1 /*
   2  * Copyright (c) 1997, 2015, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_solaris.h"
  35 #include "memory/allocation.inline.hpp"
  36 #include "memory/filemap.hpp"
  37 #include "mutex_solaris.inline.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_share_solaris.hpp"
  40 #include "os_solaris.inline.hpp"
  41 #include "prims/jniFastGetField.hpp"
  42 #include "prims/jvm.h"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.inline.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.hpp"
  49 #include "runtime/java.hpp"
  50 #include "runtime/javaCalls.hpp"
  51 #include "runtime/mutexLocker.hpp"
  52 #include "runtime/objectMonitor.hpp"
  53 #include "runtime/orderAccess.inline.hpp"
  54 #include "runtime/osThread.hpp"
  55 #include "runtime/perfMemory.hpp"
  56 #include "runtime/sharedRuntime.hpp"
  57 #include "runtime/statSampler.hpp"
  58 #include "runtime/stubRoutines.hpp"
  59 #include "runtime/thread.inline.hpp"
  60 #include "runtime/threadCritical.hpp"
  61 #include "runtime/timer.hpp"
  62 #include "runtime/vm_version.hpp"
  63 #include "services/attachListener.hpp"
  64 #include "services/memTracker.hpp"
  65 #include "services/runtimeService.hpp"
  66 #include "utilities/decoder.hpp"
  67 #include "utilities/defaultStream.hpp"
  68 #include "utilities/events.hpp"
  69 #include "utilities/growableArray.hpp"
  70 #include "utilities/vmError.hpp"
  71 
  72 // put OS-includes here
  73 # include <dlfcn.h>
  74 # include <errno.h>
  75 # include <exception>
  76 # include <link.h>
  77 # include <poll.h>
  78 # include <pthread.h>
  79 # include <pwd.h>
  80 # include <schedctl.h>
  81 # include <setjmp.h>
  82 # include <signal.h>
  83 # include <stdio.h>
  84 # include <alloca.h>
  85 # include <sys/filio.h>
  86 # include <sys/ipc.h>
  87 # include <sys/lwp.h>
  88 # include <sys/machelf.h>     // for elf Sym structure used by dladdr1
  89 # include <sys/mman.h>
  90 # include <sys/processor.h>
  91 # include <sys/procset.h>
  92 # include <sys/pset.h>
  93 # include <sys/resource.h>
  94 # include <sys/shm.h>
  95 # include <sys/socket.h>
  96 # include <sys/stat.h>
  97 # include <sys/systeminfo.h>
  98 # include <sys/time.h>
  99 # include <sys/times.h>
 100 # include <sys/types.h>
 101 # include <sys/wait.h>
 102 # include <sys/utsname.h>
 103 # include <thread.h>
 104 # include <unistd.h>
 105 # include <sys/priocntl.h>
 106 # include <sys/rtpriocntl.h>
 107 # include <sys/tspriocntl.h>
 108 # include <sys/iapriocntl.h>
 109 # include <sys/fxpriocntl.h>
 110 # include <sys/loadavg.h>
 111 # include <string.h>
 112 # include <stdio.h>
 113 
 114 # define _STRUCTURED_PROC 1  //  this gets us the new structured proc interfaces of 5.6 & later
 115 # include <sys/procfs.h>     //  see comment in <sys/procfs.h>
 116 
 117 #define MAX_PATH (2 * K)
 118 
 119 // for timer info max values which include all bits
 120 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 121 
 122 
 123 // Here are some liblgrp types from sys/lgrp_user.h to be able to
 124 // compile on older systems without this header file.
 125 
 126 #ifndef MADV_ACCESS_LWP
 127   #define  MADV_ACCESS_LWP   7       /* next LWP to access heavily */
 128 #endif
 129 #ifndef MADV_ACCESS_MANY
 130   #define  MADV_ACCESS_MANY  8       /* many processes to access heavily */
 131 #endif
 132 
 133 #ifndef LGRP_RSRC_CPU
 134   #define LGRP_RSRC_CPU      0       /* CPU resources */
 135 #endif
 136 #ifndef LGRP_RSRC_MEM
 137   #define LGRP_RSRC_MEM      1       /* memory resources */
 138 #endif
 139 
 140 // see thr_setprio(3T) for the basis of these numbers
 141 #define MinimumPriority 0
 142 #define NormalPriority  64
 143 #define MaximumPriority 127
 144 
 145 // Values for ThreadPriorityPolicy == 1
 146 int prio_policy1[CriticalPriority+1] = {
 147   -99999,  0, 16,  32,  48,  64,
 148           80, 96, 112, 124, 127, 127 };
 149 
 150 // System parameters used internally
 151 static clock_t clock_tics_per_sec = 100;
 152 
 153 // Track if we have called enable_extended_FILE_stdio (on Solaris 10u4+)
 154 static bool enabled_extended_FILE_stdio = false;
 155 
 156 // For diagnostics to print a message once. see run_periodic_checks
 157 static bool check_addr0_done = false;
 158 static sigset_t check_signal_done;
 159 static bool check_signals = true;
 160 
 161 address os::Solaris::handler_start;  // start pc of thr_sighndlrinfo
 162 address os::Solaris::handler_end;    // end pc of thr_sighndlrinfo
 163 
 164 address os::Solaris::_main_stack_base = NULL;  // 4352906 workaround
 165 
 166 
 167 // "default" initializers for missing libc APIs
 168 extern "C" {
 169   static int lwp_mutex_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
 170   static int lwp_mutex_destroy(mutex_t *mx)                 { return 0; }
 171 
 172   static int lwp_cond_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
 173   static int lwp_cond_destroy(cond_t *cv)                   { return 0; }
 174 }
 175 
 176 // "default" initializers for pthread-based synchronization
 177 extern "C" {
 178   static int pthread_mutex_default_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
 179   static int pthread_cond_default_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
 180 }
 181 
 182 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 183 
 184 // Thread Local Storage
 185 // This is common to all Solaris platforms so it is defined here,
 186 // in this common file.
 187 // The declarations are in the os_cpu threadLS*.hpp files.
 188 //
 189 // Static member initialization for TLS
 190 Thread* ThreadLocalStorage::_get_thread_cache[ThreadLocalStorage::_pd_cache_size] = {NULL};
 191 
 192 #ifndef PRODUCT
 193   #define _PCT(n,d)       ((100.0*(double)(n))/(double)(d))
 194 
 195 int ThreadLocalStorage::_tcacheHit = 0;
 196 int ThreadLocalStorage::_tcacheMiss = 0;
 197 
 198 void ThreadLocalStorage::print_statistics() {
 199   int total = _tcacheMiss+_tcacheHit;
 200   tty->print_cr("Thread cache hits %d misses %d total %d percent %f\n",
 201                 _tcacheHit, _tcacheMiss, total, _PCT(_tcacheHit, total));
 202 }
 203   #undef _PCT
 204 #endif // PRODUCT
 205 
 206 Thread* ThreadLocalStorage::get_thread_via_cache_slowly(uintptr_t raw_id,
 207                                                         int index) {
 208   Thread *thread = get_thread_slow();
 209   if (thread != NULL) {
 210     address sp = os::current_stack_pointer();
 211     guarantee(thread->_stack_base == NULL ||
 212               (sp <= thread->_stack_base &&
 213               sp >= thread->_stack_base - thread->_stack_size) ||
 214               is_error_reported(),
 215               "sp must be inside of selected thread stack");
 216 
 217     thread->set_self_raw_id(raw_id);  // mark for quick retrieval
 218     _get_thread_cache[index] = thread;
 219   }
 220   return thread;
 221 }
 222 
 223 
 224 static const double all_zero[sizeof(Thread) / sizeof(double) + 1] = {0};
 225 #define NO_CACHED_THREAD ((Thread*)all_zero)
 226 
 227 void ThreadLocalStorage::pd_set_thread(Thread* thread) {
 228 
 229   // Store the new value before updating the cache to prevent a race
 230   // between get_thread_via_cache_slowly() and this store operation.
 231   os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
 232 
 233   // Update thread cache with new thread if setting on thread create,
 234   // or NO_CACHED_THREAD (zeroed) thread if resetting thread on exit.
 235   uintptr_t raw = pd_raw_thread_id();
 236   int ix = pd_cache_index(raw);
 237   _get_thread_cache[ix] = thread == NULL ? NO_CACHED_THREAD : thread;
 238 }
 239 
 240 void ThreadLocalStorage::pd_init() {
 241   for (int i = 0; i < _pd_cache_size; i++) {
 242     _get_thread_cache[i] = NO_CACHED_THREAD;
 243   }
 244 }
 245 
 246 // Invalidate all the caches (happens to be the same as pd_init).
 247 void ThreadLocalStorage::pd_invalidate_all() { pd_init(); }
 248 
 249 #undef NO_CACHED_THREAD
 250 
 251 // END Thread Local Storage
 252 
 253 static inline size_t adjust_stack_size(address base, size_t size) {
 254   if ((ssize_t)size < 0) {
 255     // 4759953: Compensate for ridiculous stack size.
 256     size = max_intx;
 257   }
 258   if (size > (size_t)base) {
 259     // 4812466: Make sure size doesn't allow the stack to wrap the address space.
 260     size = (size_t)base;
 261   }
 262   return size;
 263 }
 264 
 265 static inline stack_t get_stack_info() {
 266   stack_t st;
 267   int retval = thr_stksegment(&st);
 268   st.ss_size = adjust_stack_size((address)st.ss_sp, st.ss_size);
 269   assert(retval == 0, "incorrect return value from thr_stksegment");
 270   assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
 271   assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
 272   return st;
 273 }
 274 
 275 address os::current_stack_base() {
 276   int r = thr_main();
 277   guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
 278   bool is_primordial_thread = r;
 279 
 280   // Workaround 4352906, avoid calls to thr_stksegment by
 281   // thr_main after the first one (it looks like we trash
 282   // some data, causing the value for ss_sp to be incorrect).
 283   if (!is_primordial_thread || os::Solaris::_main_stack_base == NULL) {
 284     stack_t st = get_stack_info();
 285     if (is_primordial_thread) {
 286       // cache initial value of stack base
 287       os::Solaris::_main_stack_base = (address)st.ss_sp;
 288     }
 289     return (address)st.ss_sp;
 290   } else {
 291     guarantee(os::Solaris::_main_stack_base != NULL, "Attempt to use null cached stack base");
 292     return os::Solaris::_main_stack_base;
 293   }
 294 }
 295 
 296 size_t os::current_stack_size() {
 297   size_t size;
 298 
 299   int r = thr_main();
 300   guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
 301   if (!r) {
 302     size = get_stack_info().ss_size;
 303   } else {
 304     struct rlimit limits;
 305     getrlimit(RLIMIT_STACK, &limits);
 306     size = adjust_stack_size(os::Solaris::_main_stack_base, (size_t)limits.rlim_cur);
 307   }
 308   // base may not be page aligned
 309   address base = current_stack_base();
 310   address bottom = (address)align_size_up((intptr_t)(base - size), os::vm_page_size());;
 311   return (size_t)(base - bottom);
 312 }
 313 
 314 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
 315   return localtime_r(clock, res);
 316 }
 317 
 318 void os::Solaris::try_enable_extended_io() {
 319   typedef int (*enable_extended_FILE_stdio_t)(int, int);
 320 
 321   if (!UseExtendedFileIO) {
 322     return;
 323   }
 324 
 325   enable_extended_FILE_stdio_t enabler =
 326     (enable_extended_FILE_stdio_t) dlsym(RTLD_DEFAULT,
 327                                          "enable_extended_FILE_stdio");
 328   if (enabler) {
 329     enabler(-1, -1);
 330   }
 331 }
 332 
 333 static int _processors_online = 0;
 334 
 335 jint os::Solaris::_os_thread_limit = 0;
 336 volatile jint os::Solaris::_os_thread_count = 0;
 337 
 338 julong os::available_memory() {
 339   return Solaris::available_memory();
 340 }
 341 
 342 julong os::Solaris::available_memory() {
 343   return (julong)sysconf(_SC_AVPHYS_PAGES) * os::vm_page_size();
 344 }
 345 
 346 julong os::Solaris::_physical_memory = 0;
 347 
 348 julong os::physical_memory() {
 349   return Solaris::physical_memory();
 350 }
 351 
 352 static hrtime_t first_hrtime = 0;
 353 static const hrtime_t hrtime_hz = 1000*1000*1000;
 354 static volatile hrtime_t max_hrtime = 0;
 355 
 356 
 357 void os::Solaris::initialize_system_info() {
 358   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 359   _processors_online = sysconf(_SC_NPROCESSORS_ONLN);
 360   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) *
 361                                      (julong)sysconf(_SC_PAGESIZE);
 362 }
 363 
 364 int os::active_processor_count() {
 365   int online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
 366   pid_t pid = getpid();
 367   psetid_t pset = PS_NONE;
 368   // Are we running in a processor set or is there any processor set around?
 369   if (pset_bind(PS_QUERY, P_PID, pid, &pset) == 0) {
 370     uint_t pset_cpus;
 371     // Query the number of cpus available to us.
 372     if (pset_info(pset, NULL, &pset_cpus, NULL) == 0) {
 373       assert(pset_cpus > 0 && pset_cpus <= online_cpus, "sanity check");
 374       _processors_online = pset_cpus;
 375       return pset_cpus;
 376     }
 377   }
 378   // Otherwise return number of online cpus
 379   return online_cpus;
 380 }
 381 
 382 static bool find_processors_in_pset(psetid_t        pset,
 383                                     processorid_t** id_array,
 384                                     uint_t*         id_length) {
 385   bool result = false;
 386   // Find the number of processors in the processor set.
 387   if (pset_info(pset, NULL, id_length, NULL) == 0) {
 388     // Make up an array to hold their ids.
 389     *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
 390     // Fill in the array with their processor ids.
 391     if (pset_info(pset, NULL, id_length, *id_array) == 0) {
 392       result = true;
 393     }
 394   }
 395   return result;
 396 }
 397 
 398 // Callers of find_processors_online() must tolerate imprecise results --
 399 // the system configuration can change asynchronously because of DR
 400 // or explicit psradm operations.
 401 //
 402 // We also need to take care that the loop (below) terminates as the
 403 // number of processors online can change between the _SC_NPROCESSORS_ONLN
 404 // request and the loop that builds the list of processor ids.   Unfortunately
 405 // there's no reliable way to determine the maximum valid processor id,
 406 // so we use a manifest constant, MAX_PROCESSOR_ID, instead.  See p_online
 407 // man pages, which claim the processor id set is "sparse, but
 408 // not too sparse".  MAX_PROCESSOR_ID is used to ensure that we eventually
 409 // exit the loop.
 410 //
 411 // In the future we'll be able to use sysconf(_SC_CPUID_MAX), but that's
 412 // not available on S8.0.
 413 
 414 static bool find_processors_online(processorid_t** id_array,
 415                                    uint*           id_length) {
 416   const processorid_t MAX_PROCESSOR_ID = 100000;
 417   // Find the number of processors online.
 418   *id_length = sysconf(_SC_NPROCESSORS_ONLN);
 419   // Make up an array to hold their ids.
 420   *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
 421   // Processors need not be numbered consecutively.
 422   long found = 0;
 423   processorid_t next = 0;
 424   while (found < *id_length && next < MAX_PROCESSOR_ID) {
 425     processor_info_t info;
 426     if (processor_info(next, &info) == 0) {
 427       // NB, PI_NOINTR processors are effectively online ...
 428       if (info.pi_state == P_ONLINE || info.pi_state == P_NOINTR) {
 429         (*id_array)[found] = next;
 430         found += 1;
 431       }
 432     }
 433     next += 1;
 434   }
 435   if (found < *id_length) {
 436     // The loop above didn't identify the expected number of processors.
 437     // We could always retry the operation, calling sysconf(_SC_NPROCESSORS_ONLN)
 438     // and re-running the loop, above, but there's no guarantee of progress
 439     // if the system configuration is in flux.  Instead, we just return what
 440     // we've got.  Note that in the worst case find_processors_online() could
 441     // return an empty set.  (As a fall-back in the case of the empty set we
 442     // could just return the ID of the current processor).
 443     *id_length = found;
 444   }
 445 
 446   return true;
 447 }
 448 
 449 static bool assign_distribution(processorid_t* id_array,
 450                                 uint           id_length,
 451                                 uint*          distribution,
 452                                 uint           distribution_length) {
 453   // We assume we can assign processorid_t's to uint's.
 454   assert(sizeof(processorid_t) == sizeof(uint),
 455          "can't convert processorid_t to uint");
 456   // Quick check to see if we won't succeed.
 457   if (id_length < distribution_length) {
 458     return false;
 459   }
 460   // Assign processor ids to the distribution.
 461   // Try to shuffle processors to distribute work across boards,
 462   // assuming 4 processors per board.
 463   const uint processors_per_board = ProcessDistributionStride;
 464   // Find the maximum processor id.
 465   processorid_t max_id = 0;
 466   for (uint m = 0; m < id_length; m += 1) {
 467     max_id = MAX2(max_id, id_array[m]);
 468   }
 469   // The next id, to limit loops.
 470   const processorid_t limit_id = max_id + 1;
 471   // Make up markers for available processors.
 472   bool* available_id = NEW_C_HEAP_ARRAY(bool, limit_id, mtInternal);
 473   for (uint c = 0; c < limit_id; c += 1) {
 474     available_id[c] = false;
 475   }
 476   for (uint a = 0; a < id_length; a += 1) {
 477     available_id[id_array[a]] = true;
 478   }
 479   // Step by "boards", then by "slot", copying to "assigned".
 480   // NEEDS_CLEANUP: The assignment of processors should be stateful,
 481   //                remembering which processors have been assigned by
 482   //                previous calls, etc., so as to distribute several
 483   //                independent calls of this method.  What we'd like is
 484   //                It would be nice to have an API that let us ask
 485   //                how many processes are bound to a processor,
 486   //                but we don't have that, either.
 487   //                In the short term, "board" is static so that
 488   //                subsequent distributions don't all start at board 0.
 489   static uint board = 0;
 490   uint assigned = 0;
 491   // Until we've found enough processors ....
 492   while (assigned < distribution_length) {
 493     // ... find the next available processor in the board.
 494     for (uint slot = 0; slot < processors_per_board; slot += 1) {
 495       uint try_id = board * processors_per_board + slot;
 496       if ((try_id < limit_id) && (available_id[try_id] == true)) {
 497         distribution[assigned] = try_id;
 498         available_id[try_id] = false;
 499         assigned += 1;
 500         break;
 501       }
 502     }
 503     board += 1;
 504     if (board * processors_per_board + 0 >= limit_id) {
 505       board = 0;
 506     }
 507   }
 508   if (available_id != NULL) {
 509     FREE_C_HEAP_ARRAY(bool, available_id);
 510   }
 511   return true;
 512 }
 513 
 514 void os::set_native_thread_name(const char *name) {
 515   // Not yet implemented.
 516   return;
 517 }
 518 
 519 bool os::distribute_processes(uint length, uint* distribution) {
 520   bool result = false;
 521   // Find the processor id's of all the available CPUs.
 522   processorid_t* id_array  = NULL;
 523   uint           id_length = 0;
 524   // There are some races between querying information and using it,
 525   // since processor sets can change dynamically.
 526   psetid_t pset = PS_NONE;
 527   // Are we running in a processor set?
 528   if ((pset_bind(PS_QUERY, P_PID, P_MYID, &pset) == 0) && pset != PS_NONE) {
 529     result = find_processors_in_pset(pset, &id_array, &id_length);
 530   } else {
 531     result = find_processors_online(&id_array, &id_length);
 532   }
 533   if (result == true) {
 534     if (id_length >= length) {
 535       result = assign_distribution(id_array, id_length, distribution, length);
 536     } else {
 537       result = false;
 538     }
 539   }
 540   if (id_array != NULL) {
 541     FREE_C_HEAP_ARRAY(processorid_t, id_array);
 542   }
 543   return result;
 544 }
 545 
 546 bool os::bind_to_processor(uint processor_id) {
 547   // We assume that a processorid_t can be stored in a uint.
 548   assert(sizeof(uint) == sizeof(processorid_t),
 549          "can't convert uint to processorid_t");
 550   int bind_result =
 551     processor_bind(P_LWPID,                       // bind LWP.
 552                    P_MYID,                        // bind current LWP.
 553                    (processorid_t) processor_id,  // id.
 554                    NULL);                         // don't return old binding.
 555   return (bind_result == 0);
 556 }
 557 
 558 // Return true if user is running as root.
 559 
 560 bool os::have_special_privileges() {
 561   static bool init = false;
 562   static bool privileges = false;
 563   if (!init) {
 564     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 565     init = true;
 566   }
 567   return privileges;
 568 }
 569 
 570 
 571 void os::init_system_properties_values() {
 572   // The next steps are taken in the product version:
 573   //
 574   // Obtain the JAVA_HOME value from the location of libjvm.so.
 575   // This library should be located at:
 576   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 577   //
 578   // If "/jre/lib/" appears at the right place in the path, then we
 579   // assume libjvm.so is installed in a JDK and we use this path.
 580   //
 581   // Otherwise exit with message: "Could not create the Java virtual machine."
 582   //
 583   // The following extra steps are taken in the debugging version:
 584   //
 585   // If "/jre/lib/" does NOT appear at the right place in the path
 586   // instead of exit check for $JAVA_HOME environment variable.
 587   //
 588   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 589   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 590   // it looks like libjvm.so is installed there
 591   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 592   //
 593   // Otherwise exit.
 594   //
 595   // Important note: if the location of libjvm.so changes this
 596   // code needs to be changed accordingly.
 597 
 598 // Base path of extensions installed on the system.
 599 #define SYS_EXT_DIR     "/usr/jdk/packages"
 600 #define EXTENSIONS_DIR  "/lib/ext"
 601 
 602   char cpu_arch[12];
 603   // Buffer that fits several sprintfs.
 604   // Note that the space for the colon and the trailing null are provided
 605   // by the nulls included by the sizeof operator.
 606   const size_t bufsize =
 607     MAX3((size_t)MAXPATHLEN,  // For dll_dir & friends.
 608          sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch), // invariant ld_library_path
 609          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 610   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 611 
 612   // sysclasspath, java_home, dll_dir
 613   {
 614     char *pslash;
 615     os::jvm_path(buf, bufsize);
 616 
 617     // Found the full path to libjvm.so.
 618     // Now cut the path to <java_home>/jre if we can.
 619     *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
 620     pslash = strrchr(buf, '/');
 621     if (pslash != NULL) {
 622       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 623     }
 624     Arguments::set_dll_dir(buf);
 625 
 626     if (pslash != NULL) {
 627       pslash = strrchr(buf, '/');
 628       if (pslash != NULL) {
 629         *pslash = '\0';          // Get rid of /<arch>.
 630         pslash = strrchr(buf, '/');
 631         if (pslash != NULL) {
 632           *pslash = '\0';        // Get rid of /lib.
 633         }
 634       }
 635     }
 636     Arguments::set_java_home(buf);
 637     set_boot_path('/', ':');
 638   }
 639 
 640   // Where to look for native libraries.
 641   {
 642     // Use dlinfo() to determine the correct java.library.path.
 643     //
 644     // If we're launched by the Java launcher, and the user
 645     // does not set java.library.path explicitly on the commandline,
 646     // the Java launcher sets LD_LIBRARY_PATH for us and unsets
 647     // LD_LIBRARY_PATH_32 and LD_LIBRARY_PATH_64.  In this case
 648     // dlinfo returns LD_LIBRARY_PATH + crle settings (including
 649     // /usr/lib), which is exactly what we want.
 650     //
 651     // If the user does set java.library.path, it completely
 652     // overwrites this setting, and always has.
 653     //
 654     // If we're not launched by the Java launcher, we may
 655     // get here with any/all of the LD_LIBRARY_PATH[_32|64]
 656     // settings.  Again, dlinfo does exactly what we want.
 657 
 658     Dl_serinfo     info_sz, *info = &info_sz;
 659     Dl_serpath     *path;
 660     char           *library_path;
 661     char           *common_path = buf;
 662 
 663     // Determine search path count and required buffer size.
 664     if (dlinfo(RTLD_SELF, RTLD_DI_SERINFOSIZE, (void *)info) == -1) {
 665       FREE_C_HEAP_ARRAY(char, buf);
 666       vm_exit_during_initialization("dlinfo SERINFOSIZE request", dlerror());
 667     }
 668 
 669     // Allocate new buffer and initialize.
 670     info = (Dl_serinfo*)NEW_C_HEAP_ARRAY(char, info_sz.dls_size, mtInternal);
 671     info->dls_size = info_sz.dls_size;
 672     info->dls_cnt = info_sz.dls_cnt;
 673 
 674     // Obtain search path information.
 675     if (dlinfo(RTLD_SELF, RTLD_DI_SERINFO, (void *)info) == -1) {
 676       FREE_C_HEAP_ARRAY(char, buf);
 677       FREE_C_HEAP_ARRAY(char, info);
 678       vm_exit_during_initialization("dlinfo SERINFO request", dlerror());
 679     }
 680 
 681     path = &info->dls_serpath[0];
 682 
 683     // Note: Due to a legacy implementation, most of the library path
 684     // is set in the launcher. This was to accomodate linking restrictions
 685     // on legacy Solaris implementations (which are no longer supported).
 686     // Eventually, all the library path setting will be done here.
 687     //
 688     // However, to prevent the proliferation of improperly built native
 689     // libraries, the new path component /usr/jdk/packages is added here.
 690 
 691     // Determine the actual CPU architecture.
 692     sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
 693 #ifdef _LP64
 694     // If we are a 64-bit vm, perform the following translations:
 695     //   sparc   -> sparcv9
 696     //   i386    -> amd64
 697     if (strcmp(cpu_arch, "sparc") == 0) {
 698       strcat(cpu_arch, "v9");
 699     } else if (strcmp(cpu_arch, "i386") == 0) {
 700       strcpy(cpu_arch, "amd64");
 701     }
 702 #endif
 703 
 704     // Construct the invariant part of ld_library_path.
 705     sprintf(common_path, SYS_EXT_DIR "/lib/%s", cpu_arch);
 706 
 707     // Struct size is more than sufficient for the path components obtained
 708     // through the dlinfo() call, so only add additional space for the path
 709     // components explicitly added here.
 710     size_t library_path_size = info->dls_size + strlen(common_path);
 711     library_path = (char *)NEW_C_HEAP_ARRAY(char, library_path_size, mtInternal);
 712     library_path[0] = '\0';
 713 
 714     // Construct the desired Java library path from the linker's library
 715     // search path.
 716     //
 717     // For compatibility, it is optimal that we insert the additional path
 718     // components specific to the Java VM after those components specified
 719     // in LD_LIBRARY_PATH (if any) but before those added by the ld.so
 720     // infrastructure.
 721     if (info->dls_cnt == 0) { // Not sure this can happen, but allow for it.
 722       strcpy(library_path, common_path);
 723     } else {
 724       int inserted = 0;
 725       int i;
 726       for (i = 0; i < info->dls_cnt; i++, path++) {
 727         uint_t flags = path->dls_flags & LA_SER_MASK;
 728         if (((flags & LA_SER_LIBPATH) == 0) && !inserted) {
 729           strcat(library_path, common_path);
 730           strcat(library_path, os::path_separator());
 731           inserted = 1;
 732         }
 733         strcat(library_path, path->dls_name);
 734         strcat(library_path, os::path_separator());
 735       }
 736       // Eliminate trailing path separator.
 737       library_path[strlen(library_path)-1] = '\0';
 738     }
 739 
 740     // happens before argument parsing - can't use a trace flag
 741     // tty->print_raw("init_system_properties_values: native lib path: ");
 742     // tty->print_raw_cr(library_path);
 743 
 744     // Callee copies into its own buffer.
 745     Arguments::set_library_path(library_path);
 746 
 747     FREE_C_HEAP_ARRAY(char, library_path);
 748     FREE_C_HEAP_ARRAY(char, info);
 749   }
 750 
 751   // Extensions directories.
 752   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 753   Arguments::set_ext_dirs(buf);
 754 
 755   FREE_C_HEAP_ARRAY(char, buf);
 756 
 757 #undef SYS_EXT_DIR
 758 #undef EXTENSIONS_DIR
 759 }
 760 
 761 void os::breakpoint() {
 762   BREAKPOINT;
 763 }
 764 
 765 bool os::obsolete_option(const JavaVMOption *option) {
 766   if (!strncmp(option->optionString, "-Xt", 3)) {
 767     return true;
 768   } else if (!strncmp(option->optionString, "-Xtm", 4)) {
 769     return true;
 770   } else if (!strncmp(option->optionString, "-Xverifyheap", 12)) {
 771     return true;
 772   } else if (!strncmp(option->optionString, "-Xmaxjitcodesize", 16)) {
 773     return true;
 774   }
 775   return false;
 776 }
 777 
 778 bool os::Solaris::valid_stack_address(Thread* thread, address sp) {
 779   address  stackStart  = (address)thread->stack_base();
 780   address  stackEnd    = (address)(stackStart - (address)thread->stack_size());
 781   if (sp < stackStart && sp >= stackEnd) return true;
 782   return false;
 783 }
 784 
 785 extern "C" void breakpoint() {
 786   // use debugger to set breakpoint here
 787 }
 788 
 789 static thread_t main_thread;
 790 
 791 // Thread start routine for all new Java threads
 792 extern "C" void* java_start(void* thread_addr) {
 793   // Try to randomize the cache line index of hot stack frames.
 794   // This helps when threads of the same stack traces evict each other's
 795   // cache lines. The threads can be either from the same JVM instance, or
 796   // from different JVM instances. The benefit is especially true for
 797   // processors with hyperthreading technology.
 798   static int counter = 0;
 799   int pid = os::current_process_id();
 800   alloca(((pid ^ counter++) & 7) * 128);
 801 
 802   int prio;
 803   Thread* thread = (Thread*)thread_addr;
 804   OSThread* osthr = thread->osthread();
 805 
 806   osthr->set_lwp_id(_lwp_self());  // Store lwp in case we are bound
 807   thread->_schedctl = (void *) schedctl_init();
 808 
 809   if (UseNUMA) {
 810     int lgrp_id = os::numa_get_group_id();
 811     if (lgrp_id != -1) {
 812       thread->set_lgrp_id(lgrp_id);
 813     }
 814   }
 815 
 816   // If the creator called set priority before we started,
 817   // we need to call set_native_priority now that we have an lwp.
 818   // We used to get the priority from thr_getprio (we called
 819   // thr_setprio way back in create_thread) and pass it to
 820   // set_native_priority, but Solaris scales the priority
 821   // in java_to_os_priority, so when we read it back here,
 822   // we pass trash to set_native_priority instead of what's
 823   // in java_to_os_priority. So we save the native priority
 824   // in the osThread and recall it here.
 825 
 826   if (osthr->thread_id() != -1) {
 827     if (UseThreadPriorities) {
 828       int prio = osthr->native_priority();
 829       if (ThreadPriorityVerbose) {
 830         tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is "
 831                       INTPTR_FORMAT ", setting priority: %d\n",
 832                       osthr->thread_id(), osthr->lwp_id(), prio);
 833       }
 834       os::set_native_priority(thread, prio);
 835     }
 836   } else if (ThreadPriorityVerbose) {
 837     warning("Can't set priority in _start routine, thread id hasn't been set\n");
 838   }
 839 
 840   assert(osthr->get_state() == RUNNABLE, "invalid os thread state");
 841 
 842   // initialize signal mask for this thread
 843   os::Solaris::hotspot_sigmask(thread);
 844 
 845   thread->run();
 846 
 847   // One less thread is executing
 848   // When the VMThread gets here, the main thread may have already exited
 849   // which frees the CodeHeap containing the Atomic::dec code
 850   if (thread != VMThread::vm_thread() && VMThread::vm_thread() != NULL) {
 851     Atomic::dec(&os::Solaris::_os_thread_count);
 852   }
 853 
 854   if (UseDetachedThreads) {
 855     thr_exit(NULL);
 856     ShouldNotReachHere();
 857   }
 858   return NULL;
 859 }
 860 
 861 static OSThread* create_os_thread(Thread* thread, thread_t thread_id) {
 862   // Allocate the OSThread object
 863   OSThread* osthread = new OSThread(NULL, NULL);
 864   if (osthread == NULL) return NULL;
 865 
 866   // Store info on the Solaris thread into the OSThread
 867   osthread->set_thread_id(thread_id);
 868   osthread->set_lwp_id(_lwp_self());
 869   thread->_schedctl = (void *) schedctl_init();
 870 
 871   if (UseNUMA) {
 872     int lgrp_id = os::numa_get_group_id();
 873     if (lgrp_id != -1) {
 874       thread->set_lgrp_id(lgrp_id);
 875     }
 876   }
 877 
 878   if (ThreadPriorityVerbose) {
 879     tty->print_cr("In create_os_thread, Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT "\n",
 880                   osthread->thread_id(), osthread->lwp_id());
 881   }
 882 
 883   // Initial thread state is INITIALIZED, not SUSPENDED
 884   osthread->set_state(INITIALIZED);
 885 
 886   return osthread;
 887 }
 888 
 889 void os::Solaris::hotspot_sigmask(Thread* thread) {
 890   //Save caller's signal mask
 891   sigset_t sigmask;
 892   thr_sigsetmask(SIG_SETMASK, NULL, &sigmask);
 893   OSThread *osthread = thread->osthread();
 894   osthread->set_caller_sigmask(sigmask);
 895 
 896   thr_sigsetmask(SIG_UNBLOCK, os::Solaris::unblocked_signals(), NULL);
 897   if (!ReduceSignalUsage) {
 898     if (thread->is_VM_thread()) {
 899       // Only the VM thread handles BREAK_SIGNAL ...
 900       thr_sigsetmask(SIG_UNBLOCK, vm_signals(), NULL);
 901     } else {
 902       // ... all other threads block BREAK_SIGNAL
 903       assert(!sigismember(vm_signals(), SIGINT), "SIGINT should not be blocked");
 904       thr_sigsetmask(SIG_BLOCK, vm_signals(), NULL);
 905     }
 906   }
 907 }
 908 
 909 bool os::create_attached_thread(JavaThread* thread) {
 910 #ifdef ASSERT
 911   thread->verify_not_published();
 912 #endif
 913   OSThread* osthread = create_os_thread(thread, thr_self());
 914   if (osthread == NULL) {
 915     return false;
 916   }
 917 
 918   // Initial thread state is RUNNABLE
 919   osthread->set_state(RUNNABLE);
 920   thread->set_osthread(osthread);
 921 
 922   // initialize signal mask for this thread
 923   // and save the caller's signal mask
 924   os::Solaris::hotspot_sigmask(thread);
 925 
 926   return true;
 927 }
 928 
 929 bool os::create_main_thread(JavaThread* thread) {
 930 #ifdef ASSERT
 931   thread->verify_not_published();
 932 #endif
 933   if (_starting_thread == NULL) {
 934     _starting_thread = create_os_thread(thread, main_thread);
 935     if (_starting_thread == NULL) {
 936       return false;
 937     }
 938   }
 939 
 940   // The primodial thread is runnable from the start
 941   _starting_thread->set_state(RUNNABLE);
 942 
 943   thread->set_osthread(_starting_thread);
 944 
 945   // initialize signal mask for this thread
 946   // and save the caller's signal mask
 947   os::Solaris::hotspot_sigmask(thread);
 948 
 949   return true;
 950 }
 951 
 952 
 953 bool os::create_thread(Thread* thread, ThreadType thr_type,
 954                        size_t stack_size) {
 955   // Allocate the OSThread object
 956   OSThread* osthread = new OSThread(NULL, NULL);
 957   if (osthread == NULL) {
 958     return false;
 959   }
 960 
 961   if (ThreadPriorityVerbose) {
 962     char *thrtyp;
 963     switch (thr_type) {
 964     case vm_thread:
 965       thrtyp = (char *)"vm";
 966       break;
 967     case cgc_thread:
 968       thrtyp = (char *)"cgc";
 969       break;
 970     case pgc_thread:
 971       thrtyp = (char *)"pgc";
 972       break;
 973     case java_thread:
 974       thrtyp = (char *)"java";
 975       break;
 976     case compiler_thread:
 977       thrtyp = (char *)"compiler";
 978       break;
 979     case watcher_thread:
 980       thrtyp = (char *)"watcher";
 981       break;
 982     default:
 983       thrtyp = (char *)"unknown";
 984       break;
 985     }
 986     tty->print_cr("In create_thread, creating a %s thread\n", thrtyp);
 987   }
 988 
 989   // Calculate stack size if it's not specified by caller.
 990   if (stack_size == 0) {
 991     // The default stack size 1M (2M for LP64).
 992     stack_size = (BytesPerWord >> 2) * K * K;
 993 
 994     switch (thr_type) {
 995     case os::java_thread:
 996       // Java threads use ThreadStackSize which default value can be changed with the flag -Xss
 997       if (JavaThread::stack_size_at_create() > 0) stack_size = JavaThread::stack_size_at_create();
 998       break;
 999     case os::compiler_thread:
1000       if (CompilerThreadStackSize > 0) {
1001         stack_size = (size_t)(CompilerThreadStackSize * K);
1002         break;
1003       } // else fall through:
1004         // use VMThreadStackSize if CompilerThreadStackSize is not defined
1005     case os::vm_thread:
1006     case os::pgc_thread:
1007     case os::cgc_thread:
1008     case os::watcher_thread:
1009       if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
1010       break;
1011     }
1012   }
1013   stack_size = MAX2(stack_size, os::Solaris::min_stack_allowed);
1014 
1015   // Initial state is ALLOCATED but not INITIALIZED
1016   osthread->set_state(ALLOCATED);
1017 
1018   if (os::Solaris::_os_thread_count > os::Solaris::_os_thread_limit) {
1019     // We got lots of threads. Check if we still have some address space left.
1020     // Need to be at least 5Mb of unreserved address space. We do check by
1021     // trying to reserve some.
1022     const size_t VirtualMemoryBangSize = 20*K*K;
1023     char* mem = os::reserve_memory(VirtualMemoryBangSize);
1024     if (mem == NULL) {
1025       delete osthread;
1026       return false;
1027     } else {
1028       // Release the memory again
1029       os::release_memory(mem, VirtualMemoryBangSize);
1030     }
1031   }
1032 
1033   // Setup osthread because the child thread may need it.
1034   thread->set_osthread(osthread);
1035 
1036   // Create the Solaris thread
1037   thread_t tid = 0;
1038   long     flags = (UseDetachedThreads ? THR_DETACHED : 0) | THR_SUSPENDED;
1039   int      status;
1040 
1041   // Mark that we don't have an lwp or thread id yet.
1042   // In case we attempt to set the priority before the thread starts.
1043   osthread->set_lwp_id(-1);
1044   osthread->set_thread_id(-1);
1045 
1046   status = thr_create(NULL, stack_size, java_start, thread, flags, &tid);
1047   if (status != 0) {
1048     if (PrintMiscellaneous && (Verbose || WizardMode)) {
1049       perror("os::create_thread");
1050     }
1051     thread->set_osthread(NULL);
1052     // Need to clean up stuff we've allocated so far
1053     delete osthread;
1054     return false;
1055   }
1056 
1057   Atomic::inc(&os::Solaris::_os_thread_count);
1058 
1059   // Store info on the Solaris thread into the OSThread
1060   osthread->set_thread_id(tid);
1061 
1062   // Remember that we created this thread so we can set priority on it
1063   osthread->set_vm_created();
1064 
1065   // Initial thread state is INITIALIZED, not SUSPENDED
1066   osthread->set_state(INITIALIZED);
1067 
1068   // The thread is returned suspended (in state INITIALIZED), and is started higher up in the call chain
1069   return true;
1070 }
1071 
1072 // defined for >= Solaris 10. This allows builds on earlier versions
1073 // of Solaris to take advantage of the newly reserved Solaris JVM signals
1074 // With SIGJVM1, SIGJVM2, INTERRUPT_SIGNAL is SIGJVM1, ASYNC_SIGNAL is SIGJVM2
1075 // and -XX:+UseAltSigs does nothing since these should have no conflict
1076 //
1077 #if !defined(SIGJVM1)
1078   #define SIGJVM1 39
1079   #define SIGJVM2 40
1080 #endif
1081 
1082 debug_only(static bool signal_sets_initialized = false);
1083 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
1084 int os::Solaris::_SIGinterrupt = INTERRUPT_SIGNAL;
1085 int os::Solaris::_SIGasync = ASYNC_SIGNAL;
1086 
1087 bool os::Solaris::is_sig_ignored(int sig) {
1088   struct sigaction oact;
1089   sigaction(sig, (struct sigaction*)NULL, &oact);
1090   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
1091                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
1092   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
1093     return true;
1094   } else {
1095     return false;
1096   }
1097 }
1098 
1099 // Note: SIGRTMIN is a macro that calls sysconf() so it will
1100 // dynamically detect SIGRTMIN value for the system at runtime, not buildtime
1101 static bool isJVM1available() {
1102   return SIGJVM1 < SIGRTMIN;
1103 }
1104 
1105 void os::Solaris::signal_sets_init() {
1106   // Should also have an assertion stating we are still single-threaded.
1107   assert(!signal_sets_initialized, "Already initialized");
1108   // Fill in signals that are necessarily unblocked for all threads in
1109   // the VM. Currently, we unblock the following signals:
1110   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
1111   //                         by -Xrs (=ReduceSignalUsage));
1112   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
1113   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
1114   // the dispositions or masks wrt these signals.
1115   // Programs embedding the VM that want to use the above signals for their
1116   // own purposes must, at this time, use the "-Xrs" option to prevent
1117   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
1118   // (See bug 4345157, and other related bugs).
1119   // In reality, though, unblocking these signals is really a nop, since
1120   // these signals are not blocked by default.
1121   sigemptyset(&unblocked_sigs);
1122   sigemptyset(&allowdebug_blocked_sigs);
1123   sigaddset(&unblocked_sigs, SIGILL);
1124   sigaddset(&unblocked_sigs, SIGSEGV);
1125   sigaddset(&unblocked_sigs, SIGBUS);
1126   sigaddset(&unblocked_sigs, SIGFPE);
1127 
1128   if (isJVM1available) {
1129     os::Solaris::set_SIGinterrupt(SIGJVM1);
1130     os::Solaris::set_SIGasync(SIGJVM2);
1131   } else if (UseAltSigs) {
1132     os::Solaris::set_SIGinterrupt(ALT_INTERRUPT_SIGNAL);
1133     os::Solaris::set_SIGasync(ALT_ASYNC_SIGNAL);
1134   } else {
1135     os::Solaris::set_SIGinterrupt(INTERRUPT_SIGNAL);
1136     os::Solaris::set_SIGasync(ASYNC_SIGNAL);
1137   }
1138 
1139   sigaddset(&unblocked_sigs, os::Solaris::SIGinterrupt());
1140   sigaddset(&unblocked_sigs, os::Solaris::SIGasync());
1141 
1142   if (!ReduceSignalUsage) {
1143     if (!os::Solaris::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
1144       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
1145       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
1146     }
1147     if (!os::Solaris::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
1148       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
1149       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
1150     }
1151     if (!os::Solaris::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
1152       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
1153       sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
1154     }
1155   }
1156   // Fill in signals that are blocked by all but the VM thread.
1157   sigemptyset(&vm_sigs);
1158   if (!ReduceSignalUsage) {
1159     sigaddset(&vm_sigs, BREAK_SIGNAL);
1160   }
1161   debug_only(signal_sets_initialized = true);
1162 
1163   // For diagnostics only used in run_periodic_checks
1164   sigemptyset(&check_signal_done);
1165 }
1166 
1167 // These are signals that are unblocked while a thread is running Java.
1168 // (For some reason, they get blocked by default.)
1169 sigset_t* os::Solaris::unblocked_signals() {
1170   assert(signal_sets_initialized, "Not initialized");
1171   return &unblocked_sigs;
1172 }
1173 
1174 // These are the signals that are blocked while a (non-VM) thread is
1175 // running Java. Only the VM thread handles these signals.
1176 sigset_t* os::Solaris::vm_signals() {
1177   assert(signal_sets_initialized, "Not initialized");
1178   return &vm_sigs;
1179 }
1180 
1181 // These are signals that are blocked during cond_wait to allow debugger in
1182 sigset_t* os::Solaris::allowdebug_blocked_signals() {
1183   assert(signal_sets_initialized, "Not initialized");
1184   return &allowdebug_blocked_sigs;
1185 }
1186 
1187 
1188 void _handle_uncaught_cxx_exception() {
1189   VMError err("An uncaught C++ exception");
1190   err.report_and_die();
1191 }
1192 
1193 
1194 // First crack at OS-specific initialization, from inside the new thread.
1195 void os::initialize_thread(Thread* thr) {
1196   int r = thr_main();
1197   guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
1198   if (r) {
1199     JavaThread* jt = (JavaThread *)thr;
1200     assert(jt != NULL, "Sanity check");
1201     size_t stack_size;
1202     address base = jt->stack_base();
1203     if (Arguments::created_by_java_launcher()) {
1204       // Use 2MB to allow for Solaris 7 64 bit mode.
1205       stack_size = JavaThread::stack_size_at_create() == 0
1206         ? 2048*K : JavaThread::stack_size_at_create();
1207 
1208       // There are rare cases when we may have already used more than
1209       // the basic stack size allotment before this method is invoked.
1210       // Attempt to allow for a normally sized java_stack.
1211       size_t current_stack_offset = (size_t)(base - (address)&stack_size);
1212       stack_size += ReservedSpace::page_align_size_down(current_stack_offset);
1213     } else {
1214       // 6269555: If we were not created by a Java launcher, i.e. if we are
1215       // running embedded in a native application, treat the primordial thread
1216       // as much like a native attached thread as possible.  This means using
1217       // the current stack size from thr_stksegment(), unless it is too large
1218       // to reliably setup guard pages.  A reasonable max size is 8MB.
1219       size_t current_size = current_stack_size();
1220       // This should never happen, but just in case....
1221       if (current_size == 0) current_size = 2 * K * K;
1222       stack_size = current_size > (8 * K * K) ? (8 * K * K) : current_size;
1223     }
1224     address bottom = (address)align_size_up((intptr_t)(base - stack_size), os::vm_page_size());;
1225     stack_size = (size_t)(base - bottom);
1226 
1227     assert(stack_size > 0, "Stack size calculation problem");
1228 
1229     if (stack_size > jt->stack_size()) {
1230 #ifndef PRODUCT
1231       struct rlimit limits;
1232       getrlimit(RLIMIT_STACK, &limits);
1233       size_t size = adjust_stack_size(base, (size_t)limits.rlim_cur);
1234       assert(size >= jt->stack_size(), "Stack size problem in main thread");
1235 #endif
1236       tty->print_cr("Stack size of %d Kb exceeds current limit of %d Kb.\n"
1237                     "(Stack sizes are rounded up to a multiple of the system page size.)\n"
1238                     "See limit(1) to increase the stack size limit.",
1239                     stack_size / K, jt->stack_size() / K);
1240       vm_exit(1);
1241     }
1242     assert(jt->stack_size() >= stack_size,
1243            "Attempt to map more stack than was allocated");
1244     jt->set_stack_size(stack_size);
1245   }
1246 
1247   // With the T2 libthread (T1 is no longer supported) threads are always bound
1248   // and we use stackbanging in all cases.
1249 
1250   os::Solaris::init_thread_fpu_state();
1251   std::set_terminate(_handle_uncaught_cxx_exception);
1252 }
1253 
1254 
1255 
1256 // Free Solaris resources related to the OSThread
1257 void os::free_thread(OSThread* osthread) {
1258   assert(osthread != NULL, "os::free_thread but osthread not set");
1259 
1260 
1261   // We are told to free resources of the argument thread,
1262   // but we can only really operate on the current thread.
1263   // The main thread must take the VMThread down synchronously
1264   // before the main thread exits and frees up CodeHeap
1265   guarantee((Thread::current()->osthread() == osthread
1266              || (osthread == VMThread::vm_thread()->osthread())), "os::free_thread but not current thread");
1267   if (Thread::current()->osthread() == osthread) {
1268     // Restore caller's signal mask
1269     sigset_t sigmask = osthread->caller_sigmask();
1270     thr_sigsetmask(SIG_SETMASK, &sigmask, NULL);
1271   }
1272   delete osthread;
1273 }
1274 
1275 void os::pd_start_thread(Thread* thread) {
1276   int status = thr_continue(thread->osthread()->thread_id());
1277   assert_status(status == 0, status, "thr_continue failed");
1278 }
1279 
1280 
1281 intx os::current_thread_id() {
1282   return (intx)thr_self();
1283 }
1284 
1285 static pid_t _initial_pid = 0;
1286 
1287 int os::current_process_id() {
1288   return (int)(_initial_pid ? _initial_pid : getpid());
1289 }
1290 
1291 int os::allocate_thread_local_storage() {
1292   // %%%       in Win32 this allocates a memory segment pointed to by a
1293   //           register.  Dan Stein can implement a similar feature in
1294   //           Solaris.  Alternatively, the VM can do the same thing
1295   //           explicitly: malloc some storage and keep the pointer in a
1296   //           register (which is part of the thread's context) (or keep it
1297   //           in TLS).
1298   // %%%       In current versions of Solaris, thr_self and TSD can
1299   //           be accessed via short sequences of displaced indirections.
1300   //           The value of thr_self is available as %g7(36).
1301   //           The value of thr_getspecific(k) is stored in %g7(12)(4)(k*4-4),
1302   //           assuming that the current thread already has a value bound to k.
1303   //           It may be worth experimenting with such access patterns,
1304   //           and later having the parameters formally exported from a Solaris
1305   //           interface.  I think, however, that it will be faster to
1306   //           maintain the invariant that %g2 always contains the
1307   //           JavaThread in Java code, and have stubs simply
1308   //           treat %g2 as a caller-save register, preserving it in a %lN.
1309   thread_key_t tk;
1310   if (thr_keycreate(&tk, NULL)) {
1311     fatal(err_msg("os::allocate_thread_local_storage: thr_keycreate failed "
1312                   "(%s)", strerror(errno)));
1313   }
1314   return int(tk);
1315 }
1316 
1317 void os::free_thread_local_storage(int index) {
1318   // %%% don't think we need anything here
1319   // if (pthread_key_delete((pthread_key_t) tk)) {
1320   //   fatal("os::free_thread_local_storage: pthread_key_delete failed");
1321   // }
1322 }
1323 
1324 // libthread allocate for tsd_common is a version specific
1325 // small number - point is NO swap space available
1326 #define SMALLINT 32
1327 void os::thread_local_storage_at_put(int index, void* value) {
1328   // %%% this is used only in threadLocalStorage.cpp
1329   if (thr_setspecific((thread_key_t)index, value)) {
1330     if (errno == ENOMEM) {
1331       vm_exit_out_of_memory(SMALLINT, OOM_MALLOC_ERROR,
1332                             "thr_setspecific: out of swap space");
1333     } else {
1334       fatal(err_msg("os::thread_local_storage_at_put: thr_setspecific failed "
1335                     "(%s)", strerror(errno)));
1336     }
1337   } else {
1338     ThreadLocalStorage::set_thread_in_slot((Thread *) value);
1339   }
1340 }
1341 
1342 // This function could be called before TLS is initialized, for example, when
1343 // VM receives an async signal or when VM causes a fatal error during
1344 // initialization. Return NULL if thr_getspecific() fails.
1345 void* os::thread_local_storage_at(int index) {
1346   // %%% this is used only in threadLocalStorage.cpp
1347   void* r = NULL;
1348   return thr_getspecific((thread_key_t)index, &r) != 0 ? NULL : r;
1349 }
1350 
1351 
1352 // gethrtime() should be monotonic according to the documentation,
1353 // but some virtualized platforms are known to break this guarantee.
1354 // getTimeNanos() must be guaranteed not to move backwards, so we
1355 // are forced to add a check here.
1356 inline hrtime_t getTimeNanos() {
1357   const hrtime_t now = gethrtime();
1358   const hrtime_t prev = max_hrtime;
1359   if (now <= prev) {
1360     return prev;   // same or retrograde time;
1361   }
1362   const hrtime_t obsv = Atomic::cmpxchg(now, (volatile jlong*)&max_hrtime, prev);
1363   assert(obsv >= prev, "invariant");   // Monotonicity
1364   // If the CAS succeeded then we're done and return "now".
1365   // If the CAS failed and the observed value "obsv" is >= now then
1366   // we should return "obsv".  If the CAS failed and now > obsv > prv then
1367   // some other thread raced this thread and installed a new value, in which case
1368   // we could either (a) retry the entire operation, (b) retry trying to install now
1369   // or (c) just return obsv.  We use (c).   No loop is required although in some cases
1370   // we might discard a higher "now" value in deference to a slightly lower but freshly
1371   // installed obsv value.   That's entirely benign -- it admits no new orderings compared
1372   // to (a) or (b) -- and greatly reduces coherence traffic.
1373   // We might also condition (c) on the magnitude of the delta between obsv and now.
1374   // Avoiding excessive CAS operations to hot RW locations is critical.
1375   // See https://blogs.oracle.com/dave/entry/cas_and_cache_trivia_invalidate
1376   return (prev == obsv) ? now : obsv;
1377 }
1378 
1379 // Time since start-up in seconds to a fine granularity.
1380 // Used by VMSelfDestructTimer and the MemProfiler.
1381 double os::elapsedTime() {
1382   return (double)(getTimeNanos() - first_hrtime) / (double)hrtime_hz;
1383 }
1384 
1385 jlong os::elapsed_counter() {
1386   return (jlong)(getTimeNanos() - first_hrtime);
1387 }
1388 
1389 jlong os::elapsed_frequency() {
1390   return hrtime_hz;
1391 }
1392 
1393 // Return the real, user, and system times in seconds from an
1394 // arbitrary fixed point in the past.
1395 bool os::getTimesSecs(double* process_real_time,
1396                       double* process_user_time,
1397                       double* process_system_time) {
1398   struct tms ticks;
1399   clock_t real_ticks = times(&ticks);
1400 
1401   if (real_ticks == (clock_t) (-1)) {
1402     return false;
1403   } else {
1404     double ticks_per_second = (double) clock_tics_per_sec;
1405     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1406     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1407     // For consistency return the real time from getTimeNanos()
1408     // converted to seconds.
1409     *process_real_time = ((double) getTimeNanos()) / ((double) NANOUNITS);
1410 
1411     return true;
1412   }
1413 }
1414 
1415 bool os::supports_vtime() { return true; }
1416 
1417 bool os::enable_vtime() {
1418   int fd = ::open("/proc/self/ctl", O_WRONLY);
1419   if (fd == -1) {
1420     return false;
1421   }
1422 
1423   long cmd[] = { PCSET, PR_MSACCT };
1424   int res = ::write(fd, cmd, sizeof(long) * 2);
1425   ::close(fd);
1426   if (res != sizeof(long) * 2) {
1427     return false;
1428   }
1429   return true;
1430 }
1431 
1432 bool os::vtime_enabled() {
1433   int fd = ::open("/proc/self/status", O_RDONLY);
1434   if (fd == -1) {
1435     return false;
1436   }
1437 
1438   pstatus_t status;
1439   int res = os::read(fd, (void*) &status, sizeof(pstatus_t));
1440   ::close(fd);
1441   if (res != sizeof(pstatus_t)) {
1442     return false;
1443   }
1444   return status.pr_flags & PR_MSACCT;
1445 }
1446 
1447 double os::elapsedVTime() {
1448   return (double)gethrvtime() / (double)hrtime_hz;
1449 }
1450 
1451 // Used internally for comparisons only
1452 // getTimeMillis guaranteed to not move backwards on Solaris
1453 jlong getTimeMillis() {
1454   jlong nanotime = getTimeNanos();
1455   return (jlong)(nanotime / NANOSECS_PER_MILLISEC);
1456 }
1457 
1458 // Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1459 jlong os::javaTimeMillis() {
1460   timeval t;
1461   if (gettimeofday(&t, NULL) == -1) {
1462     fatal(err_msg("os::javaTimeMillis: gettimeofday (%s)", strerror(errno)));
1463   }
1464   return jlong(t.tv_sec) * 1000  +  jlong(t.tv_usec) / 1000;
1465 }
1466 
1467 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1468   timeval t;
1469   if (gettimeofday(&t, NULL) == -1) {
1470     fatal(err_msg("os::javaTimeSystemUTC: gettimeofday (%s)", strerror(errno)));
1471   }
1472   seconds = jlong(t.tv_sec);
1473   nanos = jlong(t.tv_usec) * 1000;
1474 }
1475 
1476 
1477 jlong os::javaTimeNanos() {
1478   return (jlong)getTimeNanos();
1479 }
1480 
1481 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1482   info_ptr->max_value = ALL_64_BITS;      // gethrtime() uses all 64 bits
1483   info_ptr->may_skip_backward = false;    // not subject to resetting or drifting
1484   info_ptr->may_skip_forward = false;     // not subject to resetting or drifting
1485   info_ptr->kind = JVMTI_TIMER_ELAPSED;   // elapsed not CPU time
1486 }
1487 
1488 char * os::local_time_string(char *buf, size_t buflen) {
1489   struct tm t;
1490   time_t long_time;
1491   time(&long_time);
1492   localtime_r(&long_time, &t);
1493   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1494                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1495                t.tm_hour, t.tm_min, t.tm_sec);
1496   return buf;
1497 }
1498 
1499 // Note: os::shutdown() might be called very early during initialization, or
1500 // called from signal handler. Before adding something to os::shutdown(), make
1501 // sure it is async-safe and can handle partially initialized VM.
1502 void os::shutdown() {
1503 
1504   // allow PerfMemory to attempt cleanup of any persistent resources
1505   perfMemory_exit();
1506 
1507   // needs to remove object in file system
1508   AttachListener::abort();
1509 
1510   // flush buffered output, finish log files
1511   ostream_abort();
1512 
1513   // Check for abort hook
1514   abort_hook_t abort_hook = Arguments::abort_hook();
1515   if (abort_hook != NULL) {
1516     abort_hook();
1517   }
1518 }
1519 
1520 // Note: os::abort() might be called very early during initialization, or
1521 // called from signal handler. Before adding something to os::abort(), make
1522 // sure it is async-safe and can handle partially initialized VM.
1523 void os::abort(bool dump_core) {
1524   os::shutdown();
1525   if (dump_core) {
1526 #ifndef PRODUCT
1527     fdStream out(defaultStream::output_fd());
1528     out.print_raw("Current thread is ");
1529     char buf[16];
1530     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1531     out.print_raw_cr(buf);
1532     out.print_raw_cr("Dumping core ...");
1533 #endif
1534     ::abort(); // dump core (for debugging)
1535   }
1536 
1537   ::exit(1);
1538 }
1539 
1540 // Die immediately, no exit hook, no abort hook, no cleanup.
1541 void os::die() {
1542   ::abort(); // dump core (for debugging)
1543 }
1544 
1545 // DLL functions
1546 
1547 const char* os::dll_file_extension() { return ".so"; }
1548 
1549 // This must be hard coded because it's the system's temporary
1550 // directory not the java application's temp directory, ala java.io.tmpdir.
1551 const char* os::get_temp_directory() { return "/tmp"; }
1552 
1553 static bool file_exists(const char* filename) {
1554   struct stat statbuf;
1555   if (filename == NULL || strlen(filename) == 0) {
1556     return false;
1557   }
1558   return os::stat(filename, &statbuf) == 0;
1559 }
1560 
1561 bool os::dll_build_name(char* buffer, size_t buflen,
1562                         const char* pname, const char* fname) {
1563   bool retval = false;
1564   const size_t pnamelen = pname ? strlen(pname) : 0;
1565 
1566   // Return error on buffer overflow.
1567   if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1568     return retval;
1569   }
1570 
1571   if (pnamelen == 0) {
1572     snprintf(buffer, buflen, "lib%s.so", fname);
1573     retval = true;
1574   } else if (strchr(pname, *os::path_separator()) != NULL) {
1575     int n;
1576     char** pelements = split_path(pname, &n);
1577     if (pelements == NULL) {
1578       return false;
1579     }
1580     for (int i = 0; i < n; i++) {
1581       // really shouldn't be NULL but what the heck, check can't hurt
1582       if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1583         continue; // skip the empty path values
1584       }
1585       snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1586       if (file_exists(buffer)) {
1587         retval = true;
1588         break;
1589       }
1590     }
1591     // release the storage
1592     for (int i = 0; i < n; i++) {
1593       if (pelements[i] != NULL) {
1594         FREE_C_HEAP_ARRAY(char, pelements[i]);
1595       }
1596     }
1597     if (pelements != NULL) {
1598       FREE_C_HEAP_ARRAY(char*, pelements);
1599     }
1600   } else {
1601     snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1602     retval = true;
1603   }
1604   return retval;
1605 }
1606 
1607 // check if addr is inside libjvm.so
1608 bool os::address_is_in_vm(address addr) {
1609   static address libjvm_base_addr;
1610   Dl_info dlinfo;
1611 
1612   if (libjvm_base_addr == NULL) {
1613     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1614       libjvm_base_addr = (address)dlinfo.dli_fbase;
1615     }
1616     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1617   }
1618 
1619   if (dladdr((void *)addr, &dlinfo) != 0) {
1620     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1621   }
1622 
1623   return false;
1624 }
1625 
1626 typedef int (*dladdr1_func_type)(void *, Dl_info *, void **, int);
1627 static dladdr1_func_type dladdr1_func = NULL;
1628 
1629 bool os::dll_address_to_function_name(address addr, char *buf,
1630                                       int buflen, int * offset) {
1631   // buf is not optional, but offset is optional
1632   assert(buf != NULL, "sanity check");
1633 
1634   Dl_info dlinfo;
1635 
1636   // dladdr1_func was initialized in os::init()
1637   if (dladdr1_func != NULL) {
1638     // yes, we have dladdr1
1639 
1640     // Support for dladdr1 is checked at runtime; it may be
1641     // available even if the vm is built on a machine that does
1642     // not have dladdr1 support.  Make sure there is a value for
1643     // RTLD_DL_SYMENT.
1644 #ifndef RTLD_DL_SYMENT
1645   #define RTLD_DL_SYMENT 1
1646 #endif
1647 #ifdef _LP64
1648     Elf64_Sym * info;
1649 #else
1650     Elf32_Sym * info;
1651 #endif
1652     if (dladdr1_func((void *)addr, &dlinfo, (void **)&info,
1653                      RTLD_DL_SYMENT) != 0) {
1654       // see if we have a matching symbol that covers our address
1655       if (dlinfo.dli_saddr != NULL &&
1656           (char *)dlinfo.dli_saddr + info->st_size > (char *)addr) {
1657         if (dlinfo.dli_sname != NULL) {
1658           if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1659             jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1660           }
1661           if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1662           return true;
1663         }
1664       }
1665       // no matching symbol so try for just file info
1666       if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1667         if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1668                             buf, buflen, offset, dlinfo.dli_fname)) {
1669           return true;
1670         }
1671       }
1672     }
1673     buf[0] = '\0';
1674     if (offset != NULL) *offset  = -1;
1675     return false;
1676   }
1677 
1678   // no, only dladdr is available
1679   if (dladdr((void *)addr, &dlinfo) != 0) {
1680     // see if we have a matching symbol
1681     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1682       if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1683         jio_snprintf(buf, buflen, dlinfo.dli_sname);
1684       }
1685       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1686       return true;
1687     }
1688     // no matching symbol so try for just file info
1689     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1690       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1691                           buf, buflen, offset, dlinfo.dli_fname)) {
1692         return true;
1693       }
1694     }
1695   }
1696   buf[0] = '\0';
1697   if (offset != NULL) *offset  = -1;
1698   return false;
1699 }
1700 
1701 bool os::dll_address_to_library_name(address addr, char* buf,
1702                                      int buflen, int* offset) {
1703   // buf is not optional, but offset is optional
1704   assert(buf != NULL, "sanity check");
1705 
1706   Dl_info dlinfo;
1707 
1708   if (dladdr((void*)addr, &dlinfo) != 0) {
1709     if (dlinfo.dli_fname != NULL) {
1710       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1711     }
1712     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1713       *offset = addr - (address)dlinfo.dli_fbase;
1714     }
1715     return true;
1716   }
1717 
1718   buf[0] = '\0';
1719   if (offset) *offset = -1;
1720   return false;
1721 }
1722 
1723 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1724   Dl_info dli;
1725   // Sanity check?
1726   if (dladdr(CAST_FROM_FN_PTR(void *, os::get_loaded_modules_info), &dli) == 0 ||
1727       dli.dli_fname == NULL) {
1728     return 1;
1729   }
1730 
1731   void * handle = dlopen(dli.dli_fname, RTLD_LAZY);
1732   if (handle == NULL) {
1733     return 1;
1734   }
1735 
1736   Link_map *map;
1737   dlinfo(handle, RTLD_DI_LINKMAP, &map);
1738   if (map == NULL) {
1739     dlclose(handle);
1740     return 1;
1741   }
1742 
1743   while (map->l_prev != NULL) {
1744     map = map->l_prev;
1745   }
1746 
1747   while (map != NULL) {
1748     // Iterate through all map entries and call callback with fields of interest
1749     if(callback(map->l_name, (address)map->l_addr, (address)0, param)) {
1750       dlclose(handle);
1751       return 1;
1752     }
1753     map = map->l_next;
1754   }
1755 
1756   dlclose(handle);
1757   return 0;
1758 }
1759 
1760 int _print_dll_info_cb(const char * name, address base_address, address top_address, void * param) {
1761   outputStream * out = (outputStream *) param;
1762   out->print_cr(PTR_FORMAT " \t%s", base_address, name);
1763   return 0;
1764 }
1765 
1766 void os::print_dll_info(outputStream * st) {
1767   st->print_cr("Dynamic libraries:"); st->flush();
1768   if (get_loaded_modules_info(_print_dll_info_cb, (void *)st)) {
1769     st->print_cr("Error: Cannot print dynamic libraries.");
1770   }
1771 }
1772 
1773 // Loads .dll/.so and
1774 // in case of error it checks if .dll/.so was built for the
1775 // same architecture as Hotspot is running on
1776 
1777 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1778   void * result= ::dlopen(filename, RTLD_LAZY);
1779   if (result != NULL) {
1780     // Successful loading
1781     return result;
1782   }
1783 
1784   Elf32_Ehdr elf_head;
1785 
1786   // Read system error message into ebuf
1787   // It may or may not be overwritten below
1788   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1789   ebuf[ebuflen-1]='\0';
1790   int diag_msg_max_length=ebuflen-strlen(ebuf);
1791   char* diag_msg_buf=ebuf+strlen(ebuf);
1792 
1793   if (diag_msg_max_length==0) {
1794     // No more space in ebuf for additional diagnostics message
1795     return NULL;
1796   }
1797 
1798 
1799   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1800 
1801   if (file_descriptor < 0) {
1802     // Can't open library, report dlerror() message
1803     return NULL;
1804   }
1805 
1806   bool failed_to_read_elf_head=
1807     (sizeof(elf_head)!=
1808      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1809 
1810   ::close(file_descriptor);
1811   if (failed_to_read_elf_head) {
1812     // file i/o error - report dlerror() msg
1813     return NULL;
1814   }
1815 
1816   typedef struct {
1817     Elf32_Half  code;         // Actual value as defined in elf.h
1818     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1819     char        elf_class;    // 32 or 64 bit
1820     char        endianess;    // MSB or LSB
1821     char*       name;         // String representation
1822   } arch_t;
1823 
1824   static const arch_t arch_array[]={
1825     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1826     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1827     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1828     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1829     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1830     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1831     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1832     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1833     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1834     {EM_ARM,         EM_ARM,     ELFCLASS32, ELFDATA2LSB, (char*)"ARM 32"}
1835   };
1836 
1837 #if  (defined IA32)
1838   static  Elf32_Half running_arch_code=EM_386;
1839 #elif   (defined AMD64)
1840   static  Elf32_Half running_arch_code=EM_X86_64;
1841 #elif  (defined IA64)
1842   static  Elf32_Half running_arch_code=EM_IA_64;
1843 #elif  (defined __sparc) && (defined _LP64)
1844   static  Elf32_Half running_arch_code=EM_SPARCV9;
1845 #elif  (defined __sparc) && (!defined _LP64)
1846   static  Elf32_Half running_arch_code=EM_SPARC;
1847 #elif  (defined __powerpc64__)
1848   static  Elf32_Half running_arch_code=EM_PPC64;
1849 #elif  (defined __powerpc__)
1850   static  Elf32_Half running_arch_code=EM_PPC;
1851 #elif (defined ARM)
1852   static  Elf32_Half running_arch_code=EM_ARM;
1853 #else
1854   #error Method os::dll_load requires that one of following is defined:\
1855        IA32, AMD64, IA64, __sparc, __powerpc__, ARM, ARM
1856 #endif
1857 
1858   // Identify compatability class for VM's architecture and library's architecture
1859   // Obtain string descriptions for architectures
1860 
1861   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1862   int running_arch_index=-1;
1863 
1864   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1865     if (running_arch_code == arch_array[i].code) {
1866       running_arch_index    = i;
1867     }
1868     if (lib_arch.code == arch_array[i].code) {
1869       lib_arch.compat_class = arch_array[i].compat_class;
1870       lib_arch.name         = arch_array[i].name;
1871     }
1872   }
1873 
1874   assert(running_arch_index != -1,
1875          "Didn't find running architecture code (running_arch_code) in arch_array");
1876   if (running_arch_index == -1) {
1877     // Even though running architecture detection failed
1878     // we may still continue with reporting dlerror() message
1879     return NULL;
1880   }
1881 
1882   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1883     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1884     return NULL;
1885   }
1886 
1887   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1888     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1889     return NULL;
1890   }
1891 
1892   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1893     if (lib_arch.name!=NULL) {
1894       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1895                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1896                  lib_arch.name, arch_array[running_arch_index].name);
1897     } else {
1898       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1899                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1900                  lib_arch.code,
1901                  arch_array[running_arch_index].name);
1902     }
1903   }
1904 
1905   return NULL;
1906 }
1907 
1908 void* os::dll_lookup(void* handle, const char* name) {
1909   return dlsym(handle, name);
1910 }
1911 
1912 void* os::get_default_process_handle() {
1913   return (void*)::dlopen(NULL, RTLD_LAZY);
1914 }
1915 
1916 int os::stat(const char *path, struct stat *sbuf) {
1917   char pathbuf[MAX_PATH];
1918   if (strlen(path) > MAX_PATH - 1) {
1919     errno = ENAMETOOLONG;
1920     return -1;
1921   }
1922   os::native_path(strcpy(pathbuf, path));
1923   return ::stat(pathbuf, sbuf);
1924 }
1925 
1926 static bool _print_ascii_file(const char* filename, outputStream* st) {
1927   int fd = ::open(filename, O_RDONLY);
1928   if (fd == -1) {
1929     return false;
1930   }
1931 
1932   char buf[32];
1933   int bytes;
1934   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1935     st->print_raw(buf, bytes);
1936   }
1937 
1938   ::close(fd);
1939 
1940   return true;
1941 }
1942 
1943 void os::print_os_info_brief(outputStream* st) {
1944   os::Solaris::print_distro_info(st);
1945 
1946   os::Posix::print_uname_info(st);
1947 
1948   os::Solaris::print_libversion_info(st);
1949 }
1950 
1951 void os::print_os_info(outputStream* st) {
1952   st->print("OS:");
1953 
1954   os::Solaris::print_distro_info(st);
1955 
1956   os::Posix::print_uname_info(st);
1957 
1958   os::Solaris::print_libversion_info(st);
1959 
1960   os::Posix::print_rlimit_info(st);
1961 
1962   os::Posix::print_load_average(st);
1963 }
1964 
1965 void os::Solaris::print_distro_info(outputStream* st) {
1966   if (!_print_ascii_file("/etc/release", st)) {
1967     st->print("Solaris");
1968   }
1969   st->cr();
1970 }
1971 
1972 void os::Solaris::print_libversion_info(outputStream* st) {
1973   st->print("  (T2 libthread)");
1974   st->cr();
1975 }
1976 
1977 static bool check_addr0(outputStream* st) {
1978   jboolean status = false;
1979   int fd = ::open("/proc/self/map",O_RDONLY);
1980   if (fd >= 0) {
1981     prmap_t p;
1982     while (::read(fd, &p, sizeof(p)) > 0) {
1983       if (p.pr_vaddr == 0x0) {
1984         st->print("Warning: Address: 0x%x, Size: %dK, ",p.pr_vaddr, p.pr_size/1024, p.pr_mapname);
1985         st->print("Mapped file: %s, ", p.pr_mapname[0] == '\0' ? "None" : p.pr_mapname);
1986         st->print("Access:");
1987         st->print("%s",(p.pr_mflags & MA_READ)  ? "r" : "-");
1988         st->print("%s",(p.pr_mflags & MA_WRITE) ? "w" : "-");
1989         st->print("%s",(p.pr_mflags & MA_EXEC)  ? "x" : "-");
1990         st->cr();
1991         status = true;
1992       }
1993     }
1994     ::close(fd);
1995   }
1996   return status;
1997 }
1998 
1999 void os::pd_print_cpu_info(outputStream* st) {
2000   // Nothing to do for now.
2001 }
2002 
2003 void os::print_memory_info(outputStream* st) {
2004   st->print("Memory:");
2005   st->print(" %dk page", os::vm_page_size()>>10);
2006   st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
2007   st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
2008   st->cr();
2009   (void) check_addr0(st);
2010 }
2011 
2012 void os::print_siginfo(outputStream* st, void* siginfo) {
2013   const siginfo_t* si = (const siginfo_t*)siginfo;
2014 
2015   os::Posix::print_siginfo_brief(st, si);
2016 
2017   if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2018       UseSharedSpaces) {
2019     FileMapInfo* mapinfo = FileMapInfo::current_info();
2020     if (mapinfo->is_in_shared_space(si->si_addr)) {
2021       st->print("\n\nError accessing class data sharing archive."   \
2022                 " Mapped file inaccessible during execution, "      \
2023                 " possible disk/network problem.");
2024     }
2025   }
2026   st->cr();
2027 }
2028 
2029 // Moved from whole group, because we need them here for diagnostic
2030 // prints.
2031 #define OLDMAXSIGNUM 32
2032 static int Maxsignum = 0;
2033 static int *ourSigFlags = NULL;
2034 
2035 extern "C" void sigINTRHandler(int, siginfo_t*, void*);
2036 
2037 int os::Solaris::get_our_sigflags(int sig) {
2038   assert(ourSigFlags!=NULL, "signal data structure not initialized");
2039   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2040   return ourSigFlags[sig];
2041 }
2042 
2043 void os::Solaris::set_our_sigflags(int sig, int flags) {
2044   assert(ourSigFlags!=NULL, "signal data structure not initialized");
2045   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
2046   ourSigFlags[sig] = flags;
2047 }
2048 
2049 
2050 static const char* get_signal_handler_name(address handler,
2051                                            char* buf, int buflen) {
2052   int offset;
2053   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
2054   if (found) {
2055     // skip directory names
2056     const char *p1, *p2;
2057     p1 = buf;
2058     size_t len = strlen(os::file_separator());
2059     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
2060     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
2061   } else {
2062     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
2063   }
2064   return buf;
2065 }
2066 
2067 static void print_signal_handler(outputStream* st, int sig,
2068                                  char* buf, size_t buflen) {
2069   struct sigaction sa;
2070 
2071   sigaction(sig, NULL, &sa);
2072 
2073   st->print("%s: ", os::exception_name(sig, buf, buflen));
2074 
2075   address handler = (sa.sa_flags & SA_SIGINFO)
2076                   ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
2077                   : CAST_FROM_FN_PTR(address, sa.sa_handler);
2078 
2079   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
2080     st->print("SIG_DFL");
2081   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
2082     st->print("SIG_IGN");
2083   } else {
2084     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
2085   }
2086 
2087   st->print(", sa_mask[0]=");
2088   os::Posix::print_signal_set_short(st, &sa.sa_mask);
2089 
2090   address rh = VMError::get_resetted_sighandler(sig);
2091   // May be, handler was resetted by VMError?
2092   if (rh != NULL) {
2093     handler = rh;
2094     sa.sa_flags = VMError::get_resetted_sigflags(sig);
2095   }
2096 
2097   st->print(", sa_flags=");
2098   os::Posix::print_sa_flags(st, sa.sa_flags);
2099 
2100   // Check: is it our handler?
2101   if (handler == CAST_FROM_FN_PTR(address, signalHandler) ||
2102       handler == CAST_FROM_FN_PTR(address, sigINTRHandler)) {
2103     // It is our signal handler
2104     // check for flags
2105     if (sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
2106       st->print(
2107                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
2108                 os::Solaris::get_our_sigflags(sig));
2109     }
2110   }
2111   st->cr();
2112 }
2113 
2114 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2115   st->print_cr("Signal Handlers:");
2116   print_signal_handler(st, SIGSEGV, buf, buflen);
2117   print_signal_handler(st, SIGBUS , buf, buflen);
2118   print_signal_handler(st, SIGFPE , buf, buflen);
2119   print_signal_handler(st, SIGPIPE, buf, buflen);
2120   print_signal_handler(st, SIGXFSZ, buf, buflen);
2121   print_signal_handler(st, SIGILL , buf, buflen);
2122   print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2123   print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
2124   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2125   print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
2126   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2127   print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
2128   print_signal_handler(st, os::Solaris::SIGinterrupt(), buf, buflen);
2129   print_signal_handler(st, os::Solaris::SIGasync(), buf, buflen);
2130 }
2131 
2132 static char saved_jvm_path[MAXPATHLEN] = { 0 };
2133 
2134 // Find the full path to the current module, libjvm.so
2135 void os::jvm_path(char *buf, jint buflen) {
2136   // Error checking.
2137   if (buflen < MAXPATHLEN) {
2138     assert(false, "must use a large-enough buffer");
2139     buf[0] = '\0';
2140     return;
2141   }
2142   // Lazy resolve the path to current module.
2143   if (saved_jvm_path[0] != 0) {
2144     strcpy(buf, saved_jvm_path);
2145     return;
2146   }
2147 
2148   Dl_info dlinfo;
2149   int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
2150   assert(ret != 0, "cannot locate libjvm");
2151   if (ret != 0 && dlinfo.dli_fname != NULL) {
2152     realpath((char *)dlinfo.dli_fname, buf);
2153   } else {
2154     buf[0] = '\0';
2155     return;
2156   }
2157 
2158   if (Arguments::sun_java_launcher_is_altjvm()) {
2159     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2160     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2161     // If "/jre/lib/" appears at the right place in the string, then
2162     // assume we are installed in a JDK and we're done.  Otherwise, check
2163     // for a JAVA_HOME environment variable and fix up the path so it
2164     // looks like libjvm.so is installed there (append a fake suffix
2165     // hotspot/libjvm.so).
2166     const char *p = buf + strlen(buf) - 1;
2167     for (int count = 0; p > buf && count < 5; ++count) {
2168       for (--p; p > buf && *p != '/'; --p)
2169         /* empty */ ;
2170     }
2171 
2172     if (strncmp(p, "/jre/lib/", 9) != 0) {
2173       // Look for JAVA_HOME in the environment.
2174       char* java_home_var = ::getenv("JAVA_HOME");
2175       if (java_home_var != NULL && java_home_var[0] != 0) {
2176         char cpu_arch[12];
2177         char* jrelib_p;
2178         int   len;
2179         sysinfo(SI_ARCHITECTURE, cpu_arch, sizeof(cpu_arch));
2180 #ifdef _LP64
2181         // If we are on sparc running a 64-bit vm, look in jre/lib/sparcv9.
2182         if (strcmp(cpu_arch, "sparc") == 0) {
2183           strcat(cpu_arch, "v9");
2184         } else if (strcmp(cpu_arch, "i386") == 0) {
2185           strcpy(cpu_arch, "amd64");
2186         }
2187 #endif
2188         // Check the current module name "libjvm.so".
2189         p = strrchr(buf, '/');
2190         assert(strstr(p, "/libjvm") == p, "invalid library name");
2191 
2192         realpath(java_home_var, buf);
2193         // determine if this is a legacy image or modules image
2194         // modules image doesn't have "jre" subdirectory
2195         len = strlen(buf);
2196         assert(len < buflen, "Ran out of buffer space");
2197         jrelib_p = buf + len;
2198         snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2199         if (0 != access(buf, F_OK)) {
2200           snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2201         }
2202 
2203         if (0 == access(buf, F_OK)) {
2204           // Use current module name "libjvm.so"
2205           len = strlen(buf);
2206           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2207         } else {
2208           // Go back to path of .so
2209           realpath((char *)dlinfo.dli_fname, buf);
2210         }
2211       }
2212     }
2213   }
2214 
2215   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2216   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2217 }
2218 
2219 
2220 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2221   // no prefix required, not even "_"
2222 }
2223 
2224 
2225 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2226   // no suffix required
2227 }
2228 
2229 // This method is a copy of JDK's sysGetLastErrorString
2230 // from src/solaris/hpi/src/system_md.c
2231 
2232 size_t os::lasterror(char *buf, size_t len) {
2233   if (errno == 0)  return 0;
2234 
2235   const char *s = ::strerror(errno);
2236   size_t n = ::strlen(s);
2237   if (n >= len) {
2238     n = len - 1;
2239   }
2240   ::strncpy(buf, s, n);
2241   buf[n] = '\0';
2242   return n;
2243 }
2244 
2245 
2246 // sun.misc.Signal
2247 
2248 extern "C" {
2249   static void UserHandler(int sig, void *siginfo, void *context) {
2250     // Ctrl-C is pressed during error reporting, likely because the error
2251     // handler fails to abort. Let VM die immediately.
2252     if (sig == SIGINT && is_error_reported()) {
2253       os::die();
2254     }
2255 
2256     os::signal_notify(sig);
2257     // We do not need to reinstate the signal handler each time...
2258   }
2259 }
2260 
2261 void* os::user_handler() {
2262   return CAST_FROM_FN_PTR(void*, UserHandler);
2263 }
2264 
2265 class Semaphore : public StackObj {
2266  public:
2267   Semaphore();
2268   ~Semaphore();
2269   void signal();
2270   void wait();
2271   bool trywait();
2272   bool timedwait(unsigned int sec, int nsec);
2273  private:
2274   sema_t _semaphore;
2275 };
2276 
2277 
2278 Semaphore::Semaphore() {
2279   sema_init(&_semaphore, 0, NULL, NULL);
2280 }
2281 
2282 Semaphore::~Semaphore() {
2283   sema_destroy(&_semaphore);
2284 }
2285 
2286 void Semaphore::signal() {
2287   sema_post(&_semaphore);
2288 }
2289 
2290 void Semaphore::wait() {
2291   sema_wait(&_semaphore);
2292 }
2293 
2294 bool Semaphore::trywait() {
2295   return sema_trywait(&_semaphore) == 0;
2296 }
2297 
2298 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2299   struct timespec ts;
2300   unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2301 
2302   while (1) {
2303     int result = sema_timedwait(&_semaphore, &ts);
2304     if (result == 0) {
2305       return true;
2306     } else if (errno == EINTR) {
2307       continue;
2308     } else if (errno == ETIME) {
2309       return false;
2310     } else {
2311       return false;
2312     }
2313   }
2314 }
2315 
2316 extern "C" {
2317   typedef void (*sa_handler_t)(int);
2318   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2319 }
2320 
2321 void* os::signal(int signal_number, void* handler) {
2322   struct sigaction sigAct, oldSigAct;
2323   sigfillset(&(sigAct.sa_mask));
2324   sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2325   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2326 
2327   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2328     // -1 means registration failed
2329     return (void *)-1;
2330   }
2331 
2332   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2333 }
2334 
2335 void os::signal_raise(int signal_number) {
2336   raise(signal_number);
2337 }
2338 
2339 // The following code is moved from os.cpp for making this
2340 // code platform specific, which it is by its very nature.
2341 
2342 // a counter for each possible signal value
2343 static int Sigexit = 0;
2344 static int Maxlibjsigsigs;
2345 static jint *pending_signals = NULL;
2346 static int *preinstalled_sigs = NULL;
2347 static struct sigaction *chainedsigactions = NULL;
2348 static sema_t sig_sem;
2349 typedef int (*version_getting_t)();
2350 version_getting_t os::Solaris::get_libjsig_version = NULL;
2351 static int libjsigversion = NULL;
2352 
2353 int os::sigexitnum_pd() {
2354   assert(Sigexit > 0, "signal memory not yet initialized");
2355   return Sigexit;
2356 }
2357 
2358 void os::Solaris::init_signal_mem() {
2359   // Initialize signal structures
2360   Maxsignum = SIGRTMAX;
2361   Sigexit = Maxsignum+1;
2362   assert(Maxsignum >0, "Unable to obtain max signal number");
2363 
2364   Maxlibjsigsigs = Maxsignum;
2365 
2366   // pending_signals has one int per signal
2367   // The additional signal is for SIGEXIT - exit signal to signal_thread
2368   pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1), mtInternal);
2369   memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2370 
2371   if (UseSignalChaining) {
2372     chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2373                                                    * (Maxsignum + 1), mtInternal);
2374     memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2375     preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2376     memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2377   }
2378   ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2379   memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2380 }
2381 
2382 void os::signal_init_pd() {
2383   int ret;
2384 
2385   ret = ::sema_init(&sig_sem, 0, NULL, NULL);
2386   assert(ret == 0, "sema_init() failed");
2387 }
2388 
2389 void os::signal_notify(int signal_number) {
2390   int ret;
2391 
2392   Atomic::inc(&pending_signals[signal_number]);
2393   ret = ::sema_post(&sig_sem);
2394   assert(ret == 0, "sema_post() failed");
2395 }
2396 
2397 static int check_pending_signals(bool wait_for_signal) {
2398   int ret;
2399   while (true) {
2400     for (int i = 0; i < Sigexit + 1; i++) {
2401       jint n = pending_signals[i];
2402       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2403         return i;
2404       }
2405     }
2406     if (!wait_for_signal) {
2407       return -1;
2408     }
2409     JavaThread *thread = JavaThread::current();
2410     ThreadBlockInVM tbivm(thread);
2411 
2412     bool threadIsSuspended;
2413     do {
2414       thread->set_suspend_equivalent();
2415       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2416       while ((ret = ::sema_wait(&sig_sem)) == EINTR)
2417         ;
2418       assert(ret == 0, "sema_wait() failed");
2419 
2420       // were we externally suspended while we were waiting?
2421       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2422       if (threadIsSuspended) {
2423         // The semaphore has been incremented, but while we were waiting
2424         // another thread suspended us. We don't want to continue running
2425         // while suspended because that would surprise the thread that
2426         // suspended us.
2427         ret = ::sema_post(&sig_sem);
2428         assert(ret == 0, "sema_post() failed");
2429 
2430         thread->java_suspend_self();
2431       }
2432     } while (threadIsSuspended);
2433   }
2434 }
2435 
2436 int os::signal_lookup() {
2437   return check_pending_signals(false);
2438 }
2439 
2440 int os::signal_wait() {
2441   return check_pending_signals(true);
2442 }
2443 
2444 ////////////////////////////////////////////////////////////////////////////////
2445 // Virtual Memory
2446 
2447 static int page_size = -1;
2448 
2449 // The mmap MAP_ALIGN flag is supported on Solaris 9 and later.  init_2() will
2450 // clear this var if support is not available.
2451 static bool has_map_align = true;
2452 
2453 int os::vm_page_size() {
2454   assert(page_size != -1, "must call os::init");
2455   return page_size;
2456 }
2457 
2458 // Solaris allocates memory by pages.
2459 int os::vm_allocation_granularity() {
2460   assert(page_size != -1, "must call os::init");
2461   return page_size;
2462 }
2463 
2464 static bool recoverable_mmap_error(int err) {
2465   // See if the error is one we can let the caller handle. This
2466   // list of errno values comes from the Solaris mmap(2) man page.
2467   switch (err) {
2468   case EBADF:
2469   case EINVAL:
2470   case ENOTSUP:
2471     // let the caller deal with these errors
2472     return true;
2473 
2474   default:
2475     // Any remaining errors on this OS can cause our reserved mapping
2476     // to be lost. That can cause confusion where different data
2477     // structures think they have the same memory mapped. The worst
2478     // scenario is if both the VM and a library think they have the
2479     // same memory mapped.
2480     return false;
2481   }
2482 }
2483 
2484 static void warn_fail_commit_memory(char* addr, size_t bytes, bool exec,
2485                                     int err) {
2486   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2487           ", %d) failed; error='%s' (errno=%d)", addr, bytes, exec,
2488           strerror(err), err);
2489 }
2490 
2491 static void warn_fail_commit_memory(char* addr, size_t bytes,
2492                                     size_t alignment_hint, bool exec,
2493                                     int err) {
2494   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2495           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, bytes,
2496           alignment_hint, exec, strerror(err), err);
2497 }
2498 
2499 int os::Solaris::commit_memory_impl(char* addr, size_t bytes, bool exec) {
2500   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2501   size_t size = bytes;
2502   char *res = Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2503   if (res != NULL) {
2504     if (UseNUMAInterleaving) {
2505       numa_make_global(addr, bytes);
2506     }
2507     return 0;
2508   }
2509 
2510   int err = errno;  // save errno from mmap() call in mmap_chunk()
2511 
2512   if (!recoverable_mmap_error(err)) {
2513     warn_fail_commit_memory(addr, bytes, exec, err);
2514     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "committing reserved memory.");
2515   }
2516 
2517   return err;
2518 }
2519 
2520 bool os::pd_commit_memory(char* addr, size_t bytes, bool exec) {
2521   return Solaris::commit_memory_impl(addr, bytes, exec) == 0;
2522 }
2523 
2524 void os::pd_commit_memory_or_exit(char* addr, size_t bytes, bool exec,
2525                                   const char* mesg) {
2526   assert(mesg != NULL, "mesg must be specified");
2527   int err = os::Solaris::commit_memory_impl(addr, bytes, exec);
2528   if (err != 0) {
2529     // the caller wants all commit errors to exit with the specified mesg:
2530     warn_fail_commit_memory(addr, bytes, exec, err);
2531     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, mesg);
2532   }
2533 }
2534 
2535 size_t os::Solaris::page_size_for_alignment(size_t alignment) {
2536   assert(is_size_aligned(alignment, (size_t) vm_page_size()),
2537          err_msg(SIZE_FORMAT " is not aligned to " SIZE_FORMAT,
2538                  alignment, (size_t) vm_page_size()));
2539 
2540   for (int i = 0; _page_sizes[i] != 0; i++) {
2541     if (is_size_aligned(alignment, _page_sizes[i])) {
2542       return _page_sizes[i];
2543     }
2544   }
2545 
2546   return (size_t) vm_page_size();
2547 }
2548 
2549 int os::Solaris::commit_memory_impl(char* addr, size_t bytes,
2550                                     size_t alignment_hint, bool exec) {
2551   int err = Solaris::commit_memory_impl(addr, bytes, exec);
2552   if (err == 0 && UseLargePages && alignment_hint > 0) {
2553     assert(is_size_aligned(bytes, alignment_hint),
2554            err_msg(SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, alignment_hint));
2555 
2556     // The syscall memcntl requires an exact page size (see man memcntl for details).
2557     size_t page_size = page_size_for_alignment(alignment_hint);
2558     if (page_size > (size_t) vm_page_size()) {
2559       (void)Solaris::setup_large_pages(addr, bytes, page_size);
2560     }
2561   }
2562   return err;
2563 }
2564 
2565 bool os::pd_commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2566                           bool exec) {
2567   return Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec) == 0;
2568 }
2569 
2570 void os::pd_commit_memory_or_exit(char* addr, size_t bytes,
2571                                   size_t alignment_hint, bool exec,
2572                                   const char* mesg) {
2573   assert(mesg != NULL, "mesg must be specified");
2574   int err = os::Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec);
2575   if (err != 0) {
2576     // the caller wants all commit errors to exit with the specified mesg:
2577     warn_fail_commit_memory(addr, bytes, alignment_hint, exec, err);
2578     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, mesg);
2579   }
2580 }
2581 
2582 // Uncommit the pages in a specified region.
2583 void os::pd_free_memory(char* addr, size_t bytes, size_t alignment_hint) {
2584   if (madvise(addr, bytes, MADV_FREE) < 0) {
2585     debug_only(warning("MADV_FREE failed."));
2586     return;
2587   }
2588 }
2589 
2590 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2591   return os::commit_memory(addr, size, !ExecMem);
2592 }
2593 
2594 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2595   return os::uncommit_memory(addr, size);
2596 }
2597 
2598 // Change the page size in a given range.
2599 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2600   assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2601   assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2602   if (UseLargePages) {
2603     size_t page_size = Solaris::page_size_for_alignment(alignment_hint);
2604     if (page_size > (size_t) vm_page_size()) {
2605       Solaris::setup_large_pages(addr, bytes, page_size);
2606     }
2607   }
2608 }
2609 
2610 // Tell the OS to make the range local to the first-touching LWP
2611 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2612   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2613   if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2614     debug_only(warning("MADV_ACCESS_LWP failed."));
2615   }
2616 }
2617 
2618 // Tell the OS that this range would be accessed from different LWPs.
2619 void os::numa_make_global(char *addr, size_t bytes) {
2620   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2621   if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2622     debug_only(warning("MADV_ACCESS_MANY failed."));
2623   }
2624 }
2625 
2626 // Get the number of the locality groups.
2627 size_t os::numa_get_groups_num() {
2628   size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2629   return n != -1 ? n : 1;
2630 }
2631 
2632 // Get a list of leaf locality groups. A leaf lgroup is group that
2633 // doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2634 // board. An LWP is assigned to one of these groups upon creation.
2635 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2636   if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2637     ids[0] = 0;
2638     return 1;
2639   }
2640   int result_size = 0, top = 1, bottom = 0, cur = 0;
2641   for (int k = 0; k < size; k++) {
2642     int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2643                                    (Solaris::lgrp_id_t*)&ids[top], size - top);
2644     if (r == -1) {
2645       ids[0] = 0;
2646       return 1;
2647     }
2648     if (!r) {
2649       // That's a leaf node.
2650       assert(bottom <= cur, "Sanity check");
2651       // Check if the node has memory
2652       if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2653                                   NULL, 0, LGRP_RSRC_MEM) > 0) {
2654         ids[bottom++] = ids[cur];
2655       }
2656     }
2657     top += r;
2658     cur++;
2659   }
2660   if (bottom == 0) {
2661     // Handle a situation, when the OS reports no memory available.
2662     // Assume UMA architecture.
2663     ids[0] = 0;
2664     return 1;
2665   }
2666   return bottom;
2667 }
2668 
2669 // Detect the topology change. Typically happens during CPU plugging-unplugging.
2670 bool os::numa_topology_changed() {
2671   int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2672   if (is_stale != -1 && is_stale) {
2673     Solaris::lgrp_fini(Solaris::lgrp_cookie());
2674     Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2675     assert(c != 0, "Failure to initialize LGRP API");
2676     Solaris::set_lgrp_cookie(c);
2677     return true;
2678   }
2679   return false;
2680 }
2681 
2682 // Get the group id of the current LWP.
2683 int os::numa_get_group_id() {
2684   int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2685   if (lgrp_id == -1) {
2686     return 0;
2687   }
2688   const int size = os::numa_get_groups_num();
2689   int *ids = (int*)alloca(size * sizeof(int));
2690 
2691   // Get the ids of all lgroups with memory; r is the count.
2692   int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2693                                   (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2694   if (r <= 0) {
2695     return 0;
2696   }
2697   return ids[os::random() % r];
2698 }
2699 
2700 // Request information about the page.
2701 bool os::get_page_info(char *start, page_info* info) {
2702   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2703   uint64_t addr = (uintptr_t)start;
2704   uint64_t outdata[2];
2705   uint_t validity = 0;
2706 
2707   if (os::Solaris::meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2708     return false;
2709   }
2710 
2711   info->size = 0;
2712   info->lgrp_id = -1;
2713 
2714   if ((validity & 1) != 0) {
2715     if ((validity & 2) != 0) {
2716       info->lgrp_id = outdata[0];
2717     }
2718     if ((validity & 4) != 0) {
2719       info->size = outdata[1];
2720     }
2721     return true;
2722   }
2723   return false;
2724 }
2725 
2726 // Scan the pages from start to end until a page different than
2727 // the one described in the info parameter is encountered.
2728 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2729                      page_info* page_found) {
2730   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2731   const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2732   uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT + 1];
2733   uint_t validity[MAX_MEMINFO_CNT];
2734 
2735   size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2736   uint64_t p = (uint64_t)start;
2737   while (p < (uint64_t)end) {
2738     addrs[0] = p;
2739     size_t addrs_count = 1;
2740     while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] + page_size < (uint64_t)end) {
2741       addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2742       addrs_count++;
2743     }
2744 
2745     if (os::Solaris::meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2746       return NULL;
2747     }
2748 
2749     size_t i = 0;
2750     for (; i < addrs_count; i++) {
2751       if ((validity[i] & 1) != 0) {
2752         if ((validity[i] & 4) != 0) {
2753           if (outdata[types * i + 1] != page_expected->size) {
2754             break;
2755           }
2756         } else if (page_expected->size != 0) {
2757           break;
2758         }
2759 
2760         if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
2761           if (outdata[types * i] != page_expected->lgrp_id) {
2762             break;
2763           }
2764         }
2765       } else {
2766         return NULL;
2767       }
2768     }
2769 
2770     if (i < addrs_count) {
2771       if ((validity[i] & 2) != 0) {
2772         page_found->lgrp_id = outdata[types * i];
2773       } else {
2774         page_found->lgrp_id = -1;
2775       }
2776       if ((validity[i] & 4) != 0) {
2777         page_found->size = outdata[types * i + 1];
2778       } else {
2779         page_found->size = 0;
2780       }
2781       return (char*)addrs[i];
2782     }
2783 
2784     p = addrs[addrs_count - 1] + page_size;
2785   }
2786   return end;
2787 }
2788 
2789 bool os::pd_uncommit_memory(char* addr, size_t bytes) {
2790   size_t size = bytes;
2791   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2792   // uncommitted page. Otherwise, the read/write might succeed if we
2793   // have enough swap space to back the physical page.
2794   return
2795     NULL != Solaris::mmap_chunk(addr, size,
2796                                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
2797                                 PROT_NONE);
2798 }
2799 
2800 char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
2801   char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
2802 
2803   if (b == MAP_FAILED) {
2804     return NULL;
2805   }
2806   return b;
2807 }
2808 
2809 char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes,
2810                              size_t alignment_hint, bool fixed) {
2811   char* addr = requested_addr;
2812   int flags = MAP_PRIVATE | MAP_NORESERVE;
2813 
2814   assert(!(fixed && (alignment_hint > 0)),
2815          "alignment hint meaningless with fixed mmap");
2816 
2817   if (fixed) {
2818     flags |= MAP_FIXED;
2819   } else if (has_map_align && (alignment_hint > (size_t) vm_page_size())) {
2820     flags |= MAP_ALIGN;
2821     addr = (char*) alignment_hint;
2822   }
2823 
2824   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2825   // uncommitted page. Otherwise, the read/write might succeed if we
2826   // have enough swap space to back the physical page.
2827   return mmap_chunk(addr, bytes, flags, PROT_NONE);
2828 }
2829 
2830 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2831                             size_t alignment_hint) {
2832   char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint,
2833                                   (requested_addr != NULL));
2834 
2835   guarantee(requested_addr == NULL || requested_addr == addr,
2836             "OS failed to return requested mmap address.");
2837   return addr;
2838 }
2839 
2840 // Reserve memory at an arbitrary address, only if that area is
2841 // available (and not reserved for something else).
2842 
2843 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2844   const int max_tries = 10;
2845   char* base[max_tries];
2846   size_t size[max_tries];
2847 
2848   // Solaris adds a gap between mmap'ed regions.  The size of the gap
2849   // is dependent on the requested size and the MMU.  Our initial gap
2850   // value here is just a guess and will be corrected later.
2851   bool had_top_overlap = false;
2852   bool have_adjusted_gap = false;
2853   size_t gap = 0x400000;
2854 
2855   // Assert only that the size is a multiple of the page size, since
2856   // that's all that mmap requires, and since that's all we really know
2857   // about at this low abstraction level.  If we need higher alignment,
2858   // we can either pass an alignment to this method or verify alignment
2859   // in one of the methods further up the call chain.  See bug 5044738.
2860   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2861 
2862   // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
2863   // Give it a try, if the kernel honors the hint we can return immediately.
2864   char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
2865 
2866   volatile int err = errno;
2867   if (addr == requested_addr) {
2868     return addr;
2869   } else if (addr != NULL) {
2870     pd_unmap_memory(addr, bytes);
2871   }
2872 
2873   if (PrintMiscellaneous && Verbose) {
2874     char buf[256];
2875     buf[0] = '\0';
2876     if (addr == NULL) {
2877       jio_snprintf(buf, sizeof(buf), ": %s", strerror(err));
2878     }
2879     warning("attempt_reserve_memory_at: couldn't reserve " SIZE_FORMAT " bytes at "
2880             PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
2881             "%s", bytes, requested_addr, addr, buf);
2882   }
2883 
2884   // Address hint method didn't work.  Fall back to the old method.
2885   // In theory, once SNV becomes our oldest supported platform, this
2886   // code will no longer be needed.
2887   //
2888   // Repeatedly allocate blocks until the block is allocated at the
2889   // right spot. Give up after max_tries.
2890   int i;
2891   for (i = 0; i < max_tries; ++i) {
2892     base[i] = reserve_memory(bytes);
2893 
2894     if (base[i] != NULL) {
2895       // Is this the block we wanted?
2896       if (base[i] == requested_addr) {
2897         size[i] = bytes;
2898         break;
2899       }
2900 
2901       // check that the gap value is right
2902       if (had_top_overlap && !have_adjusted_gap) {
2903         size_t actual_gap = base[i-1] - base[i] - bytes;
2904         if (gap != actual_gap) {
2905           // adjust the gap value and retry the last 2 allocations
2906           assert(i > 0, "gap adjustment code problem");
2907           have_adjusted_gap = true;  // adjust the gap only once, just in case
2908           gap = actual_gap;
2909           if (PrintMiscellaneous && Verbose) {
2910             warning("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
2911           }
2912           unmap_memory(base[i], bytes);
2913           unmap_memory(base[i-1], size[i-1]);
2914           i-=2;
2915           continue;
2916         }
2917       }
2918 
2919       // Does this overlap the block we wanted? Give back the overlapped
2920       // parts and try again.
2921       //
2922       // There is still a bug in this code: if top_overlap == bytes,
2923       // the overlap is offset from requested region by the value of gap.
2924       // In this case giving back the overlapped part will not work,
2925       // because we'll give back the entire block at base[i] and
2926       // therefore the subsequent allocation will not generate a new gap.
2927       // This could be fixed with a new algorithm that used larger
2928       // or variable size chunks to find the requested region -
2929       // but such a change would introduce additional complications.
2930       // It's rare enough that the planets align for this bug,
2931       // so we'll just wait for a fix for 6204603/5003415 which
2932       // will provide a mmap flag to allow us to avoid this business.
2933 
2934       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2935       if (top_overlap >= 0 && top_overlap < bytes) {
2936         had_top_overlap = true;
2937         unmap_memory(base[i], top_overlap);
2938         base[i] += top_overlap;
2939         size[i] = bytes - top_overlap;
2940       } else {
2941         size_t bottom_overlap = base[i] + bytes - requested_addr;
2942         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2943           if (PrintMiscellaneous && Verbose && bottom_overlap == 0) {
2944             warning("attempt_reserve_memory_at: possible alignment bug");
2945           }
2946           unmap_memory(requested_addr, bottom_overlap);
2947           size[i] = bytes - bottom_overlap;
2948         } else {
2949           size[i] = bytes;
2950         }
2951       }
2952     }
2953   }
2954 
2955   // Give back the unused reserved pieces.
2956 
2957   for (int j = 0; j < i; ++j) {
2958     if (base[j] != NULL) {
2959       unmap_memory(base[j], size[j]);
2960     }
2961   }
2962 
2963   return (i < max_tries) ? requested_addr : NULL;
2964 }
2965 
2966 bool os::pd_release_memory(char* addr, size_t bytes) {
2967   size_t size = bytes;
2968   return munmap(addr, size) == 0;
2969 }
2970 
2971 static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
2972   assert(addr == (char*)align_size_down((uintptr_t)addr, os::vm_page_size()),
2973          "addr must be page aligned");
2974   int retVal = mprotect(addr, bytes, prot);
2975   return retVal == 0;
2976 }
2977 
2978 // Protect memory (Used to pass readonly pages through
2979 // JNI GetArray<type>Elements with empty arrays.)
2980 // Also, used for serialization page and for compressed oops null pointer
2981 // checking.
2982 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2983                         bool is_committed) {
2984   unsigned int p = 0;
2985   switch (prot) {
2986   case MEM_PROT_NONE: p = PROT_NONE; break;
2987   case MEM_PROT_READ: p = PROT_READ; break;
2988   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2989   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2990   default:
2991     ShouldNotReachHere();
2992   }
2993   // is_committed is unused.
2994   return solaris_mprotect(addr, bytes, p);
2995 }
2996 
2997 // guard_memory and unguard_memory only happens within stack guard pages.
2998 // Since ISM pertains only to the heap, guard and unguard memory should not
2999 /// happen with an ISM region.
3000 bool os::guard_memory(char* addr, size_t bytes) {
3001   return solaris_mprotect(addr, bytes, PROT_NONE);
3002 }
3003 
3004 bool os::unguard_memory(char* addr, size_t bytes) {
3005   return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
3006 }
3007 
3008 // Large page support
3009 static size_t _large_page_size = 0;
3010 
3011 // Insertion sort for small arrays (descending order).
3012 static void insertion_sort_descending(size_t* array, int len) {
3013   for (int i = 0; i < len; i++) {
3014     size_t val = array[i];
3015     for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
3016       size_t tmp = array[key];
3017       array[key] = array[key - 1];
3018       array[key - 1] = tmp;
3019     }
3020   }
3021 }
3022 
3023 bool os::Solaris::mpss_sanity_check(bool warn, size_t* page_size) {
3024   const unsigned int usable_count = VM_Version::page_size_count();
3025   if (usable_count == 1) {
3026     return false;
3027   }
3028 
3029   // Find the right getpagesizes interface.  When solaris 11 is the minimum
3030   // build platform, getpagesizes() (without the '2') can be called directly.
3031   typedef int (*gps_t)(size_t[], int);
3032   gps_t gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes2"));
3033   if (gps_func == NULL) {
3034     gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes"));
3035     if (gps_func == NULL) {
3036       if (warn) {
3037         warning("MPSS is not supported by the operating system.");
3038       }
3039       return false;
3040     }
3041   }
3042 
3043   // Fill the array of page sizes.
3044   int n = (*gps_func)(_page_sizes, page_sizes_max);
3045   assert(n > 0, "Solaris bug?");
3046 
3047   if (n == page_sizes_max) {
3048     // Add a sentinel value (necessary only if the array was completely filled
3049     // since it is static (zeroed at initialization)).
3050     _page_sizes[--n] = 0;
3051     DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
3052   }
3053   assert(_page_sizes[n] == 0, "missing sentinel");
3054   trace_page_sizes("available page sizes", _page_sizes, n);
3055 
3056   if (n == 1) return false;     // Only one page size available.
3057 
3058   // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
3059   // select up to usable_count elements.  First sort the array, find the first
3060   // acceptable value, then copy the usable sizes to the top of the array and
3061   // trim the rest.  Make sure to include the default page size :-).
3062   //
3063   // A better policy could get rid of the 4M limit by taking the sizes of the
3064   // important VM memory regions (java heap and possibly the code cache) into
3065   // account.
3066   insertion_sort_descending(_page_sizes, n);
3067   const size_t size_limit =
3068     FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
3069   int beg;
3070   for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */;
3071   const int end = MIN2((int)usable_count, n) - 1;
3072   for (int cur = 0; cur < end; ++cur, ++beg) {
3073     _page_sizes[cur] = _page_sizes[beg];
3074   }
3075   _page_sizes[end] = vm_page_size();
3076   _page_sizes[end + 1] = 0;
3077 
3078   if (_page_sizes[end] > _page_sizes[end - 1]) {
3079     // Default page size is not the smallest; sort again.
3080     insertion_sort_descending(_page_sizes, end + 1);
3081   }
3082   *page_size = _page_sizes[0];
3083 
3084   trace_page_sizes("usable page sizes", _page_sizes, end + 1);
3085   return true;
3086 }
3087 
3088 void os::large_page_init() {
3089   if (UseLargePages) {
3090     // print a warning if any large page related flag is specified on command line
3091     bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
3092                            !FLAG_IS_DEFAULT(LargePageSizeInBytes);
3093 
3094     UseLargePages = Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
3095   }
3096 }
3097 
3098 bool os::Solaris::is_valid_page_size(size_t bytes) {
3099   for (int i = 0; _page_sizes[i] != 0; i++) {
3100     if (_page_sizes[i] == bytes) {
3101       return true;
3102     }
3103   }
3104   return false;
3105 }
3106 
3107 bool os::Solaris::setup_large_pages(caddr_t start, size_t bytes, size_t align) {
3108   assert(is_valid_page_size(align), err_msg(SIZE_FORMAT " is not a valid page size", align));
3109   assert(is_ptr_aligned((void*) start, align),
3110          err_msg(PTR_FORMAT " is not aligned to " SIZE_FORMAT, p2i((void*) start), align));
3111   assert(is_size_aligned(bytes, align),
3112          err_msg(SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, align));
3113 
3114   // Signal to OS that we want large pages for addresses
3115   // from addr, addr + bytes
3116   struct memcntl_mha mpss_struct;
3117   mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
3118   mpss_struct.mha_pagesize = align;
3119   mpss_struct.mha_flags = 0;
3120   // Upon successful completion, memcntl() returns 0
3121   if (memcntl(start, bytes, MC_HAT_ADVISE, (caddr_t) &mpss_struct, 0, 0)) {
3122     debug_only(warning("Attempt to use MPSS failed."));
3123     return false;
3124   }
3125   return true;
3126 }
3127 
3128 char* os::reserve_memory_special(size_t size, size_t alignment, char* addr, bool exec) {
3129   fatal("os::reserve_memory_special should not be called on Solaris.");
3130   return NULL;
3131 }
3132 
3133 bool os::release_memory_special(char* base, size_t bytes) {
3134   fatal("os::release_memory_special should not be called on Solaris.");
3135   return false;
3136 }
3137 
3138 size_t os::large_page_size() {
3139   return _large_page_size;
3140 }
3141 
3142 // MPSS allows application to commit large page memory on demand; with ISM
3143 // the entire memory region must be allocated as shared memory.
3144 bool os::can_commit_large_page_memory() {
3145   return true;
3146 }
3147 
3148 bool os::can_execute_large_page_memory() {
3149   return true;
3150 }
3151 
3152 // Read calls from inside the vm need to perform state transitions
3153 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3154   size_t res;
3155   JavaThread* thread = (JavaThread*)Thread::current();
3156   assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
3157   ThreadBlockInVM tbiv(thread);
3158   RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
3159   return res;
3160 }
3161 
3162 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3163   size_t res;
3164   JavaThread* thread = (JavaThread*)Thread::current();
3165   assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
3166   ThreadBlockInVM tbiv(thread);
3167   RESTARTABLE(::pread(fd, buf, (size_t) nBytes, offset), res);
3168   return res;
3169 }
3170 
3171 size_t os::restartable_read(int fd, void *buf, unsigned int nBytes) {
3172   size_t res;
3173   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
3174          "Assumed _thread_in_native");
3175   RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
3176   return res;
3177 }
3178 
3179 void os::naked_short_sleep(jlong ms) {
3180   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3181 
3182   // usleep is deprecated and removed from POSIX, in favour of nanosleep, but
3183   // Solaris requires -lrt for this.
3184   usleep((ms * 1000));
3185 
3186   return;
3187 }
3188 
3189 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3190 void os::infinite_sleep() {
3191   while (true) {    // sleep forever ...
3192     ::sleep(100);   // ... 100 seconds at a time
3193   }
3194 }
3195 
3196 // Used to convert frequent JVM_Yield() to nops
3197 bool os::dont_yield() {
3198   if (DontYieldALot) {
3199     static hrtime_t last_time = 0;
3200     hrtime_t diff = getTimeNanos() - last_time;
3201 
3202     if (diff < DontYieldALotInterval * 1000000) {
3203       return true;
3204     }
3205 
3206     last_time += diff;
3207 
3208     return false;
3209   } else {
3210     return false;
3211   }
3212 }
3213 
3214 // Note that yield semantics are defined by the scheduling class to which
3215 // the thread currently belongs.  Typically, yield will _not yield to
3216 // other equal or higher priority threads that reside on the dispatch queues
3217 // of other CPUs.
3218 
3219 void os::naked_yield() {
3220   thr_yield();
3221 }
3222 
3223 // Interface for setting lwp priorities.  If we are using T2 libthread,
3224 // which forces the use of BoundThreads or we manually set UseBoundThreads,
3225 // all of our threads will be assigned to real lwp's.  Using the thr_setprio
3226 // function is meaningless in this mode so we must adjust the real lwp's priority
3227 // The routines below implement the getting and setting of lwp priorities.
3228 //
3229 // Note: T2 is now the only supported libthread. UseBoundThreads flag is
3230 //       being deprecated and all threads are now BoundThreads
3231 //
3232 // Note: There are three priority scales used on Solaris.  Java priotities
3233 //       which range from 1 to 10, libthread "thr_setprio" scale which range
3234 //       from 0 to 127, and the current scheduling class of the process we
3235 //       are running in.  This is typically from -60 to +60.
3236 //       The setting of the lwp priorities in done after a call to thr_setprio
3237 //       so Java priorities are mapped to libthread priorities and we map from
3238 //       the latter to lwp priorities.  We don't keep priorities stored in
3239 //       Java priorities since some of our worker threads want to set priorities
3240 //       higher than all Java threads.
3241 //
3242 // For related information:
3243 // (1)  man -s 2 priocntl
3244 // (2)  man -s 4 priocntl
3245 // (3)  man dispadmin
3246 // =    librt.so
3247 // =    libthread/common/rtsched.c - thrp_setlwpprio().
3248 // =    ps -cL <pid> ... to validate priority.
3249 // =    sched_get_priority_min and _max
3250 //              pthread_create
3251 //              sched_setparam
3252 //              pthread_setschedparam
3253 //
3254 // Assumptions:
3255 // +    We assume that all threads in the process belong to the same
3256 //              scheduling class.   IE. an homogenous process.
3257 // +    Must be root or in IA group to change change "interactive" attribute.
3258 //              Priocntl() will fail silently.  The only indication of failure is when
3259 //              we read-back the value and notice that it hasn't changed.
3260 // +    Interactive threads enter the runq at the head, non-interactive at the tail.
3261 // +    For RT, change timeslice as well.  Invariant:
3262 //              constant "priority integral"
3263 //              Konst == TimeSlice * (60-Priority)
3264 //              Given a priority, compute appropriate timeslice.
3265 // +    Higher numerical values have higher priority.
3266 
3267 // sched class attributes
3268 typedef struct {
3269   int   schedPolicy;              // classID
3270   int   maxPrio;
3271   int   minPrio;
3272 } SchedInfo;
3273 
3274 
3275 static SchedInfo tsLimits, iaLimits, rtLimits, fxLimits;
3276 
3277 #ifdef ASSERT
3278 static int  ReadBackValidate = 1;
3279 #endif
3280 static int  myClass     = 0;
3281 static int  myMin       = 0;
3282 static int  myMax       = 0;
3283 static int  myCur       = 0;
3284 static bool priocntl_enable = false;
3285 
3286 static const int criticalPrio = 60; // FX/60 is critical thread class/priority on T4
3287 static int java_MaxPriority_to_os_priority = 0; // Saved mapping
3288 
3289 
3290 // lwp_priocntl_init
3291 //
3292 // Try to determine the priority scale for our process.
3293 //
3294 // Return errno or 0 if OK.
3295 //
3296 static int lwp_priocntl_init() {
3297   int rslt;
3298   pcinfo_t ClassInfo;
3299   pcparms_t ParmInfo;
3300   int i;
3301 
3302   if (!UseThreadPriorities) return 0;
3303 
3304   // If ThreadPriorityPolicy is 1, switch tables
3305   if (ThreadPriorityPolicy == 1) {
3306     for (i = 0; i < CriticalPriority+1; i++)
3307       os::java_to_os_priority[i] = prio_policy1[i];
3308   }
3309   if (UseCriticalJavaThreadPriority) {
3310     // MaxPriority always maps to the FX scheduling class and criticalPrio.
3311     // See set_native_priority() and set_lwp_class_and_priority().
3312     // Save original MaxPriority mapping in case attempt to
3313     // use critical priority fails.
3314     java_MaxPriority_to_os_priority = os::java_to_os_priority[MaxPriority];
3315     // Set negative to distinguish from other priorities
3316     os::java_to_os_priority[MaxPriority] = -criticalPrio;
3317   }
3318 
3319   // Get IDs for a set of well-known scheduling classes.
3320   // TODO-FIXME: GETCLINFO returns the current # of classes in the
3321   // the system.  We should have a loop that iterates over the
3322   // classID values, which are known to be "small" integers.
3323 
3324   strcpy(ClassInfo.pc_clname, "TS");
3325   ClassInfo.pc_cid = -1;
3326   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3327   if (rslt < 0) return errno;
3328   assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3329   tsLimits.schedPolicy = ClassInfo.pc_cid;
3330   tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3331   tsLimits.minPrio = -tsLimits.maxPrio;
3332 
3333   strcpy(ClassInfo.pc_clname, "IA");
3334   ClassInfo.pc_cid = -1;
3335   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3336   if (rslt < 0) return errno;
3337   assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3338   iaLimits.schedPolicy = ClassInfo.pc_cid;
3339   iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3340   iaLimits.minPrio = -iaLimits.maxPrio;
3341 
3342   strcpy(ClassInfo.pc_clname, "RT");
3343   ClassInfo.pc_cid = -1;
3344   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3345   if (rslt < 0) return errno;
3346   assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3347   rtLimits.schedPolicy = ClassInfo.pc_cid;
3348   rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3349   rtLimits.minPrio = 0;
3350 
3351   strcpy(ClassInfo.pc_clname, "FX");
3352   ClassInfo.pc_cid = -1;
3353   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3354   if (rslt < 0) return errno;
3355   assert(ClassInfo.pc_cid != -1, "cid for FX class is -1");
3356   fxLimits.schedPolicy = ClassInfo.pc_cid;
3357   fxLimits.maxPrio = ((fxinfo_t*)ClassInfo.pc_clinfo)->fx_maxupri;
3358   fxLimits.minPrio = 0;
3359 
3360   // Query our "current" scheduling class.
3361   // This will normally be IA, TS or, rarely, FX or RT.
3362   memset(&ParmInfo, 0, sizeof(ParmInfo));
3363   ParmInfo.pc_cid = PC_CLNULL;
3364   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3365   if (rslt < 0) return errno;
3366   myClass = ParmInfo.pc_cid;
3367 
3368   // We now know our scheduling classId, get specific information
3369   // about the class.
3370   ClassInfo.pc_cid = myClass;
3371   ClassInfo.pc_clname[0] = 0;
3372   rslt = priocntl((idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo);
3373   if (rslt < 0) return errno;
3374 
3375   if (ThreadPriorityVerbose) {
3376     tty->print_cr("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3377   }
3378 
3379   memset(&ParmInfo, 0, sizeof(pcparms_t));
3380   ParmInfo.pc_cid = PC_CLNULL;
3381   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3382   if (rslt < 0) return errno;
3383 
3384   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3385     myMin = rtLimits.minPrio;
3386     myMax = rtLimits.maxPrio;
3387   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3388     iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3389     myMin = iaLimits.minPrio;
3390     myMax = iaLimits.maxPrio;
3391     myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3392   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3393     tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3394     myMin = tsLimits.minPrio;
3395     myMax = tsLimits.maxPrio;
3396     myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3397   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3398     fxparms_t *fxInfo = (fxparms_t*)ParmInfo.pc_clparms;
3399     myMin = fxLimits.minPrio;
3400     myMax = fxLimits.maxPrio;
3401     myMax = MIN2(myMax, (int)fxInfo->fx_uprilim);       // clamp - restrict
3402   } else {
3403     // No clue - punt
3404     if (ThreadPriorityVerbose) {
3405       tty->print_cr("Unknown scheduling class: %s ... \n",
3406                     ClassInfo.pc_clname);
3407     }
3408     return EINVAL;      // no clue, punt
3409   }
3410 
3411   if (ThreadPriorityVerbose) {
3412     tty->print_cr("Thread priority Range: [%d..%d]\n", myMin, myMax);
3413   }
3414 
3415   priocntl_enable = true;  // Enable changing priorities
3416   return 0;
3417 }
3418 
3419 #define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3420 #define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3421 #define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3422 #define FXPRI(x)        ((fxparms_t *)((x).pc_clparms))
3423 
3424 
3425 // scale_to_lwp_priority
3426 //
3427 // Convert from the libthread "thr_setprio" scale to our current
3428 // lwp scheduling class scale.
3429 //
3430 static int scale_to_lwp_priority(int rMin, int rMax, int x) {
3431   int v;
3432 
3433   if (x == 127) return rMax;            // avoid round-down
3434   v = (((x*(rMax-rMin)))/128)+rMin;
3435   return v;
3436 }
3437 
3438 
3439 // set_lwp_class_and_priority
3440 int set_lwp_class_and_priority(int ThreadID, int lwpid,
3441                                int newPrio, int new_class, bool scale) {
3442   int rslt;
3443   int Actual, Expected, prv;
3444   pcparms_t ParmInfo;                   // for GET-SET
3445 #ifdef ASSERT
3446   pcparms_t ReadBack;                   // for readback
3447 #endif
3448 
3449   // Set priority via PC_GETPARMS, update, PC_SETPARMS
3450   // Query current values.
3451   // TODO: accelerate this by eliminating the PC_GETPARMS call.
3452   // Cache "pcparms_t" in global ParmCache.
3453   // TODO: elide set-to-same-value
3454 
3455   // If something went wrong on init, don't change priorities.
3456   if (!priocntl_enable) {
3457     if (ThreadPriorityVerbose) {
3458       tty->print_cr("Trying to set priority but init failed, ignoring");
3459     }
3460     return EINVAL;
3461   }
3462 
3463   // If lwp hasn't started yet, just return
3464   // the _start routine will call us again.
3465   if (lwpid <= 0) {
3466     if (ThreadPriorityVerbose) {
3467       tty->print_cr("deferring the set_lwp_class_and_priority of thread "
3468                     INTPTR_FORMAT " to %d, lwpid not set",
3469                     ThreadID, newPrio);
3470     }
3471     return 0;
3472   }
3473 
3474   if (ThreadPriorityVerbose) {
3475     tty->print_cr ("set_lwp_class_and_priority("
3476                    INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3477                    ThreadID, lwpid, newPrio);
3478   }
3479 
3480   memset(&ParmInfo, 0, sizeof(pcparms_t));
3481   ParmInfo.pc_cid = PC_CLNULL;
3482   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3483   if (rslt < 0) return errno;
3484 
3485   int cur_class = ParmInfo.pc_cid;
3486   ParmInfo.pc_cid = (id_t)new_class;
3487 
3488   if (new_class == rtLimits.schedPolicy) {
3489     rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3490     rtInfo->rt_pri     = scale ? scale_to_lwp_priority(rtLimits.minPrio,
3491                                                        rtLimits.maxPrio, newPrio)
3492                                : newPrio;
3493     rtInfo->rt_tqsecs  = RT_NOCHANGE;
3494     rtInfo->rt_tqnsecs = RT_NOCHANGE;
3495     if (ThreadPriorityVerbose) {
3496       tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3497     }
3498   } else if (new_class == iaLimits.schedPolicy) {
3499     iaparms_t* iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3500     int maxClamped     = MIN2(iaLimits.maxPrio,
3501                               cur_class == new_class
3502                               ? (int)iaInfo->ia_uprilim : iaLimits.maxPrio);
3503     iaInfo->ia_upri    = scale ? scale_to_lwp_priority(iaLimits.minPrio,
3504                                                        maxClamped, newPrio)
3505                                : newPrio;
3506     iaInfo->ia_uprilim = cur_class == new_class
3507                            ? IA_NOCHANGE : (pri_t)iaLimits.maxPrio;
3508     iaInfo->ia_mode    = IA_NOCHANGE;
3509     if (ThreadPriorityVerbose) {
3510       tty->print_cr("IA: [%d...%d] %d->%d\n",
3511                     iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3512     }
3513   } else if (new_class == tsLimits.schedPolicy) {
3514     tsparms_t* tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3515     int maxClamped     = MIN2(tsLimits.maxPrio,
3516                               cur_class == new_class
3517                               ? (int)tsInfo->ts_uprilim : tsLimits.maxPrio);
3518     tsInfo->ts_upri    = scale ? scale_to_lwp_priority(tsLimits.minPrio,
3519                                                        maxClamped, newPrio)
3520                                : newPrio;
3521     tsInfo->ts_uprilim = cur_class == new_class
3522                            ? TS_NOCHANGE : (pri_t)tsLimits.maxPrio;
3523     if (ThreadPriorityVerbose) {
3524       tty->print_cr("TS: [%d...%d] %d->%d\n",
3525                     tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
3526     }
3527   } else if (new_class == fxLimits.schedPolicy) {
3528     fxparms_t* fxInfo  = (fxparms_t*)ParmInfo.pc_clparms;
3529     int maxClamped     = MIN2(fxLimits.maxPrio,
3530                               cur_class == new_class
3531                               ? (int)fxInfo->fx_uprilim : fxLimits.maxPrio);
3532     fxInfo->fx_upri    = scale ? scale_to_lwp_priority(fxLimits.minPrio,
3533                                                        maxClamped, newPrio)
3534                                : newPrio;
3535     fxInfo->fx_uprilim = cur_class == new_class
3536                            ? FX_NOCHANGE : (pri_t)fxLimits.maxPrio;
3537     fxInfo->fx_tqsecs  = FX_NOCHANGE;
3538     fxInfo->fx_tqnsecs = FX_NOCHANGE;
3539     if (ThreadPriorityVerbose) {
3540       tty->print_cr("FX: [%d...%d] %d->%d\n",
3541                     fxLimits.minPrio, maxClamped, newPrio, fxInfo->fx_upri);
3542     }
3543   } else {
3544     if (ThreadPriorityVerbose) {
3545       tty->print_cr("Unknown new scheduling class %d\n", new_class);
3546     }
3547     return EINVAL;    // no clue, punt
3548   }
3549 
3550   rslt = priocntl(P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
3551   if (ThreadPriorityVerbose && rslt) {
3552     tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
3553   }
3554   if (rslt < 0) return errno;
3555 
3556 #ifdef ASSERT
3557   // Sanity check: read back what we just attempted to set.
3558   // In theory it could have changed in the interim ...
3559   //
3560   // The priocntl system call is tricky.
3561   // Sometimes it'll validate the priority value argument and
3562   // return EINVAL if unhappy.  At other times it fails silently.
3563   // Readbacks are prudent.
3564 
3565   if (!ReadBackValidate) return 0;
3566 
3567   memset(&ReadBack, 0, sizeof(pcparms_t));
3568   ReadBack.pc_cid = PC_CLNULL;
3569   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
3570   assert(rslt >= 0, "priocntl failed");
3571   Actual = Expected = 0xBAD;
3572   assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
3573   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3574     Actual   = RTPRI(ReadBack)->rt_pri;
3575     Expected = RTPRI(ParmInfo)->rt_pri;
3576   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3577     Actual   = IAPRI(ReadBack)->ia_upri;
3578     Expected = IAPRI(ParmInfo)->ia_upri;
3579   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3580     Actual   = TSPRI(ReadBack)->ts_upri;
3581     Expected = TSPRI(ParmInfo)->ts_upri;
3582   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3583     Actual   = FXPRI(ReadBack)->fx_upri;
3584     Expected = FXPRI(ParmInfo)->fx_upri;
3585   } else {
3586     if (ThreadPriorityVerbose) {
3587       tty->print_cr("set_lwp_class_and_priority: unexpected class in readback: %d\n",
3588                     ParmInfo.pc_cid);
3589     }
3590   }
3591 
3592   if (Actual != Expected) {
3593     if (ThreadPriorityVerbose) {
3594       tty->print_cr ("set_lwp_class_and_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
3595                      lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
3596     }
3597   }
3598 #endif
3599 
3600   return 0;
3601 }
3602 
3603 // Solaris only gives access to 128 real priorities at a time,
3604 // so we expand Java's ten to fill this range.  This would be better
3605 // if we dynamically adjusted relative priorities.
3606 //
3607 // The ThreadPriorityPolicy option allows us to select 2 different
3608 // priority scales.
3609 //
3610 // ThreadPriorityPolicy=0
3611 // Since the Solaris' default priority is MaximumPriority, we do not
3612 // set a priority lower than Max unless a priority lower than
3613 // NormPriority is requested.
3614 //
3615 // ThreadPriorityPolicy=1
3616 // This mode causes the priority table to get filled with
3617 // linear values.  NormPriority get's mapped to 50% of the
3618 // Maximum priority an so on.  This will cause VM threads
3619 // to get unfair treatment against other Solaris processes
3620 // which do not explicitly alter their thread priorities.
3621 
3622 int os::java_to_os_priority[CriticalPriority + 1] = {
3623   -99999,         // 0 Entry should never be used
3624 
3625   0,              // 1 MinPriority
3626   32,             // 2
3627   64,             // 3
3628 
3629   96,             // 4
3630   127,            // 5 NormPriority
3631   127,            // 6
3632 
3633   127,            // 7
3634   127,            // 8
3635   127,            // 9 NearMaxPriority
3636 
3637   127,            // 10 MaxPriority
3638 
3639   -criticalPrio   // 11 CriticalPriority
3640 };
3641 
3642 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3643   OSThread* osthread = thread->osthread();
3644 
3645   // Save requested priority in case the thread hasn't been started
3646   osthread->set_native_priority(newpri);
3647 
3648   // Check for critical priority request
3649   bool fxcritical = false;
3650   if (newpri == -criticalPrio) {
3651     fxcritical = true;
3652     newpri = criticalPrio;
3653   }
3654 
3655   assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
3656   if (!UseThreadPriorities) return OS_OK;
3657 
3658   int status = 0;
3659 
3660   if (!fxcritical) {
3661     // Use thr_setprio only if we have a priority that thr_setprio understands
3662     status = thr_setprio(thread->osthread()->thread_id(), newpri);
3663   }
3664 
3665   int lwp_status =
3666           set_lwp_class_and_priority(osthread->thread_id(),
3667                                      osthread->lwp_id(),
3668                                      newpri,
3669                                      fxcritical ? fxLimits.schedPolicy : myClass,
3670                                      !fxcritical);
3671   if (lwp_status != 0 && fxcritical) {
3672     // Try again, this time without changing the scheduling class
3673     newpri = java_MaxPriority_to_os_priority;
3674     lwp_status = set_lwp_class_and_priority(osthread->thread_id(),
3675                                             osthread->lwp_id(),
3676                                             newpri, myClass, false);
3677   }
3678   status |= lwp_status;
3679   return (status == 0) ? OS_OK : OS_ERR;
3680 }
3681 
3682 
3683 OSReturn os::get_native_priority(const Thread* const thread,
3684                                  int *priority_ptr) {
3685   int p;
3686   if (!UseThreadPriorities) {
3687     *priority_ptr = NormalPriority;
3688     return OS_OK;
3689   }
3690   int status = thr_getprio(thread->osthread()->thread_id(), &p);
3691   if (status != 0) {
3692     return OS_ERR;
3693   }
3694   *priority_ptr = p;
3695   return OS_OK;
3696 }
3697 
3698 
3699 // Hint to the underlying OS that a task switch would not be good.
3700 // Void return because it's a hint and can fail.
3701 void os::hint_no_preempt() {
3702   schedctl_start(schedctl_init());
3703 }
3704 
3705 static void resume_clear_context(OSThread *osthread) {
3706   osthread->set_ucontext(NULL);
3707 }
3708 
3709 static void suspend_save_context(OSThread *osthread, ucontext_t* context) {
3710   osthread->set_ucontext(context);
3711 }
3712 
3713 static Semaphore sr_semaphore;
3714 
3715 void os::Solaris::SR_handler(Thread* thread, ucontext_t* uc) {
3716   // Save and restore errno to avoid confusing native code with EINTR
3717   // after sigsuspend.
3718   int old_errno = errno;
3719 
3720   OSThread* osthread = thread->osthread();
3721   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3722 
3723   os::SuspendResume::State current = osthread->sr.state();
3724   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3725     suspend_save_context(osthread, uc);
3726 
3727     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3728     os::SuspendResume::State state = osthread->sr.suspended();
3729     if (state == os::SuspendResume::SR_SUSPENDED) {
3730       sigset_t suspend_set;  // signals for sigsuspend()
3731 
3732       // get current set of blocked signals and unblock resume signal
3733       thr_sigsetmask(SIG_BLOCK, NULL, &suspend_set);
3734       sigdelset(&suspend_set, os::Solaris::SIGasync());
3735 
3736       sr_semaphore.signal();
3737       // wait here until we are resumed
3738       while (1) {
3739         sigsuspend(&suspend_set);
3740 
3741         os::SuspendResume::State result = osthread->sr.running();
3742         if (result == os::SuspendResume::SR_RUNNING) {
3743           sr_semaphore.signal();
3744           break;
3745         }
3746       }
3747 
3748     } else if (state == os::SuspendResume::SR_RUNNING) {
3749       // request was cancelled, continue
3750     } else {
3751       ShouldNotReachHere();
3752     }
3753 
3754     resume_clear_context(osthread);
3755   } else if (current == os::SuspendResume::SR_RUNNING) {
3756     // request was cancelled, continue
3757   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3758     // ignore
3759   } else {
3760     // ignore
3761   }
3762 
3763   errno = old_errno;
3764 }
3765 
3766 void os::print_statistics() {
3767 }
3768 
3769 int os::message_box(const char* title, const char* message) {
3770   int i;
3771   fdStream err(defaultStream::error_fd());
3772   for (i = 0; i < 78; i++) err.print_raw("=");
3773   err.cr();
3774   err.print_raw_cr(title);
3775   for (i = 0; i < 78; i++) err.print_raw("-");
3776   err.cr();
3777   err.print_raw_cr(message);
3778   for (i = 0; i < 78; i++) err.print_raw("=");
3779   err.cr();
3780 
3781   char buf[16];
3782   // Prevent process from exiting upon "read error" without consuming all CPU
3783   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3784 
3785   return buf[0] == 'y' || buf[0] == 'Y';
3786 }
3787 
3788 static int sr_notify(OSThread* osthread) {
3789   int status = thr_kill(osthread->thread_id(), os::Solaris::SIGasync());
3790   assert_status(status == 0, status, "thr_kill");
3791   return status;
3792 }
3793 
3794 // "Randomly" selected value for how long we want to spin
3795 // before bailing out on suspending a thread, also how often
3796 // we send a signal to a thread we want to resume
3797 static const int RANDOMLY_LARGE_INTEGER = 1000000;
3798 static const int RANDOMLY_LARGE_INTEGER2 = 100;
3799 
3800 static bool do_suspend(OSThread* osthread) {
3801   assert(osthread->sr.is_running(), "thread should be running");
3802   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3803 
3804   // mark as suspended and send signal
3805   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3806     // failed to switch, state wasn't running?
3807     ShouldNotReachHere();
3808     return false;
3809   }
3810 
3811   if (sr_notify(osthread) != 0) {
3812     ShouldNotReachHere();
3813   }
3814 
3815   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3816   while (true) {
3817     if (sr_semaphore.timedwait(0, 2000 * NANOSECS_PER_MILLISEC)) {
3818       break;
3819     } else {
3820       // timeout
3821       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3822       if (cancelled == os::SuspendResume::SR_RUNNING) {
3823         return false;
3824       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3825         // make sure that we consume the signal on the semaphore as well
3826         sr_semaphore.wait();
3827         break;
3828       } else {
3829         ShouldNotReachHere();
3830         return false;
3831       }
3832     }
3833   }
3834 
3835   guarantee(osthread->sr.is_suspended(), "Must be suspended");
3836   return true;
3837 }
3838 
3839 static void do_resume(OSThread* osthread) {
3840   assert(osthread->sr.is_suspended(), "thread should be suspended");
3841   assert(!sr_semaphore.trywait(), "invalid semaphore state");
3842 
3843   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3844     // failed to switch to WAKEUP_REQUEST
3845     ShouldNotReachHere();
3846     return;
3847   }
3848 
3849   while (true) {
3850     if (sr_notify(osthread) == 0) {
3851       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3852         if (osthread->sr.is_running()) {
3853           return;
3854         }
3855       }
3856     } else {
3857       ShouldNotReachHere();
3858     }
3859   }
3860 
3861   guarantee(osthread->sr.is_running(), "Must be running!");
3862 }
3863 
3864 void os::SuspendedThreadTask::internal_do_task() {
3865   if (do_suspend(_thread->osthread())) {
3866     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
3867     do_task(context);
3868     do_resume(_thread->osthread());
3869   }
3870 }
3871 
3872 class PcFetcher : public os::SuspendedThreadTask {
3873  public:
3874   PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
3875   ExtendedPC result();
3876  protected:
3877   void do_task(const os::SuspendedThreadTaskContext& context);
3878  private:
3879   ExtendedPC _epc;
3880 };
3881 
3882 ExtendedPC PcFetcher::result() {
3883   guarantee(is_done(), "task is not done yet.");
3884   return _epc;
3885 }
3886 
3887 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
3888   Thread* thread = context.thread();
3889   OSThread* osthread = thread->osthread();
3890   if (osthread->ucontext() != NULL) {
3891     _epc = os::Solaris::ucontext_get_pc((ucontext_t *) context.ucontext());
3892   } else {
3893     // NULL context is unexpected, double-check this is the VMThread
3894     guarantee(thread->is_VM_thread(), "can only be called for VMThread");
3895   }
3896 }
3897 
3898 // A lightweight implementation that does not suspend the target thread and
3899 // thus returns only a hint. Used for profiling only!
3900 ExtendedPC os::get_thread_pc(Thread* thread) {
3901   // Make sure that it is called by the watcher and the Threads lock is owned.
3902   assert(Thread::current()->is_Watcher_thread(), "Must be watcher and own Threads_lock");
3903   // For now, is only used to profile the VM Thread
3904   assert(thread->is_VM_thread(), "Can only be called for VMThread");
3905   PcFetcher fetcher(thread);
3906   fetcher.run();
3907   return fetcher.result();
3908 }
3909 
3910 
3911 // This does not do anything on Solaris. This is basically a hook for being
3912 // able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
3913 void os::os_exception_wrapper(java_call_t f, JavaValue* value,
3914                               methodHandle* method, JavaCallArguments* args,
3915                               Thread* thread) {
3916   f(value, method, args, thread);
3917 }
3918 
3919 // This routine may be used by user applications as a "hook" to catch signals.
3920 // The user-defined signal handler must pass unrecognized signals to this
3921 // routine, and if it returns true (non-zero), then the signal handler must
3922 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3923 // routine will never retun false (zero), but instead will execute a VM panic
3924 // routine kill the process.
3925 //
3926 // If this routine returns false, it is OK to call it again.  This allows
3927 // the user-defined signal handler to perform checks either before or after
3928 // the VM performs its own checks.  Naturally, the user code would be making
3929 // a serious error if it tried to handle an exception (such as a null check
3930 // or breakpoint) that the VM was generating for its own correct operation.
3931 //
3932 // This routine may recognize any of the following kinds of signals:
3933 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
3934 // os::Solaris::SIGasync
3935 // It should be consulted by handlers for any of those signals.
3936 // It explicitly does not recognize os::Solaris::SIGinterrupt
3937 //
3938 // The caller of this routine must pass in the three arguments supplied
3939 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3940 // field of the structure passed to sigaction().  This routine assumes that
3941 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3942 //
3943 // Note that the VM will print warnings if it detects conflicting signal
3944 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3945 //
3946 extern "C" JNIEXPORT int JVM_handle_solaris_signal(int signo,
3947                                                    siginfo_t* siginfo,
3948                                                    void* ucontext,
3949                                                    int abort_if_unrecognized);
3950 
3951 
3952 void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
3953   int orig_errno = errno;  // Preserve errno value over signal handler.
3954   JVM_handle_solaris_signal(sig, info, ucVoid, true);
3955   errno = orig_errno;
3956 }
3957 
3958 // Do not delete - if guarantee is ever removed,  a signal handler (even empty)
3959 // is needed to provoke threads blocked on IO to return an EINTR
3960 // Note: this explicitly does NOT call JVM_handle_solaris_signal and
3961 // does NOT participate in signal chaining due to requirement for
3962 // NOT setting SA_RESTART to make EINTR work.
3963 extern "C" void sigINTRHandler(int sig, siginfo_t* info, void* ucVoid) {
3964   if (UseSignalChaining) {
3965     struct sigaction *actp = os::Solaris::get_chained_signal_action(sig);
3966     if (actp && actp->sa_handler) {
3967       vm_exit_during_initialization("Signal chaining detected for VM interrupt signal, try -XX:+UseAltSigs");
3968     }
3969   }
3970 }
3971 
3972 // This boolean allows users to forward their own non-matching signals
3973 // to JVM_handle_solaris_signal, harmlessly.
3974 bool os::Solaris::signal_handlers_are_installed = false;
3975 
3976 // For signal-chaining
3977 bool os::Solaris::libjsig_is_loaded = false;
3978 typedef struct sigaction *(*get_signal_t)(int);
3979 get_signal_t os::Solaris::get_signal_action = NULL;
3980 
3981 struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
3982   struct sigaction *actp = NULL;
3983 
3984   if ((libjsig_is_loaded)  && (sig <= Maxlibjsigsigs)) {
3985     // Retrieve the old signal handler from libjsig
3986     actp = (*get_signal_action)(sig);
3987   }
3988   if (actp == NULL) {
3989     // Retrieve the preinstalled signal handler from jvm
3990     actp = get_preinstalled_handler(sig);
3991   }
3992 
3993   return actp;
3994 }
3995 
3996 static bool call_chained_handler(struct sigaction *actp, int sig,
3997                                  siginfo_t *siginfo, void *context) {
3998   // Call the old signal handler
3999   if (actp->sa_handler == SIG_DFL) {
4000     // It's more reasonable to let jvm treat it as an unexpected exception
4001     // instead of taking the default action.
4002     return false;
4003   } else if (actp->sa_handler != SIG_IGN) {
4004     if ((actp->sa_flags & SA_NODEFER) == 0) {
4005       // automaticlly block the signal
4006       sigaddset(&(actp->sa_mask), sig);
4007     }
4008 
4009     sa_handler_t hand;
4010     sa_sigaction_t sa;
4011     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4012     // retrieve the chained handler
4013     if (siginfo_flag_set) {
4014       sa = actp->sa_sigaction;
4015     } else {
4016       hand = actp->sa_handler;
4017     }
4018 
4019     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4020       actp->sa_handler = SIG_DFL;
4021     }
4022 
4023     // try to honor the signal mask
4024     sigset_t oset;
4025     thr_sigsetmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4026 
4027     // call into the chained handler
4028     if (siginfo_flag_set) {
4029       (*sa)(sig, siginfo, context);
4030     } else {
4031       (*hand)(sig);
4032     }
4033 
4034     // restore the signal mask
4035     thr_sigsetmask(SIG_SETMASK, &oset, 0);
4036   }
4037   // Tell jvm's signal handler the signal is taken care of.
4038   return true;
4039 }
4040 
4041 bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4042   bool chained = false;
4043   // signal-chaining
4044   if (UseSignalChaining) {
4045     struct sigaction *actp = get_chained_signal_action(sig);
4046     if (actp != NULL) {
4047       chained = call_chained_handler(actp, sig, siginfo, context);
4048     }
4049   }
4050   return chained;
4051 }
4052 
4053 struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
4054   assert((chainedsigactions != (struct sigaction *)NULL) &&
4055          (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
4056   if (preinstalled_sigs[sig] != 0) {
4057     return &chainedsigactions[sig];
4058   }
4059   return NULL;
4060 }
4061 
4062 void os::Solaris::save_preinstalled_handler(int sig,
4063                                             struct sigaction& oldAct) {
4064   assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
4065   assert((chainedsigactions != (struct sigaction *)NULL) &&
4066          (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
4067   chainedsigactions[sig] = oldAct;
4068   preinstalled_sigs[sig] = 1;
4069 }
4070 
4071 void os::Solaris::set_signal_handler(int sig, bool set_installed,
4072                                      bool oktochain) {
4073   // Check for overwrite.
4074   struct sigaction oldAct;
4075   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4076   void* oldhand =
4077       oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4078                           : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4079   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4080       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4081       oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
4082     if (AllowUserSignalHandlers || !set_installed) {
4083       // Do not overwrite; user takes responsibility to forward to us.
4084       return;
4085     } else if (UseSignalChaining) {
4086       if (oktochain) {
4087         // save the old handler in jvm
4088         save_preinstalled_handler(sig, oldAct);
4089       } else {
4090         vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal, try -XX:+UseAltSigs.");
4091       }
4092       // libjsig also interposes the sigaction() call below and saves the
4093       // old sigaction on it own.
4094     } else {
4095       fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4096                     "%#lx for signal %d.", (long)oldhand, sig));
4097     }
4098   }
4099 
4100   struct sigaction sigAct;
4101   sigfillset(&(sigAct.sa_mask));
4102   sigAct.sa_handler = SIG_DFL;
4103 
4104   sigAct.sa_sigaction = signalHandler;
4105   // Handle SIGSEGV on alternate signal stack if
4106   // not using stack banging
4107   if (!UseStackBanging && sig == SIGSEGV) {
4108     sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
4109   } else if (sig == os::Solaris::SIGinterrupt()) {
4110     // Interruptible i/o requires SA_RESTART cleared so EINTR
4111     // is returned instead of restarting system calls
4112     sigemptyset(&sigAct.sa_mask);
4113     sigAct.sa_handler = NULL;
4114     sigAct.sa_flags = SA_SIGINFO;
4115     sigAct.sa_sigaction = sigINTRHandler;
4116   } else {
4117     sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
4118   }
4119   os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
4120 
4121   sigaction(sig, &sigAct, &oldAct);
4122 
4123   void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4124                                        : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4125   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4126 }
4127 
4128 
4129 #define DO_SIGNAL_CHECK(sig)                      \
4130   do {                                            \
4131     if (!sigismember(&check_signal_done, sig)) {  \
4132       os::Solaris::check_signal_handler(sig);     \
4133     }                                             \
4134   } while (0)
4135 
4136 // This method is a periodic task to check for misbehaving JNI applications
4137 // under CheckJNI, we can add any periodic checks here
4138 
4139 void os::run_periodic_checks() {
4140   // A big source of grief is hijacking virt. addr 0x0 on Solaris,
4141   // thereby preventing a NULL checks.
4142   if (!check_addr0_done) check_addr0_done = check_addr0(tty);
4143 
4144   if (check_signals == false) return;
4145 
4146   // SEGV and BUS if overridden could potentially prevent
4147   // generation of hs*.log in the event of a crash, debugging
4148   // such a case can be very challenging, so we absolutely
4149   // check for the following for a good measure:
4150   DO_SIGNAL_CHECK(SIGSEGV);
4151   DO_SIGNAL_CHECK(SIGILL);
4152   DO_SIGNAL_CHECK(SIGFPE);
4153   DO_SIGNAL_CHECK(SIGBUS);
4154   DO_SIGNAL_CHECK(SIGPIPE);
4155   DO_SIGNAL_CHECK(SIGXFSZ);
4156 
4157   // ReduceSignalUsage allows the user to override these handlers
4158   // see comments at the very top and jvm_solaris.h
4159   if (!ReduceSignalUsage) {
4160     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4161     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4162     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4163     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4164   }
4165 
4166   // See comments above for using JVM1/JVM2 and UseAltSigs
4167   DO_SIGNAL_CHECK(os::Solaris::SIGinterrupt());
4168   DO_SIGNAL_CHECK(os::Solaris::SIGasync());
4169 
4170 }
4171 
4172 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4173 
4174 static os_sigaction_t os_sigaction = NULL;
4175 
4176 void os::Solaris::check_signal_handler(int sig) {
4177   char buf[O_BUFLEN];
4178   address jvmHandler = NULL;
4179 
4180   struct sigaction act;
4181   if (os_sigaction == NULL) {
4182     // only trust the default sigaction, in case it has been interposed
4183     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4184     if (os_sigaction == NULL) return;
4185   }
4186 
4187   os_sigaction(sig, (struct sigaction*)NULL, &act);
4188 
4189   address thisHandler = (act.sa_flags & SA_SIGINFO)
4190     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4191     : CAST_FROM_FN_PTR(address, act.sa_handler);
4192 
4193 
4194   switch (sig) {
4195   case SIGSEGV:
4196   case SIGBUS:
4197   case SIGFPE:
4198   case SIGPIPE:
4199   case SIGXFSZ:
4200   case SIGILL:
4201     jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4202     break;
4203 
4204   case SHUTDOWN1_SIGNAL:
4205   case SHUTDOWN2_SIGNAL:
4206   case SHUTDOWN3_SIGNAL:
4207   case BREAK_SIGNAL:
4208     jvmHandler = (address)user_handler();
4209     break;
4210 
4211   default:
4212     int intrsig = os::Solaris::SIGinterrupt();
4213     int asynsig = os::Solaris::SIGasync();
4214 
4215     if (sig == intrsig) {
4216       jvmHandler = CAST_FROM_FN_PTR(address, sigINTRHandler);
4217     } else if (sig == asynsig) {
4218       jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
4219     } else {
4220       return;
4221     }
4222     break;
4223   }
4224 
4225 
4226   if (thisHandler != jvmHandler) {
4227     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4228     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4229     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4230     // No need to check this sig any longer
4231     sigaddset(&check_signal_done, sig);
4232     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4233     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4234       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4235                     exception_name(sig, buf, O_BUFLEN));
4236     }
4237   } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
4238     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4239     tty->print("expected:" PTR32_FORMAT, os::Solaris::get_our_sigflags(sig));
4240     tty->print_cr("  found:" PTR32_FORMAT, act.sa_flags);
4241     // No need to check this sig any longer
4242     sigaddset(&check_signal_done, sig);
4243   }
4244 
4245   // Print all the signal handler state
4246   if (sigismember(&check_signal_done, sig)) {
4247     print_signal_handlers(tty, buf, O_BUFLEN);
4248   }
4249 
4250 }
4251 
4252 void os::Solaris::install_signal_handlers() {
4253   bool libjsigdone = false;
4254   signal_handlers_are_installed = true;
4255 
4256   // signal-chaining
4257   typedef void (*signal_setting_t)();
4258   signal_setting_t begin_signal_setting = NULL;
4259   signal_setting_t end_signal_setting = NULL;
4260   begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4261                                         dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4262   if (begin_signal_setting != NULL) {
4263     end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4264                                         dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4265     get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4266                                        dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4267     get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
4268                                          dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
4269     libjsig_is_loaded = true;
4270     if (os::Solaris::get_libjsig_version != NULL) {
4271       libjsigversion =  (*os::Solaris::get_libjsig_version)();
4272     }
4273     assert(UseSignalChaining, "should enable signal-chaining");
4274   }
4275   if (libjsig_is_loaded) {
4276     // Tell libjsig jvm is setting signal handlers
4277     (*begin_signal_setting)();
4278   }
4279 
4280   set_signal_handler(SIGSEGV, true, true);
4281   set_signal_handler(SIGPIPE, true, true);
4282   set_signal_handler(SIGXFSZ, true, true);
4283   set_signal_handler(SIGBUS, true, true);
4284   set_signal_handler(SIGILL, true, true);
4285   set_signal_handler(SIGFPE, true, true);
4286 
4287 
4288   if (os::Solaris::SIGinterrupt() > OLDMAXSIGNUM || os::Solaris::SIGasync() > OLDMAXSIGNUM) {
4289 
4290     // Pre-1.4.1 Libjsig limited to signal chaining signals <= 32 so
4291     // can not register overridable signals which might be > 32
4292     if (libjsig_is_loaded && libjsigversion <= JSIG_VERSION_1_4_1) {
4293       // Tell libjsig jvm has finished setting signal handlers
4294       (*end_signal_setting)();
4295       libjsigdone = true;
4296     }
4297   }
4298 
4299   // Never ok to chain our SIGinterrupt
4300   set_signal_handler(os::Solaris::SIGinterrupt(), true, false);
4301   set_signal_handler(os::Solaris::SIGasync(), true, true);
4302 
4303   if (libjsig_is_loaded && !libjsigdone) {
4304     // Tell libjsig jvm finishes setting signal handlers
4305     (*end_signal_setting)();
4306   }
4307 
4308   // We don't activate signal checker if libjsig is in place, we trust ourselves
4309   // and if UserSignalHandler is installed all bets are off.
4310   // Log that signal checking is off only if -verbose:jni is specified.
4311   if (CheckJNICalls) {
4312     if (libjsig_is_loaded) {
4313       if (PrintJNIResolving) {
4314         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4315       }
4316       check_signals = false;
4317     }
4318     if (AllowUserSignalHandlers) {
4319       if (PrintJNIResolving) {
4320         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4321       }
4322       check_signals = false;
4323     }
4324   }
4325 }
4326 
4327 
4328 void report_error(const char* file_name, int line_no, const char* title,
4329                   const char* format, ...);
4330 
4331 const char * signames[] = {
4332   "SIG0",
4333   "SIGHUP", "SIGINT", "SIGQUIT", "SIGILL", "SIGTRAP",
4334   "SIGABRT", "SIGEMT", "SIGFPE", "SIGKILL", "SIGBUS",
4335   "SIGSEGV", "SIGSYS", "SIGPIPE", "SIGALRM", "SIGTERM",
4336   "SIGUSR1", "SIGUSR2", "SIGCLD", "SIGPWR", "SIGWINCH",
4337   "SIGURG", "SIGPOLL", "SIGSTOP", "SIGTSTP", "SIGCONT",
4338   "SIGTTIN", "SIGTTOU", "SIGVTALRM", "SIGPROF", "SIGXCPU",
4339   "SIGXFSZ", "SIGWAITING", "SIGLWP", "SIGFREEZE", "SIGTHAW",
4340   "SIGCANCEL", "SIGLOST"
4341 };
4342 
4343 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4344   if (0 < exception_code && exception_code <= SIGRTMAX) {
4345     // signal
4346     if (exception_code < sizeof(signames)/sizeof(const char*)) {
4347       jio_snprintf(buf, size, "%s", signames[exception_code]);
4348     } else {
4349       jio_snprintf(buf, size, "SIG%d", exception_code);
4350     }
4351     return buf;
4352   } else {
4353     return NULL;
4354   }
4355 }
4356 
4357 // (Static) wrapper for getisax(2) call.
4358 os::Solaris::getisax_func_t os::Solaris::_getisax = 0;
4359 
4360 // (Static) wrappers for the liblgrp API
4361 os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4362 os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4363 os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4364 os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4365 os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4366 os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4367 os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4368 os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4369 os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4370 
4371 // (Static) wrapper for meminfo() call.
4372 os::Solaris::meminfo_func_t os::Solaris::_meminfo = 0;
4373 
4374 static address resolve_symbol_lazy(const char* name) {
4375   address addr = (address) dlsym(RTLD_DEFAULT, name);
4376   if (addr == NULL) {
4377     // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4378     addr = (address) dlsym(RTLD_NEXT, name);
4379   }
4380   return addr;
4381 }
4382 
4383 static address resolve_symbol(const char* name) {
4384   address addr = resolve_symbol_lazy(name);
4385   if (addr == NULL) {
4386     fatal(dlerror());
4387   }
4388   return addr;
4389 }
4390 
4391 void os::Solaris::libthread_init() {
4392   address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4393 
4394   lwp_priocntl_init();
4395 
4396   // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4397   if (func == NULL) {
4398     func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4399     // Guarantee that this VM is running on an new enough OS (5.6 or
4400     // later) that it will have a new enough libthread.so.
4401     guarantee(func != NULL, "libthread.so is too old.");
4402   }
4403 
4404   int size;
4405   void (*handler_info_func)(address *, int *);
4406   handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4407   handler_info_func(&handler_start, &size);
4408   handler_end = handler_start + size;
4409 }
4410 
4411 
4412 int_fnP_mutex_tP os::Solaris::_mutex_lock;
4413 int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4414 int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4415 int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4416 int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4417 int os::Solaris::_mutex_scope = USYNC_THREAD;
4418 
4419 int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4420 int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4421 int_fnP_cond_tP os::Solaris::_cond_signal;
4422 int_fnP_cond_tP os::Solaris::_cond_broadcast;
4423 int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4424 int_fnP_cond_tP os::Solaris::_cond_destroy;
4425 int os::Solaris::_cond_scope = USYNC_THREAD;
4426 
4427 void os::Solaris::synchronization_init() {
4428   if (UseLWPSynchronization) {
4429     os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4430     os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4431     os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4432     os::Solaris::set_mutex_init(lwp_mutex_init);
4433     os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4434     os::Solaris::set_mutex_scope(USYNC_THREAD);
4435 
4436     os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4437     os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4438     os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4439     os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4440     os::Solaris::set_cond_init(lwp_cond_init);
4441     os::Solaris::set_cond_destroy(lwp_cond_destroy);
4442     os::Solaris::set_cond_scope(USYNC_THREAD);
4443   } else {
4444     os::Solaris::set_mutex_scope(USYNC_THREAD);
4445     os::Solaris::set_cond_scope(USYNC_THREAD);
4446 
4447     if (UsePthreads) {
4448       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4449       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4450       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4451       os::Solaris::set_mutex_init(pthread_mutex_default_init);
4452       os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4453 
4454       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4455       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4456       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4457       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4458       os::Solaris::set_cond_init(pthread_cond_default_init);
4459       os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4460     } else {
4461       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4462       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4463       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4464       os::Solaris::set_mutex_init(::mutex_init);
4465       os::Solaris::set_mutex_destroy(::mutex_destroy);
4466 
4467       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4468       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4469       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4470       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4471       os::Solaris::set_cond_init(::cond_init);
4472       os::Solaris::set_cond_destroy(::cond_destroy);
4473     }
4474   }
4475 }
4476 
4477 bool os::Solaris::liblgrp_init() {
4478   void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4479   if (handle != NULL) {
4480     os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4481     os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4482     os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4483     os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4484     os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4485     os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4486     os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4487     os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4488                                                       dlsym(handle, "lgrp_cookie_stale")));
4489 
4490     lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4491     set_lgrp_cookie(c);
4492     return true;
4493   }
4494   return false;
4495 }
4496 
4497 void os::Solaris::misc_sym_init() {
4498   address func;
4499 
4500   // getisax
4501   func = resolve_symbol_lazy("getisax");
4502   if (func != NULL) {
4503     os::Solaris::_getisax = CAST_TO_FN_PTR(getisax_func_t, func);
4504   }
4505 
4506   // meminfo
4507   func = resolve_symbol_lazy("meminfo");
4508   if (func != NULL) {
4509     os::Solaris::set_meminfo(CAST_TO_FN_PTR(meminfo_func_t, func));
4510   }
4511 }
4512 
4513 uint_t os::Solaris::getisax(uint32_t* array, uint_t n) {
4514   assert(_getisax != NULL, "_getisax not set");
4515   return _getisax(array, n);
4516 }
4517 
4518 // int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
4519 typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
4520 static pset_getloadavg_type pset_getloadavg_ptr = NULL;
4521 
4522 void init_pset_getloadavg_ptr(void) {
4523   pset_getloadavg_ptr =
4524     (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
4525   if (PrintMiscellaneous && Verbose && pset_getloadavg_ptr == NULL) {
4526     warning("pset_getloadavg function not found");
4527   }
4528 }
4529 
4530 int os::Solaris::_dev_zero_fd = -1;
4531 
4532 // this is called _before_ the global arguments have been parsed
4533 void os::init(void) {
4534   _initial_pid = getpid();
4535 
4536   max_hrtime = first_hrtime = gethrtime();
4537 
4538   init_random(1234567);
4539 
4540   page_size = sysconf(_SC_PAGESIZE);
4541   if (page_size == -1) {
4542     fatal(err_msg("os_solaris.cpp: os::init: sysconf failed (%s)",
4543                   strerror(errno)));
4544   }
4545   init_page_sizes((size_t) page_size);
4546 
4547   Solaris::initialize_system_info();
4548 
4549   // Initialize misc. symbols as soon as possible, so we can use them
4550   // if we need them.
4551   Solaris::misc_sym_init();
4552 
4553   int fd = ::open("/dev/zero", O_RDWR);
4554   if (fd < 0) {
4555     fatal(err_msg("os::init: cannot open /dev/zero (%s)", strerror(errno)));
4556   } else {
4557     Solaris::set_dev_zero_fd(fd);
4558 
4559     // Close on exec, child won't inherit.
4560     fcntl(fd, F_SETFD, FD_CLOEXEC);
4561   }
4562 
4563   clock_tics_per_sec = CLK_TCK;
4564 
4565   // check if dladdr1() exists; dladdr1 can provide more information than
4566   // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
4567   // and is available on linker patches for 5.7 and 5.8.
4568   // libdl.so must have been loaded, this call is just an entry lookup
4569   void * hdl = dlopen("libdl.so", RTLD_NOW);
4570   if (hdl) {
4571     dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
4572   }
4573 
4574   // (Solaris only) this switches to calls that actually do locking.
4575   ThreadCritical::initialize();
4576 
4577   main_thread = thr_self();
4578 
4579   // Constant minimum stack size allowed. It must be at least
4580   // the minimum of what the OS supports (thr_min_stack()), and
4581   // enough to allow the thread to get to user bytecode execution.
4582   Solaris::min_stack_allowed = MAX2(thr_min_stack(), Solaris::min_stack_allowed);
4583   // If the pagesize of the VM is greater than 8K determine the appropriate
4584   // number of initial guard pages.  The user can change this with the
4585   // command line arguments, if needed.
4586   if (vm_page_size() > 8*K) {
4587     StackYellowPages = 1;
4588     StackRedPages = 1;
4589     StackShadowPages = round_to((StackShadowPages*8*K), vm_page_size()) / vm_page_size();
4590   }
4591 }
4592 
4593 // To install functions for atexit system call
4594 extern "C" {
4595   static void perfMemory_exit_helper() {
4596     perfMemory_exit();
4597   }
4598 }
4599 
4600 // this is called _after_ the global arguments have been parsed
4601 jint os::init_2(void) {
4602   // try to enable extended file IO ASAP, see 6431278
4603   os::Solaris::try_enable_extended_io();
4604 
4605   // Allocate a single page and mark it as readable for safepoint polling.  Also
4606   // use this first mmap call to check support for MAP_ALIGN.
4607   address polling_page = (address)Solaris::mmap_chunk((char*)page_size,
4608                                                       page_size,
4609                                                       MAP_PRIVATE | MAP_ALIGN,
4610                                                       PROT_READ);
4611   if (polling_page == NULL) {
4612     has_map_align = false;
4613     polling_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE,
4614                                                 PROT_READ);
4615   }
4616 
4617   os::set_polling_page(polling_page);
4618 
4619 #ifndef PRODUCT
4620   if (Verbose && PrintMiscellaneous) {
4621     tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4622                (intptr_t)polling_page);
4623   }
4624 #endif
4625 
4626   if (!UseMembar) {
4627     address mem_serialize_page = (address)Solaris::mmap_chunk(NULL, page_size, MAP_PRIVATE, PROT_READ | PROT_WRITE);
4628     guarantee(mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4629     os::set_memory_serialize_page(mem_serialize_page);
4630 
4631 #ifndef PRODUCT
4632     if (Verbose && PrintMiscellaneous) {
4633       tty->print("[Memory Serialize  Page address: " INTPTR_FORMAT "]\n",
4634                  (intptr_t)mem_serialize_page);
4635     }
4636 #endif
4637   }
4638 
4639   // Check minimum allowable stack size for thread creation and to initialize
4640   // the java system classes, including StackOverflowError - depends on page
4641   // size.  Add a page for compiler2 recursion in main thread.
4642   // Add in 2*BytesPerWord times page size to account for VM stack during
4643   // class initialization depending on 32 or 64 bit VM.
4644   os::Solaris::min_stack_allowed = MAX2(os::Solaris::min_stack_allowed,
4645                                         (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4646                                         2*BytesPerWord COMPILER2_PRESENT(+1)) * page_size);
4647 
4648   size_t threadStackSizeInBytes = ThreadStackSize * K;
4649   if (threadStackSizeInBytes != 0 &&
4650       threadStackSizeInBytes < os::Solaris::min_stack_allowed) {
4651     tty->print_cr("\nThe stack size specified is too small, Specify at least %dk",
4652                   os::Solaris::min_stack_allowed/K);
4653     return JNI_ERR;
4654   }
4655 
4656   // For 64kbps there will be a 64kb page size, which makes
4657   // the usable default stack size quite a bit less.  Increase the
4658   // stack for 64kb (or any > than 8kb) pages, this increases
4659   // virtual memory fragmentation (since we're not creating the
4660   // stack on a power of 2 boundary.  The real fix for this
4661   // should be to fix the guard page mechanism.
4662 
4663   if (vm_page_size() > 8*K) {
4664     threadStackSizeInBytes = (threadStackSizeInBytes != 0)
4665        ? threadStackSizeInBytes +
4666          ((StackYellowPages + StackRedPages) * vm_page_size())
4667        : 0;
4668     ThreadStackSize = threadStackSizeInBytes/K;
4669   }
4670 
4671   // Make the stack size a multiple of the page size so that
4672   // the yellow/red zones can be guarded.
4673   JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4674                                                 vm_page_size()));
4675 
4676   Solaris::libthread_init();
4677 
4678   if (UseNUMA) {
4679     if (!Solaris::liblgrp_init()) {
4680       UseNUMA = false;
4681     } else {
4682       size_t lgrp_limit = os::numa_get_groups_num();
4683       int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit, mtInternal);
4684       size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
4685       FREE_C_HEAP_ARRAY(int, lgrp_ids);
4686       if (lgrp_num < 2) {
4687         // There's only one locality group, disable NUMA.
4688         UseNUMA = false;
4689       }
4690     }
4691     if (!UseNUMA && ForceNUMA) {
4692       UseNUMA = true;
4693     }
4694   }
4695 
4696   Solaris::signal_sets_init();
4697   Solaris::init_signal_mem();
4698   Solaris::install_signal_handlers();
4699 
4700   if (libjsigversion < JSIG_VERSION_1_4_1) {
4701     Maxlibjsigsigs = OLDMAXSIGNUM;
4702   }
4703 
4704   // initialize synchronization primitives to use either thread or
4705   // lwp synchronization (controlled by UseLWPSynchronization)
4706   Solaris::synchronization_init();
4707 
4708   if (MaxFDLimit) {
4709     // set the number of file descriptors to max. print out error
4710     // if getrlimit/setrlimit fails but continue regardless.
4711     struct rlimit nbr_files;
4712     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4713     if (status != 0) {
4714       if (PrintMiscellaneous && (Verbose || WizardMode)) {
4715         perror("os::init_2 getrlimit failed");
4716       }
4717     } else {
4718       nbr_files.rlim_cur = nbr_files.rlim_max;
4719       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4720       if (status != 0) {
4721         if (PrintMiscellaneous && (Verbose || WizardMode)) {
4722           perror("os::init_2 setrlimit failed");
4723         }
4724       }
4725     }
4726   }
4727 
4728   // Calculate theoretical max. size of Threads to guard gainst
4729   // artifical out-of-memory situations, where all available address-
4730   // space has been reserved by thread stacks. Default stack size is 1Mb.
4731   size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
4732     JavaThread::stack_size_at_create() : (1*K*K);
4733   assert(pre_thread_stack_size != 0, "Must have a stack");
4734   // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
4735   // we should start doing Virtual Memory banging. Currently when the threads will
4736   // have used all but 200Mb of space.
4737   size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
4738   Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
4739 
4740   // at-exit methods are called in the reverse order of their registration.
4741   // In Solaris 7 and earlier, atexit functions are called on return from
4742   // main or as a result of a call to exit(3C). There can be only 32 of
4743   // these functions registered and atexit() does not set errno. In Solaris
4744   // 8 and later, there is no limit to the number of functions registered
4745   // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
4746   // functions are called upon dlclose(3DL) in addition to return from main
4747   // and exit(3C).
4748 
4749   if (PerfAllowAtExitRegistration) {
4750     // only register atexit functions if PerfAllowAtExitRegistration is set.
4751     // atexit functions can be delayed until process exit time, which
4752     // can be problematic for embedded VM situations. Embedded VMs should
4753     // call DestroyJavaVM() to assure that VM resources are released.
4754 
4755     // note: perfMemory_exit_helper atexit function may be removed in
4756     // the future if the appropriate cleanup code can be added to the
4757     // VM_Exit VMOperation's doit method.
4758     if (atexit(perfMemory_exit_helper) != 0) {
4759       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4760     }
4761   }
4762 
4763   // Init pset_loadavg function pointer
4764   init_pset_getloadavg_ptr();
4765 
4766   return JNI_OK;
4767 }
4768 
4769 // Mark the polling page as unreadable
4770 void os::make_polling_page_unreadable(void) {
4771   if (mprotect((char *)_polling_page, page_size, PROT_NONE) != 0) {
4772     fatal("Could not disable polling page");
4773   }
4774 }
4775 
4776 // Mark the polling page as readable
4777 void os::make_polling_page_readable(void) {
4778   if (mprotect((char *)_polling_page, page_size, PROT_READ) != 0) {
4779     fatal("Could not enable polling page");
4780   }
4781 }
4782 
4783 // OS interface.
4784 
4785 bool os::check_heap(bool force) { return true; }
4786 
4787 // Is a (classpath) directory empty?
4788 bool os::dir_is_empty(const char* path) {
4789   DIR *dir = NULL;
4790   struct dirent *ptr;
4791 
4792   dir = opendir(path);
4793   if (dir == NULL) return true;
4794 
4795   // Scan the directory
4796   bool result = true;
4797   char buf[sizeof(struct dirent) + MAX_PATH];
4798   struct dirent *dbuf = (struct dirent *) buf;
4799   while (result && (ptr = readdir(dir, dbuf)) != NULL) {
4800     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4801       result = false;
4802     }
4803   }
4804   closedir(dir);
4805   return result;
4806 }
4807 
4808 // This code originates from JDK's sysOpen and open64_w
4809 // from src/solaris/hpi/src/system_md.c
4810 
4811 int os::open(const char *path, int oflag, int mode) {
4812   if (strlen(path) > MAX_PATH - 1) {
4813     errno = ENAMETOOLONG;
4814     return -1;
4815   }
4816   int fd;
4817 
4818   fd = ::open64(path, oflag, mode);
4819   if (fd == -1) return -1;
4820 
4821   // If the open succeeded, the file might still be a directory
4822   {
4823     struct stat64 buf64;
4824     int ret = ::fstat64(fd, &buf64);
4825     int st_mode = buf64.st_mode;
4826 
4827     if (ret != -1) {
4828       if ((st_mode & S_IFMT) == S_IFDIR) {
4829         errno = EISDIR;
4830         ::close(fd);
4831         return -1;
4832       }
4833     } else {
4834       ::close(fd);
4835       return -1;
4836     }
4837   }
4838 
4839   // 32-bit Solaris systems suffer from:
4840   //
4841   // - an historical default soft limit of 256 per-process file
4842   //   descriptors that is too low for many Java programs.
4843   //
4844   // - a design flaw where file descriptors created using stdio
4845   //   fopen must be less than 256, _even_ when the first limit above
4846   //   has been raised.  This can cause calls to fopen (but not calls to
4847   //   open, for example) to fail mysteriously, perhaps in 3rd party
4848   //   native code (although the JDK itself uses fopen).  One can hardly
4849   //   criticize them for using this most standard of all functions.
4850   //
4851   // We attempt to make everything work anyways by:
4852   //
4853   // - raising the soft limit on per-process file descriptors beyond
4854   //   256
4855   //
4856   // - As of Solaris 10u4, we can request that Solaris raise the 256
4857   //   stdio fopen limit by calling function enable_extended_FILE_stdio.
4858   //   This is done in init_2 and recorded in enabled_extended_FILE_stdio
4859   //
4860   // - If we are stuck on an old (pre 10u4) Solaris system, we can
4861   //   workaround the bug by remapping non-stdio file descriptors below
4862   //   256 to ones beyond 256, which is done below.
4863   //
4864   // See:
4865   // 1085341: 32-bit stdio routines should support file descriptors >255
4866   // 6533291: Work around 32-bit Solaris stdio limit of 256 open files
4867   // 6431278: Netbeans crash on 32 bit Solaris: need to call
4868   //          enable_extended_FILE_stdio() in VM initialisation
4869   // Giri Mandalika's blog
4870   // http://technopark02.blogspot.com/2005_05_01_archive.html
4871   //
4872 #ifndef  _LP64
4873   if ((!enabled_extended_FILE_stdio) && fd < 256) {
4874     int newfd = ::fcntl(fd, F_DUPFD, 256);
4875     if (newfd != -1) {
4876       ::close(fd);
4877       fd = newfd;
4878     }
4879   }
4880 #endif // 32-bit Solaris
4881 
4882   // All file descriptors that are opened in the JVM and not
4883   // specifically destined for a subprocess should have the
4884   // close-on-exec flag set.  If we don't set it, then careless 3rd
4885   // party native code might fork and exec without closing all
4886   // appropriate file descriptors (e.g. as we do in closeDescriptors in
4887   // UNIXProcess.c), and this in turn might:
4888   //
4889   // - cause end-of-file to fail to be detected on some file
4890   //   descriptors, resulting in mysterious hangs, or
4891   //
4892   // - might cause an fopen in the subprocess to fail on a system
4893   //   suffering from bug 1085341.
4894   //
4895   // (Yes, the default setting of the close-on-exec flag is a Unix
4896   // design flaw)
4897   //
4898   // See:
4899   // 1085341: 32-bit stdio routines should support file descriptors >255
4900   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4901   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4902   //
4903 #ifdef FD_CLOEXEC
4904   {
4905     int flags = ::fcntl(fd, F_GETFD);
4906     if (flags != -1) {
4907       ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4908     }
4909   }
4910 #endif
4911 
4912   return fd;
4913 }
4914 
4915 // create binary file, rewriting existing file if required
4916 int os::create_binary_file(const char* path, bool rewrite_existing) {
4917   int oflags = O_WRONLY | O_CREAT;
4918   if (!rewrite_existing) {
4919     oflags |= O_EXCL;
4920   }
4921   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4922 }
4923 
4924 // return current position of file pointer
4925 jlong os::current_file_offset(int fd) {
4926   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4927 }
4928 
4929 // move file pointer to the specified offset
4930 jlong os::seek_to_file_offset(int fd, jlong offset) {
4931   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4932 }
4933 
4934 jlong os::lseek(int fd, jlong offset, int whence) {
4935   return (jlong) ::lseek64(fd, offset, whence);
4936 }
4937 
4938 char * os::native_path(char *path) {
4939   return path;
4940 }
4941 
4942 int os::ftruncate(int fd, jlong length) {
4943   return ::ftruncate64(fd, length);
4944 }
4945 
4946 int os::fsync(int fd)  {
4947   RESTARTABLE_RETURN_INT(::fsync(fd));
4948 }
4949 
4950 int os::available(int fd, jlong *bytes) {
4951   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
4952          "Assumed _thread_in_native");
4953   jlong cur, end;
4954   int mode;
4955   struct stat64 buf64;
4956 
4957   if (::fstat64(fd, &buf64) >= 0) {
4958     mode = buf64.st_mode;
4959     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4960       int n,ioctl_return;
4961 
4962       RESTARTABLE(::ioctl(fd, FIONREAD, &n), ioctl_return);
4963       if (ioctl_return>= 0) {
4964         *bytes = n;
4965         return 1;
4966       }
4967     }
4968   }
4969   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4970     return 0;
4971   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4972     return 0;
4973   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4974     return 0;
4975   }
4976   *bytes = end - cur;
4977   return 1;
4978 }
4979 
4980 // Map a block of memory.
4981 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4982                         char *addr, size_t bytes, bool read_only,
4983                         bool allow_exec) {
4984   int prot;
4985   int flags;
4986 
4987   if (read_only) {
4988     prot = PROT_READ;
4989     flags = MAP_SHARED;
4990   } else {
4991     prot = PROT_READ | PROT_WRITE;
4992     flags = MAP_PRIVATE;
4993   }
4994 
4995   if (allow_exec) {
4996     prot |= PROT_EXEC;
4997   }
4998 
4999   if (addr != NULL) {
5000     flags |= MAP_FIXED;
5001   }
5002 
5003   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5004                                      fd, file_offset);
5005   if (mapped_address == MAP_FAILED) {
5006     return NULL;
5007   }
5008   return mapped_address;
5009 }
5010 
5011 
5012 // Remap a block of memory.
5013 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5014                           char *addr, size_t bytes, bool read_only,
5015                           bool allow_exec) {
5016   // same as map_memory() on this OS
5017   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5018                         allow_exec);
5019 }
5020 
5021 
5022 // Unmap a block of memory.
5023 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5024   return munmap(addr, bytes) == 0;
5025 }
5026 
5027 void os::pause() {
5028   char filename[MAX_PATH];
5029   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5030     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5031   } else {
5032     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5033   }
5034 
5035   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5036   if (fd != -1) {
5037     struct stat buf;
5038     ::close(fd);
5039     while (::stat(filename, &buf) == 0) {
5040       (void)::poll(NULL, 0, 100);
5041     }
5042   } else {
5043     jio_fprintf(stderr,
5044                 "Could not open pause file '%s', continuing immediately.\n", filename);
5045   }
5046 }
5047 
5048 #ifndef PRODUCT
5049 #ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5050 // Turn this on if you need to trace synch operations.
5051 // Set RECORD_SYNCH_LIMIT to a large-enough value,
5052 // and call record_synch_enable and record_synch_disable
5053 // around the computation of interest.
5054 
5055 void record_synch(char* name, bool returning);  // defined below
5056 
5057 class RecordSynch {
5058   char* _name;
5059  public:
5060   RecordSynch(char* name) :_name(name) { record_synch(_name, false); }
5061   ~RecordSynch()                       { record_synch(_name, true); }
5062 };
5063 
5064 #define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
5065 extern "C" ret name params {                                    \
5066   typedef ret name##_t params;                                  \
5067   static name##_t* implem = NULL;                               \
5068   static int callcount = 0;                                     \
5069   if (implem == NULL) {                                         \
5070     implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
5071     if (implem == NULL)  fatal(dlerror());                      \
5072   }                                                             \
5073   ++callcount;                                                  \
5074   RecordSynch _rs(#name);                                       \
5075   inner;                                                        \
5076   return implem args;                                           \
5077 }
5078 // in dbx, examine callcounts this way:
5079 // for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
5080 
5081 #define CHECK_POINTER_OK(p) \
5082   (!Universe::is_fully_initialized() || !Universe::is_reserved_heap((oop)(p)))
5083 #define CHECK_MU \
5084   if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
5085 #define CHECK_CV \
5086   if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
5087 #define CHECK_P(p) \
5088   if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
5089 
5090 #define CHECK_MUTEX(mutex_op) \
5091   CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
5092 
5093 CHECK_MUTEX(   mutex_lock)
5094 CHECK_MUTEX(  _mutex_lock)
5095 CHECK_MUTEX( mutex_unlock)
5096 CHECK_MUTEX(_mutex_unlock)
5097 CHECK_MUTEX( mutex_trylock)
5098 CHECK_MUTEX(_mutex_trylock)
5099 
5100 #define CHECK_COND(cond_op) \
5101   CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU; CHECK_CV);
5102 
5103 CHECK_COND( cond_wait);
5104 CHECK_COND(_cond_wait);
5105 CHECK_COND(_cond_wait_cancel);
5106 
5107 #define CHECK_COND2(cond_op) \
5108   CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU; CHECK_CV);
5109 
5110 CHECK_COND2( cond_timedwait);
5111 CHECK_COND2(_cond_timedwait);
5112 CHECK_COND2(_cond_timedwait_cancel);
5113 
5114 // do the _lwp_* versions too
5115 #define mutex_t lwp_mutex_t
5116 #define cond_t  lwp_cond_t
5117 CHECK_MUTEX(  _lwp_mutex_lock)
5118 CHECK_MUTEX(  _lwp_mutex_unlock)
5119 CHECK_MUTEX(  _lwp_mutex_trylock)
5120 CHECK_MUTEX( __lwp_mutex_lock)
5121 CHECK_MUTEX( __lwp_mutex_unlock)
5122 CHECK_MUTEX( __lwp_mutex_trylock)
5123 CHECK_MUTEX(___lwp_mutex_lock)
5124 CHECK_MUTEX(___lwp_mutex_unlock)
5125 
5126 CHECK_COND(  _lwp_cond_wait);
5127 CHECK_COND( __lwp_cond_wait);
5128 CHECK_COND(___lwp_cond_wait);
5129 
5130 CHECK_COND2(  _lwp_cond_timedwait);
5131 CHECK_COND2( __lwp_cond_timedwait);
5132 #undef mutex_t
5133 #undef cond_t
5134 
5135 CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5136 CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
5137 CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5138 CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
5139 CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5140 CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
5141 CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5142 CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
5143 
5144 
5145 // recording machinery:
5146 
5147 enum { RECORD_SYNCH_LIMIT = 200 };
5148 char* record_synch_name[RECORD_SYNCH_LIMIT];
5149 void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
5150 bool record_synch_returning[RECORD_SYNCH_LIMIT];
5151 thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
5152 int record_synch_count = 0;
5153 bool record_synch_enabled = false;
5154 
5155 // in dbx, examine recorded data this way:
5156 // for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
5157 
5158 void record_synch(char* name, bool returning) {
5159   if (record_synch_enabled) {
5160     if (record_synch_count < RECORD_SYNCH_LIMIT) {
5161       record_synch_name[record_synch_count] = name;
5162       record_synch_returning[record_synch_count] = returning;
5163       record_synch_thread[record_synch_count] = thr_self();
5164       record_synch_arg0ptr[record_synch_count] = &name;
5165       record_synch_count++;
5166     }
5167     // put more checking code here:
5168     // ...
5169   }
5170 }
5171 
5172 void record_synch_enable() {
5173   // start collecting trace data, if not already doing so
5174   if (!record_synch_enabled)  record_synch_count = 0;
5175   record_synch_enabled = true;
5176 }
5177 
5178 void record_synch_disable() {
5179   // stop collecting trace data
5180   record_synch_enabled = false;
5181 }
5182 
5183 #endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
5184 #endif // PRODUCT
5185 
5186 const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5187 const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
5188                                (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
5189 
5190 
5191 // JVMTI & JVM monitoring and management support
5192 // The thread_cpu_time() and current_thread_cpu_time() are only
5193 // supported if is_thread_cpu_time_supported() returns true.
5194 // They are not supported on Solaris T1.
5195 
5196 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5197 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5198 // of a thread.
5199 //
5200 // current_thread_cpu_time() and thread_cpu_time(Thread *)
5201 // returns the fast estimate available on the platform.
5202 
5203 // hrtime_t gethrvtime() return value includes
5204 // user time but does not include system time
5205 jlong os::current_thread_cpu_time() {
5206   return (jlong) gethrvtime();
5207 }
5208 
5209 jlong os::thread_cpu_time(Thread *thread) {
5210   // return user level CPU time only to be consistent with
5211   // what current_thread_cpu_time returns.
5212   // thread_cpu_time_info() must be changed if this changes
5213   return os::thread_cpu_time(thread, false /* user time only */);
5214 }
5215 
5216 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5217   if (user_sys_cpu_time) {
5218     return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
5219   } else {
5220     return os::current_thread_cpu_time();
5221   }
5222 }
5223 
5224 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5225   char proc_name[64];
5226   int count;
5227   prusage_t prusage;
5228   jlong lwp_time;
5229   int fd;
5230 
5231   sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
5232           getpid(),
5233           thread->osthread()->lwp_id());
5234   fd = ::open(proc_name, O_RDONLY);
5235   if (fd == -1) return -1;
5236 
5237   do {
5238     count = ::pread(fd,
5239                     (void *)&prusage.pr_utime,
5240                     thr_time_size,
5241                     thr_time_off);
5242   } while (count < 0 && errno == EINTR);
5243   ::close(fd);
5244   if (count < 0) return -1;
5245 
5246   if (user_sys_cpu_time) {
5247     // user + system CPU time
5248     lwp_time = (((jlong)prusage.pr_stime.tv_sec +
5249                  (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
5250                  (jlong)prusage.pr_stime.tv_nsec +
5251                  (jlong)prusage.pr_utime.tv_nsec;
5252   } else {
5253     // user level CPU time only
5254     lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
5255                 (jlong)prusage.pr_utime.tv_nsec;
5256   }
5257 
5258   return (lwp_time);
5259 }
5260 
5261 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5262   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5263   info_ptr->may_skip_backward = false;    // elapsed time not wall time
5264   info_ptr->may_skip_forward = false;     // elapsed time not wall time
5265   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5266 }
5267 
5268 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5269   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
5270   info_ptr->may_skip_backward = false;    // elapsed time not wall time
5271   info_ptr->may_skip_forward = false;     // elapsed time not wall time
5272   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
5273 }
5274 
5275 bool os::is_thread_cpu_time_supported() {
5276   return true;
5277 }
5278 
5279 // System loadavg support.  Returns -1 if load average cannot be obtained.
5280 // Return the load average for our processor set if the primitive exists
5281 // (Solaris 9 and later).  Otherwise just return system wide loadavg.
5282 int os::loadavg(double loadavg[], int nelem) {
5283   if (pset_getloadavg_ptr != NULL) {
5284     return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
5285   } else {
5286     return ::getloadavg(loadavg, nelem);
5287   }
5288 }
5289 
5290 //---------------------------------------------------------------------------------
5291 
5292 bool os::find(address addr, outputStream* st) {
5293   Dl_info dlinfo;
5294   memset(&dlinfo, 0, sizeof(dlinfo));
5295   if (dladdr(addr, &dlinfo) != 0) {
5296     st->print(PTR_FORMAT ": ", addr);
5297     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5298       st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
5299     } else if (dlinfo.dli_fbase != NULL) {
5300       st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
5301     } else {
5302       st->print("<absolute address>");
5303     }
5304     if (dlinfo.dli_fname != NULL) {
5305       st->print(" in %s", dlinfo.dli_fname);
5306     }
5307     if (dlinfo.dli_fbase != NULL) {
5308       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5309     }
5310     st->cr();
5311 
5312     if (Verbose) {
5313       // decode some bytes around the PC
5314       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5315       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5316       address       lowest = (address) dlinfo.dli_sname;
5317       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5318       if (begin < lowest)  begin = lowest;
5319       Dl_info dlinfo2;
5320       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5321           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5322         end = (address) dlinfo2.dli_saddr;
5323       }
5324       Disassembler::decode(begin, end, st);
5325     }
5326     return true;
5327   }
5328   return false;
5329 }
5330 
5331 // Following function has been added to support HotSparc's libjvm.so running
5332 // under Solaris production JDK 1.2.2 / 1.3.0.  These came from
5333 // src/solaris/hpi/native_threads in the EVM codebase.
5334 //
5335 // NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
5336 // libraries and should thus be removed. We will leave it behind for a while
5337 // until we no longer want to able to run on top of 1.3.0 Solaris production
5338 // JDK. See 4341971.
5339 
5340 #define STACK_SLACK 0x800
5341 
5342 extern "C" {
5343   intptr_t sysThreadAvailableStackWithSlack() {
5344     stack_t st;
5345     intptr_t retval, stack_top;
5346     retval = thr_stksegment(&st);
5347     assert(retval == 0, "incorrect return value from thr_stksegment");
5348     assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
5349     assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
5350     stack_top=(intptr_t)st.ss_sp-st.ss_size;
5351     return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
5352   }
5353 }
5354 
5355 // ObjectMonitor park-unpark infrastructure ...
5356 //
5357 // We implement Solaris and Linux PlatformEvents with the
5358 // obvious condvar-mutex-flag triple.
5359 // Another alternative that works quite well is pipes:
5360 // Each PlatformEvent consists of a pipe-pair.
5361 // The thread associated with the PlatformEvent
5362 // calls park(), which reads from the input end of the pipe.
5363 // Unpark() writes into the other end of the pipe.
5364 // The write-side of the pipe must be set NDELAY.
5365 // Unfortunately pipes consume a large # of handles.
5366 // Native solaris lwp_park() and lwp_unpark() work nicely, too.
5367 // Using pipes for the 1st few threads might be workable, however.
5368 //
5369 // park() is permitted to return spuriously.
5370 // Callers of park() should wrap the call to park() in
5371 // an appropriate loop.  A litmus test for the correct
5372 // usage of park is the following: if park() were modified
5373 // to immediately return 0 your code should still work,
5374 // albeit degenerating to a spin loop.
5375 //
5376 // In a sense, park()-unpark() just provides more polite spinning
5377 // and polling with the key difference over naive spinning being
5378 // that a parked thread needs to be explicitly unparked() in order
5379 // to wake up and to poll the underlying condition.
5380 //
5381 // Assumption:
5382 //    Only one parker can exist on an event, which is why we allocate
5383 //    them per-thread. Multiple unparkers can coexist.
5384 //
5385 // _Event transitions in park()
5386 //   -1 => -1 : illegal
5387 //    1 =>  0 : pass - return immediately
5388 //    0 => -1 : block; then set _Event to 0 before returning
5389 //
5390 // _Event transitions in unpark()
5391 //    0 => 1 : just return
5392 //    1 => 1 : just return
5393 //   -1 => either 0 or 1; must signal target thread
5394 //         That is, we can safely transition _Event from -1 to either
5395 //         0 or 1.
5396 //
5397 // _Event serves as a restricted-range semaphore.
5398 //   -1 : thread is blocked, i.e. there is a waiter
5399 //    0 : neutral: thread is running or ready,
5400 //        could have been signaled after a wait started
5401 //    1 : signaled - thread is running or ready
5402 //
5403 // Another possible encoding of _Event would be with
5404 // explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
5405 //
5406 // TODO-FIXME: add DTRACE probes for:
5407 // 1.   Tx parks
5408 // 2.   Ty unparks Tx
5409 // 3.   Tx resumes from park
5410 
5411 
5412 // value determined through experimentation
5413 #define ROUNDINGFIX 11
5414 
5415 // utility to compute the abstime argument to timedwait.
5416 // TODO-FIXME: switch from compute_abstime() to unpackTime().
5417 
5418 static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
5419   // millis is the relative timeout time
5420   // abstime will be the absolute timeout time
5421   if (millis < 0)  millis = 0;
5422   struct timeval now;
5423   int status = gettimeofday(&now, NULL);
5424   assert(status == 0, "gettimeofday");
5425   jlong seconds = millis / 1000;
5426   jlong max_wait_period;
5427 
5428   if (UseLWPSynchronization) {
5429     // forward port of fix for 4275818 (not sleeping long enough)
5430     // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
5431     // _lwp_cond_timedwait() used a round_down algorithm rather
5432     // than a round_up. For millis less than our roundfactor
5433     // it rounded down to 0 which doesn't meet the spec.
5434     // For millis > roundfactor we may return a bit sooner, but
5435     // since we can not accurately identify the patch level and
5436     // this has already been fixed in Solaris 9 and 8 we will
5437     // leave it alone rather than always rounding down.
5438 
5439     if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
5440     // It appears that when we go directly through Solaris _lwp_cond_timedwait()
5441     // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
5442     max_wait_period = 21000000;
5443   } else {
5444     max_wait_period = 50000000;
5445   }
5446   millis %= 1000;
5447   if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
5448     seconds = max_wait_period;
5449   }
5450   abstime->tv_sec = now.tv_sec  + seconds;
5451   long       usec = now.tv_usec + millis * 1000;
5452   if (usec >= 1000000) {
5453     abstime->tv_sec += 1;
5454     usec -= 1000000;
5455   }
5456   abstime->tv_nsec = usec * 1000;
5457   return abstime;
5458 }
5459 
5460 void os::PlatformEvent::park() {           // AKA: down()
5461   // Transitions for _Event:
5462   //   -1 => -1 : illegal
5463   //    1 =>  0 : pass - return immediately
5464   //    0 => -1 : block; then set _Event to 0 before returning
5465 
5466   // Invariant: Only the thread associated with the Event/PlatformEvent
5467   // may call park().
5468   assert(_nParked == 0, "invariant");
5469 
5470   int v;
5471   for (;;) {
5472     v = _Event;
5473     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5474   }
5475   guarantee(v >= 0, "invariant");
5476   if (v == 0) {
5477     // Do this the hard way by blocking ...
5478     // See http://monaco.sfbay/detail.jsf?cr=5094058.
5479     // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
5480     // Only for SPARC >= V8PlusA
5481 #if defined(__sparc) && defined(COMPILER2)
5482     if (ClearFPUAtPark) { _mark_fpu_nosave(); }
5483 #endif
5484     int status = os::Solaris::mutex_lock(_mutex);
5485     assert_status(status == 0, status, "mutex_lock");
5486     guarantee(_nParked == 0, "invariant");
5487     ++_nParked;
5488     while (_Event < 0) {
5489       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5490       // Treat this the same as if the wait was interrupted
5491       // With usr/lib/lwp going to kernel, always handle ETIME
5492       status = os::Solaris::cond_wait(_cond, _mutex);
5493       if (status == ETIME) status = EINTR;
5494       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5495     }
5496     --_nParked;
5497     _Event = 0;
5498     status = os::Solaris::mutex_unlock(_mutex);
5499     assert_status(status == 0, status, "mutex_unlock");
5500     // Paranoia to ensure our locked and lock-free paths interact
5501     // correctly with each other.
5502     OrderAccess::fence();
5503   }
5504 }
5505 
5506 int os::PlatformEvent::park(jlong millis) {
5507   // Transitions for _Event:
5508   //   -1 => -1 : illegal
5509   //    1 =>  0 : pass - return immediately
5510   //    0 => -1 : block; then set _Event to 0 before returning
5511 
5512   guarantee(_nParked == 0, "invariant");
5513   int v;
5514   for (;;) {
5515     v = _Event;
5516     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5517   }
5518   guarantee(v >= 0, "invariant");
5519   if (v != 0) return OS_OK;
5520 
5521   int ret = OS_TIMEOUT;
5522   timestruc_t abst;
5523   compute_abstime(&abst, millis);
5524 
5525   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5526   // For Solaris SPARC set fprs.FEF=0 prior to parking.
5527   // Only for SPARC >= V8PlusA
5528 #if defined(__sparc) && defined(COMPILER2)
5529   if (ClearFPUAtPark) { _mark_fpu_nosave(); }
5530 #endif
5531   int status = os::Solaris::mutex_lock(_mutex);
5532   assert_status(status == 0, status, "mutex_lock");
5533   guarantee(_nParked == 0, "invariant");
5534   ++_nParked;
5535   while (_Event < 0) {
5536     int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
5537     assert_status(status == 0 || status == EINTR ||
5538                   status == ETIME || status == ETIMEDOUT,
5539                   status, "cond_timedwait");
5540     if (!FilterSpuriousWakeups) break;                // previous semantics
5541     if (status == ETIME || status == ETIMEDOUT) break;
5542     // We consume and ignore EINTR and spurious wakeups.
5543   }
5544   --_nParked;
5545   if (_Event >= 0) ret = OS_OK;
5546   _Event = 0;
5547   status = os::Solaris::mutex_unlock(_mutex);
5548   assert_status(status == 0, status, "mutex_unlock");
5549   // Paranoia to ensure our locked and lock-free paths interact
5550   // correctly with each other.
5551   OrderAccess::fence();
5552   return ret;
5553 }
5554 
5555 void os::PlatformEvent::unpark() {
5556   // Transitions for _Event:
5557   //    0 => 1 : just return
5558   //    1 => 1 : just return
5559   //   -1 => either 0 or 1; must signal target thread
5560   //         That is, we can safely transition _Event from -1 to either
5561   //         0 or 1.
5562   // See also: "Semaphores in Plan 9" by Mullender & Cox
5563   //
5564   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5565   // that it will take two back-to-back park() calls for the owning
5566   // thread to block. This has the benefit of forcing a spurious return
5567   // from the first park() call after an unpark() call which will help
5568   // shake out uses of park() and unpark() without condition variables.
5569 
5570   if (Atomic::xchg(1, &_Event) >= 0) return;
5571 
5572   // If the thread associated with the event was parked, wake it.
5573   // Wait for the thread assoc with the PlatformEvent to vacate.
5574   int status = os::Solaris::mutex_lock(_mutex);
5575   assert_status(status == 0, status, "mutex_lock");
5576   int AnyWaiters = _nParked;
5577   status = os::Solaris::mutex_unlock(_mutex);
5578   assert_status(status == 0, status, "mutex_unlock");
5579   guarantee(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5580   if (AnyWaiters != 0) {
5581     // Note that we signal() *after* dropping the lock for "immortal" Events.
5582     // This is safe and avoids a common class of  futile wakeups.  In rare
5583     // circumstances this can cause a thread to return prematurely from
5584     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5585     // will simply re-test the condition and re-park itself.
5586     // This provides particular benefit if the underlying platform does not
5587     // provide wait morphing.
5588     status = os::Solaris::cond_signal(_cond);
5589     assert_status(status == 0, status, "cond_signal");
5590   }
5591 }
5592 
5593 // JSR166
5594 // -------------------------------------------------------
5595 
5596 // The solaris and linux implementations of park/unpark are fairly
5597 // conservative for now, but can be improved. They currently use a
5598 // mutex/condvar pair, plus _counter.
5599 // Park decrements _counter if > 0, else does a condvar wait.  Unpark
5600 // sets count to 1 and signals condvar.  Only one thread ever waits
5601 // on the condvar. Contention seen when trying to park implies that someone
5602 // is unparking you, so don't wait. And spurious returns are fine, so there
5603 // is no need to track notifications.
5604 
5605 #define MAX_SECS 100000000
5606 
5607 // This code is common to linux and solaris and will be moved to a
5608 // common place in dolphin.
5609 //
5610 // The passed in time value is either a relative time in nanoseconds
5611 // or an absolute time in milliseconds. Either way it has to be unpacked
5612 // into suitable seconds and nanoseconds components and stored in the
5613 // given timespec structure.
5614 // Given time is a 64-bit value and the time_t used in the timespec is only
5615 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5616 // overflow if times way in the future are given. Further on Solaris versions
5617 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5618 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5619 // As it will be 28 years before "now + 100000000" will overflow we can
5620 // ignore overflow and just impose a hard-limit on seconds using the value
5621 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5622 // years from "now".
5623 //
5624 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5625   assert(time > 0, "convertTime");
5626 
5627   struct timeval now;
5628   int status = gettimeofday(&now, NULL);
5629   assert(status == 0, "gettimeofday");
5630 
5631   time_t max_secs = now.tv_sec + MAX_SECS;
5632 
5633   if (isAbsolute) {
5634     jlong secs = time / 1000;
5635     if (secs > max_secs) {
5636       absTime->tv_sec = max_secs;
5637     } else {
5638       absTime->tv_sec = secs;
5639     }
5640     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5641   } else {
5642     jlong secs = time / NANOSECS_PER_SEC;
5643     if (secs >= MAX_SECS) {
5644       absTime->tv_sec = max_secs;
5645       absTime->tv_nsec = 0;
5646     } else {
5647       absTime->tv_sec = now.tv_sec + secs;
5648       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5649       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5650         absTime->tv_nsec -= NANOSECS_PER_SEC;
5651         ++absTime->tv_sec; // note: this must be <= max_secs
5652       }
5653     }
5654   }
5655   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5656   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5657   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5658   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5659 }
5660 
5661 void Parker::park(bool isAbsolute, jlong time) {
5662   // Ideally we'd do something useful while spinning, such
5663   // as calling unpackTime().
5664 
5665   // Optional fast-path check:
5666   // Return immediately if a permit is available.
5667   // We depend on Atomic::xchg() having full barrier semantics
5668   // since we are doing a lock-free update to _counter.
5669   if (Atomic::xchg(0, &_counter) > 0) return;
5670 
5671   // Optional fast-exit: Check interrupt before trying to wait
5672   Thread* thread = Thread::current();
5673   assert(thread->is_Java_thread(), "Must be JavaThread");
5674   JavaThread *jt = (JavaThread *)thread;
5675   if (Thread::is_interrupted(thread, false)) {
5676     return;
5677   }
5678 
5679   // First, demultiplex/decode time arguments
5680   timespec absTime;
5681   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5682     return;
5683   }
5684   if (time > 0) {
5685     // Warning: this code might be exposed to the old Solaris time
5686     // round-down bugs.  Grep "roundingFix" for details.
5687     unpackTime(&absTime, isAbsolute, time);
5688   }
5689 
5690   // Enter safepoint region
5691   // Beware of deadlocks such as 6317397.
5692   // The per-thread Parker:: _mutex is a classic leaf-lock.
5693   // In particular a thread must never block on the Threads_lock while
5694   // holding the Parker:: mutex.  If safepoints are pending both the
5695   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5696   ThreadBlockInVM tbivm(jt);
5697 
5698   // Don't wait if cannot get lock since interference arises from
5699   // unblocking.  Also. check interrupt before trying wait
5700   if (Thread::is_interrupted(thread, false) ||
5701       os::Solaris::mutex_trylock(_mutex) != 0) {
5702     return;
5703   }
5704 
5705   int status;
5706 
5707   if (_counter > 0)  { // no wait needed
5708     _counter = 0;
5709     status = os::Solaris::mutex_unlock(_mutex);
5710     assert(status == 0, "invariant");
5711     // Paranoia to ensure our locked and lock-free paths interact
5712     // correctly with each other and Java-level accesses.
5713     OrderAccess::fence();
5714     return;
5715   }
5716 
5717 #ifdef ASSERT
5718   // Don't catch signals while blocked; let the running threads have the signals.
5719   // (This allows a debugger to break into the running thread.)
5720   sigset_t oldsigs;
5721   sigset_t* allowdebug_blocked = os::Solaris::allowdebug_blocked_signals();
5722   thr_sigsetmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5723 #endif
5724 
5725   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5726   jt->set_suspend_equivalent();
5727   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5728 
5729   // Do this the hard way by blocking ...
5730   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5731   // TODO-FIXME: for Solaris SPARC set fprs.FEF=0 prior to parking.
5732   // Only for SPARC >= V8PlusA
5733 #if defined(__sparc) && defined(COMPILER2)
5734   if (ClearFPUAtPark) { _mark_fpu_nosave(); }
5735 #endif
5736 
5737   if (time == 0) {
5738     status = os::Solaris::cond_wait(_cond, _mutex);
5739   } else {
5740     status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
5741   }
5742   // Note that an untimed cond_wait() can sometimes return ETIME on older
5743   // versions of the Solaris.
5744   assert_status(status == 0 || status == EINTR ||
5745                 status == ETIME || status == ETIMEDOUT,
5746                 status, "cond_timedwait");
5747 
5748 #ifdef ASSERT
5749   thr_sigsetmask(SIG_SETMASK, &oldsigs, NULL);
5750 #endif
5751   _counter = 0;
5752   status = os::Solaris::mutex_unlock(_mutex);
5753   assert_status(status == 0, status, "mutex_unlock");
5754   // Paranoia to ensure our locked and lock-free paths interact
5755   // correctly with each other and Java-level accesses.
5756   OrderAccess::fence();
5757 
5758   // If externally suspended while waiting, re-suspend
5759   if (jt->handle_special_suspend_equivalent_condition()) {
5760     jt->java_suspend_self();
5761   }
5762 }
5763 
5764 void Parker::unpark() {
5765   int status = os::Solaris::mutex_lock(_mutex);
5766   assert(status == 0, "invariant");
5767   const int s = _counter;
5768   _counter = 1;
5769   status = os::Solaris::mutex_unlock(_mutex);
5770   assert(status == 0, "invariant");
5771 
5772   if (s < 1) {
5773     status = os::Solaris::cond_signal(_cond);
5774     assert(status == 0, "invariant");
5775   }
5776 }
5777 
5778 extern char** environ;
5779 
5780 // Run the specified command in a separate process. Return its exit value,
5781 // or -1 on failure (e.g. can't fork a new process).
5782 // Unlike system(), this function can be called from signal handler. It
5783 // doesn't block SIGINT et al.
5784 int os::fork_and_exec(char* cmd) {
5785   char * argv[4];
5786   argv[0] = (char *)"sh";
5787   argv[1] = (char *)"-c";
5788   argv[2] = cmd;
5789   argv[3] = NULL;
5790 
5791   // fork is async-safe, fork1 is not so can't use in signal handler
5792   pid_t pid;
5793   Thread* t = ThreadLocalStorage::get_thread_slow();
5794   if (t != NULL && t->is_inside_signal_handler()) {
5795     pid = fork();
5796   } else {
5797     pid = fork1();
5798   }
5799 
5800   if (pid < 0) {
5801     // fork failed
5802     warning("fork failed: %s", strerror(errno));
5803     return -1;
5804 
5805   } else if (pid == 0) {
5806     // child process
5807 
5808     // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
5809     execve("/usr/bin/sh", argv, environ);
5810 
5811     // execve failed
5812     _exit(-1);
5813 
5814   } else  {
5815     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5816     // care about the actual exit code, for now.
5817 
5818     int status;
5819 
5820     // Wait for the child process to exit.  This returns immediately if
5821     // the child has already exited. */
5822     while (waitpid(pid, &status, 0) < 0) {
5823       switch (errno) {
5824       case ECHILD: return 0;
5825       case EINTR: break;
5826       default: return -1;
5827       }
5828     }
5829 
5830     if (WIFEXITED(status)) {
5831       // The child exited normally; get its exit code.
5832       return WEXITSTATUS(status);
5833     } else if (WIFSIGNALED(status)) {
5834       // The child exited because of a signal
5835       // The best value to return is 0x80 + signal number,
5836       // because that is what all Unix shells do, and because
5837       // it allows callers to distinguish between process exit and
5838       // process death by signal.
5839       return 0x80 + WTERMSIG(status);
5840     } else {
5841       // Unknown exit code; pass it through
5842       return status;
5843     }
5844   }
5845 }
5846 
5847 // is_headless_jre()
5848 //
5849 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5850 // in order to report if we are running in a headless jre
5851 //
5852 // Since JDK8 xawt/libmawt.so was moved into the same directory
5853 // as libawt.so, and renamed libawt_xawt.so
5854 //
5855 bool os::is_headless_jre() {
5856   struct stat statbuf;
5857   char buf[MAXPATHLEN];
5858   char libmawtpath[MAXPATHLEN];
5859   const char *xawtstr  = "/xawt/libmawt.so";
5860   const char *new_xawtstr = "/libawt_xawt.so";
5861   char *p;
5862 
5863   // Get path to libjvm.so
5864   os::jvm_path(buf, sizeof(buf));
5865 
5866   // Get rid of libjvm.so
5867   p = strrchr(buf, '/');
5868   if (p == NULL) {
5869     return false;
5870   } else {
5871     *p = '\0';
5872   }
5873 
5874   // Get rid of client or server
5875   p = strrchr(buf, '/');
5876   if (p == NULL) {
5877     return false;
5878   } else {
5879     *p = '\0';
5880   }
5881 
5882   // check xawt/libmawt.so
5883   strcpy(libmawtpath, buf);
5884   strcat(libmawtpath, xawtstr);
5885   if (::stat(libmawtpath, &statbuf) == 0) return false;
5886 
5887   // check libawt_xawt.so
5888   strcpy(libmawtpath, buf);
5889   strcat(libmawtpath, new_xawtstr);
5890   if (::stat(libmawtpath, &statbuf) == 0) return false;
5891 
5892   return true;
5893 }
5894 
5895 size_t os::write(int fd, const void *buf, unsigned int nBytes) {
5896   size_t res;
5897   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5898          "Assumed _thread_in_native");
5899   RESTARTABLE((size_t) ::write(fd, buf, (size_t) nBytes), res);
5900   return res;
5901 }
5902 
5903 int os::close(int fd) {
5904   return ::close(fd);
5905 }
5906 
5907 int os::socket_close(int fd) {
5908   return ::close(fd);
5909 }
5910 
5911 int os::recv(int fd, char* buf, size_t nBytes, uint flags) {
5912   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5913          "Assumed _thread_in_native");
5914   RESTARTABLE_RETURN_INT((int)::recv(fd, buf, nBytes, flags));
5915 }
5916 
5917 int os::send(int fd, char* buf, size_t nBytes, uint flags) {
5918   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5919          "Assumed _thread_in_native");
5920   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5921 }
5922 
5923 int os::raw_send(int fd, char* buf, size_t nBytes, uint flags) {
5924   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5925 }
5926 
5927 // As both poll and select can be interrupted by signals, we have to be
5928 // prepared to restart the system call after updating the timeout, unless
5929 // a poll() is done with timeout == -1, in which case we repeat with this
5930 // "wait forever" value.
5931 
5932 int os::connect(int fd, struct sockaddr *him, socklen_t len) {
5933   int _result;
5934   _result = ::connect(fd, him, len);
5935 
5936   // On Solaris, when a connect() call is interrupted, the connection
5937   // can be established asynchronously (see 6343810). Subsequent calls
5938   // to connect() must check the errno value which has the semantic
5939   // described below (copied from the connect() man page). Handling
5940   // of asynchronously established connections is required for both
5941   // blocking and non-blocking sockets.
5942   //     EINTR            The  connection  attempt  was   interrupted
5943   //                      before  any data arrived by the delivery of
5944   //                      a signal. The connection, however, will  be
5945   //                      established asynchronously.
5946   //
5947   //     EINPROGRESS      The socket is non-blocking, and the connec-
5948   //                      tion  cannot  be completed immediately.
5949   //
5950   //     EALREADY         The socket is non-blocking,  and a previous
5951   //                      connection  attempt  has  not yet been com-
5952   //                      pleted.
5953   //
5954   //     EISCONN          The socket is already connected.
5955   if (_result == OS_ERR && errno == EINTR) {
5956     // restarting a connect() changes its errno semantics
5957     RESTARTABLE(::connect(fd, him, len), _result);
5958     // undo these changes
5959     if (_result == OS_ERR) {
5960       if (errno == EALREADY) {
5961         errno = EINPROGRESS; // fall through
5962       } else if (errno == EISCONN) {
5963         errno = 0;
5964         return OS_OK;
5965       }
5966     }
5967   }
5968   return _result;
5969 }
5970 
5971 // Get the default path to the core file
5972 // Returns the length of the string
5973 int os::get_core_path(char* buffer, size_t bufferSize) {
5974   const char* p = get_current_directory(buffer, bufferSize);
5975 
5976   if (p == NULL) {
5977     assert(p != NULL, "failed to get current directory");
5978     return 0;
5979   }
5980 
5981   jio_snprintf(buffer, bufferSize, "%s/core or core.%d",
5982                                               p, current_process_id());
5983 
5984   return strlen(buffer);
5985 }
5986 
5987 #ifndef PRODUCT
5988 void TestReserveMemorySpecial_test() {
5989   // No tests available for this platform
5990 }
5991 #endif