1 /*
   2  * Copyright (c) 1997, 2018, 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 "jvm.h"
  27 #include "classfile/classLoader.hpp"
  28 #include "classfile/systemDictionary.hpp"
  29 #include "classfile/vmSymbols.hpp"
  30 #include "code/icBuffer.hpp"
  31 #include "code/vtableStubs.hpp"
  32 #include "compiler/compileBroker.hpp"
  33 #include "compiler/disassembler.hpp"
  34 #include "interpreter/interpreter.hpp"
  35 #include "logging/log.hpp"
  36 #include "logging/logStream.hpp"
  37 #include "memory/allocation.inline.hpp"
  38 #include "memory/filemap.hpp"
  39 #include "oops/oop.inline.hpp"
  40 #include "os_share_solaris.hpp"
  41 #include "os_solaris.inline.hpp"
  42 #include "prims/jniFastGetField.hpp"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.inline.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.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 "semaphore_posix.hpp"
  64 #include "services/attachListener.hpp"
  65 #include "services/memTracker.hpp"
  66 #include "services/runtimeService.hpp"
  67 #include "utilities/align.hpp"
  68 #include "utilities/decoder.hpp"
  69 #include "utilities/defaultStream.hpp"
  70 #include "utilities/events.hpp"
  71 #include "utilities/growableArray.hpp"
  72 #include "utilities/macros.hpp"
  73 #include "utilities/vmError.hpp"
  74 
  75 // put OS-includes here
  76 # include <dlfcn.h>
  77 # include <errno.h>
  78 # include <exception>
  79 # include <link.h>
  80 # include <poll.h>
  81 # include <pthread.h>
  82 # include <schedctl.h>
  83 # include <setjmp.h>
  84 # include <signal.h>
  85 # include <stdio.h>
  86 # include <alloca.h>
  87 # include <sys/filio.h>
  88 # include <sys/ipc.h>
  89 # include <sys/lwp.h>
  90 # include <sys/machelf.h>     // for elf Sym structure used by dladdr1
  91 # include <sys/mman.h>
  92 # include <sys/processor.h>
  93 # include <sys/procset.h>
  94 # include <sys/pset.h>
  95 # include <sys/resource.h>
  96 # include <sys/shm.h>
  97 # include <sys/socket.h>
  98 # include <sys/stat.h>
  99 # include <sys/systeminfo.h>
 100 # include <sys/time.h>
 101 # include <sys/times.h>
 102 # include <sys/types.h>
 103 # include <sys/wait.h>
 104 # include <sys/utsname.h>
 105 # include <thread.h>
 106 # include <unistd.h>
 107 # include <sys/priocntl.h>
 108 # include <sys/rtpriocntl.h>
 109 # include <sys/tspriocntl.h>
 110 # include <sys/iapriocntl.h>
 111 # include <sys/fxpriocntl.h>
 112 # include <sys/loadavg.h>
 113 # include <string.h>
 114 # include <stdio.h>
 115 
 116 # define _STRUCTURED_PROC 1  //  this gets us the new structured proc interfaces of 5.6 & later
 117 # include <sys/procfs.h>     //  see comment in <sys/procfs.h>
 118 
 119 #define MAX_PATH (2 * K)
 120 
 121 // for timer info max values which include all bits
 122 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 123 
 124 
 125 // Here are some liblgrp types from sys/lgrp_user.h to be able to
 126 // compile on older systems without this header file.
 127 
 128 #ifndef MADV_ACCESS_LWP
 129   #define  MADV_ACCESS_LWP   7       /* next LWP to access heavily */
 130 #endif
 131 #ifndef MADV_ACCESS_MANY
 132   #define  MADV_ACCESS_MANY  8       /* many processes to access heavily */
 133 #endif
 134 
 135 #ifndef LGRP_RSRC_CPU
 136   #define LGRP_RSRC_CPU      0       /* CPU resources */
 137 #endif
 138 #ifndef LGRP_RSRC_MEM
 139   #define LGRP_RSRC_MEM      1       /* memory resources */
 140 #endif
 141 
 142 // Values for ThreadPriorityPolicy == 1
 143 int prio_policy1[CriticalPriority+1] = {
 144   -99999,  0, 16,  32,  48,  64,
 145           80, 96, 112, 124, 127, 127 };
 146 
 147 // System parameters used internally
 148 static clock_t clock_tics_per_sec = 100;
 149 
 150 // Track if we have called enable_extended_FILE_stdio (on Solaris 10u4+)
 151 static bool enabled_extended_FILE_stdio = false;
 152 
 153 // For diagnostics to print a message once. see run_periodic_checks
 154 static bool check_addr0_done = false;
 155 static sigset_t check_signal_done;
 156 static bool check_signals = true;
 157 
 158 address os::Solaris::handler_start;  // start pc of thr_sighndlrinfo
 159 address os::Solaris::handler_end;    // end pc of thr_sighndlrinfo
 160 
 161 address os::Solaris::_main_stack_base = NULL;  // 4352906 workaround
 162 
 163 os::Solaris::pthread_setname_np_func_t os::Solaris::_pthread_setname_np = NULL;
 164 
 165 // "default" initializers for missing libc APIs
 166 extern "C" {
 167   static int lwp_mutex_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
 168   static int lwp_mutex_destroy(mutex_t *mx)                 { return 0; }
 169 
 170   static int lwp_cond_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
 171   static int lwp_cond_destroy(cond_t *cv)                   { return 0; }
 172 }
 173 
 174 // "default" initializers for pthread-based synchronization
 175 extern "C" {
 176   static int pthread_mutex_default_init(mutex_t *mx, int scope, void *arg) { memset(mx, 0, sizeof(mutex_t)); return 0; }
 177   static int pthread_cond_default_init(cond_t *cv, int scope, void *arg){ memset(cv, 0, sizeof(cond_t)); return 0; }
 178 }
 179 
 180 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
 181 
 182 static inline size_t adjust_stack_size(address base, size_t size) {
 183   if ((ssize_t)size < 0) {
 184     // 4759953: Compensate for ridiculous stack size.
 185     size = max_intx;
 186   }
 187   if (size > (size_t)base) {
 188     // 4812466: Make sure size doesn't allow the stack to wrap the address space.
 189     size = (size_t)base;
 190   }
 191   return size;
 192 }
 193 
 194 static inline stack_t get_stack_info() {
 195   stack_t st;
 196   int retval = thr_stksegment(&st);
 197   st.ss_size = adjust_stack_size((address)st.ss_sp, st.ss_size);
 198   assert(retval == 0, "incorrect return value from thr_stksegment");
 199   assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
 200   assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
 201   return st;
 202 }
 203 
 204 static void _handle_uncaught_cxx_exception() {
 205   VMError::report_and_die("An uncaught C++ exception");
 206 }
 207 
 208 bool os::is_primordial_thread(void) {
 209   int r = thr_main();
 210   guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
 211   return r == 1;
 212 }
 213 
 214 address os::current_stack_base() {
 215   bool _is_primordial_thread = is_primordial_thread();
 216 
 217   // Workaround 4352906, avoid calls to thr_stksegment by
 218   // thr_main after the first one (it looks like we trash
 219   // some data, causing the value for ss_sp to be incorrect).
 220   if (!_is_primordial_thread || os::Solaris::_main_stack_base == NULL) {
 221     stack_t st = get_stack_info();
 222     if (_is_primordial_thread) {
 223       // cache initial value of stack base
 224       os::Solaris::_main_stack_base = (address)st.ss_sp;
 225     }
 226     return (address)st.ss_sp;
 227   } else {
 228     guarantee(os::Solaris::_main_stack_base != NULL, "Attempt to use null cached stack base");
 229     return os::Solaris::_main_stack_base;
 230   }
 231 }
 232 
 233 size_t os::current_stack_size() {
 234   size_t size;
 235 
 236   if (!is_primordial_thread()) {
 237     size = get_stack_info().ss_size;
 238   } else {
 239     struct rlimit limits;
 240     getrlimit(RLIMIT_STACK, &limits);
 241     size = adjust_stack_size(os::Solaris::_main_stack_base, (size_t)limits.rlim_cur);
 242   }
 243   // base may not be page aligned
 244   address base = current_stack_base();
 245   address bottom = align_up(base - size, os::vm_page_size());;
 246   return (size_t)(base - bottom);
 247 }
 248 
 249 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
 250   return localtime_r(clock, res);
 251 }
 252 
 253 void os::Solaris::try_enable_extended_io() {
 254   typedef int (*enable_extended_FILE_stdio_t)(int, int);
 255 
 256   if (!UseExtendedFileIO) {
 257     return;
 258   }
 259 
 260   enable_extended_FILE_stdio_t enabler =
 261     (enable_extended_FILE_stdio_t) dlsym(RTLD_DEFAULT,
 262                                          "enable_extended_FILE_stdio");
 263   if (enabler) {
 264     enabler(-1, -1);
 265   }
 266 }
 267 
 268 static int _processors_online = 0;
 269 
 270 jint os::Solaris::_os_thread_limit = 0;
 271 volatile jint os::Solaris::_os_thread_count = 0;
 272 
 273 julong os::available_memory() {
 274   return Solaris::available_memory();
 275 }
 276 
 277 julong os::Solaris::available_memory() {
 278   return (julong)sysconf(_SC_AVPHYS_PAGES) * os::vm_page_size();
 279 }
 280 
 281 julong os::Solaris::_physical_memory = 0;
 282 
 283 julong os::physical_memory() {
 284   return Solaris::physical_memory();
 285 }
 286 
 287 static hrtime_t first_hrtime = 0;
 288 static const hrtime_t hrtime_hz = 1000*1000*1000;
 289 static volatile hrtime_t max_hrtime = 0;
 290 
 291 
 292 void os::Solaris::initialize_system_info() {
 293   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 294   _processors_online = sysconf(_SC_NPROCESSORS_ONLN);
 295   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) *
 296                                      (julong)sysconf(_SC_PAGESIZE);
 297 }
 298 
 299 uint os::processor_id() {
 300   const processorid_t id = ::getcpuid();
 301   assert(id >= 0 && id < _processor_count, "Invalid processor id");
 302   return (uint)id;
 303 }
 304 
 305 int os::active_processor_count() {
 306   // User has overridden the number of active processors
 307   if (ActiveProcessorCount > 0) {
 308     log_trace(os)("active_processor_count: "
 309                   "active processor count set by user : %d",
 310                   ActiveProcessorCount);
 311     return ActiveProcessorCount;
 312   }
 313 
 314   int online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
 315   pid_t pid = getpid();
 316   psetid_t pset = PS_NONE;
 317   // Are we running in a processor set or is there any processor set around?
 318   if (pset_bind(PS_QUERY, P_PID, pid, &pset) == 0) {
 319     uint_t pset_cpus;
 320     // Query the number of cpus available to us.
 321     if (pset_info(pset, NULL, &pset_cpus, NULL) == 0) {
 322       assert(pset_cpus > 0 && pset_cpus <= online_cpus, "sanity check");
 323       _processors_online = pset_cpus;
 324       return pset_cpus;
 325     }
 326   }
 327   // Otherwise return number of online cpus
 328   return online_cpus;
 329 }
 330 
 331 static bool find_processors_in_pset(psetid_t        pset,
 332                                     processorid_t** id_array,
 333                                     uint_t*         id_length) {
 334   bool result = false;
 335   // Find the number of processors in the processor set.
 336   if (pset_info(pset, NULL, id_length, NULL) == 0) {
 337     // Make up an array to hold their ids.
 338     *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
 339     // Fill in the array with their processor ids.
 340     if (pset_info(pset, NULL, id_length, *id_array) == 0) {
 341       result = true;
 342     }
 343   }
 344   return result;
 345 }
 346 
 347 // Callers of find_processors_online() must tolerate imprecise results --
 348 // the system configuration can change asynchronously because of DR
 349 // or explicit psradm operations.
 350 //
 351 // We also need to take care that the loop (below) terminates as the
 352 // number of processors online can change between the _SC_NPROCESSORS_ONLN
 353 // request and the loop that builds the list of processor ids.   Unfortunately
 354 // there's no reliable way to determine the maximum valid processor id,
 355 // so we use a manifest constant, MAX_PROCESSOR_ID, instead.  See p_online
 356 // man pages, which claim the processor id set is "sparse, but
 357 // not too sparse".  MAX_PROCESSOR_ID is used to ensure that we eventually
 358 // exit the loop.
 359 //
 360 // In the future we'll be able to use sysconf(_SC_CPUID_MAX), but that's
 361 // not available on S8.0.
 362 
 363 static bool find_processors_online(processorid_t** id_array,
 364                                    uint*           id_length) {
 365   const processorid_t MAX_PROCESSOR_ID = 100000;
 366   // Find the number of processors online.
 367   *id_length = sysconf(_SC_NPROCESSORS_ONLN);
 368   // Make up an array to hold their ids.
 369   *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
 370   // Processors need not be numbered consecutively.
 371   long found = 0;
 372   processorid_t next = 0;
 373   while (found < *id_length && next < MAX_PROCESSOR_ID) {
 374     processor_info_t info;
 375     if (processor_info(next, &info) == 0) {
 376       // NB, PI_NOINTR processors are effectively online ...
 377       if (info.pi_state == P_ONLINE || info.pi_state == P_NOINTR) {
 378         (*id_array)[found] = next;
 379         found += 1;
 380       }
 381     }
 382     next += 1;
 383   }
 384   if (found < *id_length) {
 385     // The loop above didn't identify the expected number of processors.
 386     // We could always retry the operation, calling sysconf(_SC_NPROCESSORS_ONLN)
 387     // and re-running the loop, above, but there's no guarantee of progress
 388     // if the system configuration is in flux.  Instead, we just return what
 389     // we've got.  Note that in the worst case find_processors_online() could
 390     // return an empty set.  (As a fall-back in the case of the empty set we
 391     // could just return the ID of the current processor).
 392     *id_length = found;
 393   }
 394 
 395   return true;
 396 }
 397 
 398 static bool assign_distribution(processorid_t* id_array,
 399                                 uint           id_length,
 400                                 uint*          distribution,
 401                                 uint           distribution_length) {
 402   // We assume we can assign processorid_t's to uint's.
 403   assert(sizeof(processorid_t) == sizeof(uint),
 404          "can't convert processorid_t to uint");
 405   // Quick check to see if we won't succeed.
 406   if (id_length < distribution_length) {
 407     return false;
 408   }
 409   // Assign processor ids to the distribution.
 410   // Try to shuffle processors to distribute work across boards,
 411   // assuming 4 processors per board.
 412   const uint processors_per_board = ProcessDistributionStride;
 413   // Find the maximum processor id.
 414   processorid_t max_id = 0;
 415   for (uint m = 0; m < id_length; m += 1) {
 416     max_id = MAX2(max_id, id_array[m]);
 417   }
 418   // The next id, to limit loops.
 419   const processorid_t limit_id = max_id + 1;
 420   // Make up markers for available processors.
 421   bool* available_id = NEW_C_HEAP_ARRAY(bool, limit_id, mtInternal);
 422   for (uint c = 0; c < limit_id; c += 1) {
 423     available_id[c] = false;
 424   }
 425   for (uint a = 0; a < id_length; a += 1) {
 426     available_id[id_array[a]] = true;
 427   }
 428   // Step by "boards", then by "slot", copying to "assigned".
 429   // NEEDS_CLEANUP: The assignment of processors should be stateful,
 430   //                remembering which processors have been assigned by
 431   //                previous calls, etc., so as to distribute several
 432   //                independent calls of this method.  What we'd like is
 433   //                It would be nice to have an API that let us ask
 434   //                how many processes are bound to a processor,
 435   //                but we don't have that, either.
 436   //                In the short term, "board" is static so that
 437   //                subsequent distributions don't all start at board 0.
 438   static uint board = 0;
 439   uint assigned = 0;
 440   // Until we've found enough processors ....
 441   while (assigned < distribution_length) {
 442     // ... find the next available processor in the board.
 443     for (uint slot = 0; slot < processors_per_board; slot += 1) {
 444       uint try_id = board * processors_per_board + slot;
 445       if ((try_id < limit_id) && (available_id[try_id] == true)) {
 446         distribution[assigned] = try_id;
 447         available_id[try_id] = false;
 448         assigned += 1;
 449         break;
 450       }
 451     }
 452     board += 1;
 453     if (board * processors_per_board + 0 >= limit_id) {
 454       board = 0;
 455     }
 456   }
 457   if (available_id != NULL) {
 458     FREE_C_HEAP_ARRAY(bool, available_id);
 459   }
 460   return true;
 461 }
 462 
 463 void os::set_native_thread_name(const char *name) {
 464   if (Solaris::_pthread_setname_np != NULL) {
 465     // Only the first 31 bytes of 'name' are processed by pthread_setname_np
 466     // but we explicitly copy into a size-limited buffer to avoid any
 467     // possible overflow.
 468     char buf[32];
 469     snprintf(buf, sizeof(buf), "%s", name);
 470     buf[sizeof(buf) - 1] = '\0';
 471     Solaris::_pthread_setname_np(pthread_self(), buf);
 472   }
 473 }
 474 
 475 bool os::distribute_processes(uint length, uint* distribution) {
 476   bool result = false;
 477   // Find the processor id's of all the available CPUs.
 478   processorid_t* id_array  = NULL;
 479   uint           id_length = 0;
 480   // There are some races between querying information and using it,
 481   // since processor sets can change dynamically.
 482   psetid_t pset = PS_NONE;
 483   // Are we running in a processor set?
 484   if ((pset_bind(PS_QUERY, P_PID, P_MYID, &pset) == 0) && pset != PS_NONE) {
 485     result = find_processors_in_pset(pset, &id_array, &id_length);
 486   } else {
 487     result = find_processors_online(&id_array, &id_length);
 488   }
 489   if (result == true) {
 490     if (id_length >= length) {
 491       result = assign_distribution(id_array, id_length, distribution, length);
 492     } else {
 493       result = false;
 494     }
 495   }
 496   if (id_array != NULL) {
 497     FREE_C_HEAP_ARRAY(processorid_t, id_array);
 498   }
 499   return result;
 500 }
 501 
 502 bool os::bind_to_processor(uint processor_id) {
 503   // We assume that a processorid_t can be stored in a uint.
 504   assert(sizeof(uint) == sizeof(processorid_t),
 505          "can't convert uint to processorid_t");
 506   int bind_result =
 507     processor_bind(P_LWPID,                       // bind LWP.
 508                    P_MYID,                        // bind current LWP.
 509                    (processorid_t) processor_id,  // id.
 510                    NULL);                         // don't return old binding.
 511   return (bind_result == 0);
 512 }
 513 
 514 // Return true if user is running as root.
 515 
 516 bool os::have_special_privileges() {
 517   static bool init = false;
 518   static bool privileges = false;
 519   if (!init) {
 520     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 521     init = true;
 522   }
 523   return privileges;
 524 }
 525 
 526 
 527 void os::init_system_properties_values() {
 528   // The next steps are taken in the product version:
 529   //
 530   // Obtain the JAVA_HOME value from the location of libjvm.so.
 531   // This library should be located at:
 532   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 533   //
 534   // If "/jre/lib/" appears at the right place in the path, then we
 535   // assume libjvm.so is installed in a JDK and we use this path.
 536   //
 537   // Otherwise exit with message: "Could not create the Java virtual machine."
 538   //
 539   // The following extra steps are taken in the debugging version:
 540   //
 541   // If "/jre/lib/" does NOT appear at the right place in the path
 542   // instead of exit check for $JAVA_HOME environment variable.
 543   //
 544   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 545   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 546   // it looks like libjvm.so is installed there
 547   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 548   //
 549   // Otherwise exit.
 550   //
 551   // Important note: if the location of libjvm.so changes this
 552   // code needs to be changed accordingly.
 553 
 554 // Base path of extensions installed on the system.
 555 #define SYS_EXT_DIR     "/usr/jdk/packages"
 556 #define EXTENSIONS_DIR  "/lib/ext"
 557 
 558   // Buffer that fits several sprintfs.
 559   // Note that the space for the colon and the trailing null are provided
 560   // by the nulls included by the sizeof operator.
 561   const size_t bufsize =
 562     MAX3((size_t)MAXPATHLEN,  // For dll_dir & friends.
 563          sizeof(SYS_EXT_DIR) + sizeof("/lib/"), // invariant ld_library_path
 564          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 565   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 566 
 567   // sysclasspath, java_home, dll_dir
 568   {
 569     char *pslash;
 570     os::jvm_path(buf, bufsize);
 571 
 572     // Found the full path to libjvm.so.
 573     // Now cut the path to <java_home>/jre if we can.
 574     *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
 575     pslash = strrchr(buf, '/');
 576     if (pslash != NULL) {
 577       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 578     }
 579     Arguments::set_dll_dir(buf);
 580 
 581     if (pslash != NULL) {
 582       pslash = strrchr(buf, '/');
 583       if (pslash != NULL) {
 584         *pslash = '\0';        // Get rid of /lib.
 585       }
 586     }
 587     Arguments::set_java_home(buf);
 588     set_boot_path('/', ':');
 589   }
 590 
 591   // Where to look for native libraries.
 592   {
 593     // Use dlinfo() to determine the correct java.library.path.
 594     //
 595     // If we're launched by the Java launcher, and the user
 596     // does not set java.library.path explicitly on the commandline,
 597     // the Java launcher sets LD_LIBRARY_PATH for us and unsets
 598     // LD_LIBRARY_PATH_32 and LD_LIBRARY_PATH_64.  In this case
 599     // dlinfo returns LD_LIBRARY_PATH + crle settings (including
 600     // /usr/lib), which is exactly what we want.
 601     //
 602     // If the user does set java.library.path, it completely
 603     // overwrites this setting, and always has.
 604     //
 605     // If we're not launched by the Java launcher, we may
 606     // get here with any/all of the LD_LIBRARY_PATH[_32|64]
 607     // settings.  Again, dlinfo does exactly what we want.
 608 
 609     Dl_serinfo     info_sz, *info = &info_sz;
 610     Dl_serpath     *path;
 611     char           *library_path;
 612     char           *common_path = buf;
 613 
 614     // Determine search path count and required buffer size.
 615     if (dlinfo(RTLD_SELF, RTLD_DI_SERINFOSIZE, (void *)info) == -1) {
 616       FREE_C_HEAP_ARRAY(char, buf);
 617       vm_exit_during_initialization("dlinfo SERINFOSIZE request", dlerror());
 618     }
 619 
 620     // Allocate new buffer and initialize.
 621     info = (Dl_serinfo*)NEW_C_HEAP_ARRAY(char, info_sz.dls_size, mtInternal);
 622     info->dls_size = info_sz.dls_size;
 623     info->dls_cnt = info_sz.dls_cnt;
 624 
 625     // Obtain search path information.
 626     if (dlinfo(RTLD_SELF, RTLD_DI_SERINFO, (void *)info) == -1) {
 627       FREE_C_HEAP_ARRAY(char, buf);
 628       FREE_C_HEAP_ARRAY(char, info);
 629       vm_exit_during_initialization("dlinfo SERINFO request", dlerror());
 630     }
 631 
 632     path = &info->dls_serpath[0];
 633 
 634     // Note: Due to a legacy implementation, most of the library path
 635     // is set in the launcher. This was to accomodate linking restrictions
 636     // on legacy Solaris implementations (which are no longer supported).
 637     // Eventually, all the library path setting will be done here.
 638     //
 639     // However, to prevent the proliferation of improperly built native
 640     // libraries, the new path component /usr/jdk/packages is added here.
 641 
 642     // Construct the invariant part of ld_library_path.
 643     sprintf(common_path, SYS_EXT_DIR "/lib");
 644 
 645     // Struct size is more than sufficient for the path components obtained
 646     // through the dlinfo() call, so only add additional space for the path
 647     // components explicitly added here.
 648     size_t library_path_size = info->dls_size + strlen(common_path);
 649     library_path = (char *)NEW_C_HEAP_ARRAY(char, library_path_size, mtInternal);
 650     library_path[0] = '\0';
 651 
 652     // Construct the desired Java library path from the linker's library
 653     // search path.
 654     //
 655     // For compatibility, it is optimal that we insert the additional path
 656     // components specific to the Java VM after those components specified
 657     // in LD_LIBRARY_PATH (if any) but before those added by the ld.so
 658     // infrastructure.
 659     if (info->dls_cnt == 0) { // Not sure this can happen, but allow for it.
 660       strcpy(library_path, common_path);
 661     } else {
 662       int inserted = 0;
 663       int i;
 664       for (i = 0; i < info->dls_cnt; i++, path++) {
 665         uint_t flags = path->dls_flags & LA_SER_MASK;
 666         if (((flags & LA_SER_LIBPATH) == 0) && !inserted) {
 667           strcat(library_path, common_path);
 668           strcat(library_path, os::path_separator());
 669           inserted = 1;
 670         }
 671         strcat(library_path, path->dls_name);
 672         strcat(library_path, os::path_separator());
 673       }
 674       // Eliminate trailing path separator.
 675       library_path[strlen(library_path)-1] = '\0';
 676     }
 677 
 678     // happens before argument parsing - can't use a trace flag
 679     // tty->print_raw("init_system_properties_values: native lib path: ");
 680     // tty->print_raw_cr(library_path);
 681 
 682     // Callee copies into its own buffer.
 683     Arguments::set_library_path(library_path);
 684 
 685     FREE_C_HEAP_ARRAY(char, library_path);
 686     FREE_C_HEAP_ARRAY(char, info);
 687   }
 688 
 689   // Extensions directories.
 690   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 691   Arguments::set_ext_dirs(buf);
 692 
 693   FREE_C_HEAP_ARRAY(char, buf);
 694 
 695 #undef SYS_EXT_DIR
 696 #undef EXTENSIONS_DIR
 697 }
 698 
 699 void os::breakpoint() {
 700   BREAKPOINT;
 701 }
 702 
 703 bool os::obsolete_option(const JavaVMOption *option) {
 704   if (!strncmp(option->optionString, "-Xt", 3)) {
 705     return true;
 706   } else if (!strncmp(option->optionString, "-Xtm", 4)) {
 707     return true;
 708   } else if (!strncmp(option->optionString, "-Xverifyheap", 12)) {
 709     return true;
 710   } else if (!strncmp(option->optionString, "-Xmaxjitcodesize", 16)) {
 711     return true;
 712   }
 713   return false;
 714 }
 715 
 716 bool os::Solaris::valid_stack_address(Thread* thread, address sp) {
 717   address  stackStart  = (address)thread->stack_base();
 718   address  stackEnd    = (address)(stackStart - (address)thread->stack_size());
 719   if (sp < stackStart && sp >= stackEnd) return true;
 720   return false;
 721 }
 722 
 723 extern "C" void breakpoint() {
 724   // use debugger to set breakpoint here
 725 }
 726 
 727 static thread_t main_thread;
 728 
 729 // Thread start routine for all newly created threads
 730 extern "C" void* thread_native_entry(void* thread_addr) {
 731 
 732   Thread* thread = (Thread*)thread_addr;
 733 
 734   thread->record_stack_base_and_size();
 735 
 736   // Try to randomize the cache line index of hot stack frames.
 737   // This helps when threads of the same stack traces evict each other's
 738   // cache lines. The threads can be either from the same JVM instance, or
 739   // from different JVM instances. The benefit is especially true for
 740   // processors with hyperthreading technology.
 741   static int counter = 0;
 742   int pid = os::current_process_id();
 743   alloca(((pid ^ counter++) & 7) * 128);
 744 
 745   int prio;
 746 
 747   thread->initialize_thread_current();
 748 
 749   OSThread* osthr = thread->osthread();
 750 
 751   osthr->set_lwp_id(_lwp_self());  // Store lwp in case we are bound
 752   thread->_schedctl = (void *) schedctl_init();
 753 
 754   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ").",
 755     os::current_thread_id());
 756 
 757   if (UseNUMA) {
 758     int lgrp_id = os::numa_get_group_id();
 759     if (lgrp_id != -1) {
 760       thread->set_lgrp_id(lgrp_id);
 761     }
 762   }
 763 
 764   // Our priority was set when we were created, and stored in the
 765   // osthread, but couldn't be passed through to our LWP until now.
 766   // So read back the priority and set it again.
 767 
 768   if (osthr->thread_id() != -1) {
 769     if (UseThreadPriorities) {
 770       int prio = osthr->native_priority();
 771       if (ThreadPriorityVerbose) {
 772         tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is "
 773                       INTPTR_FORMAT ", setting priority: %d\n",
 774                       osthr->thread_id(), osthr->lwp_id(), prio);
 775       }
 776       os::set_native_priority(thread, prio);
 777     }
 778   } else if (ThreadPriorityVerbose) {
 779     warning("Can't set priority in _start routine, thread id hasn't been set\n");
 780   }
 781 
 782   assert(osthr->get_state() == RUNNABLE, "invalid os thread state");
 783 
 784   // initialize signal mask for this thread
 785   os::Solaris::hotspot_sigmask(thread);
 786 
 787   os::Solaris::init_thread_fpu_state();
 788   std::set_terminate(_handle_uncaught_cxx_exception);
 789 
 790   thread->call_run();
 791 
 792   // Note: at this point the thread object may already have deleted itself.
 793   // Do not dereference it from here on out.
 794 
 795   // One less thread is executing
 796   // When the VMThread gets here, the main thread may have already exited
 797   // which frees the CodeHeap containing the Atomic::dec code
 798   if (thread != VMThread::vm_thread() && VMThread::vm_thread() != NULL) {
 799     Atomic::dec(&os::Solaris::_os_thread_count);
 800   }
 801 
 802   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ").", os::current_thread_id());
 803 
 804   if (UseDetachedThreads) {
 805     thr_exit(NULL);
 806     ShouldNotReachHere();
 807   }
 808   return NULL;
 809 }
 810 
 811 static OSThread* create_os_thread(Thread* thread, thread_t thread_id) {
 812   // Allocate the OSThread object
 813   OSThread* osthread = new OSThread(NULL, NULL);
 814   if (osthread == NULL) return NULL;
 815 
 816   // Store info on the Solaris thread into the OSThread
 817   osthread->set_thread_id(thread_id);
 818   osthread->set_lwp_id(_lwp_self());
 819   thread->_schedctl = (void *) schedctl_init();
 820 
 821   if (UseNUMA) {
 822     int lgrp_id = os::numa_get_group_id();
 823     if (lgrp_id != -1) {
 824       thread->set_lgrp_id(lgrp_id);
 825     }
 826   }
 827 
 828   if (ThreadPriorityVerbose) {
 829     tty->print_cr("In create_os_thread, Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT "\n",
 830                   osthread->thread_id(), osthread->lwp_id());
 831   }
 832 
 833   // Initial thread state is INITIALIZED, not SUSPENDED
 834   osthread->set_state(INITIALIZED);
 835 
 836   return osthread;
 837 }
 838 
 839 void os::Solaris::hotspot_sigmask(Thread* thread) {
 840   //Save caller's signal mask
 841   sigset_t sigmask;
 842   pthread_sigmask(SIG_SETMASK, NULL, &sigmask);
 843   OSThread *osthread = thread->osthread();
 844   osthread->set_caller_sigmask(sigmask);
 845 
 846   pthread_sigmask(SIG_UNBLOCK, os::Solaris::unblocked_signals(), NULL);
 847   if (!ReduceSignalUsage) {
 848     if (thread->is_VM_thread()) {
 849       // Only the VM thread handles BREAK_SIGNAL ...
 850       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 851     } else {
 852       // ... all other threads block BREAK_SIGNAL
 853       assert(!sigismember(vm_signals(), SIGINT), "SIGINT should not be blocked");
 854       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 855     }
 856   }
 857 }
 858 
 859 bool os::create_attached_thread(JavaThread* thread) {
 860 #ifdef ASSERT
 861   thread->verify_not_published();
 862 #endif
 863   OSThread* osthread = create_os_thread(thread, thr_self());
 864   if (osthread == NULL) {
 865     return false;
 866   }
 867 
 868   // Initial thread state is RUNNABLE
 869   osthread->set_state(RUNNABLE);
 870   thread->set_osthread(osthread);
 871 
 872   // initialize signal mask for this thread
 873   // and save the caller's signal mask
 874   os::Solaris::hotspot_sigmask(thread);
 875 
 876   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ").",
 877     os::current_thread_id());
 878 
 879   return true;
 880 }
 881 
 882 bool os::create_main_thread(JavaThread* thread) {
 883 #ifdef ASSERT
 884   thread->verify_not_published();
 885 #endif
 886   if (_starting_thread == NULL) {
 887     _starting_thread = create_os_thread(thread, main_thread);
 888     if (_starting_thread == NULL) {
 889       return false;
 890     }
 891   }
 892 
 893   // The primodial thread is runnable from the start
 894   _starting_thread->set_state(RUNNABLE);
 895 
 896   thread->set_osthread(_starting_thread);
 897 
 898   // initialize signal mask for this thread
 899   // and save the caller's signal mask
 900   os::Solaris::hotspot_sigmask(thread);
 901 
 902   return true;
 903 }
 904 
 905 // Helper function to trace thread attributes, similar to os::Posix::describe_pthread_attr()
 906 static char* describe_thr_create_attributes(char* buf, size_t buflen,
 907                                             size_t stacksize, long flags) {
 908   stringStream ss(buf, buflen);
 909   ss.print("stacksize: " SIZE_FORMAT "k, ", stacksize / 1024);
 910   ss.print("flags: ");
 911   #define PRINT_FLAG(f) if (flags & f) ss.print( #f " ");
 912   #define ALL(X) \
 913     X(THR_SUSPENDED) \
 914     X(THR_DETACHED) \
 915     X(THR_BOUND) \
 916     X(THR_NEW_LWP) \
 917     X(THR_DAEMON)
 918   ALL(PRINT_FLAG)
 919   #undef ALL
 920   #undef PRINT_FLAG
 921   return buf;
 922 }
 923 
 924 // return default stack size for thr_type
 925 size_t os::Posix::default_stack_size(os::ThreadType thr_type) {
 926   // default stack size when not specified by caller is 1M (2M for LP64)
 927   size_t s = (BytesPerWord >> 2) * K * K;
 928   return s;
 929 }
 930 
 931 bool os::create_thread(Thread* thread, ThreadType thr_type,
 932                        size_t req_stack_size) {
 933   // Allocate the OSThread object
 934   OSThread* osthread = new OSThread(NULL, NULL);
 935   if (osthread == NULL) {
 936     return false;
 937   }
 938 
 939   if (ThreadPriorityVerbose) {
 940     char *thrtyp;
 941     switch (thr_type) {
 942     case vm_thread:
 943       thrtyp = (char *)"vm";
 944       break;
 945     case cgc_thread:
 946       thrtyp = (char *)"cgc";
 947       break;
 948     case pgc_thread:
 949       thrtyp = (char *)"pgc";
 950       break;
 951     case java_thread:
 952       thrtyp = (char *)"java";
 953       break;
 954     case compiler_thread:
 955       thrtyp = (char *)"compiler";
 956       break;
 957     case watcher_thread:
 958       thrtyp = (char *)"watcher";
 959       break;
 960     default:
 961       thrtyp = (char *)"unknown";
 962       break;
 963     }
 964     tty->print_cr("In create_thread, creating a %s thread\n", thrtyp);
 965   }
 966 
 967   // calculate stack size if it's not specified by caller
 968   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 969 
 970   // Initial state is ALLOCATED but not INITIALIZED
 971   osthread->set_state(ALLOCATED);
 972 
 973   if (os::Solaris::_os_thread_count > os::Solaris::_os_thread_limit) {
 974     // We got lots of threads. Check if we still have some address space left.
 975     // Need to be at least 5Mb of unreserved address space. We do check by
 976     // trying to reserve some.
 977     const size_t VirtualMemoryBangSize = 20*K*K;
 978     char* mem = os::reserve_memory(VirtualMemoryBangSize);
 979     if (mem == NULL) {
 980       delete osthread;
 981       return false;
 982     } else {
 983       // Release the memory again
 984       os::release_memory(mem, VirtualMemoryBangSize);
 985     }
 986   }
 987 
 988   // Setup osthread because the child thread may need it.
 989   thread->set_osthread(osthread);
 990 
 991   // Create the Solaris thread
 992   thread_t tid = 0;
 993   long     flags = (UseDetachedThreads ? THR_DETACHED : 0) | THR_SUSPENDED;
 994   int      status;
 995 
 996   // Mark that we don't have an lwp or thread id yet.
 997   // In case we attempt to set the priority before the thread starts.
 998   osthread->set_lwp_id(-1);
 999   osthread->set_thread_id(-1);
1000 
1001   status = thr_create(NULL, stack_size, thread_native_entry, thread, flags, &tid);
1002 
1003   char buf[64];
1004   if (status == 0) {
1005     log_info(os, thread)("Thread started (tid: " UINTX_FORMAT ", attributes: %s). ",
1006       (uintx) tid, describe_thr_create_attributes(buf, sizeof(buf), stack_size, flags));
1007   } else {
1008     log_warning(os, thread)("Failed to start thread - thr_create failed (%s) for attributes: %s.",
1009       os::errno_name(status), describe_thr_create_attributes(buf, sizeof(buf), stack_size, flags));
1010     // Log some OS information which might explain why creating the thread failed.
1011     log_info(os, thread)("Number of threads approx. running in the VM: %d", Threads::number_of_threads());
1012     LogStream st(Log(os, thread)::info());
1013     os::Posix::print_rlimit_info(&st);
1014     os::print_memory_info(&st);
1015   }
1016 
1017   if (status != 0) {
1018     thread->set_osthread(NULL);
1019     // Need to clean up stuff we've allocated so far
1020     delete osthread;
1021     return false;
1022   }
1023 
1024   Atomic::inc(&os::Solaris::_os_thread_count);
1025 
1026   // Store info on the Solaris thread into the OSThread
1027   osthread->set_thread_id(tid);
1028 
1029   // Remember that we created this thread so we can set priority on it
1030   osthread->set_vm_created();
1031 
1032   // Most thread types will set an explicit priority before starting the thread,
1033   // but for those that don't we need a valid value to read back in thread_native_entry.
1034   osthread->set_native_priority(NormPriority);
1035 
1036   // Initial thread state is INITIALIZED, not SUSPENDED
1037   osthread->set_state(INITIALIZED);
1038 
1039   // The thread is returned suspended (in state INITIALIZED), and is started higher up in the call chain
1040   return true;
1041 }
1042 
1043 debug_only(static bool signal_sets_initialized = false);
1044 static sigset_t unblocked_sigs, vm_sigs;
1045 
1046 void os::Solaris::signal_sets_init() {
1047   // Should also have an assertion stating we are still single-threaded.
1048   assert(!signal_sets_initialized, "Already initialized");
1049   // Fill in signals that are necessarily unblocked for all threads in
1050   // the VM. Currently, we unblock the following signals:
1051   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
1052   //                         by -Xrs (=ReduceSignalUsage));
1053   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
1054   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
1055   // the dispositions or masks wrt these signals.
1056   // Programs embedding the VM that want to use the above signals for their
1057   // own purposes must, at this time, use the "-Xrs" option to prevent
1058   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
1059   // (See bug 4345157, and other related bugs).
1060   // In reality, though, unblocking these signals is really a nop, since
1061   // these signals are not blocked by default.
1062   sigemptyset(&unblocked_sigs);
1063   sigaddset(&unblocked_sigs, SIGILL);
1064   sigaddset(&unblocked_sigs, SIGSEGV);
1065   sigaddset(&unblocked_sigs, SIGBUS);
1066   sigaddset(&unblocked_sigs, SIGFPE);
1067   sigaddset(&unblocked_sigs, ASYNC_SIGNAL);
1068 
1069   if (!ReduceSignalUsage) {
1070     if (!os::Posix::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
1071       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
1072     }
1073     if (!os::Posix::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
1074       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
1075     }
1076     if (!os::Posix::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
1077       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
1078     }
1079   }
1080   // Fill in signals that are blocked by all but the VM thread.
1081   sigemptyset(&vm_sigs);
1082   if (!ReduceSignalUsage) {
1083     sigaddset(&vm_sigs, BREAK_SIGNAL);
1084   }
1085   debug_only(signal_sets_initialized = true);
1086 
1087   // For diagnostics only used in run_periodic_checks
1088   sigemptyset(&check_signal_done);
1089 }
1090 
1091 // These are signals that are unblocked while a thread is running Java.
1092 // (For some reason, they get blocked by default.)
1093 sigset_t* os::Solaris::unblocked_signals() {
1094   assert(signal_sets_initialized, "Not initialized");
1095   return &unblocked_sigs;
1096 }
1097 
1098 // These are the signals that are blocked while a (non-VM) thread is
1099 // running Java. Only the VM thread handles these signals.
1100 sigset_t* os::Solaris::vm_signals() {
1101   assert(signal_sets_initialized, "Not initialized");
1102   return &vm_sigs;
1103 }
1104 
1105 // CR 7190089: on Solaris, primordial thread's stack needs adjusting.
1106 // Without the adjustment, stack size is incorrect if stack is set to unlimited (ulimit -s unlimited).
1107 void os::Solaris::correct_stack_boundaries_for_primordial_thread(Thread* thr) {
1108   assert(is_primordial_thread(), "Call only for primordial thread");
1109 
1110   JavaThread* jt = (JavaThread *)thr;
1111   assert(jt != NULL, "Sanity check");
1112   size_t stack_size;
1113   address base = jt->stack_base();
1114   if (Arguments::created_by_java_launcher()) {
1115     // Use 2MB to allow for Solaris 7 64 bit mode.
1116     stack_size = JavaThread::stack_size_at_create() == 0
1117       ? 2048*K : JavaThread::stack_size_at_create();
1118 
1119     // There are rare cases when we may have already used more than
1120     // the basic stack size allotment before this method is invoked.
1121     // Attempt to allow for a normally sized java_stack.
1122     size_t current_stack_offset = (size_t)(base - (address)&stack_size);
1123     stack_size += ReservedSpace::page_align_size_down(current_stack_offset);
1124   } else {
1125     // 6269555: If we were not created by a Java launcher, i.e. if we are
1126     // running embedded in a native application, treat the primordial thread
1127     // as much like a native attached thread as possible.  This means using
1128     // the current stack size from thr_stksegment(), unless it is too large
1129     // to reliably setup guard pages.  A reasonable max size is 8MB.
1130     size_t current_size = os::current_stack_size();
1131     // This should never happen, but just in case....
1132     if (current_size == 0) current_size = 2 * K * K;
1133     stack_size = current_size > (8 * K * K) ? (8 * K * K) : current_size;
1134   }
1135   address bottom = align_up(base - stack_size, os::vm_page_size());;
1136   stack_size = (size_t)(base - bottom);
1137 
1138   assert(stack_size > 0, "Stack size calculation problem");
1139 
1140   if (stack_size > jt->stack_size()) {
1141 #ifndef PRODUCT
1142     struct rlimit limits;
1143     getrlimit(RLIMIT_STACK, &limits);
1144     size_t size = adjust_stack_size(base, (size_t)limits.rlim_cur);
1145     assert(size >= jt->stack_size(), "Stack size problem in main thread");
1146 #endif
1147     tty->print_cr("Stack size of %d Kb exceeds current limit of %d Kb.\n"
1148                   "(Stack sizes are rounded up to a multiple of the system page size.)\n"
1149                   "See limit(1) to increase the stack size limit.",
1150                   stack_size / K, jt->stack_size() / K);
1151     vm_exit(1);
1152   }
1153   assert(jt->stack_size() >= stack_size,
1154          "Attempt to map more stack than was allocated");
1155   jt->set_stack_size(stack_size);
1156 
1157 }
1158 
1159 
1160 
1161 // Free Solaris resources related to the OSThread
1162 void os::free_thread(OSThread* osthread) {
1163   assert(osthread != NULL, "os::free_thread but osthread not set");
1164 
1165   // We are told to free resources of the argument thread,
1166   // but we can only really operate on the current thread.
1167   assert(Thread::current()->osthread() == osthread,
1168          "os::free_thread but not current thread");
1169 
1170   // Restore caller's signal mask
1171   sigset_t sigmask = osthread->caller_sigmask();
1172   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1173 
1174   delete osthread;
1175 }
1176 
1177 void os::pd_start_thread(Thread* thread) {
1178   int status = thr_continue(thread->osthread()->thread_id());
1179   assert_status(status == 0, status, "thr_continue failed");
1180 }
1181 
1182 
1183 intx os::current_thread_id() {
1184   return (intx)thr_self();
1185 }
1186 
1187 static pid_t _initial_pid = 0;
1188 
1189 int os::current_process_id() {
1190   return (int)(_initial_pid ? _initial_pid : getpid());
1191 }
1192 
1193 // gethrtime() should be monotonic according to the documentation,
1194 // but some virtualized platforms are known to break this guarantee.
1195 // getTimeNanos() must be guaranteed not to move backwards, so we
1196 // are forced to add a check here.
1197 inline hrtime_t getTimeNanos() {
1198   const hrtime_t now = gethrtime();
1199   const hrtime_t prev = max_hrtime;
1200   if (now <= prev) {
1201     return prev;   // same or retrograde time;
1202   }
1203   const hrtime_t obsv = Atomic::cmpxchg(now, &max_hrtime, prev);
1204   assert(obsv >= prev, "invariant");   // Monotonicity
1205   // If the CAS succeeded then we're done and return "now".
1206   // If the CAS failed and the observed value "obsv" is >= now then
1207   // we should return "obsv".  If the CAS failed and now > obsv > prv then
1208   // some other thread raced this thread and installed a new value, in which case
1209   // we could either (a) retry the entire operation, (b) retry trying to install now
1210   // or (c) just return obsv.  We use (c).   No loop is required although in some cases
1211   // we might discard a higher "now" value in deference to a slightly lower but freshly
1212   // installed obsv value.   That's entirely benign -- it admits no new orderings compared
1213   // to (a) or (b) -- and greatly reduces coherence traffic.
1214   // We might also condition (c) on the magnitude of the delta between obsv and now.
1215   // Avoiding excessive CAS operations to hot RW locations is critical.
1216   // See https://blogs.oracle.com/dave/entry/cas_and_cache_trivia_invalidate
1217   return (prev == obsv) ? now : obsv;
1218 }
1219 
1220 // Time since start-up in seconds to a fine granularity.
1221 // Used by VMSelfDestructTimer and the MemProfiler.
1222 double os::elapsedTime() {
1223   return (double)(getTimeNanos() - first_hrtime) / (double)hrtime_hz;
1224 }
1225 
1226 jlong os::elapsed_counter() {
1227   return (jlong)(getTimeNanos() - first_hrtime);
1228 }
1229 
1230 jlong os::elapsed_frequency() {
1231   return hrtime_hz;
1232 }
1233 
1234 // Return the real, user, and system times in seconds from an
1235 // arbitrary fixed point in the past.
1236 bool os::getTimesSecs(double* process_real_time,
1237                       double* process_user_time,
1238                       double* process_system_time) {
1239   struct tms ticks;
1240   clock_t real_ticks = times(&ticks);
1241 
1242   if (real_ticks == (clock_t) (-1)) {
1243     return false;
1244   } else {
1245     double ticks_per_second = (double) clock_tics_per_sec;
1246     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1247     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1248     // For consistency return the real time from getTimeNanos()
1249     // converted to seconds.
1250     *process_real_time = ((double) getTimeNanos()) / ((double) NANOUNITS);
1251 
1252     return true;
1253   }
1254 }
1255 
1256 bool os::supports_vtime() { return true; }
1257 bool os::enable_vtime() { return false; }
1258 bool os::vtime_enabled() { return false; }
1259 
1260 double os::elapsedVTime() {
1261   return (double)gethrvtime() / (double)hrtime_hz;
1262 }
1263 
1264 // Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1265 jlong os::javaTimeMillis() {
1266   timeval t;
1267   if (gettimeofday(&t, NULL) == -1) {
1268     fatal("os::javaTimeMillis: gettimeofday (%s)", os::strerror(errno));
1269   }
1270   return jlong(t.tv_sec) * 1000  +  jlong(t.tv_usec) / 1000;
1271 }
1272 
1273 // Must return seconds+nanos since Jan 1 1970. This must use the same
1274 // time source as javaTimeMillis and can't use get_nsec_fromepoch as
1275 // we need better than 1ms accuracy
1276 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1277   timeval t;
1278   if (gettimeofday(&t, NULL) == -1) {
1279     fatal("os::javaTimeSystemUTC: gettimeofday (%s)", os::strerror(errno));
1280   }
1281   seconds = jlong(t.tv_sec);
1282   nanos = jlong(t.tv_usec) * 1000;
1283 }
1284 
1285 
1286 jlong os::javaTimeNanos() {
1287   return (jlong)getTimeNanos();
1288 }
1289 
1290 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1291   info_ptr->max_value = ALL_64_BITS;      // gethrtime() uses all 64 bits
1292   info_ptr->may_skip_backward = false;    // not subject to resetting or drifting
1293   info_ptr->may_skip_forward = false;     // not subject to resetting or drifting
1294   info_ptr->kind = JVMTI_TIMER_ELAPSED;   // elapsed not CPU time
1295 }
1296 
1297 char * os::local_time_string(char *buf, size_t buflen) {
1298   struct tm t;
1299   time_t long_time;
1300   time(&long_time);
1301   localtime_r(&long_time, &t);
1302   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1303                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1304                t.tm_hour, t.tm_min, t.tm_sec);
1305   return buf;
1306 }
1307 
1308 // Note: os::shutdown() might be called very early during initialization, or
1309 // called from signal handler. Before adding something to os::shutdown(), make
1310 // sure it is async-safe and can handle partially initialized VM.
1311 void os::shutdown() {
1312 
1313   // allow PerfMemory to attempt cleanup of any persistent resources
1314   perfMemory_exit();
1315 
1316   // needs to remove object in file system
1317   AttachListener::abort();
1318 
1319   // flush buffered output, finish log files
1320   ostream_abort();
1321 
1322   // Check for abort hook
1323   abort_hook_t abort_hook = Arguments::abort_hook();
1324   if (abort_hook != NULL) {
1325     abort_hook();
1326   }
1327 }
1328 
1329 // Note: os::abort() might be called very early during initialization, or
1330 // called from signal handler. Before adding something to os::abort(), make
1331 // sure it is async-safe and can handle partially initialized VM.
1332 void os::abort(bool dump_core, void* siginfo, const void* context) {
1333   os::shutdown();
1334   if (dump_core) {
1335 #ifndef PRODUCT
1336     fdStream out(defaultStream::output_fd());
1337     out.print_raw("Current thread is ");
1338     char buf[16];
1339     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1340     out.print_raw_cr(buf);
1341     out.print_raw_cr("Dumping core ...");
1342 #endif
1343     ::abort(); // dump core (for debugging)
1344   }
1345 
1346   ::exit(1);
1347 }
1348 
1349 // Die immediately, no exit hook, no abort hook, no cleanup.
1350 void os::die() {
1351   ::abort(); // dump core (for debugging)
1352 }
1353 
1354 // DLL functions
1355 
1356 const char* os::dll_file_extension() { return ".so"; }
1357 
1358 // This must be hard coded because it's the system's temporary
1359 // directory not the java application's temp directory, ala java.io.tmpdir.
1360 const char* os::get_temp_directory() { return "/tmp"; }
1361 
1362 // check if addr is inside libjvm.so
1363 bool os::address_is_in_vm(address addr) {
1364   static address libjvm_base_addr;
1365   Dl_info dlinfo;
1366 
1367   if (libjvm_base_addr == NULL) {
1368     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1369       libjvm_base_addr = (address)dlinfo.dli_fbase;
1370     }
1371     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1372   }
1373 
1374   if (dladdr((void *)addr, &dlinfo) != 0) {
1375     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1376   }
1377 
1378   return false;
1379 }
1380 
1381 typedef int (*dladdr1_func_type)(void *, Dl_info *, void **, int);
1382 static dladdr1_func_type dladdr1_func = NULL;
1383 
1384 bool os::dll_address_to_function_name(address addr, char *buf,
1385                                       int buflen, int * offset,
1386                                       bool demangle) {
1387   // buf is not optional, but offset is optional
1388   assert(buf != NULL, "sanity check");
1389 
1390   Dl_info dlinfo;
1391 
1392   // dladdr1_func was initialized in os::init()
1393   if (dladdr1_func != NULL) {
1394     // yes, we have dladdr1
1395 
1396     // Support for dladdr1 is checked at runtime; it may be
1397     // available even if the vm is built on a machine that does
1398     // not have dladdr1 support.  Make sure there is a value for
1399     // RTLD_DL_SYMENT.
1400 #ifndef RTLD_DL_SYMENT
1401   #define RTLD_DL_SYMENT 1
1402 #endif
1403 #ifdef _LP64
1404     Elf64_Sym * info;
1405 #else
1406     Elf32_Sym * info;
1407 #endif
1408     if (dladdr1_func((void *)addr, &dlinfo, (void **)&info,
1409                      RTLD_DL_SYMENT) != 0) {
1410       // see if we have a matching symbol that covers our address
1411       if (dlinfo.dli_saddr != NULL &&
1412           (char *)dlinfo.dli_saddr + info->st_size > (char *)addr) {
1413         if (dlinfo.dli_sname != NULL) {
1414           if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1415             jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1416           }
1417           if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1418           return true;
1419         }
1420       }
1421       // no matching symbol so try for just file info
1422       if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1423         if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1424                             buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1425           return true;
1426         }
1427       }
1428     }
1429     buf[0] = '\0';
1430     if (offset != NULL) *offset  = -1;
1431     return false;
1432   }
1433 
1434   // no, only dladdr is available
1435   if (dladdr((void *)addr, &dlinfo) != 0) {
1436     // see if we have a matching symbol
1437     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1438       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1439         jio_snprintf(buf, buflen, dlinfo.dli_sname);
1440       }
1441       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1442       return true;
1443     }
1444     // no matching symbol so try for just file info
1445     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1446       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1447                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1448         return true;
1449       }
1450     }
1451   }
1452   buf[0] = '\0';
1453   if (offset != NULL) *offset  = -1;
1454   return false;
1455 }
1456 
1457 bool os::dll_address_to_library_name(address addr, char* buf,
1458                                      int buflen, int* offset) {
1459   // buf is not optional, but offset is optional
1460   assert(buf != NULL, "sanity check");
1461 
1462   Dl_info dlinfo;
1463 
1464   if (dladdr((void*)addr, &dlinfo) != 0) {
1465     if (dlinfo.dli_fname != NULL) {
1466       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1467     }
1468     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1469       *offset = addr - (address)dlinfo.dli_fbase;
1470     }
1471     return true;
1472   }
1473 
1474   buf[0] = '\0';
1475   if (offset) *offset = -1;
1476   return false;
1477 }
1478 
1479 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1480   Dl_info dli;
1481   // Sanity check?
1482   if (dladdr(CAST_FROM_FN_PTR(void *, os::get_loaded_modules_info), &dli) == 0 ||
1483       dli.dli_fname == NULL) {
1484     return 1;
1485   }
1486 
1487   void * handle = dlopen(dli.dli_fname, RTLD_LAZY);
1488   if (handle == NULL) {
1489     return 1;
1490   }
1491 
1492   Link_map *map;
1493   dlinfo(handle, RTLD_DI_LINKMAP, &map);
1494   if (map == NULL) {
1495     dlclose(handle);
1496     return 1;
1497   }
1498 
1499   while (map->l_prev != NULL) {
1500     map = map->l_prev;
1501   }
1502 
1503   while (map != NULL) {
1504     // Iterate through all map entries and call callback with fields of interest
1505     if(callback(map->l_name, (address)map->l_addr, (address)0, param)) {
1506       dlclose(handle);
1507       return 1;
1508     }
1509     map = map->l_next;
1510   }
1511 
1512   dlclose(handle);
1513   return 0;
1514 }
1515 
1516 int _print_dll_info_cb(const char * name, address base_address, address top_address, void * param) {
1517   outputStream * out = (outputStream *) param;
1518   out->print_cr(PTR_FORMAT " \t%s", base_address, name);
1519   return 0;
1520 }
1521 
1522 void os::print_dll_info(outputStream * st) {
1523   st->print_cr("Dynamic libraries:"); st->flush();
1524   if (get_loaded_modules_info(_print_dll_info_cb, (void *)st)) {
1525     st->print_cr("Error: Cannot print dynamic libraries.");
1526   }
1527 }
1528 
1529 // Loads .dll/.so and
1530 // in case of error it checks if .dll/.so was built for the
1531 // same architecture as Hotspot is running on
1532 
1533 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1534   void * result= ::dlopen(filename, RTLD_LAZY);
1535   if (result != NULL) {
1536     // Successful loading
1537     return result;
1538   }
1539 
1540   Elf32_Ehdr elf_head;
1541 
1542   // Read system error message into ebuf
1543   // It may or may not be overwritten below
1544   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1545   ebuf[ebuflen-1]='\0';
1546   int diag_msg_max_length=ebuflen-strlen(ebuf);
1547   char* diag_msg_buf=ebuf+strlen(ebuf);
1548 
1549   if (diag_msg_max_length==0) {
1550     // No more space in ebuf for additional diagnostics message
1551     return NULL;
1552   }
1553 
1554 
1555   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1556 
1557   if (file_descriptor < 0) {
1558     // Can't open library, report dlerror() message
1559     return NULL;
1560   }
1561 
1562   bool failed_to_read_elf_head=
1563     (sizeof(elf_head)!=
1564      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1565 
1566   ::close(file_descriptor);
1567   if (failed_to_read_elf_head) {
1568     // file i/o error - report dlerror() msg
1569     return NULL;
1570   }
1571 
1572   typedef struct {
1573     Elf32_Half  code;         // Actual value as defined in elf.h
1574     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1575     char        elf_class;    // 32 or 64 bit
1576     char        endianess;    // MSB or LSB
1577     char*       name;         // String representation
1578   } arch_t;
1579 
1580   static const arch_t arch_array[]={
1581     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1582     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1583     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1584     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1585     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1586     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1587     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1588     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1589     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1590     {EM_ARM,         EM_ARM,     ELFCLASS32, ELFDATA2LSB, (char*)"ARM 32"}
1591   };
1592 
1593 #if  (defined IA32)
1594   static  Elf32_Half running_arch_code=EM_386;
1595 #elif   (defined AMD64)
1596   static  Elf32_Half running_arch_code=EM_X86_64;
1597 #elif  (defined IA64)
1598   static  Elf32_Half running_arch_code=EM_IA_64;
1599 #elif  (defined __sparc) && (defined _LP64)
1600   static  Elf32_Half running_arch_code=EM_SPARCV9;
1601 #elif  (defined __sparc) && (!defined _LP64)
1602   static  Elf32_Half running_arch_code=EM_SPARC;
1603 #elif  (defined __powerpc64__)
1604   static  Elf32_Half running_arch_code=EM_PPC64;
1605 #elif  (defined __powerpc__)
1606   static  Elf32_Half running_arch_code=EM_PPC;
1607 #elif (defined ARM)
1608   static  Elf32_Half running_arch_code=EM_ARM;
1609 #else
1610   #error Method os::dll_load requires that one of following is defined:\
1611        IA32, AMD64, IA64, __sparc, __powerpc__, ARM, ARM
1612 #endif
1613 
1614   // Identify compatability class for VM's architecture and library's architecture
1615   // Obtain string descriptions for architectures
1616 
1617   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1618   int running_arch_index=-1;
1619 
1620   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1621     if (running_arch_code == arch_array[i].code) {
1622       running_arch_index    = i;
1623     }
1624     if (lib_arch.code == arch_array[i].code) {
1625       lib_arch.compat_class = arch_array[i].compat_class;
1626       lib_arch.name         = arch_array[i].name;
1627     }
1628   }
1629 
1630   assert(running_arch_index != -1,
1631          "Didn't find running architecture code (running_arch_code) in arch_array");
1632   if (running_arch_index == -1) {
1633     // Even though running architecture detection failed
1634     // we may still continue with reporting dlerror() message
1635     return NULL;
1636   }
1637 
1638   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1639     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1640     return NULL;
1641   }
1642 
1643   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1644     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1645     return NULL;
1646   }
1647 
1648   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1649     if (lib_arch.name!=NULL) {
1650       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1651                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1652                  lib_arch.name, arch_array[running_arch_index].name);
1653     } else {
1654       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1655                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1656                  lib_arch.code,
1657                  arch_array[running_arch_index].name);
1658     }
1659   }
1660 
1661   return NULL;
1662 }
1663 
1664 void* os::dll_lookup(void* handle, const char* name) {
1665   return dlsym(handle, name);
1666 }
1667 
1668 void* os::get_default_process_handle() {
1669   return (void*)::dlopen(NULL, RTLD_LAZY);
1670 }
1671 
1672 static inline time_t get_mtime(const char* filename) {
1673   struct stat st;
1674   int ret = os::stat(filename, &st);
1675   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
1676   return st.st_mtime;
1677 }
1678 
1679 int os::compare_file_modified_times(const char* file1, const char* file2) {
1680   time_t t1 = get_mtime(file1);
1681   time_t t2 = get_mtime(file2);
1682   return t1 - t2;
1683 }
1684 
1685 static bool _print_ascii_file(const char* filename, outputStream* st) {
1686   int fd = ::open(filename, O_RDONLY);
1687   if (fd == -1) {
1688     return false;
1689   }
1690 
1691   char buf[32];
1692   int bytes;
1693   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1694     st->print_raw(buf, bytes);
1695   }
1696 
1697   ::close(fd);
1698 
1699   return true;
1700 }
1701 
1702 void os::print_os_info_brief(outputStream* st) {
1703   os::Solaris::print_distro_info(st);
1704 
1705   os::Posix::print_uname_info(st);
1706 
1707   os::Solaris::print_libversion_info(st);
1708 }
1709 
1710 void os::print_os_info(outputStream* st) {
1711   st->print("OS:");
1712 
1713   os::Solaris::print_distro_info(st);
1714 
1715   os::Posix::print_uname_info(st);
1716 
1717   os::Solaris::print_libversion_info(st);
1718 
1719   os::Posix::print_rlimit_info(st);
1720 
1721   os::Posix::print_load_average(st);
1722 }
1723 
1724 void os::Solaris::print_distro_info(outputStream* st) {
1725   if (!_print_ascii_file("/etc/release", st)) {
1726     st->print("Solaris");
1727   }
1728   st->cr();
1729 }
1730 
1731 void os::get_summary_os_info(char* buf, size_t buflen) {
1732   strncpy(buf, "Solaris", buflen);  // default to plain solaris
1733   FILE* fp = fopen("/etc/release", "r");
1734   if (fp != NULL) {
1735     char tmp[256];
1736     // Only get the first line and chop out everything but the os name.
1737     if (fgets(tmp, sizeof(tmp), fp)) {
1738       char* ptr = tmp;
1739       // skip past whitespace characters
1740       while (*ptr != '\0' && (*ptr == ' ' || *ptr == '\t' || *ptr == '\n')) ptr++;
1741       if (*ptr != '\0') {
1742         char* nl = strchr(ptr, '\n');
1743         if (nl != NULL) *nl = '\0';
1744         strncpy(buf, ptr, buflen);
1745       }
1746     }
1747     fclose(fp);
1748   }
1749 }
1750 
1751 void os::Solaris::print_libversion_info(outputStream* st) {
1752   st->print("  (T2 libthread)");
1753   st->cr();
1754 }
1755 
1756 static bool check_addr0(outputStream* st) {
1757   jboolean status = false;
1758   const int read_chunk = 200;
1759   int ret = 0;
1760   int nmap = 0;
1761   int fd = ::open("/proc/self/map",O_RDONLY);
1762   if (fd >= 0) {
1763     prmap_t *p = NULL;
1764     char *mbuff = (char *) calloc(read_chunk, sizeof(prmap_t));
1765     if (NULL == mbuff) {
1766       ::close(fd);
1767       return status;
1768     }
1769     while ((ret = ::read(fd, mbuff, read_chunk*sizeof(prmap_t))) > 0) {
1770       //check if read() has not read partial data
1771       if( 0 != ret % sizeof(prmap_t)){
1772         break;
1773       }
1774       nmap = ret / sizeof(prmap_t);
1775       p = (prmap_t *)mbuff;
1776       for(int i = 0; i < nmap; i++){
1777         if (p->pr_vaddr == 0x0) {
1778           st->print("Warning: Address: " PTR_FORMAT ", Size: " SIZE_FORMAT "K, ",p->pr_vaddr, p->pr_size/1024);
1779           st->print("Mapped file: %s, ", p->pr_mapname[0] == '\0' ? "None" : p->pr_mapname);
1780           st->print("Access: ");
1781           st->print("%s",(p->pr_mflags & MA_READ)  ? "r" : "-");
1782           st->print("%s",(p->pr_mflags & MA_WRITE) ? "w" : "-");
1783           st->print("%s",(p->pr_mflags & MA_EXEC)  ? "x" : "-");
1784           st->cr();
1785           status = true;
1786         }
1787         p++;
1788       }
1789     }
1790     free(mbuff);
1791     ::close(fd);
1792   }
1793   return status;
1794 }
1795 
1796 void os::get_summary_cpu_info(char* buf, size_t buflen) {
1797   // Get MHz with system call. We don't seem to already have this.
1798   processor_info_t stats;
1799   processorid_t id = getcpuid();
1800   int clock = 0;
1801   if (processor_info(id, &stats) != -1) {
1802     clock = stats.pi_clock;  // pi_processor_type isn't more informative than below
1803   }
1804 #ifdef AMD64
1805   snprintf(buf, buflen, "x86 64 bit %d MHz", clock);
1806 #else
1807   // must be sparc
1808   snprintf(buf, buflen, "Sparcv9 64 bit %d MHz", clock);
1809 #endif
1810 }
1811 
1812 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
1813   // Nothing to do for now.
1814 }
1815 
1816 void os::print_memory_info(outputStream* st) {
1817   st->print("Memory:");
1818   st->print(" %dk page", os::vm_page_size()>>10);
1819   st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
1820   st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
1821   st->cr();
1822   (void) check_addr0(st);
1823 }
1824 
1825 // Moved from whole group, because we need them here for diagnostic
1826 // prints.
1827 static int Maxsignum = 0;
1828 static int *ourSigFlags = NULL;
1829 
1830 int os::Solaris::get_our_sigflags(int sig) {
1831   assert(ourSigFlags!=NULL, "signal data structure not initialized");
1832   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
1833   return ourSigFlags[sig];
1834 }
1835 
1836 void os::Solaris::set_our_sigflags(int sig, int flags) {
1837   assert(ourSigFlags!=NULL, "signal data structure not initialized");
1838   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
1839   ourSigFlags[sig] = flags;
1840 }
1841 
1842 
1843 static const char* get_signal_handler_name(address handler,
1844                                            char* buf, int buflen) {
1845   int offset;
1846   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
1847   if (found) {
1848     // skip directory names
1849     const char *p1, *p2;
1850     p1 = buf;
1851     size_t len = strlen(os::file_separator());
1852     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
1853     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
1854   } else {
1855     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
1856   }
1857   return buf;
1858 }
1859 
1860 static void print_signal_handler(outputStream* st, int sig,
1861                                  char* buf, size_t buflen) {
1862   struct sigaction sa;
1863 
1864   sigaction(sig, NULL, &sa);
1865 
1866   st->print("%s: ", os::exception_name(sig, buf, buflen));
1867 
1868   address handler = (sa.sa_flags & SA_SIGINFO)
1869                   ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
1870                   : CAST_FROM_FN_PTR(address, sa.sa_handler);
1871 
1872   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
1873     st->print("SIG_DFL");
1874   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
1875     st->print("SIG_IGN");
1876   } else {
1877     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
1878   }
1879 
1880   st->print(", sa_mask[0]=");
1881   os::Posix::print_signal_set_short(st, &sa.sa_mask);
1882 
1883   address rh = VMError::get_resetted_sighandler(sig);
1884   // May be, handler was resetted by VMError?
1885   if (rh != NULL) {
1886     handler = rh;
1887     sa.sa_flags = VMError::get_resetted_sigflags(sig);
1888   }
1889 
1890   st->print(", sa_flags=");
1891   os::Posix::print_sa_flags(st, sa.sa_flags);
1892 
1893   // Check: is it our handler?
1894   if (handler == CAST_FROM_FN_PTR(address, signalHandler)) {
1895     // It is our signal handler
1896     // check for flags
1897     if (sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
1898       st->print(
1899                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
1900                 os::Solaris::get_our_sigflags(sig));
1901     }
1902   }
1903   st->cr();
1904 }
1905 
1906 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
1907   st->print_cr("Signal Handlers:");
1908   print_signal_handler(st, SIGSEGV, buf, buflen);
1909   print_signal_handler(st, SIGBUS , buf, buflen);
1910   print_signal_handler(st, SIGFPE , buf, buflen);
1911   print_signal_handler(st, SIGPIPE, buf, buflen);
1912   print_signal_handler(st, SIGXFSZ, buf, buflen);
1913   print_signal_handler(st, SIGILL , buf, buflen);
1914   print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
1915   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
1916   print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
1917   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
1918   print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
1919 }
1920 
1921 static char saved_jvm_path[MAXPATHLEN] = { 0 };
1922 
1923 // Find the full path to the current module, libjvm.so
1924 void os::jvm_path(char *buf, jint buflen) {
1925   // Error checking.
1926   if (buflen < MAXPATHLEN) {
1927     assert(false, "must use a large-enough buffer");
1928     buf[0] = '\0';
1929     return;
1930   }
1931   // Lazy resolve the path to current module.
1932   if (saved_jvm_path[0] != 0) {
1933     strcpy(buf, saved_jvm_path);
1934     return;
1935   }
1936 
1937   Dl_info dlinfo;
1938   int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
1939   assert(ret != 0, "cannot locate libjvm");
1940   if (ret != 0 && dlinfo.dli_fname != NULL) {
1941     if (os::Posix::realpath((char *)dlinfo.dli_fname, buf, buflen) == NULL) {
1942       return;
1943     }
1944   } else {
1945     buf[0] = '\0';
1946     return;
1947   }
1948 
1949   if (Arguments::sun_java_launcher_is_altjvm()) {
1950     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
1951     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
1952     // If "/jre/lib/" appears at the right place in the string, then
1953     // assume we are installed in a JDK and we're done.  Otherwise, check
1954     // for a JAVA_HOME environment variable and fix up the path so it
1955     // looks like libjvm.so is installed there (append a fake suffix
1956     // hotspot/libjvm.so).
1957     const char *p = buf + strlen(buf) - 1;
1958     for (int count = 0; p > buf && count < 5; ++count) {
1959       for (--p; p > buf && *p != '/'; --p)
1960         /* empty */ ;
1961     }
1962 
1963     if (strncmp(p, "/jre/lib/", 9) != 0) {
1964       // Look for JAVA_HOME in the environment.
1965       char* java_home_var = ::getenv("JAVA_HOME");
1966       if (java_home_var != NULL && java_home_var[0] != 0) {
1967         char* jrelib_p;
1968         int   len;
1969 
1970         // Check the current module name "libjvm.so".
1971         p = strrchr(buf, '/');
1972         assert(strstr(p, "/libjvm") == p, "invalid library name");
1973 
1974         if (os::Posix::realpath(java_home_var, buf, buflen) == NULL) {
1975           return;
1976         }
1977         // determine if this is a legacy image or modules image
1978         // modules image doesn't have "jre" subdirectory
1979         len = strlen(buf);
1980         assert(len < buflen, "Ran out of buffer space");
1981         jrelib_p = buf + len;
1982         snprintf(jrelib_p, buflen-len, "/jre/lib");
1983         if (0 != access(buf, F_OK)) {
1984           snprintf(jrelib_p, buflen-len, "/lib");
1985         }
1986 
1987         if (0 == access(buf, F_OK)) {
1988           // Use current module name "libjvm.so"
1989           len = strlen(buf);
1990           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
1991         } else {
1992           // Go back to path of .so
1993           if (os::Posix::realpath((char *)dlinfo.dli_fname, buf, buflen) == NULL) {
1994             return;
1995           }
1996         }
1997       }
1998     }
1999   }
2000 
2001   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2002   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2003 }
2004 
2005 
2006 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2007   // no prefix required, not even "_"
2008 }
2009 
2010 
2011 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2012   // no suffix required
2013 }
2014 
2015 // This method is a copy of JDK's sysGetLastErrorString
2016 // from src/solaris/hpi/src/system_md.c
2017 
2018 size_t os::lasterror(char *buf, size_t len) {
2019   if (errno == 0)  return 0;
2020 
2021   const char *s = os::strerror(errno);
2022   size_t n = ::strlen(s);
2023   if (n >= len) {
2024     n = len - 1;
2025   }
2026   ::strncpy(buf, s, n);
2027   buf[n] = '\0';
2028   return n;
2029 }
2030 
2031 
2032 // sun.misc.Signal
2033 
2034 extern "C" {
2035   static void UserHandler(int sig, void *siginfo, void *context) {
2036     // Ctrl-C is pressed during error reporting, likely because the error
2037     // handler fails to abort. Let VM die immediately.
2038     if (sig == SIGINT && VMError::is_error_reported()) {
2039       os::die();
2040     }
2041 
2042     os::signal_notify(sig);
2043     // We do not need to reinstate the signal handler each time...
2044   }
2045 }
2046 
2047 void* os::user_handler() {
2048   return CAST_FROM_FN_PTR(void*, UserHandler);
2049 }
2050 
2051 static struct timespec create_semaphore_timespec(unsigned int sec, int nsec) {
2052   struct timespec ts;
2053   unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2054 
2055   return ts;
2056 }
2057 
2058 extern "C" {
2059   typedef void (*sa_handler_t)(int);
2060   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2061 }
2062 
2063 void* os::signal(int signal_number, void* handler) {
2064   struct sigaction sigAct, oldSigAct;
2065   sigfillset(&(sigAct.sa_mask));
2066   sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2067   sigAct.sa_flags |= SA_SIGINFO;
2068   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2069 
2070   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2071     // -1 means registration failed
2072     return (void *)-1;
2073   }
2074 
2075   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2076 }
2077 
2078 void os::signal_raise(int signal_number) {
2079   raise(signal_number);
2080 }
2081 
2082 // The following code is moved from os.cpp for making this
2083 // code platform specific, which it is by its very nature.
2084 
2085 // a counter for each possible signal value
2086 static int Sigexit = 0;
2087 static jint *pending_signals = NULL;
2088 static int *preinstalled_sigs = NULL;
2089 static struct sigaction *chainedsigactions = NULL;
2090 static Semaphore* sig_sem = NULL;
2091 
2092 int os::sigexitnum_pd() {
2093   assert(Sigexit > 0, "signal memory not yet initialized");
2094   return Sigexit;
2095 }
2096 
2097 void os::Solaris::init_signal_mem() {
2098   // Initialize signal structures
2099   Maxsignum = SIGRTMAX;
2100   Sigexit = Maxsignum+1;
2101   assert(Maxsignum >0, "Unable to obtain max signal number");
2102 
2103   // Initialize signal structures
2104   // pending_signals has one int per signal
2105   // The additional signal is for SIGEXIT - exit signal to signal_thread
2106   pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1), mtInternal);
2107   memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2108 
2109   if (UseSignalChaining) {
2110     chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2111                                                    * (Maxsignum + 1), mtInternal);
2112     memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2113     preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2114     memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2115   }
2116   ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2117   memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2118 }
2119 
2120 static void jdk_misc_signal_init() {
2121   // Initialize signal semaphore
2122   sig_sem = new Semaphore();
2123 }
2124 
2125 void os::signal_notify(int sig) {
2126   if (sig_sem != NULL) {
2127     Atomic::inc(&pending_signals[sig]);
2128     sig_sem->signal();
2129   } else {
2130     // Signal thread is not created with ReduceSignalUsage and jdk_misc_signal_init
2131     // initialization isn't called.
2132     assert(ReduceSignalUsage, "signal semaphore should be created");
2133   }
2134 }
2135 
2136 static int check_pending_signals() {
2137   int ret;
2138   while (true) {
2139     for (int i = 0; i < Sigexit + 1; i++) {
2140       jint n = pending_signals[i];
2141       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2142         return i;
2143       }
2144     }
2145     JavaThread *thread = JavaThread::current();
2146     ThreadBlockInVM tbivm(thread);
2147 
2148     bool threadIsSuspended;
2149     do {
2150       thread->set_suspend_equivalent();
2151       sig_sem->wait();
2152 
2153       // were we externally suspended while we were waiting?
2154       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2155       if (threadIsSuspended) {
2156         // The semaphore has been incremented, but while we were waiting
2157         // another thread suspended us. We don't want to continue running
2158         // while suspended because that would surprise the thread that
2159         // suspended us.
2160         sig_sem->signal();
2161 
2162         thread->java_suspend_self();
2163       }
2164     } while (threadIsSuspended);
2165   }
2166 }
2167 
2168 int os::signal_wait() {
2169   return check_pending_signals();
2170 }
2171 
2172 ////////////////////////////////////////////////////////////////////////////////
2173 // Virtual Memory
2174 
2175 static int page_size = -1;
2176 
2177 int os::vm_page_size() {
2178   assert(page_size != -1, "must call os::init");
2179   return page_size;
2180 }
2181 
2182 // Solaris allocates memory by pages.
2183 int os::vm_allocation_granularity() {
2184   assert(page_size != -1, "must call os::init");
2185   return page_size;
2186 }
2187 
2188 static bool recoverable_mmap_error(int err) {
2189   // See if the error is one we can let the caller handle. This
2190   // list of errno values comes from the Solaris mmap(2) man page.
2191   switch (err) {
2192   case EBADF:
2193   case EINVAL:
2194   case ENOTSUP:
2195     // let the caller deal with these errors
2196     return true;
2197 
2198   default:
2199     // Any remaining errors on this OS can cause our reserved mapping
2200     // to be lost. That can cause confusion where different data
2201     // structures think they have the same memory mapped. The worst
2202     // scenario is if both the VM and a library think they have the
2203     // same memory mapped.
2204     return false;
2205   }
2206 }
2207 
2208 static void warn_fail_commit_memory(char* addr, size_t bytes, bool exec,
2209                                     int err) {
2210   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2211           ", %d) failed; error='%s' (errno=%d)", addr, bytes, exec,
2212           os::strerror(err), err);
2213 }
2214 
2215 static void warn_fail_commit_memory(char* addr, size_t bytes,
2216                                     size_t alignment_hint, bool exec,
2217                                     int err) {
2218   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2219           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, bytes,
2220           alignment_hint, exec, os::strerror(err), err);
2221 }
2222 
2223 int os::Solaris::commit_memory_impl(char* addr, size_t bytes, bool exec) {
2224   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2225   size_t size = bytes;
2226   char *res = Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2227   if (res != NULL) {
2228     if (UseNUMAInterleaving) {
2229       numa_make_global(addr, bytes);
2230     }
2231     return 0;
2232   }
2233 
2234   int err = errno;  // save errno from mmap() call in mmap_chunk()
2235 
2236   if (!recoverable_mmap_error(err)) {
2237     warn_fail_commit_memory(addr, bytes, exec, err);
2238     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "committing reserved memory.");
2239   }
2240 
2241   return err;
2242 }
2243 
2244 bool os::pd_commit_memory(char* addr, size_t bytes, bool exec) {
2245   return Solaris::commit_memory_impl(addr, bytes, exec) == 0;
2246 }
2247 
2248 void os::pd_commit_memory_or_exit(char* addr, size_t bytes, bool exec,
2249                                   const char* mesg) {
2250   assert(mesg != NULL, "mesg must be specified");
2251   int err = os::Solaris::commit_memory_impl(addr, bytes, exec);
2252   if (err != 0) {
2253     // the caller wants all commit errors to exit with the specified mesg:
2254     warn_fail_commit_memory(addr, bytes, exec, err);
2255     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2256   }
2257 }
2258 
2259 size_t os::Solaris::page_size_for_alignment(size_t alignment) {
2260   assert(is_aligned(alignment, (size_t) vm_page_size()),
2261          SIZE_FORMAT " is not aligned to " SIZE_FORMAT,
2262          alignment, (size_t) vm_page_size());
2263 
2264   for (int i = 0; _page_sizes[i] != 0; i++) {
2265     if (is_aligned(alignment, _page_sizes[i])) {
2266       return _page_sizes[i];
2267     }
2268   }
2269 
2270   return (size_t) vm_page_size();
2271 }
2272 
2273 int os::Solaris::commit_memory_impl(char* addr, size_t bytes,
2274                                     size_t alignment_hint, bool exec) {
2275   int err = Solaris::commit_memory_impl(addr, bytes, exec);
2276   if (err == 0 && UseLargePages && alignment_hint > 0) {
2277     assert(is_aligned(bytes, alignment_hint),
2278            SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, alignment_hint);
2279 
2280     // The syscall memcntl requires an exact page size (see man memcntl for details).
2281     size_t page_size = page_size_for_alignment(alignment_hint);
2282     if (page_size > (size_t) vm_page_size()) {
2283       (void)Solaris::setup_large_pages(addr, bytes, page_size);
2284     }
2285   }
2286   return err;
2287 }
2288 
2289 bool os::pd_commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2290                           bool exec) {
2291   return Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec) == 0;
2292 }
2293 
2294 void os::pd_commit_memory_or_exit(char* addr, size_t bytes,
2295                                   size_t alignment_hint, bool exec,
2296                                   const char* mesg) {
2297   assert(mesg != NULL, "mesg must be specified");
2298   int err = os::Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec);
2299   if (err != 0) {
2300     // the caller wants all commit errors to exit with the specified mesg:
2301     warn_fail_commit_memory(addr, bytes, alignment_hint, exec, err);
2302     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2303   }
2304 }
2305 
2306 // Uncommit the pages in a specified region.
2307 void os::pd_free_memory(char* addr, size_t bytes, size_t alignment_hint) {
2308   if (madvise(addr, bytes, MADV_FREE) < 0) {
2309     debug_only(warning("MADV_FREE failed."));
2310     return;
2311   }
2312 }
2313 
2314 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2315   return os::commit_memory(addr, size, !ExecMem);
2316 }
2317 
2318 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2319   return os::uncommit_memory(addr, size);
2320 }
2321 
2322 // Change the page size in a given range.
2323 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2324   assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2325   assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2326   if (UseLargePages) {
2327     size_t page_size = Solaris::page_size_for_alignment(alignment_hint);
2328     if (page_size > (size_t) vm_page_size()) {
2329       Solaris::setup_large_pages(addr, bytes, page_size);
2330     }
2331   }
2332 }
2333 
2334 // Tell the OS to make the range local to the first-touching LWP
2335 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2336   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2337   if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2338     debug_only(warning("MADV_ACCESS_LWP failed."));
2339   }
2340 }
2341 
2342 // Tell the OS that this range would be accessed from different LWPs.
2343 void os::numa_make_global(char *addr, size_t bytes) {
2344   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2345   if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2346     debug_only(warning("MADV_ACCESS_MANY failed."));
2347   }
2348 }
2349 
2350 // Get the number of the locality groups.
2351 size_t os::numa_get_groups_num() {
2352   size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2353   return n != -1 ? n : 1;
2354 }
2355 
2356 // Get a list of leaf locality groups. A leaf lgroup is group that
2357 // doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2358 // board. An LWP is assigned to one of these groups upon creation.
2359 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2360   if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2361     ids[0] = 0;
2362     return 1;
2363   }
2364   int result_size = 0, top = 1, bottom = 0, cur = 0;
2365   for (int k = 0; k < size; k++) {
2366     int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2367                                    (Solaris::lgrp_id_t*)&ids[top], size - top);
2368     if (r == -1) {
2369       ids[0] = 0;
2370       return 1;
2371     }
2372     if (!r) {
2373       // That's a leaf node.
2374       assert(bottom <= cur, "Sanity check");
2375       // Check if the node has memory
2376       if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2377                                   NULL, 0, LGRP_RSRC_MEM) > 0) {
2378         ids[bottom++] = ids[cur];
2379       }
2380     }
2381     top += r;
2382     cur++;
2383   }
2384   if (bottom == 0) {
2385     // Handle a situation, when the OS reports no memory available.
2386     // Assume UMA architecture.
2387     ids[0] = 0;
2388     return 1;
2389   }
2390   return bottom;
2391 }
2392 
2393 // Detect the topology change. Typically happens during CPU plugging-unplugging.
2394 bool os::numa_topology_changed() {
2395   int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2396   if (is_stale != -1 && is_stale) {
2397     Solaris::lgrp_fini(Solaris::lgrp_cookie());
2398     Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2399     assert(c != 0, "Failure to initialize LGRP API");
2400     Solaris::set_lgrp_cookie(c);
2401     return true;
2402   }
2403   return false;
2404 }
2405 
2406 // Get the group id of the current LWP.
2407 int os::numa_get_group_id() {
2408   int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2409   if (lgrp_id == -1) {
2410     return 0;
2411   }
2412   const int size = os::numa_get_groups_num();
2413   int *ids = (int*)alloca(size * sizeof(int));
2414 
2415   // Get the ids of all lgroups with memory; r is the count.
2416   int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2417                                   (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2418   if (r <= 0) {
2419     return 0;
2420   }
2421   return ids[os::random() % r];
2422 }
2423 
2424 // Request information about the page.
2425 bool os::get_page_info(char *start, page_info* info) {
2426   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2427   uint64_t addr = (uintptr_t)start;
2428   uint64_t outdata[2];
2429   uint_t validity = 0;
2430 
2431   if (meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2432     return false;
2433   }
2434 
2435   info->size = 0;
2436   info->lgrp_id = -1;
2437 
2438   if ((validity & 1) != 0) {
2439     if ((validity & 2) != 0) {
2440       info->lgrp_id = outdata[0];
2441     }
2442     if ((validity & 4) != 0) {
2443       info->size = outdata[1];
2444     }
2445     return true;
2446   }
2447   return false;
2448 }
2449 
2450 // Scan the pages from start to end until a page different than
2451 // the one described in the info parameter is encountered.
2452 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2453                      page_info* page_found) {
2454   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2455   const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2456   uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT + 1];
2457   uint_t validity[MAX_MEMINFO_CNT];
2458 
2459   size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2460   uint64_t p = (uint64_t)start;
2461   while (p < (uint64_t)end) {
2462     addrs[0] = p;
2463     size_t addrs_count = 1;
2464     while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] + page_size < (uint64_t)end) {
2465       addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2466       addrs_count++;
2467     }
2468 
2469     if (meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2470       return NULL;
2471     }
2472 
2473     size_t i = 0;
2474     for (; i < addrs_count; i++) {
2475       if ((validity[i] & 1) != 0) {
2476         if ((validity[i] & 4) != 0) {
2477           if (outdata[types * i + 1] != page_expected->size) {
2478             break;
2479           }
2480         } else if (page_expected->size != 0) {
2481           break;
2482         }
2483 
2484         if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
2485           if (outdata[types * i] != page_expected->lgrp_id) {
2486             break;
2487           }
2488         }
2489       } else {
2490         return NULL;
2491       }
2492     }
2493 
2494     if (i < addrs_count) {
2495       if ((validity[i] & 2) != 0) {
2496         page_found->lgrp_id = outdata[types * i];
2497       } else {
2498         page_found->lgrp_id = -1;
2499       }
2500       if ((validity[i] & 4) != 0) {
2501         page_found->size = outdata[types * i + 1];
2502       } else {
2503         page_found->size = 0;
2504       }
2505       return (char*)addrs[i];
2506     }
2507 
2508     p = addrs[addrs_count - 1] + page_size;
2509   }
2510   return end;
2511 }
2512 
2513 bool os::pd_uncommit_memory(char* addr, size_t bytes) {
2514   size_t size = bytes;
2515   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2516   // uncommitted page. Otherwise, the read/write might succeed if we
2517   // have enough swap space to back the physical page.
2518   return
2519     NULL != Solaris::mmap_chunk(addr, size,
2520                                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
2521                                 PROT_NONE);
2522 }
2523 
2524 char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
2525   char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
2526 
2527   if (b == MAP_FAILED) {
2528     return NULL;
2529   }
2530   return b;
2531 }
2532 
2533 char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes,
2534                              size_t alignment_hint, bool fixed) {
2535   char* addr = requested_addr;
2536   int flags = MAP_PRIVATE | MAP_NORESERVE;
2537 
2538   assert(!(fixed && (alignment_hint > 0)),
2539          "alignment hint meaningless with fixed mmap");
2540 
2541   if (fixed) {
2542     flags |= MAP_FIXED;
2543   } else if (alignment_hint > (size_t) vm_page_size()) {
2544     flags |= MAP_ALIGN;
2545     addr = (char*) alignment_hint;
2546   }
2547 
2548   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2549   // uncommitted page. Otherwise, the read/write might succeed if we
2550   // have enough swap space to back the physical page.
2551   return mmap_chunk(addr, bytes, flags, PROT_NONE);
2552 }
2553 
2554 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2555                             size_t alignment_hint) {
2556   char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint,
2557                                   (requested_addr != NULL));
2558 
2559   guarantee(requested_addr == NULL || requested_addr == addr,
2560             "OS failed to return requested mmap address.");
2561   return addr;
2562 }
2563 
2564 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr, int file_desc) {
2565   assert(file_desc >= 0, "file_desc is not valid");
2566   char* result = pd_attempt_reserve_memory_at(bytes, requested_addr);
2567   if (result != NULL) {
2568     if (replace_existing_mapping_with_file_mapping(result, bytes, file_desc) == NULL) {
2569       vm_exit_during_initialization(err_msg("Error in mapping Java heap at the given filesystem directory"));
2570     }
2571   }
2572   return result;
2573 }
2574 
2575 // Reserve memory at an arbitrary address, only if that area is
2576 // available (and not reserved for something else).
2577 
2578 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2579   const int max_tries = 10;
2580   char* base[max_tries];
2581   size_t size[max_tries];
2582 
2583   // Solaris adds a gap between mmap'ed regions.  The size of the gap
2584   // is dependent on the requested size and the MMU.  Our initial gap
2585   // value here is just a guess and will be corrected later.
2586   bool had_top_overlap = false;
2587   bool have_adjusted_gap = false;
2588   size_t gap = 0x400000;
2589 
2590   // Assert only that the size is a multiple of the page size, since
2591   // that's all that mmap requires, and since that's all we really know
2592   // about at this low abstraction level.  If we need higher alignment,
2593   // we can either pass an alignment to this method or verify alignment
2594   // in one of the methods further up the call chain.  See bug 5044738.
2595   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2596 
2597   // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
2598   // Give it a try, if the kernel honors the hint we can return immediately.
2599   char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
2600 
2601   volatile int err = errno;
2602   if (addr == requested_addr) {
2603     return addr;
2604   } else if (addr != NULL) {
2605     pd_unmap_memory(addr, bytes);
2606   }
2607 
2608   if (log_is_enabled(Warning, os)) {
2609     char buf[256];
2610     buf[0] = '\0';
2611     if (addr == NULL) {
2612       jio_snprintf(buf, sizeof(buf), ": %s", os::strerror(err));
2613     }
2614     log_info(os)("attempt_reserve_memory_at: couldn't reserve " SIZE_FORMAT " bytes at "
2615             PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
2616             "%s", bytes, requested_addr, addr, buf);
2617   }
2618 
2619   // Address hint method didn't work.  Fall back to the old method.
2620   // In theory, once SNV becomes our oldest supported platform, this
2621   // code will no longer be needed.
2622   //
2623   // Repeatedly allocate blocks until the block is allocated at the
2624   // right spot. Give up after max_tries.
2625   int i;
2626   for (i = 0; i < max_tries; ++i) {
2627     base[i] = reserve_memory(bytes);
2628 
2629     if (base[i] != NULL) {
2630       // Is this the block we wanted?
2631       if (base[i] == requested_addr) {
2632         size[i] = bytes;
2633         break;
2634       }
2635 
2636       // check that the gap value is right
2637       if (had_top_overlap && !have_adjusted_gap) {
2638         size_t actual_gap = base[i-1] - base[i] - bytes;
2639         if (gap != actual_gap) {
2640           // adjust the gap value and retry the last 2 allocations
2641           assert(i > 0, "gap adjustment code problem");
2642           have_adjusted_gap = true;  // adjust the gap only once, just in case
2643           gap = actual_gap;
2644           log_info(os)("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
2645           unmap_memory(base[i], bytes);
2646           unmap_memory(base[i-1], size[i-1]);
2647           i-=2;
2648           continue;
2649         }
2650       }
2651 
2652       // Does this overlap the block we wanted? Give back the overlapped
2653       // parts and try again.
2654       //
2655       // There is still a bug in this code: if top_overlap == bytes,
2656       // the overlap is offset from requested region by the value of gap.
2657       // In this case giving back the overlapped part will not work,
2658       // because we'll give back the entire block at base[i] and
2659       // therefore the subsequent allocation will not generate a new gap.
2660       // This could be fixed with a new algorithm that used larger
2661       // or variable size chunks to find the requested region -
2662       // but such a change would introduce additional complications.
2663       // It's rare enough that the planets align for this bug,
2664       // so we'll just wait for a fix for 6204603/5003415 which
2665       // will provide a mmap flag to allow us to avoid this business.
2666 
2667       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2668       if (top_overlap >= 0 && top_overlap < bytes) {
2669         had_top_overlap = true;
2670         unmap_memory(base[i], top_overlap);
2671         base[i] += top_overlap;
2672         size[i] = bytes - top_overlap;
2673       } else {
2674         size_t bottom_overlap = base[i] + bytes - requested_addr;
2675         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2676           if (bottom_overlap == 0) {
2677             log_info(os)("attempt_reserve_memory_at: possible alignment bug");
2678           }
2679           unmap_memory(requested_addr, bottom_overlap);
2680           size[i] = bytes - bottom_overlap;
2681         } else {
2682           size[i] = bytes;
2683         }
2684       }
2685     }
2686   }
2687 
2688   // Give back the unused reserved pieces.
2689 
2690   for (int j = 0; j < i; ++j) {
2691     if (base[j] != NULL) {
2692       unmap_memory(base[j], size[j]);
2693     }
2694   }
2695 
2696   return (i < max_tries) ? requested_addr : NULL;
2697 }
2698 
2699 bool os::pd_release_memory(char* addr, size_t bytes) {
2700   size_t size = bytes;
2701   return munmap(addr, size) == 0;
2702 }
2703 
2704 static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
2705   assert(addr == (char*)align_down((uintptr_t)addr, os::vm_page_size()),
2706          "addr must be page aligned");
2707   int retVal = mprotect(addr, bytes, prot);
2708   return retVal == 0;
2709 }
2710 
2711 // Protect memory (Used to pass readonly pages through
2712 // JNI GetArray<type>Elements with empty arrays.)
2713 // Also, used for serialization page and for compressed oops null pointer
2714 // checking.
2715 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2716                         bool is_committed) {
2717   unsigned int p = 0;
2718   switch (prot) {
2719   case MEM_PROT_NONE: p = PROT_NONE; break;
2720   case MEM_PROT_READ: p = PROT_READ; break;
2721   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2722   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2723   default:
2724     ShouldNotReachHere();
2725   }
2726   // is_committed is unused.
2727   return solaris_mprotect(addr, bytes, p);
2728 }
2729 
2730 // guard_memory and unguard_memory only happens within stack guard pages.
2731 // Since ISM pertains only to the heap, guard and unguard memory should not
2732 /// happen with an ISM region.
2733 bool os::guard_memory(char* addr, size_t bytes) {
2734   return solaris_mprotect(addr, bytes, PROT_NONE);
2735 }
2736 
2737 bool os::unguard_memory(char* addr, size_t bytes) {
2738   return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
2739 }
2740 
2741 // Large page support
2742 static size_t _large_page_size = 0;
2743 
2744 // Insertion sort for small arrays (descending order).
2745 static void insertion_sort_descending(size_t* array, int len) {
2746   for (int i = 0; i < len; i++) {
2747     size_t val = array[i];
2748     for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
2749       size_t tmp = array[key];
2750       array[key] = array[key - 1];
2751       array[key - 1] = tmp;
2752     }
2753   }
2754 }
2755 
2756 bool os::Solaris::mpss_sanity_check(bool warn, size_t* page_size) {
2757   const unsigned int usable_count = VM_Version::page_size_count();
2758   if (usable_count == 1) {
2759     return false;
2760   }
2761 
2762   // Find the right getpagesizes interface.  When solaris 11 is the minimum
2763   // build platform, getpagesizes() (without the '2') can be called directly.
2764   typedef int (*gps_t)(size_t[], int);
2765   gps_t gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes2"));
2766   if (gps_func == NULL) {
2767     gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes"));
2768     if (gps_func == NULL) {
2769       if (warn) {
2770         warning("MPSS is not supported by the operating system.");
2771       }
2772       return false;
2773     }
2774   }
2775 
2776   // Fill the array of page sizes.
2777   int n = (*gps_func)(_page_sizes, page_sizes_max);
2778   assert(n > 0, "Solaris bug?");
2779 
2780   if (n == page_sizes_max) {
2781     // Add a sentinel value (necessary only if the array was completely filled
2782     // since it is static (zeroed at initialization)).
2783     _page_sizes[--n] = 0;
2784     DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
2785   }
2786   assert(_page_sizes[n] == 0, "missing sentinel");
2787   trace_page_sizes("available page sizes", _page_sizes, n);
2788 
2789   if (n == 1) return false;     // Only one page size available.
2790 
2791   // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
2792   // select up to usable_count elements.  First sort the array, find the first
2793   // acceptable value, then copy the usable sizes to the top of the array and
2794   // trim the rest.  Make sure to include the default page size :-).
2795   //
2796   // A better policy could get rid of the 4M limit by taking the sizes of the
2797   // important VM memory regions (java heap and possibly the code cache) into
2798   // account.
2799   insertion_sort_descending(_page_sizes, n);
2800   const size_t size_limit =
2801     FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
2802   int beg;
2803   for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */;
2804   const int end = MIN2((int)usable_count, n) - 1;
2805   for (int cur = 0; cur < end; ++cur, ++beg) {
2806     _page_sizes[cur] = _page_sizes[beg];
2807   }
2808   _page_sizes[end] = vm_page_size();
2809   _page_sizes[end + 1] = 0;
2810 
2811   if (_page_sizes[end] > _page_sizes[end - 1]) {
2812     // Default page size is not the smallest; sort again.
2813     insertion_sort_descending(_page_sizes, end + 1);
2814   }
2815   *page_size = _page_sizes[0];
2816 
2817   trace_page_sizes("usable page sizes", _page_sizes, end + 1);
2818   return true;
2819 }
2820 
2821 void os::large_page_init() {
2822   if (UseLargePages) {
2823     // print a warning if any large page related flag is specified on command line
2824     bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
2825                            !FLAG_IS_DEFAULT(LargePageSizeInBytes);
2826 
2827     UseLargePages = Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
2828   }
2829 }
2830 
2831 bool os::Solaris::is_valid_page_size(size_t bytes) {
2832   for (int i = 0; _page_sizes[i] != 0; i++) {
2833     if (_page_sizes[i] == bytes) {
2834       return true;
2835     }
2836   }
2837   return false;
2838 }
2839 
2840 bool os::Solaris::setup_large_pages(caddr_t start, size_t bytes, size_t align) {
2841   assert(is_valid_page_size(align), SIZE_FORMAT " is not a valid page size", align);
2842   assert(is_aligned((void*) start, align),
2843          PTR_FORMAT " is not aligned to " SIZE_FORMAT, p2i((void*) start), align);
2844   assert(is_aligned(bytes, align),
2845          SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, align);
2846 
2847   // Signal to OS that we want large pages for addresses
2848   // from addr, addr + bytes
2849   struct memcntl_mha mpss_struct;
2850   mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
2851   mpss_struct.mha_pagesize = align;
2852   mpss_struct.mha_flags = 0;
2853   // Upon successful completion, memcntl() returns 0
2854   if (memcntl(start, bytes, MC_HAT_ADVISE, (caddr_t) &mpss_struct, 0, 0)) {
2855     debug_only(warning("Attempt to use MPSS failed."));
2856     return false;
2857   }
2858   return true;
2859 }
2860 
2861 char* os::reserve_memory_special(size_t size, size_t alignment, char* addr, bool exec) {
2862   fatal("os::reserve_memory_special should not be called on Solaris.");
2863   return NULL;
2864 }
2865 
2866 bool os::release_memory_special(char* base, size_t bytes) {
2867   fatal("os::release_memory_special should not be called on Solaris.");
2868   return false;
2869 }
2870 
2871 size_t os::large_page_size() {
2872   return _large_page_size;
2873 }
2874 
2875 // MPSS allows application to commit large page memory on demand; with ISM
2876 // the entire memory region must be allocated as shared memory.
2877 bool os::can_commit_large_page_memory() {
2878   return true;
2879 }
2880 
2881 bool os::can_execute_large_page_memory() {
2882   return true;
2883 }
2884 
2885 // Read calls from inside the vm need to perform state transitions
2886 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2887   size_t res;
2888   JavaThread* thread = (JavaThread*)Thread::current();
2889   assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
2890   ThreadBlockInVM tbiv(thread);
2891   RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
2892   return res;
2893 }
2894 
2895 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
2896   size_t res;
2897   JavaThread* thread = (JavaThread*)Thread::current();
2898   assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
2899   ThreadBlockInVM tbiv(thread);
2900   RESTARTABLE(::pread(fd, buf, (size_t) nBytes, offset), res);
2901   return res;
2902 }
2903 
2904 size_t os::restartable_read(int fd, void *buf, unsigned int nBytes) {
2905   size_t res;
2906   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
2907          "Assumed _thread_in_native");
2908   RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
2909   return res;
2910 }
2911 
2912 void os::naked_short_sleep(jlong ms) {
2913   assert(ms < 1000, "Un-interruptable sleep, short time use only");
2914 
2915   // usleep is deprecated and removed from POSIX, in favour of nanosleep, but
2916   // Solaris requires -lrt for this.
2917   usleep((ms * 1000));
2918 
2919   return;
2920 }
2921 
2922 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2923 void os::infinite_sleep() {
2924   while (true) {    // sleep forever ...
2925     ::sleep(100);   // ... 100 seconds at a time
2926   }
2927 }
2928 
2929 // Used to convert frequent JVM_Yield() to nops
2930 bool os::dont_yield() {
2931   if (DontYieldALot) {
2932     static hrtime_t last_time = 0;
2933     hrtime_t diff = getTimeNanos() - last_time;
2934 
2935     if (diff < DontYieldALotInterval * 1000000) {
2936       return true;
2937     }
2938 
2939     last_time += diff;
2940 
2941     return false;
2942   } else {
2943     return false;
2944   }
2945 }
2946 
2947 // Note that yield semantics are defined by the scheduling class to which
2948 // the thread currently belongs.  Typically, yield will _not yield to
2949 // other equal or higher priority threads that reside on the dispatch queues
2950 // of other CPUs.
2951 
2952 void os::naked_yield() {
2953   thr_yield();
2954 }
2955 
2956 // Interface for setting lwp priorities.  We are using T2 libthread,
2957 // which forces the use of bound threads, so all of our threads will
2958 // be assigned to real lwp's.  Using the thr_setprio function is
2959 // meaningless in this mode so we must adjust the real lwp's priority.
2960 // The routines below implement the getting and setting of lwp priorities.
2961 //
2962 // Note: There are three priority scales used on Solaris.  Java priotities
2963 //       which range from 1 to 10, libthread "thr_setprio" scale which range
2964 //       from 0 to 127, and the current scheduling class of the process we
2965 //       are running in.  This is typically from -60 to +60.
2966 //       The setting of the lwp priorities in done after a call to thr_setprio
2967 //       so Java priorities are mapped to libthread priorities and we map from
2968 //       the latter to lwp priorities.  We don't keep priorities stored in
2969 //       Java priorities since some of our worker threads want to set priorities
2970 //       higher than all Java threads.
2971 //
2972 // For related information:
2973 // (1)  man -s 2 priocntl
2974 // (2)  man -s 4 priocntl
2975 // (3)  man dispadmin
2976 // =    librt.so
2977 // =    libthread/common/rtsched.c - thrp_setlwpprio().
2978 // =    ps -cL <pid> ... to validate priority.
2979 // =    sched_get_priority_min and _max
2980 //              pthread_create
2981 //              sched_setparam
2982 //              pthread_setschedparam
2983 //
2984 // Assumptions:
2985 // +    We assume that all threads in the process belong to the same
2986 //              scheduling class.   IE. an homogenous process.
2987 // +    Must be root or in IA group to change change "interactive" attribute.
2988 //              Priocntl() will fail silently.  The only indication of failure is when
2989 //              we read-back the value and notice that it hasn't changed.
2990 // +    Interactive threads enter the runq at the head, non-interactive at the tail.
2991 // +    For RT, change timeslice as well.  Invariant:
2992 //              constant "priority integral"
2993 //              Konst == TimeSlice * (60-Priority)
2994 //              Given a priority, compute appropriate timeslice.
2995 // +    Higher numerical values have higher priority.
2996 
2997 // sched class attributes
2998 typedef struct {
2999   int   schedPolicy;              // classID
3000   int   maxPrio;
3001   int   minPrio;
3002 } SchedInfo;
3003 
3004 
3005 static SchedInfo tsLimits, iaLimits, rtLimits, fxLimits;
3006 
3007 #ifdef ASSERT
3008 static int  ReadBackValidate = 1;
3009 #endif
3010 static int  myClass     = 0;
3011 static int  myMin       = 0;
3012 static int  myMax       = 0;
3013 static int  myCur       = 0;
3014 static bool priocntl_enable = false;
3015 
3016 static const int criticalPrio = FXCriticalPriority;
3017 static int java_MaxPriority_to_os_priority = 0; // Saved mapping
3018 
3019 
3020 // lwp_priocntl_init
3021 //
3022 // Try to determine the priority scale for our process.
3023 //
3024 // Return errno or 0 if OK.
3025 //
3026 static int lwp_priocntl_init() {
3027   int rslt;
3028   pcinfo_t ClassInfo;
3029   pcparms_t ParmInfo;
3030   int i;
3031 
3032   if (!UseThreadPriorities) return 0;
3033 
3034   // If ThreadPriorityPolicy is 1, switch tables
3035   if (ThreadPriorityPolicy == 1) {
3036     for (i = 0; i < CriticalPriority+1; i++)
3037       os::java_to_os_priority[i] = prio_policy1[i];
3038   }
3039   if (UseCriticalJavaThreadPriority) {
3040     // MaxPriority always maps to the FX scheduling class and criticalPrio.
3041     // See set_native_priority() and set_lwp_class_and_priority().
3042     // Save original MaxPriority mapping in case attempt to
3043     // use critical priority fails.
3044     java_MaxPriority_to_os_priority = os::java_to_os_priority[MaxPriority];
3045     // Set negative to distinguish from other priorities
3046     os::java_to_os_priority[MaxPriority] = -criticalPrio;
3047   }
3048 
3049   // Get IDs for a set of well-known scheduling classes.
3050   // TODO-FIXME: GETCLINFO returns the current # of classes in the
3051   // the system.  We should have a loop that iterates over the
3052   // classID values, which are known to be "small" integers.
3053 
3054   strcpy(ClassInfo.pc_clname, "TS");
3055   ClassInfo.pc_cid = -1;
3056   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3057   if (rslt < 0) return errno;
3058   assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3059   tsLimits.schedPolicy = ClassInfo.pc_cid;
3060   tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3061   tsLimits.minPrio = -tsLimits.maxPrio;
3062 
3063   strcpy(ClassInfo.pc_clname, "IA");
3064   ClassInfo.pc_cid = -1;
3065   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3066   if (rslt < 0) return errno;
3067   assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3068   iaLimits.schedPolicy = ClassInfo.pc_cid;
3069   iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3070   iaLimits.minPrio = -iaLimits.maxPrio;
3071 
3072   strcpy(ClassInfo.pc_clname, "RT");
3073   ClassInfo.pc_cid = -1;
3074   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3075   if (rslt < 0) return errno;
3076   assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3077   rtLimits.schedPolicy = ClassInfo.pc_cid;
3078   rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3079   rtLimits.minPrio = 0;
3080 
3081   strcpy(ClassInfo.pc_clname, "FX");
3082   ClassInfo.pc_cid = -1;
3083   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3084   if (rslt < 0) return errno;
3085   assert(ClassInfo.pc_cid != -1, "cid for FX class is -1");
3086   fxLimits.schedPolicy = ClassInfo.pc_cid;
3087   fxLimits.maxPrio = ((fxinfo_t*)ClassInfo.pc_clinfo)->fx_maxupri;
3088   fxLimits.minPrio = 0;
3089 
3090   // Query our "current" scheduling class.
3091   // This will normally be IA, TS or, rarely, FX or RT.
3092   memset(&ParmInfo, 0, sizeof(ParmInfo));
3093   ParmInfo.pc_cid = PC_CLNULL;
3094   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3095   if (rslt < 0) return errno;
3096   myClass = ParmInfo.pc_cid;
3097 
3098   // We now know our scheduling classId, get specific information
3099   // about the class.
3100   ClassInfo.pc_cid = myClass;
3101   ClassInfo.pc_clname[0] = 0;
3102   rslt = priocntl((idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo);
3103   if (rslt < 0) return errno;
3104 
3105   if (ThreadPriorityVerbose) {
3106     tty->print_cr("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3107   }
3108 
3109   memset(&ParmInfo, 0, sizeof(pcparms_t));
3110   ParmInfo.pc_cid = PC_CLNULL;
3111   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3112   if (rslt < 0) return errno;
3113 
3114   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3115     myMin = rtLimits.minPrio;
3116     myMax = rtLimits.maxPrio;
3117   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3118     iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3119     myMin = iaLimits.minPrio;
3120     myMax = iaLimits.maxPrio;
3121     myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3122   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3123     tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3124     myMin = tsLimits.minPrio;
3125     myMax = tsLimits.maxPrio;
3126     myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3127   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3128     fxparms_t *fxInfo = (fxparms_t*)ParmInfo.pc_clparms;
3129     myMin = fxLimits.minPrio;
3130     myMax = fxLimits.maxPrio;
3131     myMax = MIN2(myMax, (int)fxInfo->fx_uprilim);       // clamp - restrict
3132   } else {
3133     // No clue - punt
3134     if (ThreadPriorityVerbose) {
3135       tty->print_cr("Unknown scheduling class: %s ... \n",
3136                     ClassInfo.pc_clname);
3137     }
3138     return EINVAL;      // no clue, punt
3139   }
3140 
3141   if (ThreadPriorityVerbose) {
3142     tty->print_cr("Thread priority Range: [%d..%d]\n", myMin, myMax);
3143   }
3144 
3145   priocntl_enable = true;  // Enable changing priorities
3146   return 0;
3147 }
3148 
3149 #define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3150 #define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3151 #define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3152 #define FXPRI(x)        ((fxparms_t *)((x).pc_clparms))
3153 
3154 
3155 // scale_to_lwp_priority
3156 //
3157 // Convert from the libthread "thr_setprio" scale to our current
3158 // lwp scheduling class scale.
3159 //
3160 static int scale_to_lwp_priority(int rMin, int rMax, int x) {
3161   int v;
3162 
3163   if (x == 127) return rMax;            // avoid round-down
3164   v = (((x*(rMax-rMin)))/128)+rMin;
3165   return v;
3166 }
3167 
3168 
3169 // set_lwp_class_and_priority
3170 int set_lwp_class_and_priority(int ThreadID, int lwpid,
3171                                int newPrio, int new_class, bool scale) {
3172   int rslt;
3173   int Actual, Expected, prv;
3174   pcparms_t ParmInfo;                   // for GET-SET
3175 #ifdef ASSERT
3176   pcparms_t ReadBack;                   // for readback
3177 #endif
3178 
3179   // Set priority via PC_GETPARMS, update, PC_SETPARMS
3180   // Query current values.
3181   // TODO: accelerate this by eliminating the PC_GETPARMS call.
3182   // Cache "pcparms_t" in global ParmCache.
3183   // TODO: elide set-to-same-value
3184 
3185   // If something went wrong on init, don't change priorities.
3186   if (!priocntl_enable) {
3187     if (ThreadPriorityVerbose) {
3188       tty->print_cr("Trying to set priority but init failed, ignoring");
3189     }
3190     return EINVAL;
3191   }
3192 
3193   // If lwp hasn't started yet, just return
3194   // the _start routine will call us again.
3195   if (lwpid <= 0) {
3196     if (ThreadPriorityVerbose) {
3197       tty->print_cr("deferring the set_lwp_class_and_priority of thread "
3198                     INTPTR_FORMAT " to %d, lwpid not set",
3199                     ThreadID, newPrio);
3200     }
3201     return 0;
3202   }
3203 
3204   if (ThreadPriorityVerbose) {
3205     tty->print_cr ("set_lwp_class_and_priority("
3206                    INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3207                    ThreadID, lwpid, newPrio);
3208   }
3209 
3210   memset(&ParmInfo, 0, sizeof(pcparms_t));
3211   ParmInfo.pc_cid = PC_CLNULL;
3212   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3213   if (rslt < 0) return errno;
3214 
3215   int cur_class = ParmInfo.pc_cid;
3216   ParmInfo.pc_cid = (id_t)new_class;
3217 
3218   if (new_class == rtLimits.schedPolicy) {
3219     rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3220     rtInfo->rt_pri     = scale ? scale_to_lwp_priority(rtLimits.minPrio,
3221                                                        rtLimits.maxPrio, newPrio)
3222                                : newPrio;
3223     rtInfo->rt_tqsecs  = RT_NOCHANGE;
3224     rtInfo->rt_tqnsecs = RT_NOCHANGE;
3225     if (ThreadPriorityVerbose) {
3226       tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3227     }
3228   } else if (new_class == iaLimits.schedPolicy) {
3229     iaparms_t* iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3230     int maxClamped     = MIN2(iaLimits.maxPrio,
3231                               cur_class == new_class
3232                               ? (int)iaInfo->ia_uprilim : iaLimits.maxPrio);
3233     iaInfo->ia_upri    = scale ? scale_to_lwp_priority(iaLimits.minPrio,
3234                                                        maxClamped, newPrio)
3235                                : newPrio;
3236     iaInfo->ia_uprilim = cur_class == new_class
3237                            ? IA_NOCHANGE : (pri_t)iaLimits.maxPrio;
3238     iaInfo->ia_mode    = IA_NOCHANGE;
3239     if (ThreadPriorityVerbose) {
3240       tty->print_cr("IA: [%d...%d] %d->%d\n",
3241                     iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3242     }
3243   } else if (new_class == tsLimits.schedPolicy) {
3244     tsparms_t* tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3245     int maxClamped     = MIN2(tsLimits.maxPrio,
3246                               cur_class == new_class
3247                               ? (int)tsInfo->ts_uprilim : tsLimits.maxPrio);
3248     tsInfo->ts_upri    = scale ? scale_to_lwp_priority(tsLimits.minPrio,
3249                                                        maxClamped, newPrio)
3250                                : newPrio;
3251     tsInfo->ts_uprilim = cur_class == new_class
3252                            ? TS_NOCHANGE : (pri_t)tsLimits.maxPrio;
3253     if (ThreadPriorityVerbose) {
3254       tty->print_cr("TS: [%d...%d] %d->%d\n",
3255                     tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
3256     }
3257   } else if (new_class == fxLimits.schedPolicy) {
3258     fxparms_t* fxInfo  = (fxparms_t*)ParmInfo.pc_clparms;
3259     int maxClamped     = MIN2(fxLimits.maxPrio,
3260                               cur_class == new_class
3261                               ? (int)fxInfo->fx_uprilim : fxLimits.maxPrio);
3262     fxInfo->fx_upri    = scale ? scale_to_lwp_priority(fxLimits.minPrio,
3263                                                        maxClamped, newPrio)
3264                                : newPrio;
3265     fxInfo->fx_uprilim = cur_class == new_class
3266                            ? FX_NOCHANGE : (pri_t)fxLimits.maxPrio;
3267     fxInfo->fx_tqsecs  = FX_NOCHANGE;
3268     fxInfo->fx_tqnsecs = FX_NOCHANGE;
3269     if (ThreadPriorityVerbose) {
3270       tty->print_cr("FX: [%d...%d] %d->%d\n",
3271                     fxLimits.minPrio, maxClamped, newPrio, fxInfo->fx_upri);
3272     }
3273   } else {
3274     if (ThreadPriorityVerbose) {
3275       tty->print_cr("Unknown new scheduling class %d\n", new_class);
3276     }
3277     return EINVAL;    // no clue, punt
3278   }
3279 
3280   rslt = priocntl(P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
3281   if (ThreadPriorityVerbose && rslt) {
3282     tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
3283   }
3284   if (rslt < 0) return errno;
3285 
3286 #ifdef ASSERT
3287   // Sanity check: read back what we just attempted to set.
3288   // In theory it could have changed in the interim ...
3289   //
3290   // The priocntl system call is tricky.
3291   // Sometimes it'll validate the priority value argument and
3292   // return EINVAL if unhappy.  At other times it fails silently.
3293   // Readbacks are prudent.
3294 
3295   if (!ReadBackValidate) return 0;
3296 
3297   memset(&ReadBack, 0, sizeof(pcparms_t));
3298   ReadBack.pc_cid = PC_CLNULL;
3299   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
3300   assert(rslt >= 0, "priocntl failed");
3301   Actual = Expected = 0xBAD;
3302   assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
3303   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3304     Actual   = RTPRI(ReadBack)->rt_pri;
3305     Expected = RTPRI(ParmInfo)->rt_pri;
3306   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3307     Actual   = IAPRI(ReadBack)->ia_upri;
3308     Expected = IAPRI(ParmInfo)->ia_upri;
3309   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3310     Actual   = TSPRI(ReadBack)->ts_upri;
3311     Expected = TSPRI(ParmInfo)->ts_upri;
3312   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3313     Actual   = FXPRI(ReadBack)->fx_upri;
3314     Expected = FXPRI(ParmInfo)->fx_upri;
3315   } else {
3316     if (ThreadPriorityVerbose) {
3317       tty->print_cr("set_lwp_class_and_priority: unexpected class in readback: %d\n",
3318                     ParmInfo.pc_cid);
3319     }
3320   }
3321 
3322   if (Actual != Expected) {
3323     if (ThreadPriorityVerbose) {
3324       tty->print_cr ("set_lwp_class_and_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
3325                      lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
3326     }
3327   }
3328 #endif
3329 
3330   return 0;
3331 }
3332 
3333 // Solaris only gives access to 128 real priorities at a time,
3334 // so we expand Java's ten to fill this range.  This would be better
3335 // if we dynamically adjusted relative priorities.
3336 //
3337 // The ThreadPriorityPolicy option allows us to select 2 different
3338 // priority scales.
3339 //
3340 // ThreadPriorityPolicy=0
3341 // Since the Solaris' default priority is MaximumPriority, we do not
3342 // set a priority lower than Max unless a priority lower than
3343 // NormPriority is requested.
3344 //
3345 // ThreadPriorityPolicy=1
3346 // This mode causes the priority table to get filled with
3347 // linear values.  NormPriority get's mapped to 50% of the
3348 // Maximum priority an so on.  This will cause VM threads
3349 // to get unfair treatment against other Solaris processes
3350 // which do not explicitly alter their thread priorities.
3351 
3352 int os::java_to_os_priority[CriticalPriority + 1] = {
3353   -99999,         // 0 Entry should never be used
3354 
3355   0,              // 1 MinPriority
3356   32,             // 2
3357   64,             // 3
3358 
3359   96,             // 4
3360   127,            // 5 NormPriority
3361   127,            // 6
3362 
3363   127,            // 7
3364   127,            // 8
3365   127,            // 9 NearMaxPriority
3366 
3367   127,            // 10 MaxPriority
3368 
3369   -criticalPrio   // 11 CriticalPriority
3370 };
3371 
3372 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3373   OSThread* osthread = thread->osthread();
3374 
3375   // Save requested priority in case the thread hasn't been started
3376   osthread->set_native_priority(newpri);
3377 
3378   // Check for critical priority request
3379   bool fxcritical = false;
3380   if (newpri == -criticalPrio) {
3381     fxcritical = true;
3382     newpri = criticalPrio;
3383   }
3384 
3385   assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
3386   if (!UseThreadPriorities) return OS_OK;
3387 
3388   int status = 0;
3389 
3390   if (!fxcritical) {
3391     // Use thr_setprio only if we have a priority that thr_setprio understands
3392     status = thr_setprio(thread->osthread()->thread_id(), newpri);
3393   }
3394 
3395   int lwp_status =
3396           set_lwp_class_and_priority(osthread->thread_id(),
3397                                      osthread->lwp_id(),
3398                                      newpri,
3399                                      fxcritical ? fxLimits.schedPolicy : myClass,
3400                                      !fxcritical);
3401   if (lwp_status != 0 && fxcritical) {
3402     // Try again, this time without changing the scheduling class
3403     newpri = java_MaxPriority_to_os_priority;
3404     lwp_status = set_lwp_class_and_priority(osthread->thread_id(),
3405                                             osthread->lwp_id(),
3406                                             newpri, myClass, false);
3407   }
3408   status |= lwp_status;
3409   return (status == 0) ? OS_OK : OS_ERR;
3410 }
3411 
3412 
3413 OSReturn os::get_native_priority(const Thread* const thread,
3414                                  int *priority_ptr) {
3415   int p;
3416   if (!UseThreadPriorities) {
3417     *priority_ptr = NormalPriority;
3418     return OS_OK;
3419   }
3420   int status = thr_getprio(thread->osthread()->thread_id(), &p);
3421   if (status != 0) {
3422     return OS_ERR;
3423   }
3424   *priority_ptr = p;
3425   return OS_OK;
3426 }
3427 
3428 
3429 // Hint to the underlying OS that a task switch would not be good.
3430 // Void return because it's a hint and can fail.
3431 void os::hint_no_preempt() {
3432   schedctl_start(schedctl_init());
3433 }
3434 
3435 ////////////////////////////////////////////////////////////////////////////////
3436 // suspend/resume support
3437 
3438 //  The low-level signal-based suspend/resume support is a remnant from the
3439 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3440 //  within hotspot. Currently used by JFR's OSThreadSampler
3441 //
3442 //  The remaining code is greatly simplified from the more general suspension
3443 //  code that used to be used.
3444 //
3445 //  The protocol is quite simple:
3446 //  - suspend:
3447 //      - sends a signal to the target thread
3448 //      - polls the suspend state of the osthread using a yield loop
3449 //      - target thread signal handler (SR_handler) sets suspend state
3450 //        and blocks in sigsuspend until continued
3451 //  - resume:
3452 //      - sets target osthread state to continue
3453 //      - sends signal to end the sigsuspend loop in the SR_handler
3454 //
3455 //  Note that the SR_lock plays no role in this suspend/resume protocol,
3456 //  but is checked for NULL in SR_handler as a thread termination indicator.
3457 //  The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
3458 //
3459 //  Note that resume_clear_context() and suspend_save_context() are needed
3460 //  by SR_handler(), so that fetch_frame_from_ucontext() works,
3461 //  which in part is used by:
3462 //    - Forte Analyzer: AsyncGetCallTrace()
3463 //    - StackBanging: get_frame_at_stack_banging_point()
3464 //    - JFR: get_topframe()-->....-->get_valid_uc_in_signal_handler()
3465 
3466 static void resume_clear_context(OSThread *osthread) {
3467   osthread->set_ucontext(NULL);
3468 }
3469 
3470 static void suspend_save_context(OSThread *osthread, ucontext_t* context) {
3471   osthread->set_ucontext(context);
3472 }
3473 
3474 static PosixSemaphore sr_semaphore;
3475 
3476 void os::Solaris::SR_handler(Thread* thread, ucontext_t* context) {
3477   // Save and restore errno to avoid confusing native code with EINTR
3478   // after sigsuspend.
3479   int old_errno = errno;
3480 
3481   OSThread* osthread = thread->osthread();
3482   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3483 
3484   os::SuspendResume::State current = osthread->sr.state();
3485   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3486     suspend_save_context(osthread, context);
3487 
3488     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3489     os::SuspendResume::State state = osthread->sr.suspended();
3490     if (state == os::SuspendResume::SR_SUSPENDED) {
3491       sigset_t suspend_set;  // signals for sigsuspend()
3492 
3493       // get current set of blocked signals and unblock resume signal
3494       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3495       sigdelset(&suspend_set, ASYNC_SIGNAL);
3496 
3497       sr_semaphore.signal();
3498       // wait here until we are resumed
3499       while (1) {
3500         sigsuspend(&suspend_set);
3501 
3502         os::SuspendResume::State result = osthread->sr.running();
3503         if (result == os::SuspendResume::SR_RUNNING) {
3504           sr_semaphore.signal();
3505           break;
3506         }
3507       }
3508 
3509     } else if (state == os::SuspendResume::SR_RUNNING) {
3510       // request was cancelled, continue
3511     } else {
3512       ShouldNotReachHere();
3513     }
3514 
3515     resume_clear_context(osthread);
3516   } else if (current == os::SuspendResume::SR_RUNNING) {
3517     // request was cancelled, continue
3518   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3519     // ignore
3520   } else {
3521     // ignore
3522   }
3523 
3524   errno = old_errno;
3525 }
3526 
3527 void os::print_statistics() {
3528 }
3529 
3530 bool os::message_box(const char* title, const char* message) {
3531   int i;
3532   fdStream err(defaultStream::error_fd());
3533   for (i = 0; i < 78; i++) err.print_raw("=");
3534   err.cr();
3535   err.print_raw_cr(title);
3536   for (i = 0; i < 78; i++) err.print_raw("-");
3537   err.cr();
3538   err.print_raw_cr(message);
3539   for (i = 0; i < 78; i++) err.print_raw("=");
3540   err.cr();
3541 
3542   char buf[16];
3543   // Prevent process from exiting upon "read error" without consuming all CPU
3544   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3545 
3546   return buf[0] == 'y' || buf[0] == 'Y';
3547 }
3548 
3549 static int sr_notify(OSThread* osthread) {
3550   int status = thr_kill(osthread->thread_id(), ASYNC_SIGNAL);
3551   assert_status(status == 0, status, "thr_kill");
3552   return status;
3553 }
3554 
3555 // "Randomly" selected value for how long we want to spin
3556 // before bailing out on suspending a thread, also how often
3557 // we send a signal to a thread we want to resume
3558 static const int RANDOMLY_LARGE_INTEGER = 1000000;
3559 static const int RANDOMLY_LARGE_INTEGER2 = 100;
3560 
3561 static bool do_suspend(OSThread* osthread) {
3562   assert(osthread->sr.is_running(), "thread should be running");
3563   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3564 
3565   // mark as suspended and send signal
3566   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3567     // failed to switch, state wasn't running?
3568     ShouldNotReachHere();
3569     return false;
3570   }
3571 
3572   if (sr_notify(osthread) != 0) {
3573     ShouldNotReachHere();
3574   }
3575 
3576   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3577   while (true) {
3578     if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2000 * NANOSECS_PER_MILLISEC))) {
3579       break;
3580     } else {
3581       // timeout
3582       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3583       if (cancelled == os::SuspendResume::SR_RUNNING) {
3584         return false;
3585       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3586         // make sure that we consume the signal on the semaphore as well
3587         sr_semaphore.wait();
3588         break;
3589       } else {
3590         ShouldNotReachHere();
3591         return false;
3592       }
3593     }
3594   }
3595 
3596   guarantee(osthread->sr.is_suspended(), "Must be suspended");
3597   return true;
3598 }
3599 
3600 static void do_resume(OSThread* osthread) {
3601   assert(osthread->sr.is_suspended(), "thread should be suspended");
3602   assert(!sr_semaphore.trywait(), "invalid semaphore state");
3603 
3604   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3605     // failed to switch to WAKEUP_REQUEST
3606     ShouldNotReachHere();
3607     return;
3608   }
3609 
3610   while (true) {
3611     if (sr_notify(osthread) == 0) {
3612       if (sr_semaphore.timedwait(create_semaphore_timespec(0, 2 * NANOSECS_PER_MILLISEC))) {
3613         if (osthread->sr.is_running()) {
3614           return;
3615         }
3616       }
3617     } else {
3618       ShouldNotReachHere();
3619     }
3620   }
3621 
3622   guarantee(osthread->sr.is_running(), "Must be running!");
3623 }
3624 
3625 void os::SuspendedThreadTask::internal_do_task() {
3626   if (do_suspend(_thread->osthread())) {
3627     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
3628     do_task(context);
3629     do_resume(_thread->osthread());
3630   }
3631 }
3632 
3633 // This does not do anything on Solaris. This is basically a hook for being
3634 // able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
3635 void os::os_exception_wrapper(java_call_t f, JavaValue* value,
3636                               const methodHandle& method, JavaCallArguments* args,
3637                               Thread* thread) {
3638   f(value, method, args, thread);
3639 }
3640 
3641 // This routine may be used by user applications as a "hook" to catch signals.
3642 // The user-defined signal handler must pass unrecognized signals to this
3643 // routine, and if it returns true (non-zero), then the signal handler must
3644 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3645 // routine will never retun false (zero), but instead will execute a VM panic
3646 // routine kill the process.
3647 //
3648 // If this routine returns false, it is OK to call it again.  This allows
3649 // the user-defined signal handler to perform checks either before or after
3650 // the VM performs its own checks.  Naturally, the user code would be making
3651 // a serious error if it tried to handle an exception (such as a null check
3652 // or breakpoint) that the VM was generating for its own correct operation.
3653 //
3654 // This routine may recognize any of the following kinds of signals:
3655 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
3656 // ASYNC_SIGNAL.
3657 // It should be consulted by handlers for any of those signals.
3658 //
3659 // The caller of this routine must pass in the three arguments supplied
3660 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3661 // field of the structure passed to sigaction().  This routine assumes that
3662 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3663 //
3664 // Note that the VM will print warnings if it detects conflicting signal
3665 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3666 //
3667 extern "C" JNIEXPORT int JVM_handle_solaris_signal(int signo,
3668                                                    siginfo_t* siginfo,
3669                                                    void* ucontext,
3670                                                    int abort_if_unrecognized);
3671 
3672 
3673 void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
3674   int orig_errno = errno;  // Preserve errno value over signal handler.
3675   JVM_handle_solaris_signal(sig, info, ucVoid, true);
3676   errno = orig_errno;
3677 }
3678 
3679 // This boolean allows users to forward their own non-matching signals
3680 // to JVM_handle_solaris_signal, harmlessly.
3681 bool os::Solaris::signal_handlers_are_installed = false;
3682 
3683 // For signal-chaining
3684 bool os::Solaris::libjsig_is_loaded = false;
3685 typedef struct sigaction *(*get_signal_t)(int);
3686 get_signal_t os::Solaris::get_signal_action = NULL;
3687 
3688 struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
3689   struct sigaction *actp = NULL;
3690 
3691   if ((libjsig_is_loaded)  && (sig <= Maxsignum)) {
3692     // Retrieve the old signal handler from libjsig
3693     actp = (*get_signal_action)(sig);
3694   }
3695   if (actp == NULL) {
3696     // Retrieve the preinstalled signal handler from jvm
3697     actp = get_preinstalled_handler(sig);
3698   }
3699 
3700   return actp;
3701 }
3702 
3703 static bool call_chained_handler(struct sigaction *actp, int sig,
3704                                  siginfo_t *siginfo, void *context) {
3705   // Call the old signal handler
3706   if (actp->sa_handler == SIG_DFL) {
3707     // It's more reasonable to let jvm treat it as an unexpected exception
3708     // instead of taking the default action.
3709     return false;
3710   } else if (actp->sa_handler != SIG_IGN) {
3711     if ((actp->sa_flags & SA_NODEFER) == 0) {
3712       // automaticlly block the signal
3713       sigaddset(&(actp->sa_mask), sig);
3714     }
3715 
3716     sa_handler_t hand;
3717     sa_sigaction_t sa;
3718     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3719     // retrieve the chained handler
3720     if (siginfo_flag_set) {
3721       sa = actp->sa_sigaction;
3722     } else {
3723       hand = actp->sa_handler;
3724     }
3725 
3726     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3727       actp->sa_handler = SIG_DFL;
3728     }
3729 
3730     // try to honor the signal mask
3731     sigset_t oset;
3732     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3733 
3734     // call into the chained handler
3735     if (siginfo_flag_set) {
3736       (*sa)(sig, siginfo, context);
3737     } else {
3738       (*hand)(sig);
3739     }
3740 
3741     // restore the signal mask
3742     pthread_sigmask(SIG_SETMASK, &oset, 0);
3743   }
3744   // Tell jvm's signal handler the signal is taken care of.
3745   return true;
3746 }
3747 
3748 bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3749   bool chained = false;
3750   // signal-chaining
3751   if (UseSignalChaining) {
3752     struct sigaction *actp = get_chained_signal_action(sig);
3753     if (actp != NULL) {
3754       chained = call_chained_handler(actp, sig, siginfo, context);
3755     }
3756   }
3757   return chained;
3758 }
3759 
3760 struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
3761   assert((chainedsigactions != (struct sigaction *)NULL) &&
3762          (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3763   if (preinstalled_sigs[sig] != 0) {
3764     return &chainedsigactions[sig];
3765   }
3766   return NULL;
3767 }
3768 
3769 void os::Solaris::save_preinstalled_handler(int sig,
3770                                             struct sigaction& oldAct) {
3771   assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
3772   assert((chainedsigactions != (struct sigaction *)NULL) &&
3773          (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3774   chainedsigactions[sig] = oldAct;
3775   preinstalled_sigs[sig] = 1;
3776 }
3777 
3778 void os::Solaris::set_signal_handler(int sig, bool set_installed,
3779                                      bool oktochain) {
3780   // Check for overwrite.
3781   struct sigaction oldAct;
3782   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3783   void* oldhand =
3784       oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3785                           : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3786   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3787       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3788       oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
3789     if (AllowUserSignalHandlers || !set_installed) {
3790       // Do not overwrite; user takes responsibility to forward to us.
3791       return;
3792     } else if (UseSignalChaining) {
3793       if (oktochain) {
3794         // save the old handler in jvm
3795         save_preinstalled_handler(sig, oldAct);
3796       } else {
3797         vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal.");
3798       }
3799       // libjsig also interposes the sigaction() call below and saves the
3800       // old sigaction on it own.
3801     } else {
3802       fatal("Encountered unexpected pre-existing sigaction handler "
3803             "%#lx for signal %d.", (long)oldhand, sig);
3804     }
3805   }
3806 
3807   struct sigaction sigAct;
3808   sigfillset(&(sigAct.sa_mask));
3809   sigAct.sa_handler = SIG_DFL;
3810 
3811   sigAct.sa_sigaction = signalHandler;
3812   // Handle SIGSEGV on alternate signal stack if
3813   // not using stack banging
3814   if (!UseStackBanging && sig == SIGSEGV) {
3815     sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
3816   } else {
3817     sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
3818   }
3819   os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
3820 
3821   sigaction(sig, &sigAct, &oldAct);
3822 
3823   void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3824                                        : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3825   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3826 }
3827 
3828 
3829 #define DO_SIGNAL_CHECK(sig)                      \
3830   do {                                            \
3831     if (!sigismember(&check_signal_done, sig)) {  \
3832       os::Solaris::check_signal_handler(sig);     \
3833     }                                             \
3834   } while (0)
3835 
3836 // This method is a periodic task to check for misbehaving JNI applications
3837 // under CheckJNI, we can add any periodic checks here
3838 
3839 void os::run_periodic_checks() {
3840   // A big source of grief is hijacking virt. addr 0x0 on Solaris,
3841   // thereby preventing a NULL checks.
3842   if (!check_addr0_done) check_addr0_done = check_addr0(tty);
3843 
3844   if (check_signals == false) return;
3845 
3846   // SEGV and BUS if overridden could potentially prevent
3847   // generation of hs*.log in the event of a crash, debugging
3848   // such a case can be very challenging, so we absolutely
3849   // check for the following for a good measure:
3850   DO_SIGNAL_CHECK(SIGSEGV);
3851   DO_SIGNAL_CHECK(SIGILL);
3852   DO_SIGNAL_CHECK(SIGFPE);
3853   DO_SIGNAL_CHECK(SIGBUS);
3854   DO_SIGNAL_CHECK(SIGPIPE);
3855   DO_SIGNAL_CHECK(SIGXFSZ);
3856   DO_SIGNAL_CHECK(ASYNC_SIGNAL);
3857 
3858   // ReduceSignalUsage allows the user to override these handlers
3859   // see comments at the very top and jvm_solaris.h
3860   if (!ReduceSignalUsage) {
3861     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3862     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3863     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3864     DO_SIGNAL_CHECK(BREAK_SIGNAL);
3865   }
3866 }
3867 
3868 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3869 
3870 static os_sigaction_t os_sigaction = NULL;
3871 
3872 void os::Solaris::check_signal_handler(int sig) {
3873   char buf[O_BUFLEN];
3874   address jvmHandler = NULL;
3875 
3876   struct sigaction act;
3877   if (os_sigaction == NULL) {
3878     // only trust the default sigaction, in case it has been interposed
3879     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3880     if (os_sigaction == NULL) return;
3881   }
3882 
3883   os_sigaction(sig, (struct sigaction*)NULL, &act);
3884 
3885   address thisHandler = (act.sa_flags & SA_SIGINFO)
3886     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3887     : CAST_FROM_FN_PTR(address, act.sa_handler);
3888 
3889 
3890   switch (sig) {
3891   case SIGSEGV:
3892   case SIGBUS:
3893   case SIGFPE:
3894   case SIGPIPE:
3895   case SIGXFSZ:
3896   case SIGILL:
3897   case ASYNC_SIGNAL:
3898     jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
3899     break;
3900 
3901   case SHUTDOWN1_SIGNAL:
3902   case SHUTDOWN2_SIGNAL:
3903   case SHUTDOWN3_SIGNAL:
3904   case BREAK_SIGNAL:
3905     jvmHandler = (address)user_handler();
3906     break;
3907 
3908   default:
3909       return;
3910   }
3911 
3912   if (thisHandler != jvmHandler) {
3913     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3914     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3915     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3916     // No need to check this sig any longer
3917     sigaddset(&check_signal_done, sig);
3918     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
3919     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
3920       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
3921                     exception_name(sig, buf, O_BUFLEN));
3922     }
3923   } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
3924     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3925     tty->print("expected:");
3926     os::Posix::print_sa_flags(tty, os::Solaris::get_our_sigflags(sig));
3927     tty->cr();
3928     tty->print("  found:");
3929     os::Posix::print_sa_flags(tty, act.sa_flags);
3930     tty->cr();
3931     // No need to check this sig any longer
3932     sigaddset(&check_signal_done, sig);
3933   }
3934 
3935   // Print all the signal handler state
3936   if (sigismember(&check_signal_done, sig)) {
3937     print_signal_handlers(tty, buf, O_BUFLEN);
3938   }
3939 
3940 }
3941 
3942 void os::Solaris::install_signal_handlers() {
3943   signal_handlers_are_installed = true;
3944 
3945   // signal-chaining
3946   typedef void (*signal_setting_t)();
3947   signal_setting_t begin_signal_setting = NULL;
3948   signal_setting_t end_signal_setting = NULL;
3949   begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3950                                         dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3951   if (begin_signal_setting != NULL) {
3952     end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3953                                         dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3954     get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3955                                        dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3956     libjsig_is_loaded = true;
3957     assert(UseSignalChaining, "should enable signal-chaining");
3958   }
3959   if (libjsig_is_loaded) {
3960     // Tell libjsig jvm is setting signal handlers
3961     (*begin_signal_setting)();
3962   }
3963 
3964   set_signal_handler(SIGSEGV, true, true);
3965   set_signal_handler(SIGPIPE, true, true);
3966   set_signal_handler(SIGXFSZ, true, true);
3967   set_signal_handler(SIGBUS, true, true);
3968   set_signal_handler(SIGILL, true, true);
3969   set_signal_handler(SIGFPE, true, true);
3970   set_signal_handler(ASYNC_SIGNAL, true, true);
3971 
3972   if (libjsig_is_loaded) {
3973     // Tell libjsig jvm finishes setting signal handlers
3974     (*end_signal_setting)();
3975   }
3976 
3977   // We don't activate signal checker if libjsig is in place, we trust ourselves
3978   // and if UserSignalHandler is installed all bets are off.
3979   // Log that signal checking is off only if -verbose:jni is specified.
3980   if (CheckJNICalls) {
3981     if (libjsig_is_loaded) {
3982       if (PrintJNIResolving) {
3983         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3984       }
3985       check_signals = false;
3986     }
3987     if (AllowUserSignalHandlers) {
3988       if (PrintJNIResolving) {
3989         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3990       }
3991       check_signals = false;
3992     }
3993   }
3994 }
3995 
3996 
3997 void report_error(const char* file_name, int line_no, const char* title,
3998                   const char* format, ...);
3999 
4000 // (Static) wrappers for the liblgrp API
4001 os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4002 os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4003 os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4004 os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4005 os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4006 os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4007 os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4008 os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4009 os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4010 
4011 static address resolve_symbol_lazy(const char* name) {
4012   address addr = (address) dlsym(RTLD_DEFAULT, name);
4013   if (addr == NULL) {
4014     // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4015     addr = (address) dlsym(RTLD_NEXT, name);
4016   }
4017   return addr;
4018 }
4019 
4020 static address resolve_symbol(const char* name) {
4021   address addr = resolve_symbol_lazy(name);
4022   if (addr == NULL) {
4023     fatal(dlerror());
4024   }
4025   return addr;
4026 }
4027 
4028 void os::Solaris::libthread_init() {
4029   address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4030 
4031   lwp_priocntl_init();
4032 
4033   // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4034   if (func == NULL) {
4035     func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4036     // Guarantee that this VM is running on an new enough OS (5.6 or
4037     // later) that it will have a new enough libthread.so.
4038     guarantee(func != NULL, "libthread.so is too old.");
4039   }
4040 
4041   int size;
4042   void (*handler_info_func)(address *, int *);
4043   handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4044   handler_info_func(&handler_start, &size);
4045   handler_end = handler_start + size;
4046 }
4047 
4048 
4049 int_fnP_mutex_tP os::Solaris::_mutex_lock;
4050 int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4051 int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4052 int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4053 int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4054 int os::Solaris::_mutex_scope = USYNC_THREAD;
4055 
4056 int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4057 int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4058 int_fnP_cond_tP os::Solaris::_cond_signal;
4059 int_fnP_cond_tP os::Solaris::_cond_broadcast;
4060 int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4061 int_fnP_cond_tP os::Solaris::_cond_destroy;
4062 int os::Solaris::_cond_scope = USYNC_THREAD;
4063 bool os::Solaris::_synchronization_initialized;
4064 
4065 void os::Solaris::synchronization_init() {
4066   if (UseLWPSynchronization) {
4067     os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4068     os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4069     os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4070     os::Solaris::set_mutex_init(lwp_mutex_init);
4071     os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4072     os::Solaris::set_mutex_scope(USYNC_THREAD);
4073 
4074     os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4075     os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4076     os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4077     os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4078     os::Solaris::set_cond_init(lwp_cond_init);
4079     os::Solaris::set_cond_destroy(lwp_cond_destroy);
4080     os::Solaris::set_cond_scope(USYNC_THREAD);
4081   } else {
4082     os::Solaris::set_mutex_scope(USYNC_THREAD);
4083     os::Solaris::set_cond_scope(USYNC_THREAD);
4084 
4085     if (UsePthreads) {
4086       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4087       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4088       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4089       os::Solaris::set_mutex_init(pthread_mutex_default_init);
4090       os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4091 
4092       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4093       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4094       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4095       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4096       os::Solaris::set_cond_init(pthread_cond_default_init);
4097       os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4098     } else {
4099       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4100       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4101       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4102       os::Solaris::set_mutex_init(::mutex_init);
4103       os::Solaris::set_mutex_destroy(::mutex_destroy);
4104 
4105       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4106       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4107       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4108       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4109       os::Solaris::set_cond_init(::cond_init);
4110       os::Solaris::set_cond_destroy(::cond_destroy);
4111     }
4112   }
4113   _synchronization_initialized = true;
4114 }
4115 
4116 bool os::Solaris::liblgrp_init() {
4117   void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4118   if (handle != NULL) {
4119     os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4120     os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4121     os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4122     os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4123     os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4124     os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4125     os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4126     os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4127                                                       dlsym(handle, "lgrp_cookie_stale")));
4128 
4129     lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4130     set_lgrp_cookie(c);
4131     return true;
4132   }
4133   return false;
4134 }
4135 
4136 // int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
4137 typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
4138 static pset_getloadavg_type pset_getloadavg_ptr = NULL;
4139 
4140 void init_pset_getloadavg_ptr(void) {
4141   pset_getloadavg_ptr =
4142     (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
4143   if (pset_getloadavg_ptr == NULL) {
4144     log_warning(os)("pset_getloadavg function not found");
4145   }
4146 }
4147 
4148 int os::Solaris::_dev_zero_fd = -1;
4149 
4150 // this is called _before_ the global arguments have been parsed
4151 void os::init(void) {
4152   _initial_pid = getpid();
4153 
4154   max_hrtime = first_hrtime = gethrtime();
4155 
4156   init_random(1234567);
4157 
4158   page_size = sysconf(_SC_PAGESIZE);
4159   if (page_size == -1) {
4160     fatal("os_solaris.cpp: os::init: sysconf failed (%s)", os::strerror(errno));
4161   }
4162   init_page_sizes((size_t) page_size);
4163 
4164   Solaris::initialize_system_info();
4165 
4166   int fd = ::open("/dev/zero", O_RDWR);
4167   if (fd < 0) {
4168     fatal("os::init: cannot open /dev/zero (%s)", os::strerror(errno));
4169   } else {
4170     Solaris::set_dev_zero_fd(fd);
4171 
4172     // Close on exec, child won't inherit.
4173     fcntl(fd, F_SETFD, FD_CLOEXEC);
4174   }
4175 
4176   clock_tics_per_sec = CLK_TCK;
4177 
4178   // check if dladdr1() exists; dladdr1 can provide more information than
4179   // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
4180   // and is available on linker patches for 5.7 and 5.8.
4181   // libdl.so must have been loaded, this call is just an entry lookup
4182   void * hdl = dlopen("libdl.so", RTLD_NOW);
4183   if (hdl) {
4184     dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
4185   }
4186 
4187   // main_thread points to the thread that created/loaded the JVM.
4188   main_thread = thr_self();
4189 
4190   // dynamic lookup of functions that may not be available in our lowest
4191   // supported Solaris release
4192   void * handle = dlopen("libc.so.1", RTLD_LAZY);
4193   if (handle != NULL) {
4194     Solaris::_pthread_setname_np =  // from 11.3
4195         (Solaris::pthread_setname_np_func_t)dlsym(handle, "pthread_setname_np");
4196   }
4197 }
4198 
4199 // To install functions for atexit system call
4200 extern "C" {
4201   static void perfMemory_exit_helper() {
4202     perfMemory_exit();
4203   }
4204 }
4205 
4206 // this is called _after_ the global arguments have been parsed
4207 jint os::init_2(void) {
4208   // try to enable extended file IO ASAP, see 6431278
4209   os::Solaris::try_enable_extended_io();
4210 
4211   // Check and sets minimum stack sizes against command line options
4212   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4213     return JNI_ERR;
4214   }
4215 
4216   Solaris::libthread_init();
4217 
4218   if (UseNUMA) {
4219     if (!Solaris::liblgrp_init()) {
4220       UseNUMA = false;
4221     } else {
4222       size_t lgrp_limit = os::numa_get_groups_num();
4223       int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit, mtInternal);
4224       size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
4225       FREE_C_HEAP_ARRAY(int, lgrp_ids);
4226       if (lgrp_num < 2) {
4227         // There's only one locality group, disable NUMA.
4228         UseNUMA = false;
4229       }
4230     }
4231     if (!UseNUMA && ForceNUMA) {
4232       UseNUMA = true;
4233     }
4234   }
4235 
4236   Solaris::signal_sets_init();
4237   Solaris::init_signal_mem();
4238   Solaris::install_signal_handlers();
4239   // Initialize data for jdk.internal.misc.Signal
4240   if (!ReduceSignalUsage) {
4241     jdk_misc_signal_init();
4242   }
4243 
4244   // initialize synchronization primitives to use either thread or
4245   // lwp synchronization (controlled by UseLWPSynchronization)
4246   Solaris::synchronization_init();
4247 
4248   if (MaxFDLimit) {
4249     // set the number of file descriptors to max. print out error
4250     // if getrlimit/setrlimit fails but continue regardless.
4251     struct rlimit nbr_files;
4252     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4253     if (status != 0) {
4254       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4255     } else {
4256       nbr_files.rlim_cur = nbr_files.rlim_max;
4257       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4258       if (status != 0) {
4259         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4260       }
4261     }
4262   }
4263 
4264   // Calculate theoretical max. size of Threads to guard gainst
4265   // artifical out-of-memory situations, where all available address-
4266   // space has been reserved by thread stacks. Default stack size is 1Mb.
4267   size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
4268     JavaThread::stack_size_at_create() : (1*K*K);
4269   assert(pre_thread_stack_size != 0, "Must have a stack");
4270   // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
4271   // we should start doing Virtual Memory banging. Currently when the threads will
4272   // have used all but 200Mb of space.
4273   size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
4274   Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
4275 
4276   // at-exit methods are called in the reverse order of their registration.
4277   // In Solaris 7 and earlier, atexit functions are called on return from
4278   // main or as a result of a call to exit(3C). There can be only 32 of
4279   // these functions registered and atexit() does not set errno. In Solaris
4280   // 8 and later, there is no limit to the number of functions registered
4281   // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
4282   // functions are called upon dlclose(3DL) in addition to return from main
4283   // and exit(3C).
4284 
4285   if (PerfAllowAtExitRegistration) {
4286     // only register atexit functions if PerfAllowAtExitRegistration is set.
4287     // atexit functions can be delayed until process exit time, which
4288     // can be problematic for embedded VM situations. Embedded VMs should
4289     // call DestroyJavaVM() to assure that VM resources are released.
4290 
4291     // note: perfMemory_exit_helper atexit function may be removed in
4292     // the future if the appropriate cleanup code can be added to the
4293     // VM_Exit VMOperation's doit method.
4294     if (atexit(perfMemory_exit_helper) != 0) {
4295       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4296     }
4297   }
4298 
4299   // Init pset_loadavg function pointer
4300   init_pset_getloadavg_ptr();
4301 
4302   return JNI_OK;
4303 }
4304 
4305 // Mark the polling page as unreadable
4306 void os::make_polling_page_unreadable(void) {
4307   if (mprotect((char *)_polling_page, page_size, PROT_NONE) != 0) {
4308     fatal("Could not disable polling page");
4309   }
4310 }
4311 
4312 // Mark the polling page as readable
4313 void os::make_polling_page_readable(void) {
4314   if (mprotect((char *)_polling_page, page_size, PROT_READ) != 0) {
4315     fatal("Could not enable polling page");
4316   }
4317 }
4318 
4319 // Is a (classpath) directory empty?
4320 bool os::dir_is_empty(const char* path) {
4321   DIR *dir = NULL;
4322   struct dirent *ptr;
4323 
4324   dir = opendir(path);
4325   if (dir == NULL) return true;
4326 
4327   // Scan the directory
4328   bool result = true;
4329   while (result && (ptr = readdir(dir)) != NULL) {
4330     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4331       result = false;
4332     }
4333   }
4334   closedir(dir);
4335   return result;
4336 }
4337 
4338 // This code originates from JDK's sysOpen and open64_w
4339 // from src/solaris/hpi/src/system_md.c
4340 
4341 int os::open(const char *path, int oflag, int mode) {
4342   if (strlen(path) > MAX_PATH - 1) {
4343     errno = ENAMETOOLONG;
4344     return -1;
4345   }
4346   int fd;
4347 
4348   fd = ::open64(path, oflag, mode);
4349   if (fd == -1) return -1;
4350 
4351   // If the open succeeded, the file might still be a directory
4352   {
4353     struct stat64 buf64;
4354     int ret = ::fstat64(fd, &buf64);
4355     int st_mode = buf64.st_mode;
4356 
4357     if (ret != -1) {
4358       if ((st_mode & S_IFMT) == S_IFDIR) {
4359         errno = EISDIR;
4360         ::close(fd);
4361         return -1;
4362       }
4363     } else {
4364       ::close(fd);
4365       return -1;
4366     }
4367   }
4368 
4369   // 32-bit Solaris systems suffer from:
4370   //
4371   // - an historical default soft limit of 256 per-process file
4372   //   descriptors that is too low for many Java programs.
4373   //
4374   // - a design flaw where file descriptors created using stdio
4375   //   fopen must be less than 256, _even_ when the first limit above
4376   //   has been raised.  This can cause calls to fopen (but not calls to
4377   //   open, for example) to fail mysteriously, perhaps in 3rd party
4378   //   native code (although the JDK itself uses fopen).  One can hardly
4379   //   criticize them for using this most standard of all functions.
4380   //
4381   // We attempt to make everything work anyways by:
4382   //
4383   // - raising the soft limit on per-process file descriptors beyond
4384   //   256
4385   //
4386   // - As of Solaris 10u4, we can request that Solaris raise the 256
4387   //   stdio fopen limit by calling function enable_extended_FILE_stdio.
4388   //   This is done in init_2 and recorded in enabled_extended_FILE_stdio
4389   //
4390   // - If we are stuck on an old (pre 10u4) Solaris system, we can
4391   //   workaround the bug by remapping non-stdio file descriptors below
4392   //   256 to ones beyond 256, which is done below.
4393   //
4394   // See:
4395   // 1085341: 32-bit stdio routines should support file descriptors >255
4396   // 6533291: Work around 32-bit Solaris stdio limit of 256 open files
4397   // 6431278: Netbeans crash on 32 bit Solaris: need to call
4398   //          enable_extended_FILE_stdio() in VM initialisation
4399   // Giri Mandalika's blog
4400   // http://technopark02.blogspot.com/2005_05_01_archive.html
4401   //
4402 #ifndef  _LP64
4403   if ((!enabled_extended_FILE_stdio) && fd < 256) {
4404     int newfd = ::fcntl(fd, F_DUPFD, 256);
4405     if (newfd != -1) {
4406       ::close(fd);
4407       fd = newfd;
4408     }
4409   }
4410 #endif // 32-bit Solaris
4411 
4412   // All file descriptors that are opened in the JVM and not
4413   // specifically destined for a subprocess should have the
4414   // close-on-exec flag set.  If we don't set it, then careless 3rd
4415   // party native code might fork and exec without closing all
4416   // appropriate file descriptors (e.g. as we do in closeDescriptors in
4417   // UNIXProcess.c), and this in turn might:
4418   //
4419   // - cause end-of-file to fail to be detected on some file
4420   //   descriptors, resulting in mysterious hangs, or
4421   //
4422   // - might cause an fopen in the subprocess to fail on a system
4423   //   suffering from bug 1085341.
4424   //
4425   // (Yes, the default setting of the close-on-exec flag is a Unix
4426   // design flaw)
4427   //
4428   // See:
4429   // 1085341: 32-bit stdio routines should support file descriptors >255
4430   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4431   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4432   //
4433 #ifdef FD_CLOEXEC
4434   {
4435     int flags = ::fcntl(fd, F_GETFD);
4436     if (flags != -1) {
4437       ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4438     }
4439   }
4440 #endif
4441 
4442   return fd;
4443 }
4444 
4445 // create binary file, rewriting existing file if required
4446 int os::create_binary_file(const char* path, bool rewrite_existing) {
4447   int oflags = O_WRONLY | O_CREAT;
4448   if (!rewrite_existing) {
4449     oflags |= O_EXCL;
4450   }
4451   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4452 }
4453 
4454 // return current position of file pointer
4455 jlong os::current_file_offset(int fd) {
4456   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4457 }
4458 
4459 // move file pointer to the specified offset
4460 jlong os::seek_to_file_offset(int fd, jlong offset) {
4461   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4462 }
4463 
4464 jlong os::lseek(int fd, jlong offset, int whence) {
4465   return (jlong) ::lseek64(fd, offset, whence);
4466 }
4467 
4468 int os::ftruncate(int fd, jlong length) {
4469   return ::ftruncate64(fd, length);
4470 }
4471 
4472 int os::fsync(int fd)  {
4473   RESTARTABLE_RETURN_INT(::fsync(fd));
4474 }
4475 
4476 int os::available(int fd, jlong *bytes) {
4477   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
4478          "Assumed _thread_in_native");
4479   jlong cur, end;
4480   int mode;
4481   struct stat64 buf64;
4482 
4483   if (::fstat64(fd, &buf64) >= 0) {
4484     mode = buf64.st_mode;
4485     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4486       int n,ioctl_return;
4487 
4488       RESTARTABLE(::ioctl(fd, FIONREAD, &n), ioctl_return);
4489       if (ioctl_return>= 0) {
4490         *bytes = n;
4491         return 1;
4492       }
4493     }
4494   }
4495   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4496     return 0;
4497   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4498     return 0;
4499   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4500     return 0;
4501   }
4502   *bytes = end - cur;
4503   return 1;
4504 }
4505 
4506 // Map a block of memory.
4507 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4508                         char *addr, size_t bytes, bool read_only,
4509                         bool allow_exec) {
4510   int prot;
4511   int flags;
4512 
4513   if (read_only) {
4514     prot = PROT_READ;
4515     flags = MAP_SHARED;
4516   } else {
4517     prot = PROT_READ | PROT_WRITE;
4518     flags = MAP_PRIVATE;
4519   }
4520 
4521   if (allow_exec) {
4522     prot |= PROT_EXEC;
4523   }
4524 
4525   if (addr != NULL) {
4526     flags |= MAP_FIXED;
4527   }
4528 
4529   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4530                                      fd, file_offset);
4531   if (mapped_address == MAP_FAILED) {
4532     return NULL;
4533   }
4534   return mapped_address;
4535 }
4536 
4537 
4538 // Remap a block of memory.
4539 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4540                           char *addr, size_t bytes, bool read_only,
4541                           bool allow_exec) {
4542   // same as map_memory() on this OS
4543   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4544                         allow_exec);
4545 }
4546 
4547 
4548 // Unmap a block of memory.
4549 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4550   return munmap(addr, bytes) == 0;
4551 }
4552 
4553 void os::pause() {
4554   char filename[MAX_PATH];
4555   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4556     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4557   } else {
4558     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4559   }
4560 
4561   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4562   if (fd != -1) {
4563     struct stat buf;
4564     ::close(fd);
4565     while (::stat(filename, &buf) == 0) {
4566       (void)::poll(NULL, 0, 100);
4567     }
4568   } else {
4569     jio_fprintf(stderr,
4570                 "Could not open pause file '%s', continuing immediately.\n", filename);
4571   }
4572 }
4573 
4574 #ifndef PRODUCT
4575 #ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
4576 // Turn this on if you need to trace synch operations.
4577 // Set RECORD_SYNCH_LIMIT to a large-enough value,
4578 // and call record_synch_enable and record_synch_disable
4579 // around the computation of interest.
4580 
4581 void record_synch(char* name, bool returning);  // defined below
4582 
4583 class RecordSynch {
4584   char* _name;
4585  public:
4586   RecordSynch(char* name) :_name(name) { record_synch(_name, false); }
4587   ~RecordSynch()                       { record_synch(_name, true); }
4588 };
4589 
4590 #define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
4591 extern "C" ret name params {                                    \
4592   typedef ret name##_t params;                                  \
4593   static name##_t* implem = NULL;                               \
4594   static int callcount = 0;                                     \
4595   if (implem == NULL) {                                         \
4596     implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
4597     if (implem == NULL)  fatal(dlerror());                      \
4598   }                                                             \
4599   ++callcount;                                                  \
4600   RecordSynch _rs(#name);                                       \
4601   inner;                                                        \
4602   return implem args;                                           \
4603 }
4604 // in dbx, examine callcounts this way:
4605 // for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
4606 
4607 #define CHECK_POINTER_OK(p) \
4608   (!Universe::is_fully_initialized() || !Universe::is_reserved_heap((oop)(p)))
4609 #define CHECK_MU \
4610   if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
4611 #define CHECK_CV \
4612   if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
4613 #define CHECK_P(p) \
4614   if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
4615 
4616 #define CHECK_MUTEX(mutex_op) \
4617   CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
4618 
4619 CHECK_MUTEX(   mutex_lock)
4620 CHECK_MUTEX(  _mutex_lock)
4621 CHECK_MUTEX( mutex_unlock)
4622 CHECK_MUTEX(_mutex_unlock)
4623 CHECK_MUTEX( mutex_trylock)
4624 CHECK_MUTEX(_mutex_trylock)
4625 
4626 #define CHECK_COND(cond_op) \
4627   CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU; CHECK_CV);
4628 
4629 CHECK_COND( cond_wait);
4630 CHECK_COND(_cond_wait);
4631 CHECK_COND(_cond_wait_cancel);
4632 
4633 #define CHECK_COND2(cond_op) \
4634   CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU; CHECK_CV);
4635 
4636 CHECK_COND2( cond_timedwait);
4637 CHECK_COND2(_cond_timedwait);
4638 CHECK_COND2(_cond_timedwait_cancel);
4639 
4640 // do the _lwp_* versions too
4641 #define mutex_t lwp_mutex_t
4642 #define cond_t  lwp_cond_t
4643 CHECK_MUTEX(  _lwp_mutex_lock)
4644 CHECK_MUTEX(  _lwp_mutex_unlock)
4645 CHECK_MUTEX(  _lwp_mutex_trylock)
4646 CHECK_MUTEX( __lwp_mutex_lock)
4647 CHECK_MUTEX( __lwp_mutex_unlock)
4648 CHECK_MUTEX( __lwp_mutex_trylock)
4649 CHECK_MUTEX(___lwp_mutex_lock)
4650 CHECK_MUTEX(___lwp_mutex_unlock)
4651 
4652 CHECK_COND(  _lwp_cond_wait);
4653 CHECK_COND( __lwp_cond_wait);
4654 CHECK_COND(___lwp_cond_wait);
4655 
4656 CHECK_COND2(  _lwp_cond_timedwait);
4657 CHECK_COND2( __lwp_cond_timedwait);
4658 #undef mutex_t
4659 #undef cond_t
4660 
4661 CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4662 CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4663 CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4664 CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4665 CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4666 CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4667 CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4668 CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4669 
4670 
4671 // recording machinery:
4672 
4673 enum { RECORD_SYNCH_LIMIT = 200 };
4674 char* record_synch_name[RECORD_SYNCH_LIMIT];
4675 void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
4676 bool record_synch_returning[RECORD_SYNCH_LIMIT];
4677 thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
4678 int record_synch_count = 0;
4679 bool record_synch_enabled = false;
4680 
4681 // in dbx, examine recorded data this way:
4682 // for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
4683 
4684 void record_synch(char* name, bool returning) {
4685   if (record_synch_enabled) {
4686     if (record_synch_count < RECORD_SYNCH_LIMIT) {
4687       record_synch_name[record_synch_count] = name;
4688       record_synch_returning[record_synch_count] = returning;
4689       record_synch_thread[record_synch_count] = thr_self();
4690       record_synch_arg0ptr[record_synch_count] = &name;
4691       record_synch_count++;
4692     }
4693     // put more checking code here:
4694     // ...
4695   }
4696 }
4697 
4698 void record_synch_enable() {
4699   // start collecting trace data, if not already doing so
4700   if (!record_synch_enabled)  record_synch_count = 0;
4701   record_synch_enabled = true;
4702 }
4703 
4704 void record_synch_disable() {
4705   // stop collecting trace data
4706   record_synch_enabled = false;
4707 }
4708 
4709 #endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
4710 #endif // PRODUCT
4711 
4712 const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
4713 const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
4714                                (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
4715 
4716 
4717 // JVMTI & JVM monitoring and management support
4718 // The thread_cpu_time() and current_thread_cpu_time() are only
4719 // supported if is_thread_cpu_time_supported() returns true.
4720 // They are not supported on Solaris T1.
4721 
4722 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4723 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4724 // of a thread.
4725 //
4726 // current_thread_cpu_time() and thread_cpu_time(Thread *)
4727 // returns the fast estimate available on the platform.
4728 
4729 // hrtime_t gethrvtime() return value includes
4730 // user time but does not include system time
4731 jlong os::current_thread_cpu_time() {
4732   return (jlong) gethrvtime();
4733 }
4734 
4735 jlong os::thread_cpu_time(Thread *thread) {
4736   // return user level CPU time only to be consistent with
4737   // what current_thread_cpu_time returns.
4738   // thread_cpu_time_info() must be changed if this changes
4739   return os::thread_cpu_time(thread, false /* user time only */);
4740 }
4741 
4742 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4743   if (user_sys_cpu_time) {
4744     return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
4745   } else {
4746     return os::current_thread_cpu_time();
4747   }
4748 }
4749 
4750 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4751   char proc_name[64];
4752   int count;
4753   prusage_t prusage;
4754   jlong lwp_time;
4755   int fd;
4756 
4757   sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
4758           getpid(),
4759           thread->osthread()->lwp_id());
4760   fd = ::open(proc_name, O_RDONLY);
4761   if (fd == -1) return -1;
4762 
4763   do {
4764     count = ::pread(fd,
4765                     (void *)&prusage.pr_utime,
4766                     thr_time_size,
4767                     thr_time_off);
4768   } while (count < 0 && errno == EINTR);
4769   ::close(fd);
4770   if (count < 0) return -1;
4771 
4772   if (user_sys_cpu_time) {
4773     // user + system CPU time
4774     lwp_time = (((jlong)prusage.pr_stime.tv_sec +
4775                  (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
4776                  (jlong)prusage.pr_stime.tv_nsec +
4777                  (jlong)prusage.pr_utime.tv_nsec;
4778   } else {
4779     // user level CPU time only
4780     lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
4781                 (jlong)prusage.pr_utime.tv_nsec;
4782   }
4783 
4784   return (lwp_time);
4785 }
4786 
4787 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4788   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
4789   info_ptr->may_skip_backward = false;    // elapsed time not wall time
4790   info_ptr->may_skip_forward = false;     // elapsed time not wall time
4791   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
4792 }
4793 
4794 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4795   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
4796   info_ptr->may_skip_backward = false;    // elapsed time not wall time
4797   info_ptr->may_skip_forward = false;     // elapsed time not wall time
4798   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
4799 }
4800 
4801 bool os::is_thread_cpu_time_supported() {
4802   return true;
4803 }
4804 
4805 // System loadavg support.  Returns -1 if load average cannot be obtained.
4806 // Return the load average for our processor set if the primitive exists
4807 // (Solaris 9 and later).  Otherwise just return system wide loadavg.
4808 int os::loadavg(double loadavg[], int nelem) {
4809   if (pset_getloadavg_ptr != NULL) {
4810     return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
4811   } else {
4812     return ::getloadavg(loadavg, nelem);
4813   }
4814 }
4815 
4816 //---------------------------------------------------------------------------------
4817 
4818 bool os::find(address addr, outputStream* st) {
4819   Dl_info dlinfo;
4820   memset(&dlinfo, 0, sizeof(dlinfo));
4821   if (dladdr(addr, &dlinfo) != 0) {
4822     st->print(PTR_FORMAT ": ", addr);
4823     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4824       st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
4825     } else if (dlinfo.dli_fbase != NULL) {
4826       st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
4827     } else {
4828       st->print("<absolute address>");
4829     }
4830     if (dlinfo.dli_fname != NULL) {
4831       st->print(" in %s", dlinfo.dli_fname);
4832     }
4833     if (dlinfo.dli_fbase != NULL) {
4834       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4835     }
4836     st->cr();
4837 
4838     if (Verbose) {
4839       // decode some bytes around the PC
4840       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4841       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4842       address       lowest = (address) dlinfo.dli_sname;
4843       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4844       if (begin < lowest)  begin = lowest;
4845       Dl_info dlinfo2;
4846       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4847           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4848         end = (address) dlinfo2.dli_saddr;
4849       }
4850       Disassembler::decode(begin, end, st);
4851     }
4852     return true;
4853   }
4854   return false;
4855 }
4856 
4857 // Following function has been added to support HotSparc's libjvm.so running
4858 // under Solaris production JDK 1.2.2 / 1.3.0.  These came from
4859 // src/solaris/hpi/native_threads in the EVM codebase.
4860 //
4861 // NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
4862 // libraries and should thus be removed. We will leave it behind for a while
4863 // until we no longer want to able to run on top of 1.3.0 Solaris production
4864 // JDK. See 4341971.
4865 
4866 #define STACK_SLACK 0x800
4867 
4868 extern "C" {
4869   intptr_t sysThreadAvailableStackWithSlack() {
4870     stack_t st;
4871     intptr_t retval, stack_top;
4872     retval = thr_stksegment(&st);
4873     assert(retval == 0, "incorrect return value from thr_stksegment");
4874     assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
4875     assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
4876     stack_top=(intptr_t)st.ss_sp-st.ss_size;
4877     return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
4878   }
4879 }
4880 
4881 // ObjectMonitor park-unpark infrastructure ...
4882 //
4883 // We implement Solaris and Linux PlatformEvents with the
4884 // obvious condvar-mutex-flag triple.
4885 // Another alternative that works quite well is pipes:
4886 // Each PlatformEvent consists of a pipe-pair.
4887 // The thread associated with the PlatformEvent
4888 // calls park(), which reads from the input end of the pipe.
4889 // Unpark() writes into the other end of the pipe.
4890 // The write-side of the pipe must be set NDELAY.
4891 // Unfortunately pipes consume a large # of handles.
4892 // Native solaris lwp_park() and lwp_unpark() work nicely, too.
4893 // Using pipes for the 1st few threads might be workable, however.
4894 //
4895 // park() is permitted to return spuriously.
4896 // Callers of park() should wrap the call to park() in
4897 // an appropriate loop.  A litmus test for the correct
4898 // usage of park is the following: if park() were modified
4899 // to immediately return 0 your code should still work,
4900 // albeit degenerating to a spin loop.
4901 //
4902 // In a sense, park()-unpark() just provides more polite spinning
4903 // and polling with the key difference over naive spinning being
4904 // that a parked thread needs to be explicitly unparked() in order
4905 // to wake up and to poll the underlying condition.
4906 //
4907 // Assumption:
4908 //    Only one parker can exist on an event, which is why we allocate
4909 //    them per-thread. Multiple unparkers can coexist.
4910 //
4911 // _Event transitions in park()
4912 //   -1 => -1 : illegal
4913 //    1 =>  0 : pass - return immediately
4914 //    0 => -1 : block; then set _Event to 0 before returning
4915 //
4916 // _Event transitions in unpark()
4917 //    0 => 1 : just return
4918 //    1 => 1 : just return
4919 //   -1 => either 0 or 1; must signal target thread
4920 //         That is, we can safely transition _Event from -1 to either
4921 //         0 or 1.
4922 //
4923 // _Event serves as a restricted-range semaphore.
4924 //   -1 : thread is blocked, i.e. there is a waiter
4925 //    0 : neutral: thread is running or ready,
4926 //        could have been signaled after a wait started
4927 //    1 : signaled - thread is running or ready
4928 //
4929 // Another possible encoding of _Event would be with
4930 // explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
4931 //
4932 // TODO-FIXME: add DTRACE probes for:
4933 // 1.   Tx parks
4934 // 2.   Ty unparks Tx
4935 // 3.   Tx resumes from park
4936 
4937 
4938 // value determined through experimentation
4939 #define ROUNDINGFIX 11
4940 
4941 // utility to compute the abstime argument to timedwait.
4942 // TODO-FIXME: switch from compute_abstime() to unpackTime().
4943 
4944 static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
4945   // millis is the relative timeout time
4946   // abstime will be the absolute timeout time
4947   if (millis < 0)  millis = 0;
4948   struct timeval now;
4949   int status = gettimeofday(&now, NULL);
4950   assert(status == 0, "gettimeofday");
4951   jlong seconds = millis / 1000;
4952   jlong max_wait_period;
4953 
4954   if (UseLWPSynchronization) {
4955     // forward port of fix for 4275818 (not sleeping long enough)
4956     // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
4957     // _lwp_cond_timedwait() used a round_down algorithm rather
4958     // than a round_up. For millis less than our roundfactor
4959     // it rounded down to 0 which doesn't meet the spec.
4960     // For millis > roundfactor we may return a bit sooner, but
4961     // since we can not accurately identify the patch level and
4962     // this has already been fixed in Solaris 9 and 8 we will
4963     // leave it alone rather than always rounding down.
4964 
4965     if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
4966     // It appears that when we go directly through Solaris _lwp_cond_timedwait()
4967     // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
4968     max_wait_period = 21000000;
4969   } else {
4970     max_wait_period = 50000000;
4971   }
4972   millis %= 1000;
4973   if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
4974     seconds = max_wait_period;
4975   }
4976   abstime->tv_sec = now.tv_sec  + seconds;
4977   long       usec = now.tv_usec + millis * 1000;
4978   if (usec >= 1000000) {
4979     abstime->tv_sec += 1;
4980     usec -= 1000000;
4981   }
4982   abstime->tv_nsec = usec * 1000;
4983   return abstime;
4984 }
4985 
4986 void os::PlatformEvent::park() {           // AKA: down()
4987   // Transitions for _Event:
4988   //   -1 => -1 : illegal
4989   //    1 =>  0 : pass - return immediately
4990   //    0 => -1 : block; then set _Event to 0 before returning
4991 
4992   // Invariant: Only the thread associated with the Event/PlatformEvent
4993   // may call park().
4994   assert(_nParked == 0, "invariant");
4995 
4996   int v;
4997   for (;;) {
4998     v = _Event;
4999     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5000   }
5001   guarantee(v >= 0, "invariant");
5002   if (v == 0) {
5003     // Do this the hard way by blocking ...
5004     // See http://monaco.sfbay/detail.jsf?cr=5094058.
5005     int status = os::Solaris::mutex_lock(_mutex);
5006     assert_status(status == 0, status, "mutex_lock");
5007     guarantee(_nParked == 0, "invariant");
5008     ++_nParked;
5009     while (_Event < 0) {
5010       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5011       // Treat this the same as if the wait was interrupted
5012       // With usr/lib/lwp going to kernel, always handle ETIME
5013       status = os::Solaris::cond_wait(_cond, _mutex);
5014       if (status == ETIME) status = EINTR;
5015       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5016     }
5017     --_nParked;
5018     _Event = 0;
5019     status = os::Solaris::mutex_unlock(_mutex);
5020     assert_status(status == 0, status, "mutex_unlock");
5021     // Paranoia to ensure our locked and lock-free paths interact
5022     // correctly with each other.
5023     OrderAccess::fence();
5024   }
5025 }
5026 
5027 int os::PlatformEvent::park(jlong millis) {
5028   // Transitions for _Event:
5029   //   -1 => -1 : illegal
5030   //    1 =>  0 : pass - return immediately
5031   //    0 => -1 : block; then set _Event to 0 before returning
5032 
5033   guarantee(_nParked == 0, "invariant");
5034   int v;
5035   for (;;) {
5036     v = _Event;
5037     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5038   }
5039   guarantee(v >= 0, "invariant");
5040   if (v != 0) return OS_OK;
5041 
5042   int ret = OS_TIMEOUT;
5043   timestruc_t abst;
5044   compute_abstime(&abst, millis);
5045 
5046   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5047   int status = os::Solaris::mutex_lock(_mutex);
5048   assert_status(status == 0, status, "mutex_lock");
5049   guarantee(_nParked == 0, "invariant");
5050   ++_nParked;
5051   while (_Event < 0) {
5052     int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
5053     assert_status(status == 0 || status == EINTR ||
5054                   status == ETIME || status == ETIMEDOUT,
5055                   status, "cond_timedwait");
5056     if (!FilterSpuriousWakeups) break;                // previous semantics
5057     if (status == ETIME || status == ETIMEDOUT) break;
5058     // We consume and ignore EINTR and spurious wakeups.
5059   }
5060   --_nParked;
5061   if (_Event >= 0) ret = OS_OK;
5062   _Event = 0;
5063   status = os::Solaris::mutex_unlock(_mutex);
5064   assert_status(status == 0, status, "mutex_unlock");
5065   // Paranoia to ensure our locked and lock-free paths interact
5066   // correctly with each other.
5067   OrderAccess::fence();
5068   return ret;
5069 }
5070 
5071 void os::PlatformEvent::unpark() {
5072   // Transitions for _Event:
5073   //    0 => 1 : just return
5074   //    1 => 1 : just return
5075   //   -1 => either 0 or 1; must signal target thread
5076   //         That is, we can safely transition _Event from -1 to either
5077   //         0 or 1.
5078   // See also: "Semaphores in Plan 9" by Mullender & Cox
5079   //
5080   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5081   // that it will take two back-to-back park() calls for the owning
5082   // thread to block. This has the benefit of forcing a spurious return
5083   // from the first park() call after an unpark() call which will help
5084   // shake out uses of park() and unpark() without condition variables.
5085 
5086   if (Atomic::xchg(1, &_Event) >= 0) return;
5087 
5088   // If the thread associated with the event was parked, wake it.
5089   // Wait for the thread assoc with the PlatformEvent to vacate.
5090   int status = os::Solaris::mutex_lock(_mutex);
5091   assert_status(status == 0, status, "mutex_lock");
5092   int AnyWaiters = _nParked;
5093   status = os::Solaris::mutex_unlock(_mutex);
5094   assert_status(status == 0, status, "mutex_unlock");
5095   guarantee(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5096   if (AnyWaiters != 0) {
5097     // Note that we signal() *after* dropping the lock for "immortal" Events.
5098     // This is safe and avoids a common class of  futile wakeups.  In rare
5099     // circumstances this can cause a thread to return prematurely from
5100     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5101     // will simply re-test the condition and re-park itself.
5102     // This provides particular benefit if the underlying platform does not
5103     // provide wait morphing.
5104     status = os::Solaris::cond_signal(_cond);
5105     assert_status(status == 0, status, "cond_signal");
5106   }
5107 }
5108 
5109 // JSR166
5110 // -------------------------------------------------------
5111 
5112 // The solaris and linux implementations of park/unpark are fairly
5113 // conservative for now, but can be improved. They currently use a
5114 // mutex/condvar pair, plus _counter.
5115 // Park decrements _counter if > 0, else does a condvar wait.  Unpark
5116 // sets count to 1 and signals condvar.  Only one thread ever waits
5117 // on the condvar. Contention seen when trying to park implies that someone
5118 // is unparking you, so don't wait. And spurious returns are fine, so there
5119 // is no need to track notifications.
5120 
5121 #define MAX_SECS 100000000
5122 
5123 // This code is common to linux and solaris and will be moved to a
5124 // common place in dolphin.
5125 //
5126 // The passed in time value is either a relative time in nanoseconds
5127 // or an absolute time in milliseconds. Either way it has to be unpacked
5128 // into suitable seconds and nanoseconds components and stored in the
5129 // given timespec structure.
5130 // Given time is a 64-bit value and the time_t used in the timespec is only
5131 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5132 // overflow if times way in the future are given. Further on Solaris versions
5133 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5134 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5135 // As it will be 28 years before "now + 100000000" will overflow we can
5136 // ignore overflow and just impose a hard-limit on seconds using the value
5137 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5138 // years from "now".
5139 //
5140 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5141   assert(time > 0, "convertTime");
5142 
5143   struct timeval now;
5144   int status = gettimeofday(&now, NULL);
5145   assert(status == 0, "gettimeofday");
5146 
5147   time_t max_secs = now.tv_sec + MAX_SECS;
5148 
5149   if (isAbsolute) {
5150     jlong secs = time / 1000;
5151     if (secs > max_secs) {
5152       absTime->tv_sec = max_secs;
5153     } else {
5154       absTime->tv_sec = secs;
5155     }
5156     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5157   } else {
5158     jlong secs = time / NANOSECS_PER_SEC;
5159     if (secs >= MAX_SECS) {
5160       absTime->tv_sec = max_secs;
5161       absTime->tv_nsec = 0;
5162     } else {
5163       absTime->tv_sec = now.tv_sec + secs;
5164       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5165       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5166         absTime->tv_nsec -= NANOSECS_PER_SEC;
5167         ++absTime->tv_sec; // note: this must be <= max_secs
5168       }
5169     }
5170   }
5171   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5172   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5173   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5174   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5175 }
5176 
5177 void Parker::park(bool isAbsolute, jlong time) {
5178   // Ideally we'd do something useful while spinning, such
5179   // as calling unpackTime().
5180 
5181   // Optional fast-path check:
5182   // Return immediately if a permit is available.
5183   // We depend on Atomic::xchg() having full barrier semantics
5184   // since we are doing a lock-free update to _counter.
5185   if (Atomic::xchg(0, &_counter) > 0) return;
5186 
5187   // Optional fast-exit: Check interrupt before trying to wait
5188   Thread* thread = Thread::current();
5189   assert(thread->is_Java_thread(), "Must be JavaThread");
5190   JavaThread *jt = (JavaThread *)thread;
5191   if (Thread::is_interrupted(thread, false)) {
5192     return;
5193   }
5194 
5195   // First, demultiplex/decode time arguments
5196   timespec absTime;
5197   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5198     return;
5199   }
5200   if (time > 0) {
5201     // Warning: this code might be exposed to the old Solaris time
5202     // round-down bugs.  Grep "roundingFix" for details.
5203     unpackTime(&absTime, isAbsolute, time);
5204   }
5205 
5206   // Enter safepoint region
5207   // Beware of deadlocks such as 6317397.
5208   // The per-thread Parker:: _mutex is a classic leaf-lock.
5209   // In particular a thread must never block on the Threads_lock while
5210   // holding the Parker:: mutex.  If safepoints are pending both the
5211   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5212   ThreadBlockInVM tbivm(jt);
5213 
5214   // Don't wait if cannot get lock since interference arises from
5215   // unblocking.  Also. check interrupt before trying wait
5216   if (Thread::is_interrupted(thread, false) ||
5217       os::Solaris::mutex_trylock(_mutex) != 0) {
5218     return;
5219   }
5220 
5221   int status;
5222 
5223   if (_counter > 0)  { // no wait needed
5224     _counter = 0;
5225     status = os::Solaris::mutex_unlock(_mutex);
5226     assert(status == 0, "invariant");
5227     // Paranoia to ensure our locked and lock-free paths interact
5228     // correctly with each other and Java-level accesses.
5229     OrderAccess::fence();
5230     return;
5231   }
5232 
5233   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5234   jt->set_suspend_equivalent();
5235   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5236 
5237   // Do this the hard way by blocking ...
5238   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5239   if (time == 0) {
5240     status = os::Solaris::cond_wait(_cond, _mutex);
5241   } else {
5242     status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
5243   }
5244   // Note that an untimed cond_wait() can sometimes return ETIME on older
5245   // versions of the Solaris.
5246   assert_status(status == 0 || status == EINTR ||
5247                 status == ETIME || status == ETIMEDOUT,
5248                 status, "cond_timedwait");
5249 
5250   _counter = 0;
5251   status = os::Solaris::mutex_unlock(_mutex);
5252   assert_status(status == 0, status, "mutex_unlock");
5253   // Paranoia to ensure our locked and lock-free paths interact
5254   // correctly with each other and Java-level accesses.
5255   OrderAccess::fence();
5256 
5257   // If externally suspended while waiting, re-suspend
5258   if (jt->handle_special_suspend_equivalent_condition()) {
5259     jt->java_suspend_self();
5260   }
5261 }
5262 
5263 void Parker::unpark() {
5264   int status = os::Solaris::mutex_lock(_mutex);
5265   assert(status == 0, "invariant");
5266   const int s = _counter;
5267   _counter = 1;
5268   status = os::Solaris::mutex_unlock(_mutex);
5269   assert(status == 0, "invariant");
5270 
5271   if (s < 1) {
5272     status = os::Solaris::cond_signal(_cond);
5273     assert(status == 0, "invariant");
5274   }
5275 }
5276 
5277 extern char** environ;
5278 
5279 // Run the specified command in a separate process. Return its exit value,
5280 // or -1 on failure (e.g. can't fork a new process).
5281 // Unlike system(), this function can be called from signal handler. It
5282 // doesn't block SIGINT et al.
5283 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
5284   char * argv[4];
5285   argv[0] = (char *)"sh";
5286   argv[1] = (char *)"-c";
5287   argv[2] = cmd;
5288   argv[3] = NULL;
5289 
5290   // fork is async-safe, fork1 is not so can't use in signal handler
5291   pid_t pid;
5292   Thread* t = Thread::current_or_null_safe();
5293   if (t != NULL && t->is_inside_signal_handler()) {
5294     pid = fork();
5295   } else {
5296     pid = fork1();
5297   }
5298 
5299   if (pid < 0) {
5300     // fork failed
5301     warning("fork failed: %s", os::strerror(errno));
5302     return -1;
5303 
5304   } else if (pid == 0) {
5305     // child process
5306 
5307     // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
5308     execve("/usr/bin/sh", argv, environ);
5309 
5310     // execve failed
5311     _exit(-1);
5312 
5313   } else  {
5314     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5315     // care about the actual exit code, for now.
5316 
5317     int status;
5318 
5319     // Wait for the child process to exit.  This returns immediately if
5320     // the child has already exited. */
5321     while (waitpid(pid, &status, 0) < 0) {
5322       switch (errno) {
5323       case ECHILD: return 0;
5324       case EINTR: break;
5325       default: return -1;
5326       }
5327     }
5328 
5329     if (WIFEXITED(status)) {
5330       // The child exited normally; get its exit code.
5331       return WEXITSTATUS(status);
5332     } else if (WIFSIGNALED(status)) {
5333       // The child exited because of a signal
5334       // The best value to return is 0x80 + signal number,
5335       // because that is what all Unix shells do, and because
5336       // it allows callers to distinguish between process exit and
5337       // process death by signal.
5338       return 0x80 + WTERMSIG(status);
5339     } else {
5340       // Unknown exit code; pass it through
5341       return status;
5342     }
5343   }
5344 }
5345 
5346 size_t os::write(int fd, const void *buf, unsigned int nBytes) {
5347   size_t res;
5348   RESTARTABLE((size_t) ::write(fd, buf, (size_t) nBytes), res);
5349   return res;
5350 }
5351 
5352 int os::close(int fd) {
5353   return ::close(fd);
5354 }
5355 
5356 int os::socket_close(int fd) {
5357   return ::close(fd);
5358 }
5359 
5360 int os::recv(int fd, char* buf, size_t nBytes, uint flags) {
5361   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5362          "Assumed _thread_in_native");
5363   RESTARTABLE_RETURN_INT((int)::recv(fd, buf, nBytes, flags));
5364 }
5365 
5366 int os::send(int fd, char* buf, size_t nBytes, uint flags) {
5367   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5368          "Assumed _thread_in_native");
5369   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5370 }
5371 
5372 int os::raw_send(int fd, char* buf, size_t nBytes, uint flags) {
5373   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5374 }
5375 
5376 // As both poll and select can be interrupted by signals, we have to be
5377 // prepared to restart the system call after updating the timeout, unless
5378 // a poll() is done with timeout == -1, in which case we repeat with this
5379 // "wait forever" value.
5380 
5381 int os::connect(int fd, struct sockaddr *him, socklen_t len) {
5382   int _result;
5383   _result = ::connect(fd, him, len);
5384 
5385   // On Solaris, when a connect() call is interrupted, the connection
5386   // can be established asynchronously (see 6343810). Subsequent calls
5387   // to connect() must check the errno value which has the semantic
5388   // described below (copied from the connect() man page). Handling
5389   // of asynchronously established connections is required for both
5390   // blocking and non-blocking sockets.
5391   //     EINTR            The  connection  attempt  was   interrupted
5392   //                      before  any data arrived by the delivery of
5393   //                      a signal. The connection, however, will  be
5394   //                      established asynchronously.
5395   //
5396   //     EINPROGRESS      The socket is non-blocking, and the connec-
5397   //                      tion  cannot  be completed immediately.
5398   //
5399   //     EALREADY         The socket is non-blocking,  and a previous
5400   //                      connection  attempt  has  not yet been com-
5401   //                      pleted.
5402   //
5403   //     EISCONN          The socket is already connected.
5404   if (_result == OS_ERR && errno == EINTR) {
5405     // restarting a connect() changes its errno semantics
5406     RESTARTABLE(::connect(fd, him, len), _result);
5407     // undo these changes
5408     if (_result == OS_ERR) {
5409       if (errno == EALREADY) {
5410         errno = EINPROGRESS; // fall through
5411       } else if (errno == EISCONN) {
5412         errno = 0;
5413         return OS_OK;
5414       }
5415     }
5416   }
5417   return _result;
5418 }
5419 
5420 // Get the default path to the core file
5421 // Returns the length of the string
5422 int os::get_core_path(char* buffer, size_t bufferSize) {
5423   const char* p = get_current_directory(buffer, bufferSize);
5424 
5425   if (p == NULL) {
5426     assert(p != NULL, "failed to get current directory");
5427     return 0;
5428   }
5429 
5430   jio_snprintf(buffer, bufferSize, "%s/core or core.%d",
5431                                               p, current_process_id());
5432 
5433   return strlen(buffer);
5434 }
5435 
5436 #ifndef PRODUCT
5437 void TestReserveMemorySpecial_test() {
5438   // No tests available for this platform
5439 }
5440 #endif
5441 
5442 bool os::start_debugging(char *buf, int buflen) {
5443   int len = (int)strlen(buf);
5444   char *p = &buf[len];
5445 
5446   jio_snprintf(p, buflen-len,
5447                "\n\n"
5448                "Do you want to debug the problem?\n\n"
5449                "To debug, run 'dbx - %d'; then switch to thread " INTX_FORMAT "\n"
5450                "Enter 'yes' to launch dbx automatically (PATH must include dbx)\n"
5451                "Otherwise, press RETURN to abort...",
5452                os::current_process_id(), os::current_thread_id());
5453 
5454   bool yes = os::message_box("Unexpected Error", buf);
5455 
5456   if (yes) {
5457     // yes, user asked VM to launch debugger
5458     jio_snprintf(buf, sizeof(buf), "dbx - %d", os::current_process_id());
5459 
5460     os::fork_and_exec(buf);
5461     yes = false;
5462   }
5463   return yes;
5464 }