retAddr = NULL;
3407   int shmid;
3408   key_t ismKey;
3409 
3410   bool warn_on_failure = UseISM &&
3411                         (!FLAG_IS_DEFAULT(UseLargePages)         ||
3412                          !FLAG_IS_DEFAULT(UseISM)                ||
3413                          !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3414                         );
3415   char msg[128];
3416 
3417   ismKey = IPC_PRIVATE;
3418 
3419   // Create a large shared memory region to attach to based on size.
3420   // Currently, size is the total size of the heap
3421   shmid = shmget(ismKey, size, SHM_R | SHM_W | IPC_CREAT);
3422   if (shmid == -1){
3423      if (warn_on_failure) {
3424        jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3425        warning(msg);
3426      }
3427      return NULL;
3428   }
3429 
3430   // Attach to the region
3431   retAddr = (char *) shmat(shmid, 0, SHM_SHARE_MMU | SHM_R | SHM_W);
3432   int err = errno;
3433 
3434   // Remove shmid. If shmat() is successful, the actual shared memory segment
3435   // will be deleted when it's detached by shmdt() or when the process
3436   // terminates. If shmat() is not successful this will remove the shared
3437   // segment immediately.
3438   shmctl(shmid, IPC_RMID, NULL);
3439 
3440   if (retAddr == (char *) -1) {
3441     if (warn_on_failure) {
3442       jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3443       warning(msg);
3444     }
3445     return NULL;
3446   }
3447   if ((retAddr != NULL) && UseNUMAInterleaving) {
3448     numa_make_global(retAddr, size);
3449   }
3450   return retAddr;
3451 }
3452 
3453 bool os::release_memory_special(char* base, size_t bytes) {
3454   // detaching the SHM segment will also delete it, see reserve_memory_special()
3455   int rslt = shmdt(base);
3456   return rslt == 0;
3457 }
3458 
3459 size_t os::large_page_size() {
3460   return _large_page_size;
3461 }
3462 
3463 // MPSS allows application to commit large page memory on demand; with ISM
3464 // the entire memory region must be allocated as shared memory.
3465 bool os::can_commit_large_page_memory() {
3466   return UseISM ? false : true;
3467 }
3468 
3469 bool os::can_execute_large_page_memory() {
3470   return UseISM ? false : true;
3471 }
3472 
3473 static int os_sleep(jlong millis, bool interruptible) {
3474   const jlong limit = INT_MAX;
3475   jlong prevtime;
3476   int res;
3477 
3478   while (millis > limit) {
3479     if ((res = os_sleep(limit, interruptible)) != OS_OK)
3480       return res;
3481     millis -= limit;
3482   }
3483 
3484   // Restart interrupted polls with new parameters until the proper delay
3485   // has been completed.
3486 
3487   prevtime = getTimeMillis();
3488 
3489   while (millis > 0) {
3490     jlong newtime;
3491 
3492     if (!interruptible) {
3493       // Following assert fails for os::yield_all:
3494       // assert(!thread->is_Java_thread(), "must not be java thread");
3495       res = poll(NULL, 0, millis);
3496     } else {
3497       JavaThread *jt = JavaThread::current();
3498 
3499       INTERRUPTIBLE_NORESTART_VM_ALWAYS(poll(NULL, 0, millis), res, jt,
3500         os::Solaris::clear_interrupted);
3501     }
3502 
3503     // INTERRUPTIBLE_NORESTART_VM_ALWAYS returns res == OS_INTRPT for
3504     // thread.Interrupt.
3505 
3506     // See c/r 6751923. Poll can return 0 before time
3507     // has elapsed if time is set via clock_settime (as NTP does).
3508     // res == 0 if poll timed out (see man poll RETURN VALUES)
3509     // using the logic below checks that we really did
3510     // sleep at least "millis" if not we'll sleep again.
3511     if( ( res == 0 ) || ((res == OS_ERR) && (errno == EINTR))) {
3512       newtime = getTimeMillis();
3513       assert(newtime >= prevtime, "time moving backwards");
3514     /* Doing prevtime and newtime in microseconds doesn't help precision,
3515        and trying to round up to avoid lost milliseconds can result in a
3516        too-short delay. */
3517       millis -= newtime - prevtime;
3518       if(millis <= 0)
3519         return OS_OK;
3520       prevtime = newtime;
3521     } else
3522       return res;
3523   }
3524 
3525   return OS_OK;
3526 }
3527 
3528 // Read calls from inside the vm need to perform state transitions
3529 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3530   INTERRUPTIBLE_RETURN_INT_VM(::read(fd, buf, nBytes), os::Solaris::clear_interrupted);
3531 }
3532 
3533 size_t os::restartable_read(int fd, void *buf, unsigned int nBytes) {
3534   INTERRUPTIBLE_RETURN_INT(::read(fd, buf, nBytes), os::Solaris::clear_interrupted);
3535 }
3536 
3537 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3538   assert(thread == Thread::current(),  "thread consistency check");
3539 
3540   // TODO-FIXME: this should be removed.
3541   // On Solaris machines (especially 2.5.1) we found that sometimes the VM gets into a live lock
3542   // situation with a JavaThread being starved out of a lwp. The kernel doesn't seem to generate
3543   // a SIGWAITING signal which would enable the threads library to create a new lwp for the starving
3544   // thread. We suspect that because the Watcher thread keeps waking up at periodic intervals the kernel
3545   // is fooled into believing that the system is making progress. In the code below we block the
3546   // the watcher thread while safepoint is in progress so that it would not appear as though the
3547   // system is making progress.
3548   if (!Solaris::T2_libthread() &&
3549       thread->is_Watcher_thread() && SafepointSynchronize::is_synchronizing() && !Arguments::has_profile()) {
3550     // We now try to acquire the threads lock. Since this lock is held by the VM thread during
3551     // the entire safepoint, the watcher thread will  line up here during the safepoint.
3552     Threads_lock->lock_without_safepoint_check();
3553     Threads_lock->unlock();
3554   }
3555 
3556   if (thread->is_Java_thread()) {
3557     // This is a JavaThread so we honor the _thread_blocked protocol
3558     // even for sleeps of 0 milliseconds. This was originally done
3559     // as a workaround for bug 4338139. However, now we also do it
3560     // to honor the suspend-equivalent protocol.
3561 
3562     JavaThread *jt = (JavaThread *) thread;
3563     ThreadBlockInVM tbivm(jt);
3564 
3565     jt->set_suspend_equivalent();
3566     // cleared by handle_special_suspend_equivalent_condition() or
3567     // java_suspend_self() via check_and_wait_while_suspended()
3568 
3569     int ret_code;
3570     if (millis <= 0) {
3571       thr_yield();
3572       ret_code = 0;
3573     } else {
3574       // The original sleep() implementation did not create an
3575       // OSThreadWaitState helper for sleeps of 0 milliseconds.
3576       // I'm preserving that decision for now.
3577       OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3578 
3579       ret_code = os_sleep(millis, interruptible);
3580     }
3581 
3582     // were we externally suspended while we were waiting?
3583     jt->check_and_wait_while_suspended();
3584 
3585     return ret_code;
3586   }
3587 
3588   // non-JavaThread from this point on:
3589 
3590   if (millis <= 0) {
3591     thr_yield();
3592     return 0;
3593   }
3594 
3595   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3596 
3597   return os_sleep(millis, interruptible);
3598 }
3599 
3600 int os::naked_sleep() {
3601   // %% make the sleep time an integer flag. for now use 1 millisec.
3602   return os_sleep(1, false);
3603 }
3604 
3605 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3606 void os::infinite_sleep() {
3607   while (true) {    // sleep forever ...
3608     ::sleep(100);   // ... 100 seconds at a time
3609   }
3610 }
3611 
3612 // Used to convert frequent JVM_Yield() to nops
3613 bool os::dont_yield() {
3614   if (DontYieldALot) {
3615     static hrtime_t last_time = 0;
3616     hrtime_t diff = getTimeNanos() - last_time;
3617 
3618     if (diff < DontYieldALotInterval * 1000000)
3619       return true;
3620 
3621     last_time += diff;
3622 
3623     return false;
3624   }
3625   else {
3626     return false;
3627   }
3628 }
3629 
3630 // Caveat: Solaris os::yield() causes a thread-state transition whereas
3631 // the linux and win32 implementations do not.  This should be checked.
3632 
3633 void os::yield() {
3634   // Yields to all threads with same or greater priority
3635   os::sleep(Thread::current(), 0, false);
3636 }
3637 
3638 // Note that yield semantics are defined by the scheduling class to which
3639 // the thread currently belongs.  Typically, yield will _not yield to
3640 // other equal or higher priority threads that reside on the dispatch queues
3641 // of other CPUs.
3642 
3643 os::YieldResult os::NakedYield() { thr_yield(); return os::YIELD_UNKNOWN; }
3644 
3645 
3646 // On Solaris we found that yield_all doesn't always yield to all other threads.
3647 // There have been cases where there is a thread ready to execute but it doesn't
3648 // get an lwp as the VM thread continues to spin with sleeps of 1 millisecond.
3649 // The 1 millisecond wait doesn't seem long enough for the kernel to issue a
3650 // SIGWAITING signal which will cause a new lwp to be created. So we count the
3651 // number of times yield_all is called in the one loop and increase the sleep
3652 // time after 8 attempts. If this fails too we increase the concurrency level
3653 // so that the starving thread would get an lwp
3654 
3655 void os::yield_all(int attempts) {
3656   // Yields to all threads, including threads with lower priorities
3657   if (attempts == 0) {
3658     os::sleep(Thread::current(), 1, false);
3659   } else {
3660     int iterations = attempts % 30;
3661     if (iterations == 0 && !os::Solaris::T2_libthread()) {
3662       // thr_setconcurrency and _getconcurrency make sense only under T1.
3663       int noofLWPS = thr_getconcurrency();
3664       if (noofLWPS < (Threads::number_of_threads() + 2)) {
3665         thr_setconcurrency(thr_getconcurrency() + 1);
3666       }
3667     } else if (iterations < 25) {
3668       os::sleep(Thread::current(), 1, false);
3669     } else {
3670       os::sleep(Thread::current(), 10, false);
3671     }
3672   }
3673 }
3674 
3675 // Called from the tight loops to possibly influence time-sharing heuristics
3676 void os::loop_breaker(int attempts) {
3677   os::yield_all(attempts);
3678 }
3679 
3680 
3681 // Interface for setting lwp priorities.  If we are using T2 libthread,
3682 // which forces the use of BoundThreads or we manually set UseBoundThreads,
3683 // all of our threads will be assigned to real lwp's.  Using the thr_setprio
3684 // function is meaningless in this mode so we must adjust the real lwp's priority
3685 // The routines below implement the getting and setting of lwp priorities.
3686 //
3687 // Note: There are three priority scales used on Solaris.  Java priotities
3688 //       which range from 1 to 10, libthread "thr_setprio" scale which range
3689 //       from 0 to 127, and the current scheduling class of the process we
3690 //       are running in.  This is typically from -60 to +60.
3691 //       The setting of the lwp priorities in done after a call to thr_setprio
3692 //       so Java priorities are mapped to libthread priorities and we map from
3693 //       the latter to lwp priorities.  We don't keep priorities stored in
3694 //       Java priorities since some of our worker threads want to set priorities
3695 //       higher than all Java threads.
3696 //
3697 // For related information:
3698 // (1)  man -s 2 priocntl
3699 // (2)  man -s 4 priocntl
3700 // (3)  man dispadmin
3701 // =    librt.so
3702 // =    libthread/common/rtsched.c - thrp_setlwpprio().
3703 // =    ps -cL <pid> ... to validate priority.
3704 // =    sched_get_priority_min and _max
3705 //              pthread_create
3706 //              sched_setparam
3707 //              pthread_setschedparam
3708 //
3709 // Assumptions:
3710 // +    We assume that all threads in the process belong to the same
3711 //              scheduling class.   IE. an homogenous process.
3712 // +    Must be root or in IA group to change change "interactive" attribute.
3713 //              Priocntl() will fail silently.  The only indication of failure is when
3714 //              we read-back the value and notice that it hasn't changed.
3715 // +    Interactive threads enter the runq at the head, non-interactive at the tail.
3716 // +    For RT, change timeslice as well.  Invariant:
3717 //              constant "priority integral"
3718 //              Konst == TimeSlice * (60-Priority)
3719 //              Given a priority, compute appropriate timeslice.
3720 // +    Higher numerical values have higher priority.
3721 
3722 // sched class attributes
3723 typedef struct {
3724         int   schedPolicy;              // classID
3725         int   maxPrio;
3726         int   minPrio;
3727 } SchedInfo;
3728 
3729 
3730 static SchedInfo tsLimits, iaLimits, rtLimits, fxLimits;
3731 
3732 #ifdef ASSERT
3733 static int  ReadBackValidate = 1;
3734 #endif
3735 static int  myClass     = 0;
3736 static int  myMin       = 0;
3737 static int  myMax       = 0;
3738 static int  myCur       = 0;
3739 static bool priocntl_enable = false;
3740 
3741 static const int criticalPrio = 60; // FX/60 is critical thread class/priority on T4
3742 static int java_MaxPriority_to_os_priority = 0; // Saved mapping
3743 
3744 // Call the version of priocntl suitable for all supported versions
3745 // of Solaris. We need to call through this wrapper so that we can
3746 // build on Solaris 9 and run on Solaris 8, 9 and 10.
3747 //
3748 // This code should be removed if we ever stop supporting Solaris 8
3749 // and earlier releases.
3750 
3751 static long priocntl_stub(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg);
3752 typedef long (*priocntl_type)(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg);
3753 static priocntl_type priocntl_ptr = priocntl_stub;
3754 
3755 // Stub to set the value of the real pointer, and then call the real
3756 // function.
3757 
3758 static long priocntl_stub(int pcver, idtype_t idtype, id_t id, int cmd, caddr_t arg) {
3759   // Try Solaris 8- name only.
3760   priocntl_type tmp = (priocntl_type)dlsym(RTLD_DEFAULT, "__priocntl");
3761   guarantee(tmp != NULL, "priocntl function not found.");
3762   priocntl_ptr = tmp;
3763   return (*priocntl_ptr)(PC_VERSION, idtype, id, cmd, arg);
3764 }
3765 
3766 
3767 // lwp_priocntl_init
3768 //
3769 // Try to determine the priority scale for our process.
3770 //
3771 // Return errno or 0 if OK.
3772 //
3773 static
3774 int     lwp_priocntl_init ()
3775 {
3776   int rslt;
3777   pcinfo_t ClassInfo;
3778   pcparms_t ParmInfo;
3779   int i;
3780 
3781   if (!UseThreadPriorities) return 0;
3782 
3783   // We are using Bound threads, we need to determine our priority ranges
3784   if (os::Solaris::T2_libthread() || UseBoundThreads) {
3785     // If ThreadPriorityPolicy is 1, switch tables
3786     if (ThreadPriorityPolicy == 1) {
3787       for (i = 0 ; i < CriticalPriority+1; i++)
3788         os::java_to_os_priority[i] = prio_policy1[i];
3789     }
3790     if (UseCriticalJavaThreadPriority) {
3791       // MaxPriority always maps to the FX scheduling class and criticalPrio.
3792       // See set_native_priority() and set_lwp_class_and_priority().
3793       // Save original MaxPriority mapping in case attempt to
3794       // use critical priority fails.
3795       java_MaxPriority_to_os_priority = os::java_to_os_priority[MaxPriority];
3796       // Set negative to distinguish from other priorities
3797       os::java_to_os_priority[MaxPriority] = -criticalPrio;
3798     }
3799   }
3800   // Not using Bound Threads, set to ThreadPolicy 1
3801   else {
3802     for ( i = 0 ; i < CriticalPriority+1; i++ ) {
3803       os::java_to_os_priority[i] = prio_policy1[i];
3804     }
3805     return 0;
3806   }
3807 
3808   // Get IDs for a set of well-known scheduling classes.
3809   // TODO-FIXME: GETCLINFO returns the current # of classes in the
3810   // the system.  We should have a loop that iterates over the
3811   // classID values, which are known to be "small" integers.
3812 
3813   strcpy(ClassInfo.pc_clname, "TS");
3814   ClassInfo.pc_cid = -1;
3815   rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3816   if (rslt < 0) return errno;
3817   assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3818   tsLimits.schedPolicy = ClassInfo.pc_cid;
3819   tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3820   tsLimits.minPrio = -tsLimits.maxPrio;
3821 
3822   strcpy(ClassInfo.pc_clname, "IA");
3823   ClassInfo.pc_cid = -1;
3824   rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3825   if (rslt < 0) return errno;
3826   assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3827   iaLimits.schedPolicy = ClassInfo.pc_cid;
3828   iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3829   iaLimits.minPrio = -iaLimits.maxPrio;
3830 
3831   strcpy(ClassInfo.pc_clname, "RT");
3832   ClassInfo.pc_cid = -1;
3833   rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3834   if (rslt < 0) return errno;
3835   assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3836   rtLimits.schedPolicy = ClassInfo.pc_cid;
3837   rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3838   rtLimits.minPrio = 0;
3839 
3840   strcpy(ClassInfo.pc_clname, "FX");
3841   ClassInfo.pc_cid = -1;
3842   rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3843   if (rslt < 0) return errno;
3844   assert(ClassInfo.pc_cid != -1, "cid for FX class is -1");
3845   fxLimits.schedPolicy = ClassInfo.pc_cid;
3846   fxLimits.maxPrio = ((fxinfo_t*)ClassInfo.pc_clinfo)->fx_maxupri;
3847   fxLimits.minPrio = 0;
3848 
3849   // Query our "current" scheduling class.
3850   // This will normally be IA, TS or, rarely, FX or RT.
3851   memset(&ParmInfo, 0, sizeof(ParmInfo));
3852   ParmInfo.pc_cid = PC_CLNULL;
3853   rslt = (*priocntl_ptr) (PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3854   if (rslt < 0) return errno;
3855   myClass = ParmInfo.pc_cid;
3856 
3857   // We now know our scheduling classId, get specific information
3858   // about the class.
3859   ClassInfo.pc_cid = myClass;
3860   ClassInfo.pc_clname[0] = 0;
3861   rslt = (*priocntl_ptr) (PC_VERSION, (idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo);
3862   if (rslt < 0) return errno;
3863 
3864   if (ThreadPriorityVerbose) {
3865     tty->print_cr("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3866   }
3867 
3868   memset(&ParmInfo, 0, sizeof(pcparms_t));
3869   ParmInfo.pc_cid = PC_CLNULL;
3870   rslt = (*priocntl_ptr)(PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3871   if (rslt < 0) return errno;
3872 
3873   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3874     myMin = rtLimits.minPrio;
3875     myMax = rtLimits.maxPrio;
3876   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3877     iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3878     myMin = iaLimits.minPrio;
3879     myMax = iaLimits.maxPrio;
3880     myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3881   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3882     tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3883     myMin = tsLimits.minPrio;
3884     myMax = tsLimits.maxPrio;
3885     myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3886   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3887     fxparms_t *fxInfo = (fxparms_t*)ParmInfo.pc_clparms;
3888     myMin = fxLimits.minPrio;
3889     myMax = fxLimits.maxPrio;
3890     myMax = MIN2(myMax, (int)fxInfo->fx_uprilim);       // clamp - restrict
3891   } else {
3892     // No clue - punt
3893     if (ThreadPriorityVerbose)
3894       tty->print_cr ("Unknown scheduling class: %s ... \n", ClassInfo.pc_clname);
3895     return EINVAL;      // no clue, punt
3896   }
3897 
3898   if (ThreadPriorityVerbose) {
3899     tty->print_cr ("Thread priority Range: [%d..%d]\n", myMin, myMax);
3900   }
3901 
3902   priocntl_enable = true;  // Enable changing priorities
3903   return 0;
3904 }
3905 
3906 #define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3907 #define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3908 #define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3909 #define FXPRI(x)        ((fxparms_t *)((x).pc_clparms))
3910 
3911 
3912 // scale_to_lwp_priority
3913 //
3914 // Convert from the libthread "thr_setprio" scale to our current
3915 // lwp scheduling class scale.
3916 //
3917 static
3918 int     scale_to_lwp_priority (int rMin, int rMax, int x)
3919 {
3920   int v;
3921 
3922   if (x == 127) return rMax;            // avoid round-down
3923     v = (((x*(rMax-rMin)))/128)+rMin;
3924   return v;
3925 }
3926 
3927 
3928 // set_lwp_class_and_priority
3929 //
3930 // Set the class and priority of the lwp.  This call should only
3931 // be made when using bound threads (T2 threads are bound by default).
3932 //
3933 int set_lwp_class_and_priority(int ThreadID, int lwpid,
3934                                int newPrio, int new_class, bool scale) {
3935   int rslt;
3936   int Actual, Expected, prv;
3937   pcparms_t ParmInfo;                   // for GET-SET
3938 #ifdef ASSERT
3939   pcparms_t ReadBack;                   // for readback
3940 #endif
3941 
3942   // Set priority via PC_GETPARMS, update, PC_SETPARMS
3943   // Query current values.
3944   // TODO: accelerate this by eliminating the PC_GETPARMS call.
3945   // Cache "pcparms_t" in global ParmCache.
3946   // TODO: elide set-to-same-value
3947 
3948   // If something went wrong on init, don't change priorities.
3949   if ( !priocntl_enable ) {
3950     if (ThreadPriorityVerbose)
3951       tty->print_cr("Trying to set priority but init failed, ignoring");
3952     return EINVAL;
3953   }
3954 
3955   // If lwp hasn't started yet, just return
3956   // the _start routine will call us again.
3957   if ( lwpid <= 0 ) {
3958     if (ThreadPriorityVerbose) {
3959       tty->print_cr ("deferring the set_lwp_class_and_priority of thread "
3960                      INTPTR_FORMAT " to %d, lwpid not set",
3961                      ThreadID, newPrio);
3962     }
3963     return 0;
3964   }
3965 
3966   if (ThreadPriorityVerbose) {
3967     tty->print_cr ("set_lwp_class_and_priority("
3968                    INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3969                    ThreadID, lwpid, newPrio);
3970   }
3971 
3972   memset(&ParmInfo, 0, sizeof(pcparms_t));
3973   ParmInfo.pc_cid = PC_CLNULL;
3974   rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3975   if (rslt < 0) return errno;
3976 
3977   int cur_class = ParmInfo.pc_cid;
3978   ParmInfo.pc_cid = (id_t)new_class;
3979 
3980   if (new_class == rtLimits.schedPolicy) {
3981     rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3982     rtInfo->rt_pri     = scale ? scale_to_lwp_priority(rtLimits.minPrio,
3983                                                        rtLimits.maxPrio, newPrio)
3984                                : newPrio;
3985     rtInfo->rt_tqsecs  = RT_NOCHANGE;
3986     rtInfo->rt_tqnsecs = RT_NOCHANGE;
3987     if (ThreadPriorityVerbose) {
3988       tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3989     }
3990   } else if (new_class == iaLimits.schedPolicy) {
3991     iaparms_t* iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3992     int maxClamped     = MIN2(iaLimits.maxPrio,
3993                               cur_class == new_class
3994                                 ? (int)iaInfo->ia_uprilim : iaLimits.maxPrio);
3995     iaInfo->ia_upri    = scale ? scale_to_lwp_priority(iaLimits.minPrio,
3996                                                        maxClamped, newPrio)
3997                                : newPrio;
3998     iaInfo->ia_uprilim = cur_class == new_class
3999                            ? IA_NOCHANGE : (pri_t)iaLimits.maxPrio;
4000     iaInfo->ia_mode    = IA_NOCHANGE;
4001     if (ThreadPriorityVerbose) {
4002       tty->print_cr("IA: [%d...%d] %d->%d\n",
4003                     iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
4004     }
4005   } else if (new_class == tsLimits.schedPolicy) {
4006     tsparms_t* tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
4007     int maxClamped     = MIN2(tsLimits.maxPrio,
4008                               cur_class == new_class
4009                                 ? (int)tsInfo->ts_uprilim : tsLimits.maxPrio);
4010     tsInfo->ts_upri    = scale ? scale_to_lwp_priority(tsLimits.minPrio,
4011                                                        maxClamped, newPrio)
4012                                : newPrio;
4013     tsInfo->ts_uprilim = cur_class == new_class
4014                            ? TS_NOCHANGE : (pri_t)tsLimits.maxPrio;
4015     if (ThreadPriorityVerbose) {
4016       tty->print_cr("TS: [%d...%d] %d->%d\n",
4017                     tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
4018     }
4019   } else if (new_class == fxLimits.schedPolicy) {
4020     fxparms_t* fxInfo  = (fxparms_t*)ParmInfo.pc_clparms;
4021     int maxClamped     = MIN2(fxLimits.maxPrio,
4022                               cur_class == new_class
4023                                 ? (int)fxInfo->fx_uprilim : fxLimits.maxPrio);
4024     fxInfo->fx_upri    = scale ? scale_to_lwp_priority(fxLimits.minPrio,
4025                                                        maxClamped, newPrio)
4026                                : newPrio;
4027     fxInfo->fx_uprilim = cur_class == new_class
4028                            ? FX_NOCHANGE : (pri_t)fxLimits.maxPrio;
4029     fxInfo->fx_tqsecs  = FX_NOCHANGE;
4030     fxInfo->fx_tqnsecs = FX_NOCHANGE;
4031     if (ThreadPriorityVerbose) {
4032       tty->print_cr("FX: [%d...%d] %d->%d\n",
4033                     fxLimits.minPrio, maxClamped, newPrio, fxInfo->fx_upri);
4034     }
4035   } else {
4036     if (ThreadPriorityVerbose) {
4037       tty->print_cr("Unknown new scheduling class %d\n", new_class);
4038     }
4039     return EINVAL;    // no clue, punt
4040   }
4041 
4042   rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
4043   if (ThreadPriorityVerbose && rslt) {
4044     tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
4045   }
4046   if (rslt < 0) return errno;
4047 
4048 #ifdef ASSERT
4049   // Sanity check: read back what we just attempted to set.
4050   // In theory it could have changed in the interim ...
4051   //
4052   // The priocntl system call is tricky.
4053   // Sometimes it'll validate the priority value argument and
4054   // return EINVAL if unhappy.  At other times it fails silently.
4055   // Readbacks are prudent.
4056 
4057   if (!ReadBackValidate) return 0;
4058 
4059   memset(&ReadBack, 0, sizeof(pcparms_t));
4060   ReadBack.pc_cid = PC_CLNULL;
4061   rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
4062   assert(rslt >= 0, "priocntl failed");
4063   Actual = Expected = 0xBAD;
4064   assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
4065   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
4066     Actual   = RTPRI(ReadBack)->rt_pri;
4067     Expected = RTPRI(ParmInfo)->rt_pri;
4068   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
4069     Actual   = IAPRI(ReadBack)->ia_upri;
4070     Expected = IAPRI(ParmInfo)->ia_upri;
4071   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
4072     Actual   = TSPRI(ReadBack)->ts_upri;
4073     Expected = TSPRI(ParmInfo)->ts_upri;
4074   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
4075     Actual   = FXPRI(ReadBack)->fx_upri;
4076     Expected = FXPRI(ParmInfo)->fx_upri;
4077   } else {
4078     if (ThreadPriorityVerbose) {
4079       tty->print_cr("set_lwp_class_and_priority: unexpected class in readback: %d\n",
4080                     ParmInfo.pc_cid);
4081     }
4082   }
4083 
4084   if (Actual != Expected) {
4085     if (ThreadPriorityVerbose) {
4086       tty->print_cr ("set_lwp_class_and_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
4087                      lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
4088     }
4089   }
4090 #endif
4091 
4092   return 0;
4093 }
4094 
4095 // Solaris only gives access to 128 real priorities at a time,
4096 // so we expand Java's ten to fill this range.  This would be better
4097 // if we dynamically adjusted relative priorities.
4098 //
4099 // The ThreadPriorityPolicy option allows us to select 2 different
4100 // priority scales.
4101 //
4102 // ThreadPriorityPolicy=0
4103 // Since the Solaris' default priority is MaximumPriority, we do not
4104 // set a priority lower than Max unless a priority lower than
4105 // NormPriority is requested.
4106 //
4107 // ThreadPriorityPolicy=1
4108 // This mode causes the priority table to get filled with
4109 // linear values.  NormPriority get's mapped to 50% of the
4110 // Maximum priority an so on.  This will cause VM threads
4111 // to get unfair treatment against other Solaris processes
4112 // which do not explicitly alter their thread priorities.
4113 //
4114 
4115 int os::java_to_os_priority[CriticalPriority + 1] = {
4116   -99999,         // 0 Entry should never be used
4117 
4118   0,              // 1 MinPriority
4119   32,             // 2
4120   64,             // 3
4121 
4122   96,             // 4
4123   127,            // 5 NormPriority
4124   127,            // 6
4125 
4126   127,            // 7
4127   127,            // 8
4128   127,            // 9 NearMaxPriority
4129 
4130   127,            // 10 MaxPriority
4131 
4132   -criticalPrio   // 11 CriticalPriority
4133 };
4134 
4135 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4136   OSThread* osthread = thread->osthread();
4137 
4138   // Save requested priority in case the thread hasn't been started
4139   osthread->set_native_priority(newpri);
4140 
4141   // Check for critical priority request
4142   bool fxcritical = false;
4143   if (newpri == -criticalPrio) {
4144     fxcritical = true;
4145     newpri = criticalPrio;
4146   }
4147 
4148   assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
4149   if (!UseThreadPriorities) return OS_OK;
4150 
4151   int status = 0;
4152 
4153   if (!fxcritical) {
4154     // Use thr_setprio only if we have a priority that thr_setprio understands
4155     status = thr_setprio(thread->osthread()->thread_id(), newpri);
4156   }
4157 
4158   if (os::Solaris::T2_libthread() ||
4159       (UseBoundThreads && osthread->is_vm_created())) {
4160     int lwp_status =
4161       set_lwp_class_and_priority(osthread->thread_id(),
4162                                  osthread->lwp_id(),
4163                                  newpri,
4164                                  fxcritical ? fxLimits.schedPolicy : myClass,
4165                                  !fxcritical);
4166     if (lwp_status != 0 && fxcritical) {
4167       // Try again, this time without changing the scheduling class
4168       newpri = java_MaxPriority_to_os_priority;
4169       lwp_status = set_lwp_class_and_priority(osthread->thread_id(),
4170                                               osthread->lwp_id(),
4171                                               newpri, myClass, false);
4172     }
4173     status |= lwp_status;
4174   }
4175   return (status == 0) ? OS_OK : OS_ERR;
4176 }
4177 
4178 
4179 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4180   int p;
4181   if ( !UseThreadPriorities ) {
4182     *priority_ptr = NormalPriority;
4183     return OS_OK;
4184   }
4185   int status = thr_getprio(thread->osthread()->thread_id(), &p);
4186   if (status != 0) {
4187     return OS_ERR;
4188   }
4189   *priority_ptr = p;
4190   return OS_OK;
4191 }
4192 
4193 
4194 // Hint to the underlying OS that a task switch would not be good.
4195 // Void return because it's a hint and can fail.
4196 void os::hint_no_preempt() {
4197   schedctl_start(schedctl_init());
4198 }
4199 
4200 void os::interrupt(Thread* thread) {
4201   assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
4202 
4203   OSThread* osthread = thread->osthread();
4204 
4205   int isInterrupted = osthread->interrupted();
4206   if (!isInterrupted) {
4207       osthread->set_interrupted(true);
4208       OrderAccess::fence();
4209       // os::sleep() is implemented with either poll (NULL,0,timeout) or
4210       // by parking on _SleepEvent.  If the former, thr_kill will unwedge
4211       // the sleeper by SIGINTR, otherwise the unpark() will wake the sleeper.
4212       ParkEvent * const slp = thread->_SleepEvent ;
4213       if (slp != NULL) slp->unpark() ;
4214   }
4215 
4216   // For JSR166:  unpark after setting status but before thr_kill -dl
4217   if (thread->is_Java_thread()) {
4218     ((JavaThread*)thread)->parker()->unpark();
4219   }
4220 
4221   // Handle interruptible wait() ...
4222   ParkEvent * const ev = thread->_ParkEvent ;
4223   if (ev != NULL) ev->unpark() ;
4224 
4225   // When events are used everywhere for os::sleep, then this thr_kill
4226   // will only be needed if UseVMInterruptibleIO is true.
4227 
4228   if (!isInterrupted) {
4229     int status = thr_kill(osthread->thread_id(), os::Solaris::SIGinterrupt());
4230     assert_status(status == 0, status, "thr_kill");
4231 
4232     // Bump thread interruption counter
4233     RuntimeService::record_thread_interrupt_signaled_count();
4234   }
4235 }
4236 
4237 
4238 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4239   assert(Thread::current() == thread || Threads_lock->owned_by_self(), "possibility of dangling Thread pointer");
4240 
4241   OSThread* osthread = thread->osthread();
4242 
4243   bool res = osthread->interrupted();
4244 
4245   // NOTE that since there is no "lock" around these two operations,
4246   // there is the possibility that the interrupted flag will be
4247   // "false" but that the interrupt event will be set. This is
4248   // intentional. The effect of this is that Object.wait() will appear
4249   // to have a spurious wakeup, which is not harmful, and the
4250   // possibility is so rare that it is not worth the added complexity
4251   // to add yet another lock. It has also been recommended not to put
4252   // the interrupted flag into the os::Solaris::Event structure,
4253   // because it hides the issue.
4254   if (res && clear_interrupted) {
4255     osthread->set_interrupted(false);
4256   }
4257   return res;
4258 }
4259 
4260 
4261 void os::print_statistics() {
4262 }
4263 
4264 int os::message_box(const char* title, const char* message) {
4265   int i;
4266   fdStream err(defaultStream::error_fd());
4267   for (i = 0; i < 78; i++) err.print_raw("=");
4268   err.cr();
4269   err.print_raw_cr(title);
4270   for (i = 0; i < 78; i++) err.print_raw("-");
4271   err.cr();
4272   err.print_raw_cr(message);
4273   for (i = 0; i < 78; i++) err.print_raw("=");
4274   err.cr();
4275 
4276   char buf[16];
4277   // Prevent process from exiting upon "read error" without consuming all CPU
4278   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4279 
4280   return buf[0] == 'y' || buf[0] == 'Y';
4281 }
4282 
4283 // A lightweight implementation that does not suspend the target thread and
4284 // thus returns only a hint. Used for profiling only!
4285 ExtendedPC os::get_thread_pc(Thread* thread) {
4286   // Make sure that it is called by the watcher and the Threads lock is owned.
4287   assert(Thread::current()->is_Watcher_thread(), "Must be watcher and own Threads_lock");
4288   // For now, is only used to profile the VM Thread
4289   assert(thread->is_VM_thread(), "Can only be called for VMThread");
4290   ExtendedPC epc;
4291 
4292   GetThreadPC_Callback  cb(ProfileVM_lock);
4293   OSThread *osthread = thread->osthread();
4294   const int time_to_wait = 400; // 400ms wait for initial response
4295   int status = cb.interrupt(thread, time_to_wait);
4296 
4297   if (cb.is_done() ) {
4298     epc = cb.addr();
4299   } else {
4300     DEBUG_ONLY(tty->print_cr("Failed to get pc for thread: %d got %d status",
4301                               osthread->thread_id(), status););
4302     // epc is already NULL
4303   }
4304   return epc;
4305 }
4306 
4307 
4308 // This does not do anything on Solaris. This is basically a hook for being
4309 // able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
4310 void os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method, JavaCallArguments* args, Thread* thread) {
4311   f(value, method, args, thread);
4312 }
4313 
4314 // This routine may be used by user applications as a "hook" to catch signals.
4315 // The user-defined signal handler must pass unrecognized signals to this
4316 // routine, and if it returns true (non-zero), then the signal handler must
4317 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4318 // routine will never retun false (zero), but instead will execute a VM panic
4319 // routine kill the process.
4320 //
4321 // If this routine returns false, it is OK to call it again.  This allows
4322 // the user-defined signal handler to perform checks either before or after
4323 // the VM performs its own checks.  Naturally, the user code would be making
4324 // a serious error if it tried to handle an exception (such as a null check
4325 // or breakpoint) that the VM was generating for its own correct operation.
4326 //
4327 // This routine may recognize any of the following kinds of signals:
4328 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
4329 // os::Solaris::SIGasync
4330 // It should be consulted by handlers for any of those signals.
4331 // It explicitly does not recognize os::Solaris::SIGinterrupt
4332 //
4333 // The caller of this routine must pass in the three arguments supplied
4334 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4335 // field of the structure passed to sigaction().  This routine assumes that
4336 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4337 //
4338 // Note that the VM will print warnings if it detects conflicting signal
4339 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4340 //
4341 extern "C" JNIEXPORT int
4342 JVM_handle_solaris_signal(int signo, siginfo_t* siginfo, void* ucontext,
4343                           int abort_if_unrecognized);
4344 
4345 
4346 void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
4347   JVM_handle_solaris_signal(sig, info, ucVoid, true);
4348 }
4349 
4350 /* Do not delete - if guarantee is ever removed,  a signal handler (even empty)
4351    is needed to provoke threads blocked on IO to return an EINTR
4352    Note: this explicitly does NOT call JVM_handle_solaris_signal and
4353    does NOT participate in signal chaining due to requirement for
4354    NOT setting SA_RESTART to make EINTR work. */
4355 extern "C" void sigINTRHandler(int sig, siginfo_t* info, void* ucVoid) {
4356    if (UseSignalChaining) {
4357       struct sigaction *actp = os::Solaris::get_chained_signal_action(sig);
4358       if (actp && actp->sa_handler) {
4359         vm_exit_during_initialization("Signal chaining detected for VM interrupt signal, try -XX:+UseAltSigs");
4360       }
4361    }
4362 }
4363 
4364 // This boolean allows users to forward their own non-matching signals
4365 // to JVM_handle_solaris_signal, harmlessly.
4366 bool os::Solaris::signal_handlers_are_installed = false;
4367 
4368 // For signal-chaining
4369 bool os::Solaris::libjsig_is_loaded = false;
4370 typedef struct sigaction *(*get_signal_t)(int);
4371 get_signal_t os::Solaris::get_signal_action = NULL;
4372 
4373 struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
4374   struct sigaction *actp = NULL;
4375 
4376   if ((libjsig_is_loaded)  && (sig <= Maxlibjsigsigs)) {
4377     // Retrieve the old signal handler from libjsig
4378     actp = (*get_signal_action)(sig);
4379   }
4380   if (actp == NULL) {
4381     // Retrieve the preinstalled signal handler from jvm
4382     actp = get_preinstalled_handler(sig);
4383   }
4384 
4385   return actp;
4386 }
4387 
4388 static bool call_chained_handler(struct sigaction *actp, int sig,
4389                                  siginfo_t *siginfo, void *context) {
4390   // Call the old signal handler
4391   if (actp->sa_handler == SIG_DFL) {
4392     // It's more reasonable to let jvm treat it as an unexpected exception
4393     // instead of taking the default action.
4394     return false;
4395   } else if (actp->sa_handler != SIG_IGN) {
4396     if ((actp->sa_flags & SA_NODEFER) == 0) {
4397       // automaticlly block the signal
4398       sigaddset(&(actp->sa_mask), sig);
4399     }
4400 
4401     sa_handler_t hand;
4402     sa_sigaction_t sa;
4403     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4404     // retrieve the chained handler
4405     if (siginfo_flag_set) {
4406       sa = actp->sa_sigaction;
4407     } else {
4408       hand = actp->sa_handler;
4409     }
4410 
4411     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4412       actp->sa_handler = SIG_DFL;
4413     }
4414 
4415     // try to honor the signal mask
4416     sigset_t oset;
4417     thr_sigsetmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4418 
4419     // call into the chained handler
4420     if (siginfo_flag_set) {
4421       (*sa)(sig, siginfo, context);
4422     } else {
4423       (*hand)(sig);
4424     }
4425 
4426     // restore the signal mask
4427     thr_sigsetmask(SIG_SETMASK, &oset, 0);
4428   }
4429   // Tell jvm's signal handler the signal is taken care of.
4430   return true;
4431 }
4432 
4433 bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4434   bool chained = false;
4435   // signal-chaining
4436   if (UseSignalChaining) {
4437     struct sigaction *actp = get_chained_signal_action(sig);
4438     if (actp != NULL) {
4439       chained = call_chained_handler(actp, sig, siginfo, context);
4440     }
4441   }
4442   return chained;
4443 }
4444 
4445 struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
4446   assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4447   if (preinstalled_sigs[sig] != 0) {
4448     return &chainedsigactions[sig];
4449   }
4450   return NULL;
4451 }
4452 
4453 void os::Solaris::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4454 
4455   assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
4456   assert((chainedsigactions != (struct sigaction *)NULL) && (preinstalled_sigs != (int *)NULL) , "signals not yet initialized");
4457   chainedsigactions[sig] = oldAct;
4458   preinstalled_sigs[sig] = 1;
4459 }
4460 
4461 void os::Solaris::set_signal_handler(int sig, bool set_installed, bool oktochain) {
4462   // Check for overwrite.
4463   struct sigaction oldAct;
4464   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4465   void* oldhand = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4466                                       : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4467   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4468       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4469       oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
4470     if (AllowUserSignalHandlers || !set_installed) {
4471       // Do not overwrite; user takes responsibility to forward to us.
4472       return;
4473     } else if (UseSignalChaining) {
4474       if (oktochain) {
4475         // save the old handler in jvm
4476         save_preinstalled_handler(sig, oldAct);
4477       } else {
4478         vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal, try -XX:+UseAltSigs.");
4479       }
4480       // libjsig also interposes the sigaction() call below and saves the
4481       // old sigaction on it own.
4482     } else {
4483       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4484                     "%#lx for signal %d.", (long)oldhand, sig));
4485     }
4486   }
4487 
4488   struct sigaction sigAct;
4489   sigfillset(&(sigAct.sa_mask));
4490   sigAct.sa_handler = SIG_DFL;
4491 
4492   sigAct.sa_sigaction = signalHandler;
4493   // Handle SIGSEGV on alternate signal stack if
4494   // not using stack banging
4495   if (!UseStackBanging && sig == SIGSEGV) {
4496     sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
4497   // Interruptible i/o requires SA_RESTART cleared so EINTR
4498   // is returned instead of restarting system calls
4499   } else if (sig == os::Solaris::SIGinterrupt()) {
4500     sigemptyset(&sigAct.sa_mask);
4501     sigAct.sa_handler = NULL;
4502     sigAct.sa_flags = SA_SIGINFO;
4503     sigAct.sa_sigaction = sigINTRHandler;
4504   } else {
4505     sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
4506   }
4507   os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
4508 
4509   sigaction(sig, &sigAct, &oldAct);
4510 
4511   void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4512                                        : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4513   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4514 }
4515 
4516 
4517 #define DO_SIGNAL_CHECK(sig) \
4518   if (!sigismember(&check_signal_done, sig)) \
4519     os::Solaris::check_signal_handler(sig)
4520 
4521 // This method is a periodic task to check for misbehaving JNI applications
4522 // under CheckJNI, we can add any periodic checks here
4523 
4524 void os::run_periodic_checks() {
4525   // A big source of grief is hijacking virt. addr 0x0 on Solaris,
4526   // thereby preventing a NULL checks.
4527   if(!check_addr0_done) check_addr0_done = check_addr0(tty);
4528 
4529   if (check_signals == false) return;
4530 
4531   // SEGV and BUS if overridden could potentially prevent
4532   // generation of hs*.log in the event of a crash, debugging
4533   // such a case can be very challenging, so we absolutely
4534   // check for the following for a good measure:
4535   DO_SIGNAL_CHECK(SIGSEGV);
4536   DO_SIGNAL_CHECK(SIGILL);
4537   DO_SIGNAL_CHECK(SIGFPE);
4538   DO_SIGNAL_CHECK(SIGBUS);
4539   DO_SIGNAL_CHECK(SIGPIPE);
4540   DO_SIGNAL_CHECK(SIGXFSZ);
4541 
4542   // ReduceSignalUsage allows the user to override these handlers
4543   // see comments at the very top and jvm_solaris.h
4544   if (!ReduceSignalUsage) {
4545     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4546     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4547     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4548     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4549   }
4550 
4551   // See comments above for using JVM1/JVM2 and UseAltSigs
4552   DO_SIGNAL_CHECK(os::Solaris::SIGinterrupt());
4553   DO_SIGNAL_CHECK(os::Solaris::SIGasync());
4554 
4555 }
4556 
4557 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4558 
4559 static os_sigaction_t os_sigaction = NULL;
4560 
4561 void os::Solaris::check_signal_handler(int sig) {
4562   char buf[O_BUFLEN];
4563   address jvmHandler = NULL;
4564 
4565   struct sigaction act;
4566   if (os_sigaction == NULL) {
4567     // only trust the default sigaction, in case it has been interposed
4568     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4569     if (os_sigaction == NULL) return;
4570   }
4571 
4572   os_sigaction(sig, (struct sigaction*)NULL, &act);
4573 
4574   address thisHandler = (act.sa_flags & SA_SIGINFO)
4575     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4576     : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4577 
4578 
4579   switch(sig) {
4580     case SIGSEGV:
4581     case SIGBUS:
4582     case SIGFPE:
4583     case SIGPIPE:
4584     case SIGXFSZ:
4585     case SIGILL:
4586       jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4587       break;
4588 
4589     case SHUTDOWN1_SIGNAL:
4590     case SHUTDOWN2_SIGNAL:
4591     case SHUTDOWN3_SIGNAL:
4592     case BREAK_SIGNAL:
4593       jvmHandler = (address)user_handler();
4594       break;
4595 
4596     default:
4597       int intrsig = os::Solaris::SIGinterrupt();
4598       int asynsig = os::Solaris::SIGasync();
4599 
4600       if (sig == intrsig) {
4601         jvmHandler = CAST_FROM_FN_PTR(address, sigINTRHandler);
4602       } else if (sig == asynsig) {
4603         jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4604       } else {
4605         return;
4606       }
4607       break;
4608   }
4609 
4610 
4611   if (thisHandler != jvmHandler) {
4612     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4613     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4614     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4615     // No need to check this sig any longer
4616     sigaddset(&check_signal_done, sig);
4617   } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
4618     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4619     tty->print("expected:" PTR32_FORMAT, os::Solaris::get_our_sigflags(sig));
4620     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4621     // No need to check this sig any longer
4622     sigaddset(&check_signal_done, sig);
4623   }
4624 
4625   // Print all the signal handler state
4626   if (sigismember(&check_signal_done, sig)) {
4627     print_signal_handlers(tty, buf, O_BUFLEN);
4628   }
4629 
4630 }
4631 
4632 void os::Solaris::install_signal_handlers() {
4633   bool libjsigdone = false;
4634   signal_handlers_are_installed = true;
4635 
4636   // signal-chaining
4637   typedef void (*signal_setting_t)();
4638   signal_setting_t begin_signal_setting = NULL;
4639   signal_setting_t end_signal_setting = NULL;
4640   begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4641                                         dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4642   if (begin_signal_setting != NULL) {
4643     end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4644                                         dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4645     get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4646                                        dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4647     get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
4648                                          dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
4649     libjsig_is_loaded = true;
4650     if (os::Solaris::get_libjsig_version != NULL) {
4651       libjsigversion =  (*os::Solaris::get_libjsig_version)();
4652     }
4653     assert(UseSignalChaining, "should enable signal-chaining");
4654   }
4655   if (libjsig_is_loaded) {
4656     // Tell libjsig jvm is setting signal handlers
4657     (*begin_signal_setting)();
4658   }
4659 
4660   set_signal_handler(SIGSEGV, true, true);
4661   set_signal_handler(SIGPIPE, true, true);
4662   set_signal_handler(SIGXFSZ, true, true);
4663   set_signal_handler(SIGBUS, true, true);
4664   set_signal_handler(SIGILL, true, true);
4665   set_signal_handler(SIGFPE, true, true);
4666 
4667 
4668   if (os::Solaris::SIGinterrupt() > OLDMAXSIGNUM || os::Solaris::SIGasync() > OLDMAXSIGNUM) {
4669 
4670     // Pre-1.4.1 Libjsig limited to signal chaining signals <= 32 so
4671     // can not register overridable signals which might be > 32
4672     if (libjsig_is_loaded && libjsigversion <= JSIG_VERSION_1_4_1) {
4673     // Tell libjsig jvm has finished setting signal handlers
4674       (*end_signal_setting)();
4675       libjsigdone = true;
4676     }
4677   }
4678 
4679   // Never ok to chain our SIGinterrupt
4680   set_signal_handler(os::Solaris::SIGinterrupt(), true, false);
4681   set_signal_handler(os::Solaris::SIGasync(), true, true);
4682 
4683   if (libjsig_is_loaded && !libjsigdone) {
4684     // Tell libjsig jvm finishes setting signal handlers
4685     (*end_signal_setting)();
4686   }
4687 
4688   // We don't activate signal checker if libjsig is in place, we trust ourselves
4689   // and if UserSignalHandler is installed all bets are off.
4690   // Log that signal checking is off only if -verbose:jni is specified.
4691   if (CheckJNICalls) {
4692     if (libjsig_is_loaded) {
4693       if (PrintJNIResolving) {
4694         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4695       }
4696       check_signals = false;
4697     }
4698     if (AllowUserSignalHandlers) {
4699       if (PrintJNIResolving) {
4700         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4701       }
4702       check_signals = false;
4703     }
4704   }
4705 }
4706 
4707 
4708 void report_error(const char* file_name, int line_no, const char* title, const char* format, ...);
4709 
4710 const char * signames[] = {
4711   "SIG0",
4712   "SIGHUP", "SIGINT", "SIGQUIT", "SIGILL", "SIGTRAP",
4713   "SIGABRT", "SIGEMT", "SIGFPE", "SIGKILL", "SIGBUS",
4714   "SIGSEGV", "SIGSYS", "SIGPIPE", "SIGALRM", "SIGTERM",
4715   "SIGUSR1", "SIGUSR2", "SIGCLD", "SIGPWR", "SIGWINCH",
4716   "SIGURG", "SIGPOLL", "SIGSTOP", "SIGTSTP", "SIGCONT",
4717   "SIGTTIN", "SIGTTOU", "SIGVTALRM", "SIGPROF", "SIGXCPU",
4718   "SIGXFSZ", "SIGWAITING", "SIGLWP", "SIGFREEZE", "SIGTHAW",
4719   "SIGCANCEL", "SIGLOST"
4720 };
4721 
4722 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4723   if (0 < exception_code && exception_code <= SIGRTMAX) {
4724     // signal
4725     if (exception_code < sizeof(signames)/sizeof(const char*)) {
4726        jio_snprintf(buf, size, "%s", signames[exception_code]);
4727     } else {
4728        jio_snprintf(buf, size, "SIG%d", exception_code);
4729     }
4730     return buf;
4731   } else {
4732     return NULL;
4733   }
4734 }
4735 
4736 // (Static) wrappers for the new libthread API
4737 int_fnP_thread_t_iP_uP_stack_tP_gregset_t os::Solaris::_thr_getstate;
4738 int_fnP_thread_t_i_gregset_t os::Solaris::_thr_setstate;
4739 int_fnP_thread_t_i os::Solaris::_thr_setmutator;
4740 int_fnP_thread_t os::Solaris::_thr_suspend_mutator;
4741 int_fnP_thread_t os::Solaris::_thr_continue_mutator;
4742 
4743 // (Static) wrapper for getisax(2) call.
4744 os::Solaris::getisax_func_t os::Solaris::_getisax = 0;
4745 
4746 // (Static) wrappers for the liblgrp API
4747 os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4748 os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4749 os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4750 os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4751 os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4752 os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4753 os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4754 os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4755 os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4756 
4757 // (Static) wrapper for meminfo() call.
4758 os::Solaris::meminfo_func_t os::Solaris::_meminfo = 0;
4759 
4760 static address resolve_symbol_lazy(const char* name) {
4761   address addr = (address) dlsym(RTLD_DEFAULT, name);
4762   if(addr == NULL) {
4763     // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4764     addr = (address) dlsym(RTLD_NEXT, name);
4765   }
4766   return addr;
4767 }
4768 
4769 static address resolve_symbol(const char* name) {
4770   address addr = resolve_symbol_lazy(name);
4771   if(addr == NULL) {
4772     fatal(dlerror());
4773   }
4774   return addr;
4775 }
4776 
4777 
4778 
4779 // isT2_libthread()
4780 //
4781 // Routine to determine if we are currently using the new T2 libthread.
4782 //
4783 // We determine if we are using T2 by reading /proc/self/lstatus and
4784 // looking for a thread with the ASLWP bit set.  If we find this status
4785 // bit set, we must assume that we are NOT using T2.  The T2 team
4786 // has approved this algorithm.
4787 //
4788 // We need to determine if we are running with the new T2 libthread
4789 // since setting native thread priorities is handled differently
4790 // when using this library.  All threads created using T2 are bound
4791 // threads. Calling thr_setprio is meaningless in this case.
4792 //
4793 bool isT2_libthread() {
4794   static prheader_t * lwpArray = NULL;
4795   static int lwpSize = 0;
4796   static int lwpFile = -1;
4797   lwpstatus_t * that;
4798   char lwpName [128];
4799   bool isT2 = false;
4800 
4801 #define ADR(x)  ((uintptr_t)(x))
4802 #define LWPINDEX(ary,ix)   ((lwpstatus_t *)(((ary)->pr_entsize * (ix)) + (ADR((ary) + 1))))
4803 
4804   lwpFile = ::open("/proc/self/lstatus", O_RDONLY, 0);
4805   if (lwpFile < 0) {
4806       if (ThreadPriorityVerbose) warning ("Couldn't open /proc/self/lstatus\n");
4807       return false;
4808   }
4809   lwpSize = 16*1024;
4810   for (;;) {
4811     ::lseek64 (lwpFile, 0, SEEK_SET);
4812     lwpArray = (prheader_t *)NEW_C_HEAP_ARRAY(char, lwpSize);
4813     if (::read(lwpFile, lwpArray, lwpSize) < 0) {
4814       if (ThreadPriorityVerbose) warning("Error reading /proc/self/lstatus\n");
4815       break;
4816     }
4817     if ((lwpArray->pr_nent * lwpArray->pr_entsize) <= lwpSize) {
4818        // We got a good snapshot - now iterate over the list.
4819       int aslwpcount = 0;
4820       for (int i = 0; i < lwpArray->pr_nent; i++ ) {
4821         that = LWPINDEX(lwpArray,i);
4822         if (that->pr_flags & PR_ASLWP) {
4823           aslwpcount++;
4824         }
4825       }
4826       if (aslwpcount == 0) isT2 = true;
4827       break;
4828     }
4829     lwpSize = lwpArray->pr_nent * lwpArray->pr_entsize;
4830     FREE_C_HEAP_ARRAY(char, lwpArray);  // retry.
4831   }
4832 
4833   FREE_C_HEAP_ARRAY(char, lwpArray);
4834   ::close (lwpFile);
4835   if (ThreadPriorityVerbose) {
4836     if (isT2) tty->print_cr("We are running with a T2 libthread\n");
4837     else tty->print_cr("We are not running with a T2 libthread\n");
4838   }
4839   return isT2;
4840 }
4841 
4842 
4843 void os::Solaris::libthread_init() {
4844   address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4845 
4846   // Determine if we are running with the new T2 libthread
4847   os::Solaris::set_T2_libthread(isT2_libthread());
4848 
4849   lwp_priocntl_init();
4850 
4851   // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4852   if(func == NULL) {
4853     func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4854     // Guarantee that this VM is running on an new enough OS (5.6 or
4855     // later) that it will have a new enough libthread.so.
4856     guarantee(func != NULL, "libthread.so is too old.");
4857   }
4858 
4859   // Initialize the new libthread getstate API wrappers
4860   func = resolve_symbol("thr_getstate");
4861   os::Solaris::set_thr_getstate(CAST_TO_FN_PTR(int_fnP_thread_t_iP_uP_stack_tP_gregset_t, func));
4862 
4863   func = resolve_symbol("thr_setstate");
4864   os::Solaris::set_thr_setstate(CAST_TO_FN_PTR(int_fnP_thread_t_i_gregset_t, func));
4865 
4866   func = resolve_symbol("thr_setmutator");
4867   os::Solaris::set_thr_setmutator(CAST_TO_FN_PTR(int_fnP_thread_t_i, func));
4868 
4869   func = resolve_symbol("thr_suspend_mutator");
4870   os::Solaris::set_thr_suspend_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4871 
4872   func = resolve_symbol("thr_continue_mutator");
4873   os::Solaris::set_thr_continue_mutator(CAST_TO_FN_PTR(int_fnP_thread_t, func));
4874 
4875   int size;
4876   void (*handler_info_func)(address *, int *);
4877   handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4878   handler_info_func(&handler_start, &size);
4879   handler_end = handler_start + size;
4880 }
4881 
4882 
4883 int_fnP_mutex_tP os::Solaris::_mutex_lock;
4884 int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4885 int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4886 int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4887 int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4888 int os::Solaris::_mutex_scope = USYNC_THREAD;
4889 
4890 int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4891 int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4892 int_fnP_cond_tP os::Solaris::_cond_signal;
4893 int_fnP_cond_tP os::Solaris::_cond_broadcast;
4894 int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4895 int_fnP_cond_tP os::Solaris::_cond_destroy;
4896 int os::Solaris::_cond_scope = USYNC_THREAD;
4897 
4898 void os::Solaris::synchronization_init() {
4899   if(UseLWPSynchronization) {
4900     os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4901     os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4902     os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4903     os::Solaris::set_mutex_init(lwp_mutex_init);
4904     os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4905     os::Solaris::set_mutex_scope(USYNC_THREAD);
4906 
4907     os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4908     os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4909     os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4910     os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4911     os::Solaris::set_cond_init(lwp_cond_init);
4912     os::Solaris::set_cond_destroy(lwp_cond_destroy);
4913     os::Solaris::set_cond_scope(USYNC_THREAD);
4914   }
4915   else {
4916     os::Solaris::set_mutex_scope(USYNC_THREAD);
4917     os::Solaris::set_cond_scope(USYNC_THREAD);
4918 
4919     if(UsePthreads) {
4920       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4921       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4922       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4923       os::Solaris::set_mutex_init(pthread_mutex_default_init);
4924       os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4925 
4926       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4927       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4928       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4929       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4930       os::Solaris::set_cond_init(pthread_cond_default_init);
4931       os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4932     }
4933     else {
4934       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4935       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4936       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4937       os::Solaris::set_mutex_init(::mutex_init);
4938       os::Solaris::set_mutex_destroy(::mutex_destroy);
4939 
4940       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4941       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4942       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4943       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4944       os::Solaris::set_cond_init(::cond_init);
4945       os::Solaris::set_cond_destroy(::cond_destroy);
4946     }
4947   }
4948 }
4949 
4950 bool os::Solaris::liblgrp_init() {
4951   void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4952   if (handle != NULL) {
4953     os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4954     os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4955     os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4956     os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4957     os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4958     os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4959     os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4960     os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4961                                        dlsym(handle, "lgrp_cookie_stale")));
4962 
4963     lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4964     set_lgrp_cookie(c);
4965     return true;
4966   }
4967   return false;
4968 }
4969 
4970 void os::Solaris::misc_sym_init() {
4971   address func;
4972 
4973   // getisax
4974   func = resolve_symbol_lazy("getisax");
4975   if (func != NULL) {
4976     os::Solaris::_getisax = CAST_TO_FN_PTR(getisax_func_t, func);
4977   }
4978 
4979   // meminfo
4980   func = resolve_symbol_lazy("meminfo");
4981   if (func != NULL) {
4982     os::Solaris::set_meminfo(CAST_TO_FN_PTR(meminfo_func_t, func));
4983   }
4984 }
4985 
4986 uint_t os::Solaris::getisax(uint32_t* array, uint_t n) {
4987   assert(_getisax != NULL, "_getisax not set");
4988   return _getisax(array, n);
4989 }
4990 
4991 // Symbol doesn't exist in Solaris 8 pset.h
4992 #ifndef PS_MYID
4993 #define PS_MYID -3
4994 #endif
4995 
4996 // int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
4997 typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
4998 static pset_getloadavg_type pset_getloadavg_ptr = NULL;
4999 
5000 void init_pset_getloadavg_ptr(void) {
5001   pset_getloadavg_ptr =
5002     (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
5003   if (PrintMiscellaneous && Verbose && pset_getloadavg_ptr == NULL) {
5004     warning("pset_getloadavg function not found");
5005   }
5006 }
5007 
5008 int os::Solaris::_dev_zero_fd = -1;
5009 
5010 // this is called _before_ the global arguments have been parsed
5011 void os::init(void) {
5012   _initial_pid = getpid();
5013 
5014   max_hrtime = first_hrtime = gethrtime();
5015 
5016   init_random(1234567);
5017 
5018   page_size = sysconf(_SC_PAGESIZE);
5019   if (page_size == -1)
5020     fatal(err_msg("os_solaris.cpp: os::init: sysconf failed (%s)",
5021                   strerror(errno)));
5022   init_page_sizes((size_t) page_size);
5023 
5024   Solaris::initialize_system_info();
5025 
5026   // Initialize misc. symbols as soon as possible, so we can use them
5027   // if we need them.
5028   Solaris::misc_sym_init();
5029 
5030   int fd = ::open("/dev/zero", O_RDWR);
5031   if (fd < 0) {
5032     fatal(err_msg("os::init: cannot open /dev/zero (%s)", strerror(errno)));
5033   } else {
5034     Solaris::set_dev_zero_fd(fd);
5035 
5036     // Close on exec, child won't inherit.
5037     fcntl(fd, F_SETFD, FD_CLOEXEC);
5038   }
5039 
5040   clock_tics_per_sec = CLK_TCK;
5041 
5042   // check if dladdr1() exists; dladdr1 can provide more information than
5043   // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
5044   // and is available on linker patches for 5.7 and 5.8.
5045   // libdl.so must have been loaded, this call is just an entry lookup
5046   void * hdl = dlopen("libdl.so", RTLD_NOW);
5047   if (hdl)
5048     dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
5049 
5050   // (Solaris only) this switches to calls that actually do locking.
5051   ThreadCritical::initialize();
5052 
5053   main_thread = thr_self();
5054 
5055   // Constant minimum stack size allowed. It must be at least
5056   // the minimum of what the OS supports (thr_min_stack()), and
5057   // enough to allow the thread to get to user bytecode execution.
5058   Solaris::min_stack_allowed = MAX2(thr_min_stack(), Solaris::min_stack_allowed);
5059   // If the pagesize of the VM is greater than 8K determine the appropriate
5060   // number of initial guard pages.  The user can change this with the
5061   // command line arguments, if needed.
5062   if (vm_page_size() > 8*K) {
5063     StackYellowPages = 1;
5064     StackRedPages = 1;
5065     StackShadowPages = round_to((StackShadowPages*8*K), vm_page_size()) / vm_page_size();
5066   }
5067 }
5068 
5069 // To install functions for atexit system call
5070 extern "C" {
5071   static void perfMemory_exit_helper() {
5072     perfMemory_exit();
5073   }
5074 }
5075 
5076 // this is called _after_ the global arguments have been parsed
5077 jint os::init_2(void) {
5078   // try to enable extended file IO ASAP, see 6431278
5079   os::Solaris::try_enable_extended_io();
5080 
5081   // Allocate a single page and mark it as readable for safepoint polling.  Also
5082   // use this first mmap call to check support for MAP_ALIGN.
5083   address polling_page = (address)Solaris::mmap_chunk((char*)page_size,
5084                                                       page_size,
5085                                                       MAP_PRIVATE | MAP_ALIGN,
5086                                                       PROT_READ);
5087   if (polling_page == NULL) {
5088     has_map_align = false;
5089     polling_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE,
5090                                                 PROT_READ);
5091   }
5092 
5093   os::set_polling_page(polling_page);
5094 
5095 #ifndef PRODUCT
5096   if( Verbose && PrintMiscellaneous )
5097     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
5098 #endif
5099 
5100   if (!UseMembar) {
5101     address mem_serialize_page = (address)Solaris::mmap_chunk( NULL, page_size, MAP_PRIVATE, PROT_READ | PROT_WRITE );
5102     guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
5103     os::set_memory_serialize_page( mem_serialize_page );
5104 
5105 #ifndef PRODUCT
5106     if(Verbose && PrintMiscellaneous)
5107       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
5108 #endif
5109 }
5110 
5111   os::large_page_init();
5112 
5113   // Check minimum allowable stack size for thread creation and to initialize
5114   // the java system classes, including StackOverflowError - depends on page
5115   // size.  Add a page for compiler2 recursion in main thread.
5116   // Add in 2*BytesPerWord times page size to account for VM stack during
5117   // class initialization depending on 32 or 64 bit VM.
5118   os::Solaris::min_stack_allowed = MAX2(os::Solaris::min_stack_allowed,
5119             (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
5120                     2*BytesPerWord COMPILER2_PRESENT(+1)) * page_size);
5121 
5122   size_t threadStackSizeInBytes = ThreadStackSize * K;
5123   if (threadStackSizeInBytes != 0 &&
5124     threadStackSizeInBytes < os::Solaris::min_stack_allowed) {
5125     tty->print_cr("\nThe stack size specified is too small, Specify at least %dk",
5126                   os::Solaris::min_stack_allowed/K);
5127     return JNI_ERR;
5128   }
5129 
5130   // For 64kbps there will be a 64kb page size, which makes
5131   // the usable default stack size quite a bit less.  Increase the
5132   // stack for 64kb (or any > than 8kb) pages, this increases
5133   // virtual memory fragmentation (since we're not creating the
5134   // stack on a power of 2 boundary.  The real fix for this
5135   // should be to fix the guard page mechanism.
5136 
5137   if (vm_page_size() > 8*K) {
5138       threadStackSizeInBytes = (threadStackSizeInBytes != 0)
5139          ? threadStackSizeInBytes +
5140            ((StackYellowPages + StackRedPages) * vm_page_size())
5141          : 0;
5142       ThreadStackSize = threadStackSizeInBytes/K;
5143   }
5144 
5145   // Make the stack size a multiple of the page size so that
5146   // the yellow/red zones can be guarded.
5147   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
5148         vm_page_size()));
5149 
5150   Solaris::libthread_init();
5151 
5152   if (UseNUMA) {
5153     if (!Solaris::liblgrp_init()) {
5154       UseNUMA = false;
5155     } else {
5156       size_t lgrp_limit = os::numa_get_groups_num();
5157       int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit);
5158       size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
5159       FREE_C_HEAP_ARRAY(int, lgrp_ids);
5160       if (lgrp_num < 2) {
5161         // There's only one locality group, disable NUMA.
5162         UseNUMA = false;
5163       }
5164     }
5165     // ISM is not compatible with the NUMA allocator - it always allocates
5166     // pages round-robin across the lgroups.
5167     if (UseNUMA && UseLargePages && UseISM) {
5168       if (!FLAG_IS_DEFAULT(UseNUMA)) {
5169         if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseISM)) {
5170           UseLargePages = false;
5171         } else {
5172           warning("UseNUMA is not compatible with ISM large pages, disabling NUMA allocator");
5173           UseNUMA = false;
5174         }
5175       } else {
5176         UseNUMA = false;
5177       }
5178     }
5179     if (!UseNUMA && ForceNUMA) {
5180       UseNUMA = true;
5181     }
5182   }
5183 
5184   Solaris::signal_sets_init();
5185   Solaris::init_signal_mem();
5186   Solaris::install_signal_handlers();
5187 
5188   if (libjsigversion < JSIG_VERSION_1_4_1) {
5189     Maxlibjsigsigs = OLDMAXSIGNUM;
5190   }
5191 
5192   // initialize synchronization primitives to use either thread or
5193   // lwp synchronization (controlled by UseLWPSynchronization)
5194   Solaris::synchronization_init();
5195 
5196   if (MaxFDLimit) {
5197     // set the number of file descriptors to max. print out error
5198     // if getrlimit/setrlimit fails but continue regardless.
5199     struct rlimit nbr_files;
5200     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5201     if (status != 0) {
5202       if (PrintMiscellaneous && (Verbose || WizardMode))
5203         perror("os::init_2 getrlimit failed");
5204     } else {
5205       nbr_files.rlim_cur = nbr_files.rlim_max;
5206       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5207       if (status != 0) {
5208         if (PrintMiscellaneous && (Verbose || WizardMode))
5209           perror("os::init_2 setrlimit failed");
5210       }
5211     }
5212   }
5213 
5214   // Calculate theoretical max. size of Threads to guard gainst
5215   // artifical out-of-memory situations, where all available address-
5216   // space has been reserved by thread stacks. Default stack size is 1Mb.
5217   size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
5218     JavaThread::stack_size_at_create() : (1*K*K);
5219   assert(pre_thread_stack_size != 0, "Must have a stack");
5220   // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
5221   // we should start doing Virtual Memory banging. Currently when the threads will
5222   // have used all but 200Mb of space.
5223   size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
5224   Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
5225 
5226   // at-exit methods are called in the reverse order of their registration.
5227   // In Solaris 7 and earlier, atexit functions are called on return from
5228   // main or as a result of a call to exit(3C). There can be only 32 of
5229   // these functions registered and atexit() does not set errno. In Solaris
5230   // 8 and later, there is no limit to the number of functions registered
5231   // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
5232   // functions are called upon dlclose(3DL) in addition to return from main
5233   // and exit(3C).
5234 
5235   if (PerfAllowAtExitRegistration) {
5236     // only register atexit functions if PerfAllowAtExitRegistration is set.
5237     // atexit functions can be delayed until process exit time, which
5238     // can be problematic for embedded VM situations. Embedded VMs should
5239     // call DestroyJavaVM() to assure that VM resources are released.
5240 
5241     // note: perfMemory_exit_helper atexit function may be removed in
5242     // the future if the appropriate cleanup code can be added to the
5243     // VM_Exit VMOperation's doit method.
5244     if (atexit(perfMemory_exit_helper) != 0) {
5245       warning("os::init2 atexit(perfMemory_exit_helper) failed");
5246     }
5247   }
5248 
5249   // Init pset_loadavg function pointer
5250   init_pset_getloadavg_ptr();
5251 
5252   return JNI_OK;
5253 }
5254 
5255 void os::init_3(void) {
5256   return;
5257 }
5258 
5259 // Mark the polling page as unreadable
5260 void os::make_polling_page_unreadable(void) {
5261   if( mprotect((char *)_polling_page, page_size, PROT_NONE) != 0 )
5262     fatal("Could not disable polling page");
5263 };
5264 
5265 // Mark the polling page as readable
5266 void os::make_polling_page_readable(void) {
5267   if( mprotect((char *)_polling_page, page_size, PROT_READ) != 0 )
5268     fatal("Could not enable polling page");
5269 };
5270 
5271 // OS interface.
5272 
5273 bool os::check_heap(bool force) { return true; }
5274 
5275 typedef int (*vsnprintf_t)(char* buf, size_t count, const char* fmt, va_list argptr);
5276 static vsnprintf_t sol_vsnprintf = NULL;
5277 
5278 int local_vsnprintf(char* buf, size_t count, const char* fmt, va_list argptr) {
5279   if (!sol_vsnprintf) {
5280     //search  for the named symbol in the objects that were loaded after libjvm
5281     void* where = RTLD_NEXT;
5282     if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5283         sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5284     if (!sol_vsnprintf){
5285       //search  for the named symbol in the objects that were loaded before libjvm
5286       where = RTLD_DEFAULT;
5287       if ((sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "__vsnprintf"))) == NULL)
5288         sol_vsnprintf = CAST_TO_FN_PTR(vsnprintf_t, dlsym(where, "vsnprintf"));
5289       assert(sol_vsnprintf != NULL, "vsnprintf not found");
5290     }
5291   }
5292   return (*sol_vsnprintf)(buf, count, fmt, argptr);
5293 }
5294 
5295 
5296 // Is a (classpath) directory empty?
5297 bool os::dir_is_empty(const char* path) {
5298   DIR *dir = NULL;
5299   struct dirent *ptr;
5300 
5301   dir = opendir(path);
5302   if (dir == NULL) return true;
5303 
5304   /* Scan the directory */
5305   bool result = true;
5306   char buf[sizeof(struct dirent) + MAX_PATH];
5307   struct dirent *dbuf = (struct dirent *) buf;
5308   while (result && (ptr = readdir(dir, dbuf)) != NULL) {
5309     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5310       result = false;
5311     }
5312   }
5313   closedir(dir);
5314   return result;
5315 }
5316 
5317 // This code originates from JDK's sysOpen and open64_w
5318 // from src/solaris/hpi/src/system_md.c
5319 
5320 #ifndef O_DELETE
5321 #define O_DELETE 0x10000
5322 #endif
5323 
5324 // Open a file. Unlink the file immediately after open returns
5325 // if the specified oflag has the O_DELETE flag set.
5326 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5327 
5328 int os::open(const char *path, int oflag, int mode) {
5329   if (strlen(path) > MAX_PATH - 1) {
5330     errno = ENAMETOOLONG;
5331     return -1;
5332   }
5333   int fd;
5334   int o_delete = (oflag & O_DELETE);
5335   oflag = oflag & ~O_DELETE;
5336 
5337   fd = ::open64(path, oflag, mode);
5338   if (fd == -1) return -1;
5339 
5340   //If the open succeeded, the file might still be a directory
5341   {
5342     struct stat64 buf64;
5343     int ret = ::fstat64(fd, &buf64);
5344     int st_mode = buf64.st_mode;
5345 
5346     if (ret != -1) {
5347       if ((st_mode & S_IFMT) == S_IFDIR) {
5348         errno = EISDIR;
5349         ::close(fd);
5350         return -1;
5351       }
5352     } else {
5353       ::close(fd);
5354       return -1;
5355     }
5356   }
5357     /*
5358      * 32-bit Solaris systems suffer from:
5359      *
5360      * - an historical default soft limit of 256 per-process file
5361      *   descriptors that is too low for many Java programs.
5362      *
5363      * - a design flaw where file descriptors created using stdio
5364      *   fopen must be less than 256, _even_ when the first limit above
5365      *   has been raised.  This can cause calls to fopen (but not calls to
5366      *   open, for example) to fail mysteriously, perhaps in 3rd party
5367      *   native code (although the JDK itself uses fopen).  One can hardly
5368      *   criticize them for using this most standard of all functions.
5369      *
5370      * We attempt to make everything work anyways by:
5371      *
5372      * - raising the soft limit on per-process file descriptors beyond
5373      *   256
5374      *
5375      * - As of Solaris 10u4, we can request that Solaris raise the 256
5376      *   stdio fopen limit by calling function enable_extended_FILE_stdio.
5377      *   This is done in init_2 and recorded in enabled_extended_FILE_stdio
5378      *
5379      * - If we are stuck on an old (pre 10u4) Solaris system, we can
5380      *   workaround the bug by remapping non-stdio file descriptors below
5381      *   256 to ones beyond 256, which is done below.
5382      *
5383      * See:
5384      * 1085341: 32-bit stdio routines should support file descriptors >255
5385      * 6533291: Work around 32-bit Solaris stdio limit of 256 open files
5386      * 6431278: Netbeans crash on 32 bit Solaris: need to call
5387      *          enable_extended_FILE_stdio() in VM initialisation
5388      * Giri Mandalika's blog
5389      * http://technopark02.blogspot.com/2005_05_01_archive.html
5390      */
5391 #ifndef  _LP64
5392      if ((!enabled_extended_FILE_stdio) && fd < 256) {
5393          int newfd = ::fcntl(fd, F_DUPFD, 256);
5394          if (newfd != -1) {
5395              ::close(fd);
5396              fd = newfd;
5397          }
5398      }
5399 #endif // 32-bit Solaris
5400     /*
5401      * All file descriptors that are opened in the JVM and not
5402      * specifically destined for a subprocess should have the
5403      * close-on-exec flag set.  If we don't set it, then careless 3rd
5404      * party native code might fork and exec without closing all
5405      * appropriate file descriptors (e.g. as we do in closeDescriptors in
5406      * UNIXProcess.c), and this in turn might:
5407      *
5408      * - cause end-of-file to fail to be detected on some file
5409      *   descriptors, resulting in mysterious hangs, or
5410      *
5411      * - might cause an fopen in the subprocess to fail on a system
5412      *   suffering from bug 1085341.
5413      *
5414      * (Yes, the default setting of the close-on-exec flag is a Unix
5415      * design flaw)
5416      *
5417      * See:
5418      * 1085341: 32-bit stdio routines should support file descriptors >255
5419      * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5420      * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5421      */
5422 #ifdef FD_CLOEXEC
5423     {
5424         int flags = ::fcntl(fd, F_GETFD);
5425         if (flags != -1)
5426             ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5427     }
5428 #endif
5429 
5430   if (o_delete != 0) {
5431     ::unlink(path);
5432   }
5433   return fd;
5434 }
5435 
5436 // create binary file, rewriting existing file if required
5437 int os::create_binary_file(const char* path, bool rewrite_existing) {
5438   int oflags = O_WRONLY | O_CREAT;
5439   if (!rewrite_existing) {
5440     oflags |= O_EXCL;
5441   }
5442   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5443 }
5444 
5445 // return current position of file pointer
5446 jlong os::current_file_offset(int fd) {
5447   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5448 }
5449 
5450 // move file pointer to the specified offset
5451 jlong os::seek_to_file_offset(int fd, jlong offset) {
5452   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5453 }
5454 
5455 jlong os::lseek(int fd, jlong offset, int whence) {
5456   return (jlong) ::lseek64(fd, offset, whence);
5457 }
5458 
5459 char * os::native_path(char *path) {
5460   return path;
5461 }
5462 
5463 int os::ftruncate(int fd, jlong length) {
5464   return ::ftruncate64(fd, length);
5465 }
5466 
5467 int os::fsync(int fd)  {
5468   RESTARTABLE_RETURN_INT(::fsync(fd));
5469 }
5470 
5471 int os::available(int fd, jlong *bytes) {
5472   jlong cur, end;
5473   int mode;
5474   struct stat64 buf64;
5475 
5476   if (::fstat64(fd, &buf64) >= 0) {
5477     mode = buf64.st_mode;
5478     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5479       /*
5480       * XXX: is the following call interruptible? If so, this might
5481       * need to go through the INTERRUPT_IO() wrapper as for other
5482       * blocking, interruptible calls in this file.
5483       */
5484       int n,ioctl_return;
5485 
5486       INTERRUPTIBLE(::ioctl(fd, FIONREAD, &n),ioctl_return,os::Solaris::clear_interrupted);
5487       if (ioctl_return>= 0) {
5488           *bytes = n;
5489         return 1;
5490       }
5491     }
5492   }
5493   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5494     return 0;
5495   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5496     return 0;
5497   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5498     return 0;
5499   }
5500   *bytes = end - cur;
5501   return 1;
5502 }
5503 
5504 // Map a block of memory.
5505 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
5506                      char *addr, size_t bytes, bool read_only,
5507                      bool allow_exec) {
5508   int prot;
5509   int flags;
5510 
5511   if (read_only) {
5512     prot = PROT_READ;
5513     flags = MAP_SHARED;
5514   } else {
5515     prot = PROT_READ | PROT_WRITE;
5516     flags = MAP_PRIVATE;
5517   }
5518 
5519   if (allow_exec) {
5520     prot |= PROT_EXEC;
5521   }
5522 
5523   if (addr != NULL) {
5524     flags |= MAP_FIXED;
5525   }
5526 
5527   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5528                                      fd, file_offset);
5529   if (mapped_address == MAP_FAILED) {
5530     return NULL;
5531   }
5532   return mapped_address;
5533 }
5534 
5535 
5536 // Remap a block of memory.
5537 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
5538                        char *addr, size_t bytes, bool read_only,
5539                        bool allow_exec) {
5540   // same as map_memory() on this OS
5541   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5542                         allow_exec);
5543 }
5544 
5545 
5546 // Unmap a block of memory.
5547 bool os::unmap_memory(char* addr, size_t bytes) {
5548   return munmap(addr, bytes) == 0;
5549 }
5550 
5551 void os::pause() {
5552   char filename[MAX_PATH];
5553   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5554     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5555   } else {
5556     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5557   }
5558 
5559   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5560   if (fd != -1) {
5561     struct stat buf;
5562     ::close(fd);
5563     while (::stat(filename, &buf) == 0) {
5564       (void)::poll(NULL, 0, 100);
5565     }
5566   } else {
5567     jio_fprintf(stderr,
5568       "Could not open pause file '%s', continuing immediately.\n", filename);
5569   }
5570 }
5571 
5572 #ifndef PRODUCT
5573 #ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5574 // Turn this on if you need to trace synch operations.
5575 // Set RECORD_SYNCH_LIMIT to a large-enough value,
5576 // and call record_synch_enable and record_synch_disable
5577 // around the computation of interest.
5578 
5579 void record_synch(char* name, bool returning);  // defined below
5580 
5581 class RecordSynch {
5582   char* _name;
5583  public:
5584   RecordSynch(char* name) :_name(name)
5585                  { record_synch(_name, false); }
5586   ~RecordSynch() { record_synch(_name,   true);  }
5587 };
5588 
5589 #define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
5590 extern "C" ret name params {                                    \
5591   typedef ret name##_t params;                                  \
5592   static name##_t* implem = NULL;                               \
5593   static int callcount = 0;                                     \
5594   if (implem == NULL) {                                         \
5595     implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
5596     if (implem == NULL)  fatal(dlerror());                      \
5597   }                                                             \
5598   ++callcount;                                                  \
5599   RecordSynch _rs(#name);                                       \
5600   inner;                                                        \
5601   return implem args;                                           \
5602 }
5603 // in dbx, examine callcounts this way:
5604 // for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
5605 
5606 #define CHECK_POINTER_OK(p) \
5607   (Universe::perm_gen() == NULL || !Universe::is_reserved_heap((oop)(p)))
5608 #define CHECK_MU \
5609   if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
5610 #define CHECK_CV \
5611   if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
5612 #define CHECK_P(p) \
5613   if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
5614 
5615 #define CHECK_MUTEX(mutex_op) \
5616 CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
5617 
5618 CHECK_MUTEX(   mutex_lock)
5619 CHECK_MUTEX(  _mutex_lock)
5620 CHECK_MUTEX( mutex_unlock)
5621 CHECK_MUTEX(_mutex_unlock)
5622 CHECK_MUTEX( mutex_trylock)
5623 CHECK_MUTEX(_mutex_trylock)
5624 
5625 #define CHECK_COND(cond_op) \
5626 CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU;CHECK_CV);
5627 
5628 CHECK_COND( cond_wait);
5629 CHECK_COND(_cond_wait);
5630 CHECK_COND(_cond_wait_cancel);
5631 
5632 #define CHECK_COND2(cond_op) \
5633 CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU;CHECK_CV);
5634 
5635 CHECK_COND2( cond_timedwait);
5636 CHECK_COND2(_cond_timedwait);
5637 CHECK_COND2(_cond_timedwait_cancel);
5638 
5639 // do the _lwp_* versions too
5640 #define mutex_t lwp_mutex_t
5641 #define cond_t  lwp_cond_t
5642 CHECK_MUTEX(  _lwp_mutex_lock)
5643 CHECK_MUTEX(  _lwp_mutex_unlock)
5644 CHECK_MUTEX(  _lwp_mutex_trylock)
5645 CHECK_MUTEX( __lwp_mutex_lock)
5646 CHECK_MUTEX( __lwp_mutex_unlock)
5647 CHECK_MUTEX( __lwp_mutex_trylock)
5648 CHECK_MUTEX(___lwp_mutex_lock)
5649 CHECK_MUTEX(___lwp_mutex_unlock)
5650 
5651 CHECK_COND(  _lwp_cond_wait);
5652 CHECK_COND( __lwp_cond_wait);
5653 CHECK_COND(___lwp_cond_wait);
5654 
5655 CHECK_COND2(  _lwp_cond_timedwait);
5656 CHECK_COND2( __lwp_cond_timedwait);
5657 #undef mutex_t
5658 #undef cond_t
5659 
5660 CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5661 CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5662 CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5663 CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5664 CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5665 CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5666 CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5667 CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5668 
5669 
5670 // recording machinery:
5671 
5672 enum { RECORD_SYNCH_LIMIT = 200 };
5673 char* record_synch_name[RECORD_SYNCH_LIMIT];
5674 void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
5675 bool record_synch_returning[RECORD_SYNCH_LIMIT];
5676 thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
5677 int record_synch_count = 0;
5678 bool record_synch_enabled = false;
5679 
5680 // in dbx, examine recorded data this way:
5681 // for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
5682 
5683 void record_synch(char* name, bool returning) {
5684   if (record_synch_enabled) {
5685     if (record_synch_count < RECORD_SYNCH_LIMIT) {
5686       record_synch_name[record_synch_count] = name;
5687       record_synch_returning[record_synch_count] = returning;
5688       record_synch_thread[record_synch_count] = thr_self();
5689       record_synch_arg0ptr[record_synch_count] = &name;
5690       record_synch_count++;
5691     }
5692     // put more checking code here:
5693     // ...
5694   }
5695 }
5696 
5697 void record_synch_enable() {
5698   // start collecting trace data, if not already doing so
5699   if (!record_synch_enabled)  record_synch_count = 0;
5700   record_synch_enabled = true;
5701 }
5702 
5703 void record_synch_disable() {
5704   // stop collecting trace data
5705   record_synch_enabled = false;
5706 }
5707 
5708 #endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5709 #endif // PRODUCT
5710 
5711 const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5712 const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
5713                                (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5714 
5715 
5716 // JVMTI & JVM monitoring and management support
5717 // The thread_cpu_time() and current_thread_cpu_time() are only
5718 // supported if is_thread_cpu_time_supported() returns true.
5719 // They are not supported on Solaris T1.
5720 
5721 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5722 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5723 // of a thread.
5724 //
5725 // current_thread_cpu_time() and thread_cpu_time(Thread *)
5726 // returns the fast estimate available on the platform.
5727 
5728 // hrtime_t gethrvtime() return value includes
5729 // user time but does not include system time
5730 jlong os::current_thread_cpu_time() {
5731   return (jlong) gethrvtime();
5732 }
5733 
5734 jlong os::thread_cpu_time(Thread *thread) {
5735   // return user level CPU time only to be consistent with
5736   // what current_thread_cpu_time returns.
5737   // thread_cpu_time_info() must be changed if this changes
5738   return os::thread_cpu_time(thread, false /* user time only */);
5739 }
5740 
5741 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5742   if (user_sys_cpu_time) {
5743     return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5744   } else {
5745     return os::current_thread_cpu_time();
5746   }
5747 }
5748 
5749 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5750   char proc_name[64];
5751   int count;
5752   prusage_t prusage;
5753   jlong lwp_time;
5754   int fd;
5755 
5756   sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
5757                      getpid(),
5758                      thread->osthread()->lwp_id());
5759   fd = ::open(proc_name, O_RDONLY);
5760   if ( fd == -1 ) return -1;
5761 
5762   do {
5763     count = ::pread(fd,
5764                   (void *)&prusage.pr_utime,
5765                   thr_time_size,
5766                   thr_time_off);
5767   } while (count < 0 && errno == EINTR);
5768   ::close(fd);
5769   if ( count < 0 ) return -1;
5770 
5771   if (user_sys_cpu_time) {
5772     // user + system CPU time
5773     lwp_time = (((jlong)prusage.pr_stime.tv_sec +
5774                  (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
5775                  (jlong)prusage.pr_stime.tv_nsec +
5776                  (jlong)prusage.pr_utime.tv_nsec;
5777   } else {
5778     // user level CPU time only
5779     lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
5780                 (jlong)prusage.pr_utime.tv_nsec;
5781   }
5782 
5783   return(lwp_time);
5784 }
5785 
5786 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5787   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5788   info_ptr->may_skip_backward = false;    // elapsed time not wall time
5789   info_ptr->may_skip_forward = false;     // elapsed time not wall time
5790   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5791 }
5792 
5793 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5794   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5795   info_ptr->may_skip_backward = false;    // elapsed time not wall time
5796   info_ptr->may_skip_forward = false;     // elapsed time not wall time
5797   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5798 }
5799 
5800 bool os::is_thread_cpu_time_supported() {
5801   if ( os::Solaris::T2_libthread() || UseBoundThreads ) {
5802     return true;
5803   } else {
5804     return false;
5805   }
5806 }
5807 
5808 // System loadavg support.  Returns -1 if load average cannot be obtained.
5809 // Return the load average for our processor set if the primitive exists
5810 // (Solaris 9 and later).  Otherwise just return system wide loadavg.
5811 int os::loadavg(double loadavg[], int nelem) {
5812   if (pset_getloadavg_ptr != NULL) {
5813     return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
5814   } else {
5815     return ::getloadavg(loadavg, nelem);
5816   }
5817 }
5818 
5819 //---------------------------------------------------------------------------------
5820 
5821 static address same_page(address x, address y) {
5822   intptr_t page_bits = -os::vm_page_size();
5823   if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
5824     return x;
5825   else if (x > y)
5826     return (address)(intptr_t(y) | ~page_bits) + 1;
5827   else
5828     return (address)(intptr_t(y) & page_bits);
5829 }
5830 
5831 bool os::find(address addr, outputStream* st) {
5832   Dl_info dlinfo;
5833   memset(&dlinfo, 0, sizeof(dlinfo));
5834   if (dladdr(addr, &dlinfo)) {
5835 #ifdef _LP64
5836     st->print("0x%016lx: ", addr);
5837 #else
5838     st->print("0x%08x: ", addr);
5839 #endif
5840     if (dlinfo.dli_sname != NULL)
5841       st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
5842     else if (dlinfo.dli_fname)
5843       st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
5844     else
5845       st->print("<absolute address>");
5846     if (dlinfo.dli_fname)  st->print(" in %s", dlinfo.dli_fname);
5847 #ifdef _LP64
5848     if (dlinfo.dli_fbase)  st->print(" at 0x%016lx", dlinfo.dli_fbase);
5849 #else
5850     if (dlinfo.dli_fbase)  st->print(" at 0x%08x", dlinfo.dli_fbase);
5851 #endif
5852     st->cr();
5853 
5854     if (Verbose) {
5855       // decode some bytes around the PC
5856       address begin = same_page(addr-40, addr);
5857       address end   = same_page(addr+40, addr);
5858       address       lowest = (address) dlinfo.dli_sname;
5859       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5860       if (begin < lowest)  begin = lowest;
5861       Dl_info dlinfo2;
5862       if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
5863           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5864         end = (address) dlinfo2.dli_saddr;
5865       Disassembler::decode(begin, end, st);
5866     }
5867     return true;
5868   }
5869   return false;
5870 }
5871 
5872 // Following function has been added to support HotSparc's libjvm.so running
5873 // under Solaris production JDK 1.2.2 / 1.3.0.  These came from
5874 // src/solaris/hpi/native_threads in the EVM codebase.
5875 //
5876 // NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
5877 // libraries and should thus be removed. We will leave it behind for a while
5878 // until we no longer want to able to run on top of 1.3.0 Solaris production
5879 // JDK. See 4341971.
5880 
5881 #define STACK_SLACK 0x800
5882 
5883 extern "C" {
5884   intptr_t sysThreadAvailableStackWithSlack() {
5885     stack_t st;
5886     intptr_t retval, stack_top;
5887     retval = thr_stksegment(&st);
5888     assert(retval == 0, "incorrect return value from thr_stksegment");
5889     assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
5890     assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
5891     stack_top=(intptr_t)st.ss_sp-st.ss_size;
5892     return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
5893   }
5894 }
5895 
5896 // Just to get the Kernel build to link on solaris for testing.
5897 
5898 extern "C" {
5899 class ASGCT_CallTrace;
5900 void AsyncGetCallTrace(ASGCT_CallTrace *trace, jint depth, void* ucontext)
5901   KERNEL_RETURN;
5902 }
5903 
5904 
5905 // ObjectMonitor park-unpark infrastructure ...
5906 //
5907 // We implement Solaris and Linux PlatformEvents with the
5908 // obvious condvar-mutex-flag triple.
5909 // Another alternative that works quite well is pipes:
5910 // Each PlatformEvent consists of a pipe-pair.
5911 // The thread associated with the PlatformEvent
5912 // calls park(), which reads from the input end of the pipe.
5913 // Unpark() writes into the other end of the pipe.
5914 // The write-side of the pipe must be set NDELAY.
5915 // Unfortunately pipes consume a large # of handles.
5916 // Native solaris lwp_park() and lwp_unpark() work nicely, too.
5917 // Using pipes for the 1st few threads might be workable, however.
5918 //
5919 // park() is permitted to return spuriously.
5920 // Callers of park() should wrap the call to park() in
5921 // an appropriate loop.  A litmus test for the correct
5922 // usage of park is the following: if park() were modified
5923 // to immediately return 0 your code should still work,
5924 // albeit degenerating to a spin loop.
5925 //
5926 // An interesting optimization for park() is to use a trylock()
5927 // to attempt to acquire the mutex.  If the trylock() fails
5928 // then we know that a concurrent unpark() operation is in-progress.
5929 // in that case the park() code could simply set _count to 0
5930 // and return immediately.  The subsequent park() operation *might*
5931 // return immediately.  That's harmless as the caller of park() is
5932 // expected to loop.  By using trylock() we will have avoided a
5933 // avoided a context switch caused by contention on the per-thread mutex.
5934 //
5935 // TODO-FIXME:
5936 // 1.  Reconcile Doug's JSR166 j.u.c park-unpark with the
5937 //     objectmonitor implementation.
5938 // 2.  Collapse the JSR166 parker event, and the
5939 //     objectmonitor ParkEvent into a single "Event" construct.
5940 // 3.  In park() and unpark() add:
5941 //     assert (Thread::current() == AssociatedWith).
5942 // 4.  add spurious wakeup injection on a -XX:EarlyParkReturn=N switch.
5943 //     1-out-of-N park() operations will return immediately.
5944 //
5945 // _Event transitions in park()
5946 //   -1 => -1 : illegal
5947 //    1 =>  0 : pass - return immediately
5948 //    0 => -1 : block
5949 //
5950 // _Event serves as a restricted-range semaphore.
5951 //
5952 // Another possible encoding of _Event would be with
5953 // explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
5954 //
5955 // TODO-FIXME: add DTRACE probes for:
5956 // 1.   Tx parks
5957 // 2.   Ty unparks Tx
5958 // 3.   Tx resumes from park
5959 
5960 
5961 // value determined through experimentation
5962 #define ROUNDINGFIX 11
5963 
5964 // utility to compute the abstime argument to timedwait.
5965 // TODO-FIXME: switch from compute_abstime() to unpackTime().
5966 
5967 static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
5968   // millis is the relative timeout time
5969   // abstime will be the absolute timeout time
5970   if (millis < 0)  millis = 0;
5971   struct timeval now;
5972   int status = gettimeofday(&now, NULL);
5973   assert(status == 0, "gettimeofday");
5974   jlong seconds = millis / 1000;
5975   jlong max_wait_period;
5976 
5977   if (UseLWPSynchronization) {
5978     // forward port of fix for 4275818 (not sleeping long enough)
5979     // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
5980     // _lwp_cond_timedwait() used a round_down algorithm rather
5981     // than a round_up. For millis less than our roundfactor
5982     // it rounded down to 0 which doesn't meet the spec.
5983     // For millis > roundfactor we may return a bit sooner, but
5984     // since we can not accurately identify the patch level and
5985     // this has already been fixed in Solaris 9 and 8 we will
5986     // leave it alone rather than always rounding down.
5987 
5988     if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
5989        // It appears that when we go directly through Solaris _lwp_cond_timedwait()
5990            // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
5991            max_wait_period = 21000000;
5992   } else {
5993     max_wait_period = 50000000;
5994   }
5995   millis %= 1000;
5996   if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
5997      seconds = max_wait_period;
5998   }
5999   abstime->tv_sec = now.tv_sec  + seconds;
6000   long       usec = now.tv_usec + millis * 1000;
6001   if (usec >= 1000000) {
6002     abstime->tv_sec += 1;
6003     usec -= 1000000;
6004   }
6005   abstime->tv_nsec = usec * 1000;
6006   return abstime;
6007 }
6008 
6009 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
6010 // Conceptually TryPark() should be equivalent to park(0).
6011 
6012 int os::PlatformEvent::TryPark() {
6013   for (;;) {
6014     const int v = _Event ;
6015     guarantee ((v == 0) || (v == 1), "invariant") ;
6016     if (Atomic::cmpxchg (0, &_Event, v) == v) return v  ;
6017   }
6018 }
6019 
6020 void os::PlatformEvent::park() {           // AKA: down()
6021   // Invariant: Only the thread associated with the Event/PlatformEvent
6022   // may call park().
6023   int v ;
6024   for (;;) {
6025       v = _Event ;
6026       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6027   }
6028   guarantee (v >= 0, "invariant") ;
6029   if (v == 0) {
6030      // Do this the hard way by blocking ...
6031      // See http://monaco.sfbay/detail.jsf?cr=5094058.
6032      // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
6033      // Only for SPARC >= V8PlusA
6034 #if defined(__sparc) && defined(COMPILER2)
6035      if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6036 #endif
6037      int status = os::Solaris::mutex_lock(_mutex);
6038      assert_status(status == 0, status,  "mutex_lock");
6039      guarantee (_nParked == 0, "invariant") ;
6040      ++ _nParked ;
6041      while (_Event < 0) {
6042         // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
6043         // Treat this the same as if the wait was interrupted
6044         // With usr/lib/lwp going to kernel, always handle ETIME
6045         status = os::Solaris::cond_wait(_cond, _mutex);
6046         if (status == ETIME) status = EINTR ;
6047         assert_status(status == 0 || status == EINTR, status, "cond_wait");
6048      }
6049      -- _nParked ;
6050      _Event = 0 ;
6051      status = os::Solaris::mutex_unlock(_mutex);
6052      assert_status(status == 0, status, "mutex_unlock");
6053   }
6054 }
6055 
6056 int os::PlatformEvent::park(jlong millis) {
6057   guarantee (_nParked == 0, "invariant") ;
6058   int v ;
6059   for (;;) {
6060       v = _Event ;
6061       if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6062   }
6063   guarantee (v >= 0, "invariant") ;
6064   if (v != 0) return OS_OK ;
6065 
6066   int ret = OS_TIMEOUT;
6067   timestruc_t abst;
6068   compute_abstime (&abst, millis);
6069 
6070   // See http://monaco.sfbay/detail.jsf?cr=5094058.
6071   // For Solaris SPARC set fprs.FEF=0 prior to parking.
6072   // Only for SPARC >= V8PlusA
6073 #if defined(__sparc) && defined(COMPILER2)
6074  if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6075 #endif
6076   int status = os::Solaris::mutex_lock(_mutex);
6077   assert_status(status == 0, status, "mutex_lock");
6078   guarantee (_nParked == 0, "invariant") ;
6079   ++ _nParked ;
6080   while (_Event < 0) {
6081      int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
6082      assert_status(status == 0 || status == EINTR ||
6083                    status == ETIME || status == ETIMEDOUT,
6084                    status, "cond_timedwait");
6085      if (!FilterSpuriousWakeups) break ;                // previous semantics
6086      if (status == ETIME || status == ETIMEDOUT) break ;
6087      // We consume and ignore EINTR and spurious wakeups.
6088   }
6089   -- _nParked ;
6090   if (_Event >= 0) ret = OS_OK ;
6091   _Event = 0 ;
6092   status = os::Solaris::mutex_unlock(_mutex);
6093   assert_status(status == 0, status, "mutex_unlock");
6094   return ret;
6095 }
6096 
6097 void os::PlatformEvent::unpark() {
6098   int v, AnyWaiters;
6099 
6100   // Increment _Event.
6101   // Another acceptable implementation would be to simply swap 1
6102   // into _Event:
6103   //   if (Swap (&_Event, 1) < 0) {
6104   //      mutex_lock (_mutex) ; AnyWaiters = nParked; mutex_unlock (_mutex) ;
6105   //      if (AnyWaiters) cond_signal (_cond) ;
6106   //   }
6107 
6108   for (;;) {
6109     v = _Event ;
6110     if (v > 0) {
6111        // The LD of _Event could have reordered or be satisfied
6112        // by a read-aside from this processor's write buffer.
6113        // To avoid problems execute a barrier and then
6114        // ratify the value.  A degenerate CAS() would also work.
6115        // Viz., CAS (v+0, &_Event, v) == v).
6116        OrderAccess::fence() ;
6117        if (_Event == v) return ;
6118        continue ;
6119     }
6120     if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
6121   }
6122 
6123   // If the thread associated with the event was parked, wake it.
6124   if (v < 0) {
6125      int status ;
6126      // Wait for the thread assoc with the PlatformEvent to vacate.
6127      status = os::Solaris::mutex_lock(_mutex);
6128      assert_status(status == 0, status, "mutex_lock");
6129      AnyWaiters = _nParked ;
6130      status = os::Solaris::mutex_unlock(_mutex);
6131      assert_status(status == 0, status, "mutex_unlock");
6132      guarantee (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
6133      if (AnyWaiters != 0) {
6134        // We intentional signal *after* dropping the lock
6135        // to avoid a common class of futile wakeups.
6136        status = os::Solaris::cond_signal(_cond);
6137        assert_status(status == 0, status, "cond_signal");
6138      }
6139   }
6140 }
6141 
6142 // JSR166
6143 // -------------------------------------------------------
6144 
6145 /*
6146  * The solaris and linux implementations of park/unpark are fairly
6147  * conservative for now, but can be improved. They currently use a
6148  * mutex/condvar pair, plus _counter.
6149  * Park decrements _counter if > 0, else does a condvar wait.  Unpark
6150  * sets count to 1 and signals condvar.  Only one thread ever waits
6151  * on the condvar. Contention seen when trying to park implies that someone
6152  * is unparking you, so don't wait. And spurious returns are fine, so there
6153  * is no need to track notifications.
6154  */
6155 
6156 #define MAX_SECS 100000000
6157 /*
6158  * This code is common to linux and solaris and will be moved to a
6159  * common place in dolphin.
6160  *
6161  * The passed in time value is either a relative time in nanoseconds
6162  * or an absolute time in milliseconds. Either way it has to be unpacked
6163  * into suitable seconds and nanoseconds components and stored in the
6164  * given timespec structure.
6165  * Given time is a 64-bit value and the time_t used in the timespec is only
6166  * a signed-32-bit value (except on 64-bit Linux) we have to watch for
6167  * overflow if times way in the future are given. Further on Solaris versions
6168  * prior to 10 there is a restriction (see cond_timedwait) that the specified
6169  * number of seconds, in abstime, is less than current_time  + 100,000,000.
6170  * As it will be 28 years before "now + 100000000" will overflow we can
6171  * ignore overflow and just impose a hard-limit on seconds using the value
6172  * of "now + 100,000,000". This places a limit on the timeout of about 3.17
6173  * years from "now".
6174  */
6175 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
6176   assert (time > 0, "convertTime");
6177 
6178   struct timeval now;
6179   int status = gettimeofday(&now, NULL);
6180   assert(status == 0, "gettimeofday");
6181 
6182   time_t max_secs = now.tv_sec + MAX_SECS;
6183 
6184   if (isAbsolute) {
6185     jlong secs = time / 1000;
6186     if (secs > max_secs) {
6187       absTime->tv_sec = max_secs;
6188     }
6189     else {
6190       absTime->tv_sec = secs;
6191     }
6192     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
6193   }
6194   else {
6195     jlong secs = time / NANOSECS_PER_SEC;
6196     if (secs >= MAX_SECS) {
6197       absTime->tv_sec = max_secs;
6198       absTime->tv_nsec = 0;
6199     }
6200     else {
6201       absTime->tv_sec = now.tv_sec + secs;
6202       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
6203       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6204         absTime->tv_nsec -= NANOSECS_PER_SEC;
6205         ++absTime->tv_sec; // note: this must be <= max_secs
6206       }
6207     }
6208   }
6209   assert(absTime->tv_sec >= 0, "tv_sec < 0");
6210   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
6211   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
6212   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
6213 }
6214 
6215 void Parker::park(bool isAbsolute, jlong time) {
6216 
6217   // Optional fast-path check:
6218   // Return immediately if a permit is available.
6219   if (_counter > 0) {
6220       _counter = 0 ;
6221       OrderAccess::fence();
6222       return ;
6223   }
6224 
6225   // Optional fast-exit: Check interrupt before trying to wait
6226   Thread* thread = Thread::current();
6227   assert(thread->is_Java_thread(), "Must be JavaThread");
6228   JavaThread *jt = (JavaThread *)thread;
6229   if (Thread::is_interrupted(thread, false)) {
6230     return;
6231   }
6232 
6233   // First, demultiplex/decode time arguments
6234   timespec absTime;
6235   if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
6236     return;
6237   }
6238   if (time > 0) {
6239     // Warning: this code might be exposed to the old Solaris time
6240     // round-down bugs.  Grep "roundingFix" for details.
6241     unpackTime(&absTime, isAbsolute, time);
6242   }
6243 
6244   // Enter safepoint region
6245   // Beware of deadlocks such as 6317397.
6246   // The per-thread Parker:: _mutex is a classic leaf-lock.
6247   // In particular a thread must never block on the Threads_lock while
6248   // holding the Parker:: mutex.  If safepoints are pending both the
6249   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6250   ThreadBlockInVM tbivm(jt);
6251 
6252   // Don't wait if cannot get lock since interference arises from
6253   // unblocking.  Also. check interrupt before trying wait
6254   if (Thread::is_interrupted(thread, false) ||
6255       os::Solaris::mutex_trylock(_mutex) != 0) {
6256     return;
6257   }
6258 
6259   int status ;
6260 
6261   if (_counter > 0)  { // no wait needed
6262     _counter = 0;
6263     status = os::Solaris::mutex_unlock(_mutex);
6264     assert (status == 0, "invariant") ;
6265     OrderAccess::fence();
6266     return;
6267   }
6268 
6269 #ifdef ASSERT
6270   // Don't catch signals while blocked; let the running threads have the signals.
6271   // (This allows a debugger to break into the running thread.)
6272   sigset_t oldsigs;
6273   sigset_t* allowdebug_blocked = os::Solaris::allowdebug_blocked_signals();
6274   thr_sigsetmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6275 #endif
6276 
6277   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6278   jt->set_suspend_equivalent();
6279   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6280 
6281   // Do this the hard way by blocking ...
6282   // See http://monaco.sfbay/detail.jsf?cr=5094058.
6283   // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
6284   // Only for SPARC >= V8PlusA
6285 #if defined(__sparc) && defined(COMPILER2)
6286   if (ClearFPUAtPark) { _mark_fpu_nosave() ; }
6287 #endif
6288 
6289   if (time == 0) {
6290     status = os::Solaris::cond_wait (_cond, _mutex) ;
6291   } else {
6292     status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
6293   }
6294   // Note that an untimed cond_wait() can sometimes return ETIME on older
6295   // versions of the Solaris.
6296   assert_status(status == 0 || status == EINTR ||
6297                 status == ETIME || status == ETIMEDOUT,
6298                 status, "cond_timedwait");
6299 
6300 #ifdef ASSERT
6301   thr_sigsetmask(SIG_SETMASK, &oldsigs, NULL);
6302 #endif
6303   _counter = 0 ;
6304   status = os::Solaris::mutex_unlock(_mutex);
6305   assert_status(status == 0, status, "mutex_unlock") ;
6306 
6307   // If externally suspended while waiting, re-suspend
6308   if (jt->handle_special_suspend_equivalent_condition()) {
6309     jt->java_suspend_self();
6310   }
6311   OrderAccess::fence();
6312 }
6313 
6314 void Parker::unpark() {
6315   int s, status ;
6316   status = os::Solaris::mutex_lock (_mutex) ;
6317   assert (status == 0, "invariant") ;
6318   s = _counter;
6319   _counter = 1;
6320   status = os::Solaris::mutex_unlock (_mutex) ;
6321   assert (status == 0, "invariant") ;
6322 
6323   if (s < 1) {
6324     status = os::Solaris::cond_signal (_cond) ;
6325     assert (status == 0, "invariant") ;
6326   }
6327 }
6328 
6329 extern char** environ;
6330 
6331 // Run the specified command in a separate process. Return its exit value,
6332 // or -1 on failure (e.g. can't fork a new process).
6333 // Unlike system(), this function can be called from signal handler. It
6334 // doesn't block SIGINT et al.
6335 int os::fork_and_exec(char* cmd) {
6336   char * argv[4];
6337   argv[0] = (char *)"sh";
6338   argv[1] = (char *)"-c";
6339   argv[2] = cmd;
6340   argv[3] = NULL;
6341 
6342   // fork is async-safe, fork1 is not so can't use in signal handler
6343   pid_t pid;
6344   Thread* t = ThreadLocalStorage::get_thread_slow();
6345   if (t != NULL && t->is_inside_signal_handler()) {
6346     pid = fork();
6347   } else {
6348     pid = fork1();
6349   }
6350 
6351   if (pid < 0) {
6352     // fork failed
6353     warning("fork failed: %s", strerror(errno));
6354     return -1;
6355 
6356   } else if (pid == 0) {
6357     // child process
6358 
6359     // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
6360     execve("/usr/bin/sh", argv, environ);
6361 
6362     // execve failed
6363     _exit(-1);
6364 
6365   } else  {
6366     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6367     // care about the actual exit code, for now.
6368 
6369     int status;
6370 
6371     // Wait for the child process to exit.  This returns immediately if
6372     // the child has already exited. */
6373     while (waitpid(pid, &status, 0) < 0) {
6374         switch (errno) {
6375         case ECHILD: return 0;
6376         case EINTR: break;
6377         default: return -1;
6378         }
6379     }
6380 
6381     if (WIFEXITED(status)) {
6382        // The child exited normally; get its exit code.
6383        return WEXITSTATUS(status);
6384     } else if (WIFSIGNALED(status)) {
6385        // The child exited because of a signal
6386        // The best value to return is 0x80 + signal number,
6387        // because that is what all Unix shells do, and because
6388        // it allows callers to distinguish between process exit and
6389        // process death by signal.
6390        return 0x80 + WTERMSIG(status);
6391     } else {
6392        // Unknown exit code; pass it through
6393        return status;
6394     }
6395   }
6396 }
6397 
6398 // is_headless_jre()
6399 //
6400 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6401 // in order to report if we are running in a headless jre
6402 //
6403 // Since JDK8 xawt/libmawt.so was moved into the same directory
6404 // as libawt.so, and renamed libawt_xawt.so
6405 //
6406 bool os::is_headless_jre() {
6407     struct stat statbuf;
6408     char buf[MAXPATHLEN];
6409     char libmawtpath[MAXPATHLEN];
6410     const char *xawtstr  = "/xawt/libmawt.so";
6411     const char *new_xawtstr = "/libawt_xawt.so";
6412     char *p;
6413 
6414     // Get path to libjvm.so
6415     os::jvm_path(buf, sizeof(buf));
6416 
6417     // Get rid of libjvm.so
6418     p = strrchr(buf, '/');
6419     if (p == NULL) return false;
6420     else *p = '\0';
6421 
6422     // Get rid of client or server
6423     p = strrchr(buf, '/');
6424     if (p == NULL) return false;
6425     else *p = '\0';
6426 
6427     // check xawt/libmawt.so
6428     strcpy(libmawtpath, buf);
6429     strcat(libmawtpath, xawtstr);
6430     if (::stat(libmawtpath, &statbuf) == 0) return false;
6431 
6432     // check libawt_xawt.so
6433     strcpy(libmawtpath, buf);
6434     strcat(libmawtpath, new_xawtstr);
6435     if (::stat(libmawtpath, &statbuf) == 0) return false;
6436 
6437     return true;
6438 }
6439 
6440 size_t os::write(int fd, const void *buf, unsigned int nBytes) {
6441   INTERRUPTIBLE_RETURN_INT(::write(fd, buf, nBytes), os::Solaris::clear_interrupted);
6442 }
6443 
6444 int os::close(int fd) {
6445   RESTARTABLE_RETURN_INT(::close(fd));
6446 }
6447 
6448 int os::socket_close(int fd) {
6449   RESTARTABLE_RETURN_INT(::close(fd));
6450 }
6451 
6452 int os::recv(int fd, char* buf, int nBytes, int flags) {
6453   INTERRUPTIBLE_RETURN_INT((int)::recv(fd, buf, nBytes, flags), os::Solaris::clear_interrupted);
6454 }
6455 
6456 int os::send(int fd, char* buf, int nBytes, int flags) {
6457   INTERRUPTIBLE_RETURN_INT((int)::send(fd, buf, nBytes, flags), os::Solaris::clear_interrupted);
6458 }
6459 
6460 int os::raw_send(int fd, char* buf, int nBytes, int flags) {
6461   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
6462 }
6463 
6464 // As both poll and select can be interrupted by signals, we have to be
6465 // prepared to restart the system call after updating the timeout, unless
6466 // a poll() is done with timeout == -1, in which case we repeat with this
6467 // "wait forever" value.
6468 
6469 int os::timeout(int fd, long timeout) {
6470   int res;
6471   struct timeval t;
6472   julong prevtime, newtime;
6473   static const char* aNull = 0;
6474   struct pollfd pfd;
6475   pfd.fd = fd;
6476   pfd.events = POLLIN;
6477 
6478   gettimeofday(&t, &aNull);
6479   prevtime = ((julong)t.tv_sec * 1000)  +  t.tv_usec / 1000;
6480 
6481   for(;;) {
6482     INTERRUPTIBLE_NORESTART(::poll(&pfd, 1, timeout), res, os::Solaris::clear_interrupted);
6483     if(res == OS_ERR && errno == EINTR) {
6484         if(timeout != -1) {
6485           gettimeofday(&t, &aNull);
6486           newtime = ((julong)t.tv_sec * 1000)  +  t.tv_usec /1000;
6487           timeout -= newtime - prevtime;
6488           if(timeout <= 0)
6489             return OS_OK;
6490           prevtime = newtime;
6491         }
6492     } else return res;
6493   }
6494 }
6495 
6496 int os::connect(int fd, struct sockaddr *him, int len) {
6497   int _result;
6498   INTERRUPTIBLE_NORESTART(::connect(fd, him, len), _result,\
6499                           os::Solaris::clear_interrupted);
6500 
6501   // Depending on when thread interruption is reset, _result could be
6502   // one of two values when errno == EINTR
6503 
6504   if (((_result == OS_INTRPT) || (_result == OS_ERR))
6505       && (errno == EINTR)) {
6506      /* restarting a connect() changes its errno semantics */
6507      INTERRUPTIBLE(::connect(fd, him, len), _result,\
6508                    os::Solaris::clear_interrupted);
6509      /* undo these changes */
6510      if (_result == OS_ERR) {
6511        if (errno == EALREADY) {
6512          errno = EINPROGRESS; /* fall through */
6513        } else if (errno == EISCONN) {
6514          errno = 0;
6515          return OS_OK;
6516        }
6517      }
6518    }
6519    return _result;
6520  }
6521 
6522 int os::accept(int fd, struct sockaddr* him, int* len) {
6523   if (fd < 0) {
6524     return OS_ERR;
6525   }
6526   INTERRUPTIBLE_RETURN_INT((int)::accept(fd, him, len),\
6527                            os::Solaris::clear_interrupted);
6528 }
6529 
6530 int os::recvfrom(int fd, char* buf, int nBytes, int flags,
6531                  sockaddr* from, int* fromlen) {
6532   INTERRUPTIBLE_RETURN_INT((int)::recvfrom(fd, buf, nBytes, flags, from, fromlen),\
6533                            os::Solaris::clear_interrupted);
6534 }
6535 
6536 int os::sendto(int fd, char* buf, int len, int flags,
6537                struct sockaddr* to, int tolen) {
6538   INTERRUPTIBLE_RETURN_INT((int)::sendto(fd, buf, len, flags, to, tolen),\
6539                            os::Solaris::clear_interrupted);
6540 }
6541 
6542 int os::socket_available(int fd, jint *pbytes) {
6543   if (fd < 0) {
6544     return OS_OK;
6545   }
6546   int ret;
6547   RESTARTABLE(::ioctl(fd, FIONREAD, pbytes), ret);
6548   // note: ioctl can return 0 when successful, JVM_SocketAvailable
6549   // is expected to return 0 on failure and 1 on success to the jdk.
6550   return (ret == OS_ERR) ? 0 : 1;
6551 }
6552 
6553 int os::bind(int fd, struct sockaddr* him, int len) {
6554    INTERRUPTIBLE_RETURN_INT_NORESTART(::bind(fd, him, len),\
6555                                       os::Solaris::clear_interrupted);
6556 }