1 /*
   2  * Copyright (c) 1997, 2017, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "classfile/classLoader.hpp"
  27 #include "classfile/systemDictionary.hpp"
  28 #include "classfile/vmSymbols.hpp"
  29 #include "code/icBuffer.hpp"
  30 #include "code/vtableStubs.hpp"
  31 #include "compiler/compileBroker.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "jvm_solaris.h"
  35 #include "logging/log.hpp"
  36 #include "memory/allocation.inline.hpp"
  37 #include "memory/filemap.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_share_solaris.hpp"
  40 #include "os_solaris.inline.hpp"
  41 #include "prims/jniFastGetField.hpp"
  42 #include "prims/jvm.h"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.hpp"
  49 #include "runtime/java.hpp"
  50 #include "runtime/javaCalls.hpp"
  51 #include "runtime/mutexLocker.hpp"
  52 #include "runtime/objectMonitor.hpp"
  53 #include "runtime/orderAccess.inline.hpp"
  54 #include "runtime/osThread.hpp"
  55 #include "runtime/perfMemory.hpp"
  56 #include "runtime/sharedRuntime.hpp"
  57 #include "runtime/statSampler.hpp"
  58 #include "runtime/stubRoutines.hpp"
  59 #include "runtime/thread.inline.hpp"
  60 #include "runtime/threadCritical.hpp"
  61 #include "runtime/timer.hpp"
  62 #include "runtime/vm_version.hpp"
  63 #include "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 address os::current_stack_base() {
 205   int r = thr_main();
 206   guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
 207   bool is_primordial_thread = r;
 208 
 209   // Workaround 4352906, avoid calls to thr_stksegment by
 210   // thr_main after the first one (it looks like we trash
 211   // some data, causing the value for ss_sp to be incorrect).
 212   if (!is_primordial_thread || os::Solaris::_main_stack_base == NULL) {
 213     stack_t st = get_stack_info();
 214     if (is_primordial_thread) {
 215       // cache initial value of stack base
 216       os::Solaris::_main_stack_base = (address)st.ss_sp;
 217     }
 218     return (address)st.ss_sp;
 219   } else {
 220     guarantee(os::Solaris::_main_stack_base != NULL, "Attempt to use null cached stack base");
 221     return os::Solaris::_main_stack_base;
 222   }
 223 }
 224 
 225 size_t os::current_stack_size() {
 226   size_t size;
 227 
 228   int r = thr_main();
 229   guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
 230   if (!r) {
 231     size = get_stack_info().ss_size;
 232   } else {
 233     struct rlimit limits;
 234     getrlimit(RLIMIT_STACK, &limits);
 235     size = adjust_stack_size(os::Solaris::_main_stack_base, (size_t)limits.rlim_cur);
 236   }
 237   // base may not be page aligned
 238   address base = current_stack_base();
 239   address bottom = align_up(base - size, os::vm_page_size());;
 240   return (size_t)(base - bottom);
 241 }
 242 
 243 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
 244   return localtime_r(clock, res);
 245 }
 246 
 247 void os::Solaris::try_enable_extended_io() {
 248   typedef int (*enable_extended_FILE_stdio_t)(int, int);
 249 
 250   if (!UseExtendedFileIO) {
 251     return;
 252   }
 253 
 254   enable_extended_FILE_stdio_t enabler =
 255     (enable_extended_FILE_stdio_t) dlsym(RTLD_DEFAULT,
 256                                          "enable_extended_FILE_stdio");
 257   if (enabler) {
 258     enabler(-1, -1);
 259   }
 260 }
 261 
 262 static int _processors_online = 0;
 263 
 264 jint os::Solaris::_os_thread_limit = 0;
 265 volatile jint os::Solaris::_os_thread_count = 0;
 266 
 267 julong os::available_memory() {
 268   return Solaris::available_memory();
 269 }
 270 
 271 julong os::Solaris::available_memory() {
 272   return (julong)sysconf(_SC_AVPHYS_PAGES) * os::vm_page_size();
 273 }
 274 
 275 julong os::Solaris::_physical_memory = 0;
 276 
 277 julong os::physical_memory() {
 278   return Solaris::physical_memory();
 279 }
 280 
 281 static hrtime_t first_hrtime = 0;
 282 static const hrtime_t hrtime_hz = 1000*1000*1000;
 283 static volatile hrtime_t max_hrtime = 0;
 284 
 285 
 286 void os::Solaris::initialize_system_info() {
 287   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 288   _processors_online = sysconf(_SC_NPROCESSORS_ONLN);
 289   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) *
 290                                      (julong)sysconf(_SC_PAGESIZE);
 291 }
 292 
 293 int os::active_processor_count() {
 294   int online_cpus = sysconf(_SC_NPROCESSORS_ONLN);
 295   pid_t pid = getpid();
 296   psetid_t pset = PS_NONE;
 297   // Are we running in a processor set or is there any processor set around?
 298   if (pset_bind(PS_QUERY, P_PID, pid, &pset) == 0) {
 299     uint_t pset_cpus;
 300     // Query the number of cpus available to us.
 301     if (pset_info(pset, NULL, &pset_cpus, NULL) == 0) {
 302       assert(pset_cpus > 0 && pset_cpus <= online_cpus, "sanity check");
 303       _processors_online = pset_cpus;
 304       return pset_cpus;
 305     }
 306   }
 307   // Otherwise return number of online cpus
 308   return online_cpus;
 309 }
 310 
 311 static bool find_processors_in_pset(psetid_t        pset,
 312                                     processorid_t** id_array,
 313                                     uint_t*         id_length) {
 314   bool result = false;
 315   // Find the number of processors in the processor set.
 316   if (pset_info(pset, NULL, id_length, NULL) == 0) {
 317     // Make up an array to hold their ids.
 318     *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
 319     // Fill in the array with their processor ids.
 320     if (pset_info(pset, NULL, id_length, *id_array) == 0) {
 321       result = true;
 322     }
 323   }
 324   return result;
 325 }
 326 
 327 // Callers of find_processors_online() must tolerate imprecise results --
 328 // the system configuration can change asynchronously because of DR
 329 // or explicit psradm operations.
 330 //
 331 // We also need to take care that the loop (below) terminates as the
 332 // number of processors online can change between the _SC_NPROCESSORS_ONLN
 333 // request and the loop that builds the list of processor ids.   Unfortunately
 334 // there's no reliable way to determine the maximum valid processor id,
 335 // so we use a manifest constant, MAX_PROCESSOR_ID, instead.  See p_online
 336 // man pages, which claim the processor id set is "sparse, but
 337 // not too sparse".  MAX_PROCESSOR_ID is used to ensure that we eventually
 338 // exit the loop.
 339 //
 340 // In the future we'll be able to use sysconf(_SC_CPUID_MAX), but that's
 341 // not available on S8.0.
 342 
 343 static bool find_processors_online(processorid_t** id_array,
 344                                    uint*           id_length) {
 345   const processorid_t MAX_PROCESSOR_ID = 100000;
 346   // Find the number of processors online.
 347   *id_length = sysconf(_SC_NPROCESSORS_ONLN);
 348   // Make up an array to hold their ids.
 349   *id_array = NEW_C_HEAP_ARRAY(processorid_t, *id_length, mtInternal);
 350   // Processors need not be numbered consecutively.
 351   long found = 0;
 352   processorid_t next = 0;
 353   while (found < *id_length && next < MAX_PROCESSOR_ID) {
 354     processor_info_t info;
 355     if (processor_info(next, &info) == 0) {
 356       // NB, PI_NOINTR processors are effectively online ...
 357       if (info.pi_state == P_ONLINE || info.pi_state == P_NOINTR) {
 358         (*id_array)[found] = next;
 359         found += 1;
 360       }
 361     }
 362     next += 1;
 363   }
 364   if (found < *id_length) {
 365     // The loop above didn't identify the expected number of processors.
 366     // We could always retry the operation, calling sysconf(_SC_NPROCESSORS_ONLN)
 367     // and re-running the loop, above, but there's no guarantee of progress
 368     // if the system configuration is in flux.  Instead, we just return what
 369     // we've got.  Note that in the worst case find_processors_online() could
 370     // return an empty set.  (As a fall-back in the case of the empty set we
 371     // could just return the ID of the current processor).
 372     *id_length = found;
 373   }
 374 
 375   return true;
 376 }
 377 
 378 static bool assign_distribution(processorid_t* id_array,
 379                                 uint           id_length,
 380                                 uint*          distribution,
 381                                 uint           distribution_length) {
 382   // We assume we can assign processorid_t's to uint's.
 383   assert(sizeof(processorid_t) == sizeof(uint),
 384          "can't convert processorid_t to uint");
 385   // Quick check to see if we won't succeed.
 386   if (id_length < distribution_length) {
 387     return false;
 388   }
 389   // Assign processor ids to the distribution.
 390   // Try to shuffle processors to distribute work across boards,
 391   // assuming 4 processors per board.
 392   const uint processors_per_board = ProcessDistributionStride;
 393   // Find the maximum processor id.
 394   processorid_t max_id = 0;
 395   for (uint m = 0; m < id_length; m += 1) {
 396     max_id = MAX2(max_id, id_array[m]);
 397   }
 398   // The next id, to limit loops.
 399   const processorid_t limit_id = max_id + 1;
 400   // Make up markers for available processors.
 401   bool* available_id = NEW_C_HEAP_ARRAY(bool, limit_id, mtInternal);
 402   for (uint c = 0; c < limit_id; c += 1) {
 403     available_id[c] = false;
 404   }
 405   for (uint a = 0; a < id_length; a += 1) {
 406     available_id[id_array[a]] = true;
 407   }
 408   // Step by "boards", then by "slot", copying to "assigned".
 409   // NEEDS_CLEANUP: The assignment of processors should be stateful,
 410   //                remembering which processors have been assigned by
 411   //                previous calls, etc., so as to distribute several
 412   //                independent calls of this method.  What we'd like is
 413   //                It would be nice to have an API that let us ask
 414   //                how many processes are bound to a processor,
 415   //                but we don't have that, either.
 416   //                In the short term, "board" is static so that
 417   //                subsequent distributions don't all start at board 0.
 418   static uint board = 0;
 419   uint assigned = 0;
 420   // Until we've found enough processors ....
 421   while (assigned < distribution_length) {
 422     // ... find the next available processor in the board.
 423     for (uint slot = 0; slot < processors_per_board; slot += 1) {
 424       uint try_id = board * processors_per_board + slot;
 425       if ((try_id < limit_id) && (available_id[try_id] == true)) {
 426         distribution[assigned] = try_id;
 427         available_id[try_id] = false;
 428         assigned += 1;
 429         break;
 430       }
 431     }
 432     board += 1;
 433     if (board * processors_per_board + 0 >= limit_id) {
 434       board = 0;
 435     }
 436   }
 437   if (available_id != NULL) {
 438     FREE_C_HEAP_ARRAY(bool, available_id);
 439   }
 440   return true;
 441 }
 442 
 443 void os::set_native_thread_name(const char *name) {
 444   if (Solaris::_pthread_setname_np != NULL) {
 445     // Only the first 31 bytes of 'name' are processed by pthread_setname_np
 446     // but we explicitly copy into a size-limited buffer to avoid any
 447     // possible overflow.
 448     char buf[32];
 449     snprintf(buf, sizeof(buf), "%s", name);
 450     buf[sizeof(buf) - 1] = '\0';
 451     Solaris::_pthread_setname_np(pthread_self(), buf);
 452   }
 453 }
 454 
 455 bool os::distribute_processes(uint length, uint* distribution) {
 456   bool result = false;
 457   // Find the processor id's of all the available CPUs.
 458   processorid_t* id_array  = NULL;
 459   uint           id_length = 0;
 460   // There are some races between querying information and using it,
 461   // since processor sets can change dynamically.
 462   psetid_t pset = PS_NONE;
 463   // Are we running in a processor set?
 464   if ((pset_bind(PS_QUERY, P_PID, P_MYID, &pset) == 0) && pset != PS_NONE) {
 465     result = find_processors_in_pset(pset, &id_array, &id_length);
 466   } else {
 467     result = find_processors_online(&id_array, &id_length);
 468   }
 469   if (result == true) {
 470     if (id_length >= length) {
 471       result = assign_distribution(id_array, id_length, distribution, length);
 472     } else {
 473       result = false;
 474     }
 475   }
 476   if (id_array != NULL) {
 477     FREE_C_HEAP_ARRAY(processorid_t, id_array);
 478   }
 479   return result;
 480 }
 481 
 482 bool os::bind_to_processor(uint processor_id) {
 483   // We assume that a processorid_t can be stored in a uint.
 484   assert(sizeof(uint) == sizeof(processorid_t),
 485          "can't convert uint to processorid_t");
 486   int bind_result =
 487     processor_bind(P_LWPID,                       // bind LWP.
 488                    P_MYID,                        // bind current LWP.
 489                    (processorid_t) processor_id,  // id.
 490                    NULL);                         // don't return old binding.
 491   return (bind_result == 0);
 492 }
 493 
 494 // Return true if user is running as root.
 495 
 496 bool os::have_special_privileges() {
 497   static bool init = false;
 498   static bool privileges = false;
 499   if (!init) {
 500     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 501     init = true;
 502   }
 503   return privileges;
 504 }
 505 
 506 
 507 void os::init_system_properties_values() {
 508   // The next steps are taken in the product version:
 509   //
 510   // Obtain the JAVA_HOME value from the location of libjvm.so.
 511   // This library should be located at:
 512   // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
 513   //
 514   // If "/jre/lib/" appears at the right place in the path, then we
 515   // assume libjvm.so is installed in a JDK and we use this path.
 516   //
 517   // Otherwise exit with message: "Could not create the Java virtual machine."
 518   //
 519   // The following extra steps are taken in the debugging version:
 520   //
 521   // If "/jre/lib/" does NOT appear at the right place in the path
 522   // instead of exit check for $JAVA_HOME environment variable.
 523   //
 524   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 525   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 526   // it looks like libjvm.so is installed there
 527   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 528   //
 529   // Otherwise exit.
 530   //
 531   // Important note: if the location of libjvm.so changes this
 532   // code needs to be changed accordingly.
 533 
 534 // Base path of extensions installed on the system.
 535 #define SYS_EXT_DIR     "/usr/jdk/packages"
 536 #define EXTENSIONS_DIR  "/lib/ext"
 537 
 538   // Buffer that fits several sprintfs.
 539   // Note that the space for the colon and the trailing null are provided
 540   // by the nulls included by the sizeof operator.
 541   const size_t bufsize =
 542     MAX3((size_t)MAXPATHLEN,  // For dll_dir & friends.
 543          sizeof(SYS_EXT_DIR) + sizeof("/lib/"), // invariant ld_library_path
 544          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 545   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 546 
 547   // sysclasspath, java_home, dll_dir
 548   {
 549     char *pslash;
 550     os::jvm_path(buf, bufsize);
 551 
 552     // Found the full path to libjvm.so.
 553     // Now cut the path to <java_home>/jre if we can.
 554     *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
 555     pslash = strrchr(buf, '/');
 556     if (pslash != NULL) {
 557       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 558     }
 559     Arguments::set_dll_dir(buf);
 560 
 561     if (pslash != NULL) {
 562       pslash = strrchr(buf, '/');
 563       if (pslash != NULL) {
 564         *pslash = '\0';        // Get rid of /lib.
 565       }
 566     }
 567     Arguments::set_java_home(buf);
 568     set_boot_path('/', ':');
 569   }
 570 
 571   // Where to look for native libraries.
 572   {
 573     // Use dlinfo() to determine the correct java.library.path.
 574     //
 575     // If we're launched by the Java launcher, and the user
 576     // does not set java.library.path explicitly on the commandline,
 577     // the Java launcher sets LD_LIBRARY_PATH for us and unsets
 578     // LD_LIBRARY_PATH_32 and LD_LIBRARY_PATH_64.  In this case
 579     // dlinfo returns LD_LIBRARY_PATH + crle settings (including
 580     // /usr/lib), which is exactly what we want.
 581     //
 582     // If the user does set java.library.path, it completely
 583     // overwrites this setting, and always has.
 584     //
 585     // If we're not launched by the Java launcher, we may
 586     // get here with any/all of the LD_LIBRARY_PATH[_32|64]
 587     // settings.  Again, dlinfo does exactly what we want.
 588 
 589     Dl_serinfo     info_sz, *info = &info_sz;
 590     Dl_serpath     *path;
 591     char           *library_path;
 592     char           *common_path = buf;
 593 
 594     // Determine search path count and required buffer size.
 595     if (dlinfo(RTLD_SELF, RTLD_DI_SERINFOSIZE, (void *)info) == -1) {
 596       FREE_C_HEAP_ARRAY(char, buf);
 597       vm_exit_during_initialization("dlinfo SERINFOSIZE request", dlerror());
 598     }
 599 
 600     // Allocate new buffer and initialize.
 601     info = (Dl_serinfo*)NEW_C_HEAP_ARRAY(char, info_sz.dls_size, mtInternal);
 602     info->dls_size = info_sz.dls_size;
 603     info->dls_cnt = info_sz.dls_cnt;
 604 
 605     // Obtain search path information.
 606     if (dlinfo(RTLD_SELF, RTLD_DI_SERINFO, (void *)info) == -1) {
 607       FREE_C_HEAP_ARRAY(char, buf);
 608       FREE_C_HEAP_ARRAY(char, info);
 609       vm_exit_during_initialization("dlinfo SERINFO request", dlerror());
 610     }
 611 
 612     path = &info->dls_serpath[0];
 613 
 614     // Note: Due to a legacy implementation, most of the library path
 615     // is set in the launcher. This was to accomodate linking restrictions
 616     // on legacy Solaris implementations (which are no longer supported).
 617     // Eventually, all the library path setting will be done here.
 618     //
 619     // However, to prevent the proliferation of improperly built native
 620     // libraries, the new path component /usr/jdk/packages is added here.
 621 
 622     // Construct the invariant part of ld_library_path.
 623     sprintf(common_path, SYS_EXT_DIR "/lib");
 624 
 625     // Struct size is more than sufficient for the path components obtained
 626     // through the dlinfo() call, so only add additional space for the path
 627     // components explicitly added here.
 628     size_t library_path_size = info->dls_size + strlen(common_path);
 629     library_path = (char *)NEW_C_HEAP_ARRAY(char, library_path_size, mtInternal);
 630     library_path[0] = '\0';
 631 
 632     // Construct the desired Java library path from the linker's library
 633     // search path.
 634     //
 635     // For compatibility, it is optimal that we insert the additional path
 636     // components specific to the Java VM after those components specified
 637     // in LD_LIBRARY_PATH (if any) but before those added by the ld.so
 638     // infrastructure.
 639     if (info->dls_cnt == 0) { // Not sure this can happen, but allow for it.
 640       strcpy(library_path, common_path);
 641     } else {
 642       int inserted = 0;
 643       int i;
 644       for (i = 0; i < info->dls_cnt; i++, path++) {
 645         uint_t flags = path->dls_flags & LA_SER_MASK;
 646         if (((flags & LA_SER_LIBPATH) == 0) && !inserted) {
 647           strcat(library_path, common_path);
 648           strcat(library_path, os::path_separator());
 649           inserted = 1;
 650         }
 651         strcat(library_path, path->dls_name);
 652         strcat(library_path, os::path_separator());
 653       }
 654       // Eliminate trailing path separator.
 655       library_path[strlen(library_path)-1] = '\0';
 656     }
 657 
 658     // happens before argument parsing - can't use a trace flag
 659     // tty->print_raw("init_system_properties_values: native lib path: ");
 660     // tty->print_raw_cr(library_path);
 661 
 662     // Callee copies into its own buffer.
 663     Arguments::set_library_path(library_path);
 664 
 665     FREE_C_HEAP_ARRAY(char, library_path);
 666     FREE_C_HEAP_ARRAY(char, info);
 667   }
 668 
 669   // Extensions directories.
 670   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 671   Arguments::set_ext_dirs(buf);
 672 
 673   FREE_C_HEAP_ARRAY(char, buf);
 674 
 675 #undef SYS_EXT_DIR
 676 #undef EXTENSIONS_DIR
 677 }
 678 
 679 void os::breakpoint() {
 680   BREAKPOINT;
 681 }
 682 
 683 bool os::obsolete_option(const JavaVMOption *option) {
 684   if (!strncmp(option->optionString, "-Xt", 3)) {
 685     return true;
 686   } else if (!strncmp(option->optionString, "-Xtm", 4)) {
 687     return true;
 688   } else if (!strncmp(option->optionString, "-Xverifyheap", 12)) {
 689     return true;
 690   } else if (!strncmp(option->optionString, "-Xmaxjitcodesize", 16)) {
 691     return true;
 692   }
 693   return false;
 694 }
 695 
 696 bool os::Solaris::valid_stack_address(Thread* thread, address sp) {
 697   address  stackStart  = (address)thread->stack_base();
 698   address  stackEnd    = (address)(stackStart - (address)thread->stack_size());
 699   if (sp < stackStart && sp >= stackEnd) return true;
 700   return false;
 701 }
 702 
 703 extern "C" void breakpoint() {
 704   // use debugger to set breakpoint here
 705 }
 706 
 707 static thread_t main_thread;
 708 
 709 // Thread start routine for all newly created threads
 710 extern "C" void* thread_native_entry(void* thread_addr) {
 711   // Try to randomize the cache line index of hot stack frames.
 712   // This helps when threads of the same stack traces evict each other's
 713   // cache lines. The threads can be either from the same JVM instance, or
 714   // from different JVM instances. The benefit is especially true for
 715   // processors with hyperthreading technology.
 716   static int counter = 0;
 717   int pid = os::current_process_id();
 718   alloca(((pid ^ counter++) & 7) * 128);
 719 
 720   int prio;
 721   Thread* thread = (Thread*)thread_addr;
 722 
 723   thread->initialize_thread_current();
 724 
 725   OSThread* osthr = thread->osthread();
 726 
 727   osthr->set_lwp_id(_lwp_self());  // Store lwp in case we are bound
 728   thread->_schedctl = (void *) schedctl_init();
 729 
 730   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ").",
 731     os::current_thread_id());
 732 
 733   if (UseNUMA) {
 734     int lgrp_id = os::numa_get_group_id();
 735     if (lgrp_id != -1) {
 736       thread->set_lgrp_id(lgrp_id);
 737     }
 738   }
 739 
 740   // Our priority was set when we were created, and stored in the
 741   // osthread, but couldn't be passed through to our LWP until now.
 742   // So read back the priority and set it again.
 743 
 744   if (osthr->thread_id() != -1) {
 745     if (UseThreadPriorities) {
 746       int prio = osthr->native_priority();
 747       if (ThreadPriorityVerbose) {
 748         tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is "
 749                       INTPTR_FORMAT ", setting priority: %d\n",
 750                       osthr->thread_id(), osthr->lwp_id(), prio);
 751       }
 752       os::set_native_priority(thread, prio);
 753     }
 754   } else if (ThreadPriorityVerbose) {
 755     warning("Can't set priority in _start routine, thread id hasn't been set\n");
 756   }
 757 
 758   assert(osthr->get_state() == RUNNABLE, "invalid os thread state");
 759 
 760   // initialize signal mask for this thread
 761   os::Solaris::hotspot_sigmask(thread);
 762 
 763   thread->run();
 764 
 765   // One less thread is executing
 766   // When the VMThread gets here, the main thread may have already exited
 767   // which frees the CodeHeap containing the Atomic::dec code
 768   if (thread != VMThread::vm_thread() && VMThread::vm_thread() != NULL) {
 769     Atomic::dec(&os::Solaris::_os_thread_count);
 770   }
 771 
 772   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ").", os::current_thread_id());
 773 
 774   // If a thread has not deleted itself ("delete this") as part of its
 775   // termination sequence, we have to ensure thread-local-storage is
 776   // cleared before we actually terminate. No threads should ever be
 777   // deleted asynchronously with respect to their termination.
 778   if (Thread::current_or_null_safe() != NULL) {
 779     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 780     thread->clear_thread_current();
 781   }
 782 
 783   if (UseDetachedThreads) {
 784     thr_exit(NULL);
 785     ShouldNotReachHere();
 786   }
 787   return NULL;
 788 }
 789 
 790 static OSThread* create_os_thread(Thread* thread, thread_t thread_id) {
 791   // Allocate the OSThread object
 792   OSThread* osthread = new OSThread(NULL, NULL);
 793   if (osthread == NULL) return NULL;
 794 
 795   // Store info on the Solaris thread into the OSThread
 796   osthread->set_thread_id(thread_id);
 797   osthread->set_lwp_id(_lwp_self());
 798   thread->_schedctl = (void *) schedctl_init();
 799 
 800   if (UseNUMA) {
 801     int lgrp_id = os::numa_get_group_id();
 802     if (lgrp_id != -1) {
 803       thread->set_lgrp_id(lgrp_id);
 804     }
 805   }
 806 
 807   if (ThreadPriorityVerbose) {
 808     tty->print_cr("In create_os_thread, Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT "\n",
 809                   osthread->thread_id(), osthread->lwp_id());
 810   }
 811 
 812   // Initial thread state is INITIALIZED, not SUSPENDED
 813   osthread->set_state(INITIALIZED);
 814 
 815   return osthread;
 816 }
 817 
 818 void os::Solaris::hotspot_sigmask(Thread* thread) {
 819   //Save caller's signal mask
 820   sigset_t sigmask;
 821   pthread_sigmask(SIG_SETMASK, NULL, &sigmask);
 822   OSThread *osthread = thread->osthread();
 823   osthread->set_caller_sigmask(sigmask);
 824 
 825   pthread_sigmask(SIG_UNBLOCK, os::Solaris::unblocked_signals(), NULL);
 826   if (!ReduceSignalUsage) {
 827     if (thread->is_VM_thread()) {
 828       // Only the VM thread handles BREAK_SIGNAL ...
 829       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 830     } else {
 831       // ... all other threads block BREAK_SIGNAL
 832       assert(!sigismember(vm_signals(), SIGINT), "SIGINT should not be blocked");
 833       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 834     }
 835   }
 836 }
 837 
 838 bool os::create_attached_thread(JavaThread* thread) {
 839 #ifdef ASSERT
 840   thread->verify_not_published();
 841 #endif
 842   OSThread* osthread = create_os_thread(thread, thr_self());
 843   if (osthread == NULL) {
 844     return false;
 845   }
 846 
 847   // Initial thread state is RUNNABLE
 848   osthread->set_state(RUNNABLE);
 849   thread->set_osthread(osthread);
 850 
 851   // initialize signal mask for this thread
 852   // and save the caller's signal mask
 853   os::Solaris::hotspot_sigmask(thread);
 854 
 855   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ").",
 856     os::current_thread_id());
 857 
 858   return true;
 859 }
 860 
 861 bool os::create_main_thread(JavaThread* thread) {
 862 #ifdef ASSERT
 863   thread->verify_not_published();
 864 #endif
 865   if (_starting_thread == NULL) {
 866     _starting_thread = create_os_thread(thread, main_thread);
 867     if (_starting_thread == NULL) {
 868       return false;
 869     }
 870   }
 871 
 872   // The primodial thread is runnable from the start
 873   _starting_thread->set_state(RUNNABLE);
 874 
 875   thread->set_osthread(_starting_thread);
 876 
 877   // initialize signal mask for this thread
 878   // and save the caller's signal mask
 879   os::Solaris::hotspot_sigmask(thread);
 880 
 881   return true;
 882 }
 883 
 884 // Helper function to trace thread attributes, similar to os::Posix::describe_pthread_attr()
 885 static char* describe_thr_create_attributes(char* buf, size_t buflen,
 886                                             size_t stacksize, long flags) {
 887   stringStream ss(buf, buflen);
 888   ss.print("stacksize: " SIZE_FORMAT "k, ", stacksize / 1024);
 889   ss.print("flags: ");
 890   #define PRINT_FLAG(f) if (flags & f) ss.print( #f " ");
 891   #define ALL(X) \
 892     X(THR_SUSPENDED) \
 893     X(THR_DETACHED) \
 894     X(THR_BOUND) \
 895     X(THR_NEW_LWP) \
 896     X(THR_DAEMON)
 897   ALL(PRINT_FLAG)
 898   #undef ALL
 899   #undef PRINT_FLAG
 900   return buf;
 901 }
 902 
 903 // return default stack size for thr_type
 904 size_t os::Posix::default_stack_size(os::ThreadType thr_type) {
 905   // default stack size when not specified by caller is 1M (2M for LP64)
 906   size_t s = (BytesPerWord >> 2) * K * K;
 907   return s;
 908 }
 909 
 910 bool os::create_thread(Thread* thread, ThreadType thr_type,
 911                        size_t req_stack_size) {
 912   // Allocate the OSThread object
 913   OSThread* osthread = new OSThread(NULL, NULL);
 914   if (osthread == NULL) {
 915     return false;
 916   }
 917 
 918   if (ThreadPriorityVerbose) {
 919     char *thrtyp;
 920     switch (thr_type) {
 921     case vm_thread:
 922       thrtyp = (char *)"vm";
 923       break;
 924     case cgc_thread:
 925       thrtyp = (char *)"cgc";
 926       break;
 927     case pgc_thread:
 928       thrtyp = (char *)"pgc";
 929       break;
 930     case java_thread:
 931       thrtyp = (char *)"java";
 932       break;
 933     case compiler_thread:
 934       thrtyp = (char *)"compiler";
 935       break;
 936     case watcher_thread:
 937       thrtyp = (char *)"watcher";
 938       break;
 939     default:
 940       thrtyp = (char *)"unknown";
 941       break;
 942     }
 943     tty->print_cr("In create_thread, creating a %s thread\n", thrtyp);
 944   }
 945 
 946   // calculate stack size if it's not specified by caller
 947   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 948 
 949   // Initial state is ALLOCATED but not INITIALIZED
 950   osthread->set_state(ALLOCATED);
 951 
 952   if (os::Solaris::_os_thread_count > os::Solaris::_os_thread_limit) {
 953     // We got lots of threads. Check if we still have some address space left.
 954     // Need to be at least 5Mb of unreserved address space. We do check by
 955     // trying to reserve some.
 956     const size_t VirtualMemoryBangSize = 20*K*K;
 957     char* mem = os::reserve_memory(VirtualMemoryBangSize);
 958     if (mem == NULL) {
 959       delete osthread;
 960       return false;
 961     } else {
 962       // Release the memory again
 963       os::release_memory(mem, VirtualMemoryBangSize);
 964     }
 965   }
 966 
 967   // Setup osthread because the child thread may need it.
 968   thread->set_osthread(osthread);
 969 
 970   // Create the Solaris thread
 971   thread_t tid = 0;
 972   long     flags = (UseDetachedThreads ? THR_DETACHED : 0) | THR_SUSPENDED;
 973   int      status;
 974 
 975   // Mark that we don't have an lwp or thread id yet.
 976   // In case we attempt to set the priority before the thread starts.
 977   osthread->set_lwp_id(-1);
 978   osthread->set_thread_id(-1);
 979 
 980   status = thr_create(NULL, stack_size, thread_native_entry, thread, flags, &tid);
 981 
 982   char buf[64];
 983   if (status == 0) {
 984     log_info(os, thread)("Thread started (tid: " UINTX_FORMAT ", attributes: %s). ",
 985       (uintx) tid, describe_thr_create_attributes(buf, sizeof(buf), stack_size, flags));
 986   } else {
 987     log_warning(os, thread)("Failed to start thread - thr_create failed (%s) for attributes: %s.",
 988       os::errno_name(status), describe_thr_create_attributes(buf, sizeof(buf), stack_size, flags));
 989   }
 990 
 991   if (status != 0) {
 992     thread->set_osthread(NULL);
 993     // Need to clean up stuff we've allocated so far
 994     delete osthread;
 995     return false;
 996   }
 997 
 998   Atomic::inc(&os::Solaris::_os_thread_count);
 999 
1000   // Store info on the Solaris thread into the OSThread
1001   osthread->set_thread_id(tid);
1002 
1003   // Remember that we created this thread so we can set priority on it
1004   osthread->set_vm_created();
1005 
1006   // Most thread types will set an explicit priority before starting the thread,
1007   // but for those that don't we need a valid value to read back in thread_native_entry.
1008   osthread->set_native_priority(NormPriority);
1009 
1010   // Initial thread state is INITIALIZED, not SUSPENDED
1011   osthread->set_state(INITIALIZED);
1012 
1013   // The thread is returned suspended (in state INITIALIZED), and is started higher up in the call chain
1014   return true;
1015 }
1016 
1017 debug_only(static bool signal_sets_initialized = false);
1018 static sigset_t unblocked_sigs, vm_sigs;
1019 
1020 bool os::Solaris::is_sig_ignored(int sig) {
1021   struct sigaction oact;
1022   sigaction(sig, (struct sigaction*)NULL, &oact);
1023   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
1024                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
1025   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
1026     return true;
1027   } else {
1028     return false;
1029   }
1030 }
1031 
1032 void os::Solaris::signal_sets_init() {
1033   // Should also have an assertion stating we are still single-threaded.
1034   assert(!signal_sets_initialized, "Already initialized");
1035   // Fill in signals that are necessarily unblocked for all threads in
1036   // the VM. Currently, we unblock the following signals:
1037   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
1038   //                         by -Xrs (=ReduceSignalUsage));
1039   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
1040   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
1041   // the dispositions or masks wrt these signals.
1042   // Programs embedding the VM that want to use the above signals for their
1043   // own purposes must, at this time, use the "-Xrs" option to prevent
1044   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
1045   // (See bug 4345157, and other related bugs).
1046   // In reality, though, unblocking these signals is really a nop, since
1047   // these signals are not blocked by default.
1048   sigemptyset(&unblocked_sigs);
1049   sigaddset(&unblocked_sigs, SIGILL);
1050   sigaddset(&unblocked_sigs, SIGSEGV);
1051   sigaddset(&unblocked_sigs, SIGBUS);
1052   sigaddset(&unblocked_sigs, SIGFPE);
1053   sigaddset(&unblocked_sigs, ASYNC_SIGNAL);
1054 
1055   if (!ReduceSignalUsage) {
1056     if (!os::Solaris::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
1057       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
1058     }
1059     if (!os::Solaris::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
1060       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
1061     }
1062     if (!os::Solaris::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
1063       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
1064     }
1065   }
1066   // Fill in signals that are blocked by all but the VM thread.
1067   sigemptyset(&vm_sigs);
1068   if (!ReduceSignalUsage) {
1069     sigaddset(&vm_sigs, BREAK_SIGNAL);
1070   }
1071   debug_only(signal_sets_initialized = true);
1072 
1073   // For diagnostics only used in run_periodic_checks
1074   sigemptyset(&check_signal_done);
1075 }
1076 
1077 // These are signals that are unblocked while a thread is running Java.
1078 // (For some reason, they get blocked by default.)
1079 sigset_t* os::Solaris::unblocked_signals() {
1080   assert(signal_sets_initialized, "Not initialized");
1081   return &unblocked_sigs;
1082 }
1083 
1084 // These are the signals that are blocked while a (non-VM) thread is
1085 // running Java. Only the VM thread handles these signals.
1086 sigset_t* os::Solaris::vm_signals() {
1087   assert(signal_sets_initialized, "Not initialized");
1088   return &vm_sigs;
1089 }
1090 
1091 void _handle_uncaught_cxx_exception() {
1092   VMError::report_and_die("An uncaught C++ exception");
1093 }
1094 
1095 
1096 // First crack at OS-specific initialization, from inside the new thread.
1097 void os::initialize_thread(Thread* thr) {
1098   int r = thr_main();
1099   guarantee(r == 0 || r == 1, "CR6501650 or CR6493689");
1100   if (r) {
1101     JavaThread* jt = (JavaThread *)thr;
1102     assert(jt != NULL, "Sanity check");
1103     size_t stack_size;
1104     address base = jt->stack_base();
1105     if (Arguments::created_by_java_launcher()) {
1106       // Use 2MB to allow for Solaris 7 64 bit mode.
1107       stack_size = JavaThread::stack_size_at_create() == 0
1108         ? 2048*K : JavaThread::stack_size_at_create();
1109 
1110       // There are rare cases when we may have already used more than
1111       // the basic stack size allotment before this method is invoked.
1112       // Attempt to allow for a normally sized java_stack.
1113       size_t current_stack_offset = (size_t)(base - (address)&stack_size);
1114       stack_size += ReservedSpace::page_align_size_down(current_stack_offset);
1115     } else {
1116       // 6269555: If we were not created by a Java launcher, i.e. if we are
1117       // running embedded in a native application, treat the primordial thread
1118       // as much like a native attached thread as possible.  This means using
1119       // the current stack size from thr_stksegment(), unless it is too large
1120       // to reliably setup guard pages.  A reasonable max size is 8MB.
1121       size_t current_size = current_stack_size();
1122       // This should never happen, but just in case....
1123       if (current_size == 0) current_size = 2 * K * K;
1124       stack_size = current_size > (8 * K * K) ? (8 * K * K) : current_size;
1125     }
1126     address bottom = align_up(base - stack_size, os::vm_page_size());;
1127     stack_size = (size_t)(base - bottom);
1128 
1129     assert(stack_size > 0, "Stack size calculation problem");
1130 
1131     if (stack_size > jt->stack_size()) {
1132 #ifndef PRODUCT
1133       struct rlimit limits;
1134       getrlimit(RLIMIT_STACK, &limits);
1135       size_t size = adjust_stack_size(base, (size_t)limits.rlim_cur);
1136       assert(size >= jt->stack_size(), "Stack size problem in main thread");
1137 #endif
1138       tty->print_cr("Stack size of %d Kb exceeds current limit of %d Kb.\n"
1139                     "(Stack sizes are rounded up to a multiple of the system page size.)\n"
1140                     "See limit(1) to increase the stack size limit.",
1141                     stack_size / K, jt->stack_size() / K);
1142       vm_exit(1);
1143     }
1144     assert(jt->stack_size() >= stack_size,
1145            "Attempt to map more stack than was allocated");
1146     jt->set_stack_size(stack_size);
1147   }
1148 
1149   // With the T2 libthread (T1 is no longer supported) threads are always bound
1150   // and we use stackbanging in all cases.
1151 
1152   os::Solaris::init_thread_fpu_state();
1153   std::set_terminate(_handle_uncaught_cxx_exception);
1154 }
1155 
1156 
1157 
1158 // Free Solaris resources related to the OSThread
1159 void os::free_thread(OSThread* osthread) {
1160   assert(osthread != NULL, "os::free_thread but osthread not set");
1161 
1162   // We are told to free resources of the argument thread,
1163   // but we can only really operate on the current thread.
1164   assert(Thread::current()->osthread() == osthread,
1165          "os::free_thread but not current thread");
1166 
1167   // Restore caller's signal mask
1168   sigset_t sigmask = osthread->caller_sigmask();
1169   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1170 
1171   delete osthread;
1172 }
1173 
1174 void os::pd_start_thread(Thread* thread) {
1175   int status = thr_continue(thread->osthread()->thread_id());
1176   assert_status(status == 0, status, "thr_continue failed");
1177 }
1178 
1179 
1180 intx os::current_thread_id() {
1181   return (intx)thr_self();
1182 }
1183 
1184 static pid_t _initial_pid = 0;
1185 
1186 int os::current_process_id() {
1187   return (int)(_initial_pid ? _initial_pid : getpid());
1188 }
1189 
1190 // gethrtime() should be monotonic according to the documentation,
1191 // but some virtualized platforms are known to break this guarantee.
1192 // getTimeNanos() must be guaranteed not to move backwards, so we
1193 // are forced to add a check here.
1194 inline hrtime_t getTimeNanos() {
1195   const hrtime_t now = gethrtime();
1196   const hrtime_t prev = max_hrtime;
1197   if (now <= prev) {
1198     return prev;   // same or retrograde time;
1199   }
1200   const hrtime_t obsv = Atomic::cmpxchg(now, &max_hrtime, prev);
1201   assert(obsv >= prev, "invariant");   // Monotonicity
1202   // If the CAS succeeded then we're done and return "now".
1203   // If the CAS failed and the observed value "obsv" is >= now then
1204   // we should return "obsv".  If the CAS failed and now > obsv > prv then
1205   // some other thread raced this thread and installed a new value, in which case
1206   // we could either (a) retry the entire operation, (b) retry trying to install now
1207   // or (c) just return obsv.  We use (c).   No loop is required although in some cases
1208   // we might discard a higher "now" value in deference to a slightly lower but freshly
1209   // installed obsv value.   That's entirely benign -- it admits no new orderings compared
1210   // to (a) or (b) -- and greatly reduces coherence traffic.
1211   // We might also condition (c) on the magnitude of the delta between obsv and now.
1212   // Avoiding excessive CAS operations to hot RW locations is critical.
1213   // See https://blogs.oracle.com/dave/entry/cas_and_cache_trivia_invalidate
1214   return (prev == obsv) ? now : obsv;
1215 }
1216 
1217 // Time since start-up in seconds to a fine granularity.
1218 // Used by VMSelfDestructTimer and the MemProfiler.
1219 double os::elapsedTime() {
1220   return (double)(getTimeNanos() - first_hrtime) / (double)hrtime_hz;
1221 }
1222 
1223 jlong os::elapsed_counter() {
1224   return (jlong)(getTimeNanos() - first_hrtime);
1225 }
1226 
1227 jlong os::elapsed_frequency() {
1228   return hrtime_hz;
1229 }
1230 
1231 // Return the real, user, and system times in seconds from an
1232 // arbitrary fixed point in the past.
1233 bool os::getTimesSecs(double* process_real_time,
1234                       double* process_user_time,
1235                       double* process_system_time) {
1236   struct tms ticks;
1237   clock_t real_ticks = times(&ticks);
1238 
1239   if (real_ticks == (clock_t) (-1)) {
1240     return false;
1241   } else {
1242     double ticks_per_second = (double) clock_tics_per_sec;
1243     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1244     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1245     // For consistency return the real time from getTimeNanos()
1246     // converted to seconds.
1247     *process_real_time = ((double) getTimeNanos()) / ((double) NANOUNITS);
1248 
1249     return true;
1250   }
1251 }
1252 
1253 bool os::supports_vtime() { return true; }
1254 bool os::enable_vtime() { return false; }
1255 bool os::vtime_enabled() { return false; }
1256 
1257 double os::elapsedVTime() {
1258   return (double)gethrvtime() / (double)hrtime_hz;
1259 }
1260 
1261 // Must return millis since Jan 1 1970 for JVM_CurrentTimeMillis
1262 jlong os::javaTimeMillis() {
1263   timeval t;
1264   if (gettimeofday(&t, NULL) == -1) {
1265     fatal("os::javaTimeMillis: gettimeofday (%s)", os::strerror(errno));
1266   }
1267   return jlong(t.tv_sec) * 1000  +  jlong(t.tv_usec) / 1000;
1268 }
1269 
1270 // Must return seconds+nanos since Jan 1 1970. This must use the same
1271 // time source as javaTimeMillis and can't use get_nsec_fromepoch as
1272 // we need better than 1ms accuracy
1273 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1274   timeval t;
1275   if (gettimeofday(&t, NULL) == -1) {
1276     fatal("os::javaTimeSystemUTC: gettimeofday (%s)", os::strerror(errno));
1277   }
1278   seconds = jlong(t.tv_sec);
1279   nanos = jlong(t.tv_usec) * 1000;
1280 }
1281 
1282 
1283 jlong os::javaTimeNanos() {
1284   return (jlong)getTimeNanos();
1285 }
1286 
1287 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1288   info_ptr->max_value = ALL_64_BITS;      // gethrtime() uses all 64 bits
1289   info_ptr->may_skip_backward = false;    // not subject to resetting or drifting
1290   info_ptr->may_skip_forward = false;     // not subject to resetting or drifting
1291   info_ptr->kind = JVMTI_TIMER_ELAPSED;   // elapsed not CPU time
1292 }
1293 
1294 char * os::local_time_string(char *buf, size_t buflen) {
1295   struct tm t;
1296   time_t long_time;
1297   time(&long_time);
1298   localtime_r(&long_time, &t);
1299   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1300                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1301                t.tm_hour, t.tm_min, t.tm_sec);
1302   return buf;
1303 }
1304 
1305 // Note: os::shutdown() might be called very early during initialization, or
1306 // called from signal handler. Before adding something to os::shutdown(), make
1307 // sure it is async-safe and can handle partially initialized VM.
1308 void os::shutdown() {
1309 
1310   // allow PerfMemory to attempt cleanup of any persistent resources
1311   perfMemory_exit();
1312 
1313   // needs to remove object in file system
1314   AttachListener::abort();
1315 
1316   // flush buffered output, finish log files
1317   ostream_abort();
1318 
1319   // Check for abort hook
1320   abort_hook_t abort_hook = Arguments::abort_hook();
1321   if (abort_hook != NULL) {
1322     abort_hook();
1323   }
1324 }
1325 
1326 // Note: os::abort() might be called very early during initialization, or
1327 // called from signal handler. Before adding something to os::abort(), make
1328 // sure it is async-safe and can handle partially initialized VM.
1329 void os::abort(bool dump_core, void* siginfo, const void* context) {
1330   os::shutdown();
1331   if (dump_core) {
1332 #ifndef PRODUCT
1333     fdStream out(defaultStream::output_fd());
1334     out.print_raw("Current thread is ");
1335     char buf[16];
1336     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1337     out.print_raw_cr(buf);
1338     out.print_raw_cr("Dumping core ...");
1339 #endif
1340     ::abort(); // dump core (for debugging)
1341   }
1342 
1343   ::exit(1);
1344 }
1345 
1346 // Die immediately, no exit hook, no abort hook, no cleanup.
1347 void os::die() {
1348   ::abort(); // dump core (for debugging)
1349 }
1350 
1351 // DLL functions
1352 
1353 const char* os::dll_file_extension() { return ".so"; }
1354 
1355 // This must be hard coded because it's the system's temporary
1356 // directory not the java application's temp directory, ala java.io.tmpdir.
1357 const char* os::get_temp_directory() { return "/tmp"; }
1358 
1359 // check if addr is inside libjvm.so
1360 bool os::address_is_in_vm(address addr) {
1361   static address libjvm_base_addr;
1362   Dl_info dlinfo;
1363 
1364   if (libjvm_base_addr == NULL) {
1365     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1366       libjvm_base_addr = (address)dlinfo.dli_fbase;
1367     }
1368     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1369   }
1370 
1371   if (dladdr((void *)addr, &dlinfo) != 0) {
1372     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1373   }
1374 
1375   return false;
1376 }
1377 
1378 typedef int (*dladdr1_func_type)(void *, Dl_info *, void **, int);
1379 static dladdr1_func_type dladdr1_func = NULL;
1380 
1381 bool os::dll_address_to_function_name(address addr, char *buf,
1382                                       int buflen, int * offset,
1383                                       bool demangle) {
1384   // buf is not optional, but offset is optional
1385   assert(buf != NULL, "sanity check");
1386 
1387   Dl_info dlinfo;
1388 
1389   // dladdr1_func was initialized in os::init()
1390   if (dladdr1_func != NULL) {
1391     // yes, we have dladdr1
1392 
1393     // Support for dladdr1 is checked at runtime; it may be
1394     // available even if the vm is built on a machine that does
1395     // not have dladdr1 support.  Make sure there is a value for
1396     // RTLD_DL_SYMENT.
1397 #ifndef RTLD_DL_SYMENT
1398   #define RTLD_DL_SYMENT 1
1399 #endif
1400 #ifdef _LP64
1401     Elf64_Sym * info;
1402 #else
1403     Elf32_Sym * info;
1404 #endif
1405     if (dladdr1_func((void *)addr, &dlinfo, (void **)&info,
1406                      RTLD_DL_SYMENT) != 0) {
1407       // see if we have a matching symbol that covers our address
1408       if (dlinfo.dli_saddr != NULL &&
1409           (char *)dlinfo.dli_saddr + info->st_size > (char *)addr) {
1410         if (dlinfo.dli_sname != NULL) {
1411           if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1412             jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1413           }
1414           if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1415           return true;
1416         }
1417       }
1418       // no matching symbol so try for just file info
1419       if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1420         if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1421                             buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1422           return true;
1423         }
1424       }
1425     }
1426     buf[0] = '\0';
1427     if (offset != NULL) *offset  = -1;
1428     return false;
1429   }
1430 
1431   // no, only dladdr is available
1432   if (dladdr((void *)addr, &dlinfo) != 0) {
1433     // see if we have a matching symbol
1434     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1435       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1436         jio_snprintf(buf, buflen, dlinfo.dli_sname);
1437       }
1438       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1439       return true;
1440     }
1441     // no matching symbol so try for just file info
1442     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1443       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1444                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1445         return true;
1446       }
1447     }
1448   }
1449   buf[0] = '\0';
1450   if (offset != NULL) *offset  = -1;
1451   return false;
1452 }
1453 
1454 bool os::dll_address_to_library_name(address addr, char* buf,
1455                                      int buflen, int* offset) {
1456   // buf is not optional, but offset is optional
1457   assert(buf != NULL, "sanity check");
1458 
1459   Dl_info dlinfo;
1460 
1461   if (dladdr((void*)addr, &dlinfo) != 0) {
1462     if (dlinfo.dli_fname != NULL) {
1463       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1464     }
1465     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1466       *offset = addr - (address)dlinfo.dli_fbase;
1467     }
1468     return true;
1469   }
1470 
1471   buf[0] = '\0';
1472   if (offset) *offset = -1;
1473   return false;
1474 }
1475 
1476 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1477   Dl_info dli;
1478   // Sanity check?
1479   if (dladdr(CAST_FROM_FN_PTR(void *, os::get_loaded_modules_info), &dli) == 0 ||
1480       dli.dli_fname == NULL) {
1481     return 1;
1482   }
1483 
1484   void * handle = dlopen(dli.dli_fname, RTLD_LAZY);
1485   if (handle == NULL) {
1486     return 1;
1487   }
1488 
1489   Link_map *map;
1490   dlinfo(handle, RTLD_DI_LINKMAP, &map);
1491   if (map == NULL) {
1492     dlclose(handle);
1493     return 1;
1494   }
1495 
1496   while (map->l_prev != NULL) {
1497     map = map->l_prev;
1498   }
1499 
1500   while (map != NULL) {
1501     // Iterate through all map entries and call callback with fields of interest
1502     if(callback(map->l_name, (address)map->l_addr, (address)0, param)) {
1503       dlclose(handle);
1504       return 1;
1505     }
1506     map = map->l_next;
1507   }
1508 
1509   dlclose(handle);
1510   return 0;
1511 }
1512 
1513 int _print_dll_info_cb(const char * name, address base_address, address top_address, void * param) {
1514   outputStream * out = (outputStream *) param;
1515   out->print_cr(PTR_FORMAT " \t%s", base_address, name);
1516   return 0;
1517 }
1518 
1519 void os::print_dll_info(outputStream * st) {
1520   st->print_cr("Dynamic libraries:"); st->flush();
1521   if (get_loaded_modules_info(_print_dll_info_cb, (void *)st)) {
1522     st->print_cr("Error: Cannot print dynamic libraries.");
1523   }
1524 }
1525 
1526 // Loads .dll/.so and
1527 // in case of error it checks if .dll/.so was built for the
1528 // same architecture as Hotspot is running on
1529 
1530 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1531   void * result= ::dlopen(filename, RTLD_LAZY);
1532   if (result != NULL) {
1533     // Successful loading
1534     return result;
1535   }
1536 
1537   Elf32_Ehdr elf_head;
1538 
1539   // Read system error message into ebuf
1540   // It may or may not be overwritten below
1541   ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1542   ebuf[ebuflen-1]='\0';
1543   int diag_msg_max_length=ebuflen-strlen(ebuf);
1544   char* diag_msg_buf=ebuf+strlen(ebuf);
1545 
1546   if (diag_msg_max_length==0) {
1547     // No more space in ebuf for additional diagnostics message
1548     return NULL;
1549   }
1550 
1551 
1552   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1553 
1554   if (file_descriptor < 0) {
1555     // Can't open library, report dlerror() message
1556     return NULL;
1557   }
1558 
1559   bool failed_to_read_elf_head=
1560     (sizeof(elf_head)!=
1561      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1562 
1563   ::close(file_descriptor);
1564   if (failed_to_read_elf_head) {
1565     // file i/o error - report dlerror() msg
1566     return NULL;
1567   }
1568 
1569   typedef struct {
1570     Elf32_Half  code;         // Actual value as defined in elf.h
1571     Elf32_Half  compat_class; // Compatibility of archs at VM's sense
1572     char        elf_class;    // 32 or 64 bit
1573     char        endianess;    // MSB or LSB
1574     char*       name;         // String representation
1575   } arch_t;
1576 
1577   static const arch_t arch_array[]={
1578     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1579     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1580     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1581     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1582     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1583     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1584     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1585     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1586     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1587     {EM_ARM,         EM_ARM,     ELFCLASS32, ELFDATA2LSB, (char*)"ARM 32"}
1588   };
1589 
1590 #if  (defined IA32)
1591   static  Elf32_Half running_arch_code=EM_386;
1592 #elif   (defined AMD64)
1593   static  Elf32_Half running_arch_code=EM_X86_64;
1594 #elif  (defined IA64)
1595   static  Elf32_Half running_arch_code=EM_IA_64;
1596 #elif  (defined __sparc) && (defined _LP64)
1597   static  Elf32_Half running_arch_code=EM_SPARCV9;
1598 #elif  (defined __sparc) && (!defined _LP64)
1599   static  Elf32_Half running_arch_code=EM_SPARC;
1600 #elif  (defined __powerpc64__)
1601   static  Elf32_Half running_arch_code=EM_PPC64;
1602 #elif  (defined __powerpc__)
1603   static  Elf32_Half running_arch_code=EM_PPC;
1604 #elif (defined ARM)
1605   static  Elf32_Half running_arch_code=EM_ARM;
1606 #else
1607   #error Method os::dll_load requires that one of following is defined:\
1608        IA32, AMD64, IA64, __sparc, __powerpc__, ARM, ARM
1609 #endif
1610 
1611   // Identify compatability class for VM's architecture and library's architecture
1612   // Obtain string descriptions for architectures
1613 
1614   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1615   int running_arch_index=-1;
1616 
1617   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1618     if (running_arch_code == arch_array[i].code) {
1619       running_arch_index    = i;
1620     }
1621     if (lib_arch.code == arch_array[i].code) {
1622       lib_arch.compat_class = arch_array[i].compat_class;
1623       lib_arch.name         = arch_array[i].name;
1624     }
1625   }
1626 
1627   assert(running_arch_index != -1,
1628          "Didn't find running architecture code (running_arch_code) in arch_array");
1629   if (running_arch_index == -1) {
1630     // Even though running architecture detection failed
1631     // we may still continue with reporting dlerror() message
1632     return NULL;
1633   }
1634 
1635   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1636     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1637     return NULL;
1638   }
1639 
1640   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1641     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1642     return NULL;
1643   }
1644 
1645   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1646     if (lib_arch.name!=NULL) {
1647       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1648                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1649                  lib_arch.name, arch_array[running_arch_index].name);
1650     } else {
1651       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1652                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1653                  lib_arch.code,
1654                  arch_array[running_arch_index].name);
1655     }
1656   }
1657 
1658   return NULL;
1659 }
1660 
1661 void* os::dll_lookup(void* handle, const char* name) {
1662   return dlsym(handle, name);
1663 }
1664 
1665 void* os::get_default_process_handle() {
1666   return (void*)::dlopen(NULL, RTLD_LAZY);
1667 }
1668 
1669 int os::stat(const char *path, struct stat *sbuf) {
1670   char pathbuf[MAX_PATH];
1671   if (strlen(path) > MAX_PATH - 1) {
1672     errno = ENAMETOOLONG;
1673     return -1;
1674   }
1675   os::native_path(strcpy(pathbuf, path));
1676   return ::stat(pathbuf, sbuf);
1677 }
1678 
1679 static inline time_t get_mtime(const char* filename) {
1680   struct stat st;
1681   int ret = os::stat(filename, &st);
1682   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
1683   return st.st_mtime;
1684 }
1685 
1686 int os::compare_file_modified_times(const char* file1, const char* file2) {
1687   time_t t1 = get_mtime(file1);
1688   time_t t2 = get_mtime(file2);
1689   return t1 - t2;
1690 }
1691 
1692 static bool _print_ascii_file(const char* filename, outputStream* st) {
1693   int fd = ::open(filename, O_RDONLY);
1694   if (fd == -1) {
1695     return false;
1696   }
1697 
1698   char buf[32];
1699   int bytes;
1700   while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1701     st->print_raw(buf, bytes);
1702   }
1703 
1704   ::close(fd);
1705 
1706   return true;
1707 }
1708 
1709 void os::print_os_info_brief(outputStream* st) {
1710   os::Solaris::print_distro_info(st);
1711 
1712   os::Posix::print_uname_info(st);
1713 
1714   os::Solaris::print_libversion_info(st);
1715 }
1716 
1717 void os::print_os_info(outputStream* st) {
1718   st->print("OS:");
1719 
1720   os::Solaris::print_distro_info(st);
1721 
1722   os::Posix::print_uname_info(st);
1723 
1724   os::Solaris::print_libversion_info(st);
1725 
1726   os::Posix::print_rlimit_info(st);
1727 
1728   os::Posix::print_load_average(st);
1729 }
1730 
1731 void os::Solaris::print_distro_info(outputStream* st) {
1732   if (!_print_ascii_file("/etc/release", st)) {
1733     st->print("Solaris");
1734   }
1735   st->cr();
1736 }
1737 
1738 void os::get_summary_os_info(char* buf, size_t buflen) {
1739   strncpy(buf, "Solaris", buflen);  // default to plain solaris
1740   FILE* fp = fopen("/etc/release", "r");
1741   if (fp != NULL) {
1742     char tmp[256];
1743     // Only get the first line and chop out everything but the os name.
1744     if (fgets(tmp, sizeof(tmp), fp)) {
1745       char* ptr = tmp;
1746       // skip past whitespace characters
1747       while (*ptr != '\0' && (*ptr == ' ' || *ptr == '\t' || *ptr == '\n')) ptr++;
1748       if (*ptr != '\0') {
1749         char* nl = strchr(ptr, '\n');
1750         if (nl != NULL) *nl = '\0';
1751         strncpy(buf, ptr, buflen);
1752       }
1753     }
1754     fclose(fp);
1755   }
1756 }
1757 
1758 void os::Solaris::print_libversion_info(outputStream* st) {
1759   st->print("  (T2 libthread)");
1760   st->cr();
1761 }
1762 
1763 static bool check_addr0(outputStream* st) {
1764   jboolean status = false;
1765   const int read_chunk = 200;
1766   int ret = 0;
1767   int nmap = 0;
1768   int fd = ::open("/proc/self/map",O_RDONLY);
1769   if (fd >= 0) {
1770     prmap_t *p = NULL;
1771     char *mbuff = (char *) calloc(read_chunk, sizeof(prmap_t));
1772     if (NULL == mbuff) {
1773       ::close(fd);
1774       return status;
1775     }
1776     while ((ret = ::read(fd, mbuff, read_chunk*sizeof(prmap_t))) > 0) {
1777       //check if read() has not read partial data
1778       if( 0 != ret % sizeof(prmap_t)){
1779         break;
1780       }
1781       nmap = ret / sizeof(prmap_t);
1782       p = (prmap_t *)mbuff;
1783       for(int i = 0; i < nmap; i++){
1784         if (p->pr_vaddr == 0x0) {
1785           st->print("Warning: Address: " PTR_FORMAT ", Size: " SIZE_FORMAT "K, ",p->pr_vaddr, p->pr_size/1024);
1786           st->print("Mapped file: %s, ", p->pr_mapname[0] == '\0' ? "None" : p->pr_mapname);
1787           st->print("Access: ");
1788           st->print("%s",(p->pr_mflags & MA_READ)  ? "r" : "-");
1789           st->print("%s",(p->pr_mflags & MA_WRITE) ? "w" : "-");
1790           st->print("%s",(p->pr_mflags & MA_EXEC)  ? "x" : "-");
1791           st->cr();
1792           status = true;
1793         }
1794         p++;
1795       }
1796     }
1797     free(mbuff);
1798     ::close(fd);
1799   }
1800   return status;
1801 }
1802 
1803 void os::get_summary_cpu_info(char* buf, size_t buflen) {
1804   // Get MHz with system call. We don't seem to already have this.
1805   processor_info_t stats;
1806   processorid_t id = getcpuid();
1807   int clock = 0;
1808   if (processor_info(id, &stats) != -1) {
1809     clock = stats.pi_clock;  // pi_processor_type isn't more informative than below
1810   }
1811 #ifdef AMD64
1812   snprintf(buf, buflen, "x86 64 bit %d MHz", clock);
1813 #else
1814   // must be sparc
1815   snprintf(buf, buflen, "Sparcv9 64 bit %d MHz", clock);
1816 #endif
1817 }
1818 
1819 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
1820   // Nothing to do for now.
1821 }
1822 
1823 void os::print_memory_info(outputStream* st) {
1824   st->print("Memory:");
1825   st->print(" %dk page", os::vm_page_size()>>10);
1826   st->print(", physical " UINT64_FORMAT "k", os::physical_memory()>>10);
1827   st->print("(" UINT64_FORMAT "k free)", os::available_memory() >> 10);
1828   st->cr();
1829   (void) check_addr0(st);
1830 }
1831 
1832 // Moved from whole group, because we need them here for diagnostic
1833 // prints.
1834 static int Maxsignum = 0;
1835 static int *ourSigFlags = NULL;
1836 
1837 int os::Solaris::get_our_sigflags(int sig) {
1838   assert(ourSigFlags!=NULL, "signal data structure not initialized");
1839   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
1840   return ourSigFlags[sig];
1841 }
1842 
1843 void os::Solaris::set_our_sigflags(int sig, int flags) {
1844   assert(ourSigFlags!=NULL, "signal data structure not initialized");
1845   assert(sig > 0 && sig < Maxsignum, "vm signal out of expected range");
1846   ourSigFlags[sig] = flags;
1847 }
1848 
1849 
1850 static const char* get_signal_handler_name(address handler,
1851                                            char* buf, int buflen) {
1852   int offset;
1853   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
1854   if (found) {
1855     // skip directory names
1856     const char *p1, *p2;
1857     p1 = buf;
1858     size_t len = strlen(os::file_separator());
1859     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
1860     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
1861   } else {
1862     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
1863   }
1864   return buf;
1865 }
1866 
1867 static void print_signal_handler(outputStream* st, int sig,
1868                                  char* buf, size_t buflen) {
1869   struct sigaction sa;
1870 
1871   sigaction(sig, NULL, &sa);
1872 
1873   st->print("%s: ", os::exception_name(sig, buf, buflen));
1874 
1875   address handler = (sa.sa_flags & SA_SIGINFO)
1876                   ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
1877                   : CAST_FROM_FN_PTR(address, sa.sa_handler);
1878 
1879   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
1880     st->print("SIG_DFL");
1881   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
1882     st->print("SIG_IGN");
1883   } else {
1884     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
1885   }
1886 
1887   st->print(", sa_mask[0]=");
1888   os::Posix::print_signal_set_short(st, &sa.sa_mask);
1889 
1890   address rh = VMError::get_resetted_sighandler(sig);
1891   // May be, handler was resetted by VMError?
1892   if (rh != NULL) {
1893     handler = rh;
1894     sa.sa_flags = VMError::get_resetted_sigflags(sig);
1895   }
1896 
1897   st->print(", sa_flags=");
1898   os::Posix::print_sa_flags(st, sa.sa_flags);
1899 
1900   // Check: is it our handler?
1901   if (handler == CAST_FROM_FN_PTR(address, signalHandler)) {
1902     // It is our signal handler
1903     // check for flags
1904     if (sa.sa_flags != os::Solaris::get_our_sigflags(sig)) {
1905       st->print(
1906                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
1907                 os::Solaris::get_our_sigflags(sig));
1908     }
1909   }
1910   st->cr();
1911 }
1912 
1913 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
1914   st->print_cr("Signal Handlers:");
1915   print_signal_handler(st, SIGSEGV, buf, buflen);
1916   print_signal_handler(st, SIGBUS , buf, buflen);
1917   print_signal_handler(st, SIGFPE , buf, buflen);
1918   print_signal_handler(st, SIGPIPE, buf, buflen);
1919   print_signal_handler(st, SIGXFSZ, buf, buflen);
1920   print_signal_handler(st, SIGILL , buf, buflen);
1921   print_signal_handler(st, ASYNC_SIGNAL, buf, buflen);
1922   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
1923   print_signal_handler(st, SHUTDOWN1_SIGNAL , buf, buflen);
1924   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
1925   print_signal_handler(st, SHUTDOWN3_SIGNAL, buf, buflen);
1926 }
1927 
1928 static char saved_jvm_path[MAXPATHLEN] = { 0 };
1929 
1930 // Find the full path to the current module, libjvm.so
1931 void os::jvm_path(char *buf, jint buflen) {
1932   // Error checking.
1933   if (buflen < MAXPATHLEN) {
1934     assert(false, "must use a large-enough buffer");
1935     buf[0] = '\0';
1936     return;
1937   }
1938   // Lazy resolve the path to current module.
1939   if (saved_jvm_path[0] != 0) {
1940     strcpy(buf, saved_jvm_path);
1941     return;
1942   }
1943 
1944   Dl_info dlinfo;
1945   int ret = dladdr(CAST_FROM_FN_PTR(void *, os::jvm_path), &dlinfo);
1946   assert(ret != 0, "cannot locate libjvm");
1947   if (ret != 0 && dlinfo.dli_fname != NULL) {
1948     if (os::Posix::realpath((char *)dlinfo.dli_fname, buf, buflen) == NULL) {
1949       return;
1950     }
1951   } else {
1952     buf[0] = '\0';
1953     return;
1954   }
1955 
1956   if (Arguments::sun_java_launcher_is_altjvm()) {
1957     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
1958     // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
1959     // If "/jre/lib/" appears at the right place in the string, then
1960     // assume we are installed in a JDK and we're done.  Otherwise, check
1961     // for a JAVA_HOME environment variable and fix up the path so it
1962     // looks like libjvm.so is installed there (append a fake suffix
1963     // hotspot/libjvm.so).
1964     const char *p = buf + strlen(buf) - 1;
1965     for (int count = 0; p > buf && count < 5; ++count) {
1966       for (--p; p > buf && *p != '/'; --p)
1967         /* empty */ ;
1968     }
1969 
1970     if (strncmp(p, "/jre/lib/", 9) != 0) {
1971       // Look for JAVA_HOME in the environment.
1972       char* java_home_var = ::getenv("JAVA_HOME");
1973       if (java_home_var != NULL && java_home_var[0] != 0) {
1974         char* jrelib_p;
1975         int   len;
1976 
1977         // Check the current module name "libjvm.so".
1978         p = strrchr(buf, '/');
1979         assert(strstr(p, "/libjvm") == p, "invalid library name");
1980 
1981         if (os::Posix::realpath(java_home_var, buf, buflen) == NULL) {
1982           return;
1983         }
1984         // determine if this is a legacy image or modules image
1985         // modules image doesn't have "jre" subdirectory
1986         len = strlen(buf);
1987         assert(len < buflen, "Ran out of buffer space");
1988         jrelib_p = buf + len;
1989         snprintf(jrelib_p, buflen-len, "/jre/lib");
1990         if (0 != access(buf, F_OK)) {
1991           snprintf(jrelib_p, buflen-len, "/lib");
1992         }
1993 
1994         if (0 == access(buf, F_OK)) {
1995           // Use current module name "libjvm.so"
1996           len = strlen(buf);
1997           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
1998         } else {
1999           // Go back to path of .so
2000           if (os::Posix::realpath((char *)dlinfo.dli_fname, buf, buflen) == NULL) {
2001             return;
2002           }
2003         }
2004       }
2005     }
2006   }
2007 
2008   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2009   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2010 }
2011 
2012 
2013 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2014   // no prefix required, not even "_"
2015 }
2016 
2017 
2018 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2019   // no suffix required
2020 }
2021 
2022 // This method is a copy of JDK's sysGetLastErrorString
2023 // from src/solaris/hpi/src/system_md.c
2024 
2025 size_t os::lasterror(char *buf, size_t len) {
2026   if (errno == 0)  return 0;
2027 
2028   const char *s = os::strerror(errno);
2029   size_t n = ::strlen(s);
2030   if (n >= len) {
2031     n = len - 1;
2032   }
2033   ::strncpy(buf, s, n);
2034   buf[n] = '\0';
2035   return n;
2036 }
2037 
2038 
2039 // sun.misc.Signal
2040 
2041 extern "C" {
2042   static void UserHandler(int sig, void *siginfo, void *context) {
2043     // Ctrl-C is pressed during error reporting, likely because the error
2044     // handler fails to abort. Let VM die immediately.
2045     if (sig == SIGINT && VMError::is_error_reported()) {
2046       os::die();
2047     }
2048 
2049     os::signal_notify(sig);
2050     // We do not need to reinstate the signal handler each time...
2051   }
2052 }
2053 
2054 void* os::user_handler() {
2055   return CAST_FROM_FN_PTR(void*, UserHandler);
2056 }
2057 
2058 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2059   struct timespec ts;
2060   unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2061 
2062   return ts;
2063 }
2064 
2065 extern "C" {
2066   typedef void (*sa_handler_t)(int);
2067   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2068 }
2069 
2070 void* os::signal(int signal_number, void* handler) {
2071   struct sigaction sigAct, oldSigAct;
2072   sigfillset(&(sigAct.sa_mask));
2073   sigAct.sa_flags = SA_RESTART & ~SA_RESETHAND;
2074   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2075 
2076   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2077     // -1 means registration failed
2078     return (void *)-1;
2079   }
2080 
2081   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2082 }
2083 
2084 void os::signal_raise(int signal_number) {
2085   raise(signal_number);
2086 }
2087 
2088 // The following code is moved from os.cpp for making this
2089 // code platform specific, which it is by its very nature.
2090 
2091 // a counter for each possible signal value
2092 static int Sigexit = 0;
2093 static jint *pending_signals = NULL;
2094 static int *preinstalled_sigs = NULL;
2095 static struct sigaction *chainedsigactions = NULL;
2096 static sema_t sig_sem;
2097 typedef int (*version_getting_t)();
2098 version_getting_t os::Solaris::get_libjsig_version = NULL;
2099 
2100 int os::sigexitnum_pd() {
2101   assert(Sigexit > 0, "signal memory not yet initialized");
2102   return Sigexit;
2103 }
2104 
2105 void os::Solaris::init_signal_mem() {
2106   // Initialize signal structures
2107   Maxsignum = SIGRTMAX;
2108   Sigexit = Maxsignum+1;
2109   assert(Maxsignum >0, "Unable to obtain max signal number");
2110 
2111   // pending_signals has one int per signal
2112   // The additional signal is for SIGEXIT - exit signal to signal_thread
2113   pending_signals = (jint *)os::malloc(sizeof(jint) * (Sigexit+1), mtInternal);
2114   memset(pending_signals, 0, (sizeof(jint) * (Sigexit+1)));
2115 
2116   if (UseSignalChaining) {
2117     chainedsigactions = (struct sigaction *)malloc(sizeof(struct sigaction)
2118                                                    * (Maxsignum + 1), mtInternal);
2119     memset(chainedsigactions, 0, (sizeof(struct sigaction) * (Maxsignum + 1)));
2120     preinstalled_sigs = (int *)os::malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2121     memset(preinstalled_sigs, 0, (sizeof(int) * (Maxsignum + 1)));
2122   }
2123   ourSigFlags = (int*)malloc(sizeof(int) * (Maxsignum + 1), mtInternal);
2124   memset(ourSigFlags, 0, sizeof(int) * (Maxsignum + 1));
2125 }
2126 
2127 void os::signal_init_pd() {
2128   int ret;
2129 
2130   ret = ::sema_init(&sig_sem, 0, NULL, NULL);
2131   assert(ret == 0, "sema_init() failed");
2132 }
2133 
2134 void os::signal_notify(int signal_number) {
2135   int ret;
2136 
2137   Atomic::inc(&pending_signals[signal_number]);
2138   ret = ::sema_post(&sig_sem);
2139   assert(ret == 0, "sema_post() failed");
2140 }
2141 
2142 static int check_pending_signals(bool wait_for_signal) {
2143   int ret;
2144   while (true) {
2145     for (int i = 0; i < Sigexit + 1; i++) {
2146       jint n = pending_signals[i];
2147       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2148         return i;
2149       }
2150     }
2151     if (!wait_for_signal) {
2152       return -1;
2153     }
2154     JavaThread *thread = JavaThread::current();
2155     ThreadBlockInVM tbivm(thread);
2156 
2157     bool threadIsSuspended;
2158     do {
2159       thread->set_suspend_equivalent();
2160       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2161       while ((ret = ::sema_wait(&sig_sem)) == EINTR)
2162         ;
2163       assert(ret == 0, "sema_wait() failed");
2164 
2165       // were we externally suspended while we were waiting?
2166       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2167       if (threadIsSuspended) {
2168         // The semaphore has been incremented, but while we were waiting
2169         // another thread suspended us. We don't want to continue running
2170         // while suspended because that would surprise the thread that
2171         // suspended us.
2172         ret = ::sema_post(&sig_sem);
2173         assert(ret == 0, "sema_post() failed");
2174 
2175         thread->java_suspend_self();
2176       }
2177     } while (threadIsSuspended);
2178   }
2179 }
2180 
2181 int os::signal_lookup() {
2182   return check_pending_signals(false);
2183 }
2184 
2185 int os::signal_wait() {
2186   return check_pending_signals(true);
2187 }
2188 
2189 ////////////////////////////////////////////////////////////////////////////////
2190 // Virtual Memory
2191 
2192 static int page_size = -1;
2193 
2194 // The mmap MAP_ALIGN flag is supported on Solaris 9 and later.  init_2() will
2195 // clear this var if support is not available.
2196 static bool has_map_align = true;
2197 
2198 int os::vm_page_size() {
2199   assert(page_size != -1, "must call os::init");
2200   return page_size;
2201 }
2202 
2203 // Solaris allocates memory by pages.
2204 int os::vm_allocation_granularity() {
2205   assert(page_size != -1, "must call os::init");
2206   return page_size;
2207 }
2208 
2209 static bool recoverable_mmap_error(int err) {
2210   // See if the error is one we can let the caller handle. This
2211   // list of errno values comes from the Solaris mmap(2) man page.
2212   switch (err) {
2213   case EBADF:
2214   case EINVAL:
2215   case ENOTSUP:
2216     // let the caller deal with these errors
2217     return true;
2218 
2219   default:
2220     // Any remaining errors on this OS can cause our reserved mapping
2221     // to be lost. That can cause confusion where different data
2222     // structures think they have the same memory mapped. The worst
2223     // scenario is if both the VM and a library think they have the
2224     // same memory mapped.
2225     return false;
2226   }
2227 }
2228 
2229 static void warn_fail_commit_memory(char* addr, size_t bytes, bool exec,
2230                                     int err) {
2231   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2232           ", %d) failed; error='%s' (errno=%d)", addr, bytes, exec,
2233           os::strerror(err), err);
2234 }
2235 
2236 static void warn_fail_commit_memory(char* addr, size_t bytes,
2237                                     size_t alignment_hint, bool exec,
2238                                     int err) {
2239   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2240           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, bytes,
2241           alignment_hint, exec, os::strerror(err), err);
2242 }
2243 
2244 int os::Solaris::commit_memory_impl(char* addr, size_t bytes, bool exec) {
2245   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2246   size_t size = bytes;
2247   char *res = Solaris::mmap_chunk(addr, size, MAP_PRIVATE|MAP_FIXED, prot);
2248   if (res != NULL) {
2249     if (UseNUMAInterleaving) {
2250       numa_make_global(addr, bytes);
2251     }
2252     return 0;
2253   }
2254 
2255   int err = errno;  // save errno from mmap() call in mmap_chunk()
2256 
2257   if (!recoverable_mmap_error(err)) {
2258     warn_fail_commit_memory(addr, bytes, exec, err);
2259     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "committing reserved memory.");
2260   }
2261 
2262   return err;
2263 }
2264 
2265 bool os::pd_commit_memory(char* addr, size_t bytes, bool exec) {
2266   return Solaris::commit_memory_impl(addr, bytes, exec) == 0;
2267 }
2268 
2269 void os::pd_commit_memory_or_exit(char* addr, size_t bytes, bool exec,
2270                                   const char* mesg) {
2271   assert(mesg != NULL, "mesg must be specified");
2272   int err = os::Solaris::commit_memory_impl(addr, bytes, exec);
2273   if (err != 0) {
2274     // the caller wants all commit errors to exit with the specified mesg:
2275     warn_fail_commit_memory(addr, bytes, exec, err);
2276     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2277   }
2278 }
2279 
2280 size_t os::Solaris::page_size_for_alignment(size_t alignment) {
2281   assert(is_aligned(alignment, (size_t) vm_page_size()),
2282          SIZE_FORMAT " is not aligned to " SIZE_FORMAT,
2283          alignment, (size_t) vm_page_size());
2284 
2285   for (int i = 0; _page_sizes[i] != 0; i++) {
2286     if (is_aligned(alignment, _page_sizes[i])) {
2287       return _page_sizes[i];
2288     }
2289   }
2290 
2291   return (size_t) vm_page_size();
2292 }
2293 
2294 int os::Solaris::commit_memory_impl(char* addr, size_t bytes,
2295                                     size_t alignment_hint, bool exec) {
2296   int err = Solaris::commit_memory_impl(addr, bytes, exec);
2297   if (err == 0 && UseLargePages && alignment_hint > 0) {
2298     assert(is_aligned(bytes, alignment_hint),
2299            SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, alignment_hint);
2300 
2301     // The syscall memcntl requires an exact page size (see man memcntl for details).
2302     size_t page_size = page_size_for_alignment(alignment_hint);
2303     if (page_size > (size_t) vm_page_size()) {
2304       (void)Solaris::setup_large_pages(addr, bytes, page_size);
2305     }
2306   }
2307   return err;
2308 }
2309 
2310 bool os::pd_commit_memory(char* addr, size_t bytes, size_t alignment_hint,
2311                           bool exec) {
2312   return Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec) == 0;
2313 }
2314 
2315 void os::pd_commit_memory_or_exit(char* addr, size_t bytes,
2316                                   size_t alignment_hint, bool exec,
2317                                   const char* mesg) {
2318   assert(mesg != NULL, "mesg must be specified");
2319   int err = os::Solaris::commit_memory_impl(addr, bytes, alignment_hint, exec);
2320   if (err != 0) {
2321     // the caller wants all commit errors to exit with the specified mesg:
2322     warn_fail_commit_memory(addr, bytes, alignment_hint, exec, err);
2323     vm_exit_out_of_memory(bytes, OOM_MMAP_ERROR, "%s", mesg);
2324   }
2325 }
2326 
2327 // Uncommit the pages in a specified region.
2328 void os::pd_free_memory(char* addr, size_t bytes, size_t alignment_hint) {
2329   if (madvise(addr, bytes, MADV_FREE) < 0) {
2330     debug_only(warning("MADV_FREE failed."));
2331     return;
2332   }
2333 }
2334 
2335 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2336   return os::commit_memory(addr, size, !ExecMem);
2337 }
2338 
2339 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2340   return os::uncommit_memory(addr, size);
2341 }
2342 
2343 // Change the page size in a given range.
2344 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2345   assert((intptr_t)addr % alignment_hint == 0, "Address should be aligned.");
2346   assert((intptr_t)(addr + bytes) % alignment_hint == 0, "End should be aligned.");
2347   if (UseLargePages) {
2348     size_t page_size = Solaris::page_size_for_alignment(alignment_hint);
2349     if (page_size > (size_t) vm_page_size()) {
2350       Solaris::setup_large_pages(addr, bytes, page_size);
2351     }
2352   }
2353 }
2354 
2355 // Tell the OS to make the range local to the first-touching LWP
2356 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2357   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2358   if (madvise(addr, bytes, MADV_ACCESS_LWP) < 0) {
2359     debug_only(warning("MADV_ACCESS_LWP failed."));
2360   }
2361 }
2362 
2363 // Tell the OS that this range would be accessed from different LWPs.
2364 void os::numa_make_global(char *addr, size_t bytes) {
2365   assert((intptr_t)addr % os::vm_page_size() == 0, "Address should be page-aligned.");
2366   if (madvise(addr, bytes, MADV_ACCESS_MANY) < 0) {
2367     debug_only(warning("MADV_ACCESS_MANY failed."));
2368   }
2369 }
2370 
2371 // Get the number of the locality groups.
2372 size_t os::numa_get_groups_num() {
2373   size_t n = Solaris::lgrp_nlgrps(Solaris::lgrp_cookie());
2374   return n != -1 ? n : 1;
2375 }
2376 
2377 // Get a list of leaf locality groups. A leaf lgroup is group that
2378 // doesn't have any children. Typical leaf group is a CPU or a CPU/memory
2379 // board. An LWP is assigned to one of these groups upon creation.
2380 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2381   if ((ids[0] = Solaris::lgrp_root(Solaris::lgrp_cookie())) == -1) {
2382     ids[0] = 0;
2383     return 1;
2384   }
2385   int result_size = 0, top = 1, bottom = 0, cur = 0;
2386   for (int k = 0; k < size; k++) {
2387     int r = Solaris::lgrp_children(Solaris::lgrp_cookie(), ids[cur],
2388                                    (Solaris::lgrp_id_t*)&ids[top], size - top);
2389     if (r == -1) {
2390       ids[0] = 0;
2391       return 1;
2392     }
2393     if (!r) {
2394       // That's a leaf node.
2395       assert(bottom <= cur, "Sanity check");
2396       // Check if the node has memory
2397       if (Solaris::lgrp_resources(Solaris::lgrp_cookie(), ids[cur],
2398                                   NULL, 0, LGRP_RSRC_MEM) > 0) {
2399         ids[bottom++] = ids[cur];
2400       }
2401     }
2402     top += r;
2403     cur++;
2404   }
2405   if (bottom == 0) {
2406     // Handle a situation, when the OS reports no memory available.
2407     // Assume UMA architecture.
2408     ids[0] = 0;
2409     return 1;
2410   }
2411   return bottom;
2412 }
2413 
2414 // Detect the topology change. Typically happens during CPU plugging-unplugging.
2415 bool os::numa_topology_changed() {
2416   int is_stale = Solaris::lgrp_cookie_stale(Solaris::lgrp_cookie());
2417   if (is_stale != -1 && is_stale) {
2418     Solaris::lgrp_fini(Solaris::lgrp_cookie());
2419     Solaris::lgrp_cookie_t c = Solaris::lgrp_init(Solaris::LGRP_VIEW_CALLER);
2420     assert(c != 0, "Failure to initialize LGRP API");
2421     Solaris::set_lgrp_cookie(c);
2422     return true;
2423   }
2424   return false;
2425 }
2426 
2427 // Get the group id of the current LWP.
2428 int os::numa_get_group_id() {
2429   int lgrp_id = Solaris::lgrp_home(P_LWPID, P_MYID);
2430   if (lgrp_id == -1) {
2431     return 0;
2432   }
2433   const int size = os::numa_get_groups_num();
2434   int *ids = (int*)alloca(size * sizeof(int));
2435 
2436   // Get the ids of all lgroups with memory; r is the count.
2437   int r = Solaris::lgrp_resources(Solaris::lgrp_cookie(), lgrp_id,
2438                                   (Solaris::lgrp_id_t*)ids, size, LGRP_RSRC_MEM);
2439   if (r <= 0) {
2440     return 0;
2441   }
2442   return ids[os::random() % r];
2443 }
2444 
2445 // Request information about the page.
2446 bool os::get_page_info(char *start, page_info* info) {
2447   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2448   uint64_t addr = (uintptr_t)start;
2449   uint64_t outdata[2];
2450   uint_t validity = 0;
2451 
2452   if (meminfo(&addr, 1, info_types, 2, outdata, &validity) < 0) {
2453     return false;
2454   }
2455 
2456   info->size = 0;
2457   info->lgrp_id = -1;
2458 
2459   if ((validity & 1) != 0) {
2460     if ((validity & 2) != 0) {
2461       info->lgrp_id = outdata[0];
2462     }
2463     if ((validity & 4) != 0) {
2464       info->size = outdata[1];
2465     }
2466     return true;
2467   }
2468   return false;
2469 }
2470 
2471 // Scan the pages from start to end until a page different than
2472 // the one described in the info parameter is encountered.
2473 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2474                      page_info* page_found) {
2475   const uint_t info_types[] = { MEMINFO_VLGRP, MEMINFO_VPAGESIZE };
2476   const size_t types = sizeof(info_types) / sizeof(info_types[0]);
2477   uint64_t addrs[MAX_MEMINFO_CNT], outdata[types * MAX_MEMINFO_CNT + 1];
2478   uint_t validity[MAX_MEMINFO_CNT];
2479 
2480   size_t page_size = MAX2((size_t)os::vm_page_size(), page_expected->size);
2481   uint64_t p = (uint64_t)start;
2482   while (p < (uint64_t)end) {
2483     addrs[0] = p;
2484     size_t addrs_count = 1;
2485     while (addrs_count < MAX_MEMINFO_CNT && addrs[addrs_count - 1] + page_size < (uint64_t)end) {
2486       addrs[addrs_count] = addrs[addrs_count - 1] + page_size;
2487       addrs_count++;
2488     }
2489 
2490     if (meminfo(addrs, addrs_count, info_types, types, outdata, validity) < 0) {
2491       return NULL;
2492     }
2493 
2494     size_t i = 0;
2495     for (; i < addrs_count; i++) {
2496       if ((validity[i] & 1) != 0) {
2497         if ((validity[i] & 4) != 0) {
2498           if (outdata[types * i + 1] != page_expected->size) {
2499             break;
2500           }
2501         } else if (page_expected->size != 0) {
2502           break;
2503         }
2504 
2505         if ((validity[i] & 2) != 0 && page_expected->lgrp_id > 0) {
2506           if (outdata[types * i] != page_expected->lgrp_id) {
2507             break;
2508           }
2509         }
2510       } else {
2511         return NULL;
2512       }
2513     }
2514 
2515     if (i < addrs_count) {
2516       if ((validity[i] & 2) != 0) {
2517         page_found->lgrp_id = outdata[types * i];
2518       } else {
2519         page_found->lgrp_id = -1;
2520       }
2521       if ((validity[i] & 4) != 0) {
2522         page_found->size = outdata[types * i + 1];
2523       } else {
2524         page_found->size = 0;
2525       }
2526       return (char*)addrs[i];
2527     }
2528 
2529     p = addrs[addrs_count - 1] + page_size;
2530   }
2531   return end;
2532 }
2533 
2534 bool os::pd_uncommit_memory(char* addr, size_t bytes) {
2535   size_t size = bytes;
2536   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2537   // uncommitted page. Otherwise, the read/write might succeed if we
2538   // have enough swap space to back the physical page.
2539   return
2540     NULL != Solaris::mmap_chunk(addr, size,
2541                                 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE,
2542                                 PROT_NONE);
2543 }
2544 
2545 char* os::Solaris::mmap_chunk(char *addr, size_t size, int flags, int prot) {
2546   char *b = (char *)mmap(addr, size, prot, flags, os::Solaris::_dev_zero_fd, 0);
2547 
2548   if (b == MAP_FAILED) {
2549     return NULL;
2550   }
2551   return b;
2552 }
2553 
2554 char* os::Solaris::anon_mmap(char* requested_addr, size_t bytes,
2555                              size_t alignment_hint, bool fixed) {
2556   char* addr = requested_addr;
2557   int flags = MAP_PRIVATE | MAP_NORESERVE;
2558 
2559   assert(!(fixed && (alignment_hint > 0)),
2560          "alignment hint meaningless with fixed mmap");
2561 
2562   if (fixed) {
2563     flags |= MAP_FIXED;
2564   } else if (has_map_align && (alignment_hint > (size_t) vm_page_size())) {
2565     flags |= MAP_ALIGN;
2566     addr = (char*) alignment_hint;
2567   }
2568 
2569   // Map uncommitted pages PROT_NONE so we fail early if we touch an
2570   // uncommitted page. Otherwise, the read/write might succeed if we
2571   // have enough swap space to back the physical page.
2572   return mmap_chunk(addr, bytes, flags, PROT_NONE);
2573 }
2574 
2575 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2576                             size_t alignment_hint) {
2577   char* addr = Solaris::anon_mmap(requested_addr, bytes, alignment_hint,
2578                                   (requested_addr != NULL));
2579 
2580   guarantee(requested_addr == NULL || requested_addr == addr,
2581             "OS failed to return requested mmap address.");
2582   return addr;
2583 }
2584 
2585 // Reserve memory at an arbitrary address, only if that area is
2586 // available (and not reserved for something else).
2587 
2588 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
2589   const int max_tries = 10;
2590   char* base[max_tries];
2591   size_t size[max_tries];
2592 
2593   // Solaris adds a gap between mmap'ed regions.  The size of the gap
2594   // is dependent on the requested size and the MMU.  Our initial gap
2595   // value here is just a guess and will be corrected later.
2596   bool had_top_overlap = false;
2597   bool have_adjusted_gap = false;
2598   size_t gap = 0x400000;
2599 
2600   // Assert only that the size is a multiple of the page size, since
2601   // that's all that mmap requires, and since that's all we really know
2602   // about at this low abstraction level.  If we need higher alignment,
2603   // we can either pass an alignment to this method or verify alignment
2604   // in one of the methods further up the call chain.  See bug 5044738.
2605   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
2606 
2607   // Since snv_84, Solaris attempts to honor the address hint - see 5003415.
2608   // Give it a try, if the kernel honors the hint we can return immediately.
2609   char* addr = Solaris::anon_mmap(requested_addr, bytes, 0, false);
2610 
2611   volatile int err = errno;
2612   if (addr == requested_addr) {
2613     return addr;
2614   } else if (addr != NULL) {
2615     pd_unmap_memory(addr, bytes);
2616   }
2617 
2618   if (log_is_enabled(Warning, os)) {
2619     char buf[256];
2620     buf[0] = '\0';
2621     if (addr == NULL) {
2622       jio_snprintf(buf, sizeof(buf), ": %s", os::strerror(err));
2623     }
2624     log_info(os)("attempt_reserve_memory_at: couldn't reserve " SIZE_FORMAT " bytes at "
2625             PTR_FORMAT ": reserve_memory_helper returned " PTR_FORMAT
2626             "%s", bytes, requested_addr, addr, buf);
2627   }
2628 
2629   // Address hint method didn't work.  Fall back to the old method.
2630   // In theory, once SNV becomes our oldest supported platform, this
2631   // code will no longer be needed.
2632   //
2633   // Repeatedly allocate blocks until the block is allocated at the
2634   // right spot. Give up after max_tries.
2635   int i;
2636   for (i = 0; i < max_tries; ++i) {
2637     base[i] = reserve_memory(bytes);
2638 
2639     if (base[i] != NULL) {
2640       // Is this the block we wanted?
2641       if (base[i] == requested_addr) {
2642         size[i] = bytes;
2643         break;
2644       }
2645 
2646       // check that the gap value is right
2647       if (had_top_overlap && !have_adjusted_gap) {
2648         size_t actual_gap = base[i-1] - base[i] - bytes;
2649         if (gap != actual_gap) {
2650           // adjust the gap value and retry the last 2 allocations
2651           assert(i > 0, "gap adjustment code problem");
2652           have_adjusted_gap = true;  // adjust the gap only once, just in case
2653           gap = actual_gap;
2654           log_info(os)("attempt_reserve_memory_at: adjusted gap to 0x%lx", gap);
2655           unmap_memory(base[i], bytes);
2656           unmap_memory(base[i-1], size[i-1]);
2657           i-=2;
2658           continue;
2659         }
2660       }
2661 
2662       // Does this overlap the block we wanted? Give back the overlapped
2663       // parts and try again.
2664       //
2665       // There is still a bug in this code: if top_overlap == bytes,
2666       // the overlap is offset from requested region by the value of gap.
2667       // In this case giving back the overlapped part will not work,
2668       // because we'll give back the entire block at base[i] and
2669       // therefore the subsequent allocation will not generate a new gap.
2670       // This could be fixed with a new algorithm that used larger
2671       // or variable size chunks to find the requested region -
2672       // but such a change would introduce additional complications.
2673       // It's rare enough that the planets align for this bug,
2674       // so we'll just wait for a fix for 6204603/5003415 which
2675       // will provide a mmap flag to allow us to avoid this business.
2676 
2677       size_t top_overlap = requested_addr + (bytes + gap) - base[i];
2678       if (top_overlap >= 0 && top_overlap < bytes) {
2679         had_top_overlap = true;
2680         unmap_memory(base[i], top_overlap);
2681         base[i] += top_overlap;
2682         size[i] = bytes - top_overlap;
2683       } else {
2684         size_t bottom_overlap = base[i] + bytes - requested_addr;
2685         if (bottom_overlap >= 0 && bottom_overlap < bytes) {
2686           if (bottom_overlap == 0) {
2687             log_info(os)("attempt_reserve_memory_at: possible alignment bug");
2688           }
2689           unmap_memory(requested_addr, bottom_overlap);
2690           size[i] = bytes - bottom_overlap;
2691         } else {
2692           size[i] = bytes;
2693         }
2694       }
2695     }
2696   }
2697 
2698   // Give back the unused reserved pieces.
2699 
2700   for (int j = 0; j < i; ++j) {
2701     if (base[j] != NULL) {
2702       unmap_memory(base[j], size[j]);
2703     }
2704   }
2705 
2706   return (i < max_tries) ? requested_addr : NULL;
2707 }
2708 
2709 bool os::pd_release_memory(char* addr, size_t bytes) {
2710   size_t size = bytes;
2711   return munmap(addr, size) == 0;
2712 }
2713 
2714 static bool solaris_mprotect(char* addr, size_t bytes, int prot) {
2715   assert(addr == (char*)align_down((uintptr_t)addr, os::vm_page_size()),
2716          "addr must be page aligned");
2717   int retVal = mprotect(addr, bytes, prot);
2718   return retVal == 0;
2719 }
2720 
2721 // Protect memory (Used to pass readonly pages through
2722 // JNI GetArray<type>Elements with empty arrays.)
2723 // Also, used for serialization page and for compressed oops null pointer
2724 // checking.
2725 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2726                         bool is_committed) {
2727   unsigned int p = 0;
2728   switch (prot) {
2729   case MEM_PROT_NONE: p = PROT_NONE; break;
2730   case MEM_PROT_READ: p = PROT_READ; break;
2731   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
2732   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2733   default:
2734     ShouldNotReachHere();
2735   }
2736   // is_committed is unused.
2737   return solaris_mprotect(addr, bytes, p);
2738 }
2739 
2740 // guard_memory and unguard_memory only happens within stack guard pages.
2741 // Since ISM pertains only to the heap, guard and unguard memory should not
2742 /// happen with an ISM region.
2743 bool os::guard_memory(char* addr, size_t bytes) {
2744   return solaris_mprotect(addr, bytes, PROT_NONE);
2745 }
2746 
2747 bool os::unguard_memory(char* addr, size_t bytes) {
2748   return solaris_mprotect(addr, bytes, PROT_READ|PROT_WRITE);
2749 }
2750 
2751 // Large page support
2752 static size_t _large_page_size = 0;
2753 
2754 // Insertion sort for small arrays (descending order).
2755 static void insertion_sort_descending(size_t* array, int len) {
2756   for (int i = 0; i < len; i++) {
2757     size_t val = array[i];
2758     for (size_t key = i; key > 0 && array[key - 1] < val; --key) {
2759       size_t tmp = array[key];
2760       array[key] = array[key - 1];
2761       array[key - 1] = tmp;
2762     }
2763   }
2764 }
2765 
2766 bool os::Solaris::mpss_sanity_check(bool warn, size_t* page_size) {
2767   const unsigned int usable_count = VM_Version::page_size_count();
2768   if (usable_count == 1) {
2769     return false;
2770   }
2771 
2772   // Find the right getpagesizes interface.  When solaris 11 is the minimum
2773   // build platform, getpagesizes() (without the '2') can be called directly.
2774   typedef int (*gps_t)(size_t[], int);
2775   gps_t gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes2"));
2776   if (gps_func == NULL) {
2777     gps_func = CAST_TO_FN_PTR(gps_t, dlsym(RTLD_DEFAULT, "getpagesizes"));
2778     if (gps_func == NULL) {
2779       if (warn) {
2780         warning("MPSS is not supported by the operating system.");
2781       }
2782       return false;
2783     }
2784   }
2785 
2786   // Fill the array of page sizes.
2787   int n = (*gps_func)(_page_sizes, page_sizes_max);
2788   assert(n > 0, "Solaris bug?");
2789 
2790   if (n == page_sizes_max) {
2791     // Add a sentinel value (necessary only if the array was completely filled
2792     // since it is static (zeroed at initialization)).
2793     _page_sizes[--n] = 0;
2794     DEBUG_ONLY(warning("increase the size of the os::_page_sizes array.");)
2795   }
2796   assert(_page_sizes[n] == 0, "missing sentinel");
2797   trace_page_sizes("available page sizes", _page_sizes, n);
2798 
2799   if (n == 1) return false;     // Only one page size available.
2800 
2801   // Skip sizes larger than 4M (or LargePageSizeInBytes if it was set) and
2802   // select up to usable_count elements.  First sort the array, find the first
2803   // acceptable value, then copy the usable sizes to the top of the array and
2804   // trim the rest.  Make sure to include the default page size :-).
2805   //
2806   // A better policy could get rid of the 4M limit by taking the sizes of the
2807   // important VM memory regions (java heap and possibly the code cache) into
2808   // account.
2809   insertion_sort_descending(_page_sizes, n);
2810   const size_t size_limit =
2811     FLAG_IS_DEFAULT(LargePageSizeInBytes) ? 4 * M : LargePageSizeInBytes;
2812   int beg;
2813   for (beg = 0; beg < n && _page_sizes[beg] > size_limit; ++beg) /* empty */;
2814   const int end = MIN2((int)usable_count, n) - 1;
2815   for (int cur = 0; cur < end; ++cur, ++beg) {
2816     _page_sizes[cur] = _page_sizes[beg];
2817   }
2818   _page_sizes[end] = vm_page_size();
2819   _page_sizes[end + 1] = 0;
2820 
2821   if (_page_sizes[end] > _page_sizes[end - 1]) {
2822     // Default page size is not the smallest; sort again.
2823     insertion_sort_descending(_page_sizes, end + 1);
2824   }
2825   *page_size = _page_sizes[0];
2826 
2827   trace_page_sizes("usable page sizes", _page_sizes, end + 1);
2828   return true;
2829 }
2830 
2831 void os::large_page_init() {
2832   if (UseLargePages) {
2833     // print a warning if any large page related flag is specified on command line
2834     bool warn_on_failure = !FLAG_IS_DEFAULT(UseLargePages)        ||
2835                            !FLAG_IS_DEFAULT(LargePageSizeInBytes);
2836 
2837     UseLargePages = Solaris::mpss_sanity_check(warn_on_failure, &_large_page_size);
2838   }
2839 }
2840 
2841 bool os::Solaris::is_valid_page_size(size_t bytes) {
2842   for (int i = 0; _page_sizes[i] != 0; i++) {
2843     if (_page_sizes[i] == bytes) {
2844       return true;
2845     }
2846   }
2847   return false;
2848 }
2849 
2850 bool os::Solaris::setup_large_pages(caddr_t start, size_t bytes, size_t align) {
2851   assert(is_valid_page_size(align), SIZE_FORMAT " is not a valid page size", align);
2852   assert(is_aligned((void*) start, align),
2853          PTR_FORMAT " is not aligned to " SIZE_FORMAT, p2i((void*) start), align);
2854   assert(is_aligned(bytes, align),
2855          SIZE_FORMAT " is not aligned to " SIZE_FORMAT, bytes, align);
2856 
2857   // Signal to OS that we want large pages for addresses
2858   // from addr, addr + bytes
2859   struct memcntl_mha mpss_struct;
2860   mpss_struct.mha_cmd = MHA_MAPSIZE_VA;
2861   mpss_struct.mha_pagesize = align;
2862   mpss_struct.mha_flags = 0;
2863   // Upon successful completion, memcntl() returns 0
2864   if (memcntl(start, bytes, MC_HAT_ADVISE, (caddr_t) &mpss_struct, 0, 0)) {
2865     debug_only(warning("Attempt to use MPSS failed."));
2866     return false;
2867   }
2868   return true;
2869 }
2870 
2871 char* os::reserve_memory_special(size_t size, size_t alignment, char* addr, bool exec) {
2872   fatal("os::reserve_memory_special should not be called on Solaris.");
2873   return NULL;
2874 }
2875 
2876 bool os::release_memory_special(char* base, size_t bytes) {
2877   fatal("os::release_memory_special should not be called on Solaris.");
2878   return false;
2879 }
2880 
2881 size_t os::large_page_size() {
2882   return _large_page_size;
2883 }
2884 
2885 // MPSS allows application to commit large page memory on demand; with ISM
2886 // the entire memory region must be allocated as shared memory.
2887 bool os::can_commit_large_page_memory() {
2888   return true;
2889 }
2890 
2891 bool os::can_execute_large_page_memory() {
2892   return true;
2893 }
2894 
2895 // Read calls from inside the vm need to perform state transitions
2896 size_t os::read(int fd, void *buf, unsigned int nBytes) {
2897   size_t res;
2898   JavaThread* thread = (JavaThread*)Thread::current();
2899   assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
2900   ThreadBlockInVM tbiv(thread);
2901   RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
2902   return res;
2903 }
2904 
2905 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
2906   size_t res;
2907   JavaThread* thread = (JavaThread*)Thread::current();
2908   assert(thread->thread_state() == _thread_in_vm, "Assumed _thread_in_vm");
2909   ThreadBlockInVM tbiv(thread);
2910   RESTARTABLE(::pread(fd, buf, (size_t) nBytes, offset), res);
2911   return res;
2912 }
2913 
2914 size_t os::restartable_read(int fd, void *buf, unsigned int nBytes) {
2915   size_t res;
2916   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
2917          "Assumed _thread_in_native");
2918   RESTARTABLE(::read(fd, buf, (size_t) nBytes), res);
2919   return res;
2920 }
2921 
2922 void os::naked_short_sleep(jlong ms) {
2923   assert(ms < 1000, "Un-interruptable sleep, short time use only");
2924 
2925   // usleep is deprecated and removed from POSIX, in favour of nanosleep, but
2926   // Solaris requires -lrt for this.
2927   usleep((ms * 1000));
2928 
2929   return;
2930 }
2931 
2932 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
2933 void os::infinite_sleep() {
2934   while (true) {    // sleep forever ...
2935     ::sleep(100);   // ... 100 seconds at a time
2936   }
2937 }
2938 
2939 // Used to convert frequent JVM_Yield() to nops
2940 bool os::dont_yield() {
2941   if (DontYieldALot) {
2942     static hrtime_t last_time = 0;
2943     hrtime_t diff = getTimeNanos() - last_time;
2944 
2945     if (diff < DontYieldALotInterval * 1000000) {
2946       return true;
2947     }
2948 
2949     last_time += diff;
2950 
2951     return false;
2952   } else {
2953     return false;
2954   }
2955 }
2956 
2957 // Note that yield semantics are defined by the scheduling class to which
2958 // the thread currently belongs.  Typically, yield will _not yield to
2959 // other equal or higher priority threads that reside on the dispatch queues
2960 // of other CPUs.
2961 
2962 void os::naked_yield() {
2963   thr_yield();
2964 }
2965 
2966 // Interface for setting lwp priorities.  We are using T2 libthread,
2967 // which forces the use of bound threads, so all of our threads will
2968 // be assigned to real lwp's.  Using the thr_setprio function is
2969 // meaningless in this mode so we must adjust the real lwp's priority.
2970 // The routines below implement the getting and setting of lwp priorities.
2971 //
2972 // Note: There are three priority scales used on Solaris.  Java priotities
2973 //       which range from 1 to 10, libthread "thr_setprio" scale which range
2974 //       from 0 to 127, and the current scheduling class of the process we
2975 //       are running in.  This is typically from -60 to +60.
2976 //       The setting of the lwp priorities in done after a call to thr_setprio
2977 //       so Java priorities are mapped to libthread priorities and we map from
2978 //       the latter to lwp priorities.  We don't keep priorities stored in
2979 //       Java priorities since some of our worker threads want to set priorities
2980 //       higher than all Java threads.
2981 //
2982 // For related information:
2983 // (1)  man -s 2 priocntl
2984 // (2)  man -s 4 priocntl
2985 // (3)  man dispadmin
2986 // =    librt.so
2987 // =    libthread/common/rtsched.c - thrp_setlwpprio().
2988 // =    ps -cL <pid> ... to validate priority.
2989 // =    sched_get_priority_min and _max
2990 //              pthread_create
2991 //              sched_setparam
2992 //              pthread_setschedparam
2993 //
2994 // Assumptions:
2995 // +    We assume that all threads in the process belong to the same
2996 //              scheduling class.   IE. an homogenous process.
2997 // +    Must be root or in IA group to change change "interactive" attribute.
2998 //              Priocntl() will fail silently.  The only indication of failure is when
2999 //              we read-back the value and notice that it hasn't changed.
3000 // +    Interactive threads enter the runq at the head, non-interactive at the tail.
3001 // +    For RT, change timeslice as well.  Invariant:
3002 //              constant "priority integral"
3003 //              Konst == TimeSlice * (60-Priority)
3004 //              Given a priority, compute appropriate timeslice.
3005 // +    Higher numerical values have higher priority.
3006 
3007 // sched class attributes
3008 typedef struct {
3009   int   schedPolicy;              // classID
3010   int   maxPrio;
3011   int   minPrio;
3012 } SchedInfo;
3013 
3014 
3015 static SchedInfo tsLimits, iaLimits, rtLimits, fxLimits;
3016 
3017 #ifdef ASSERT
3018 static int  ReadBackValidate = 1;
3019 #endif
3020 static int  myClass     = 0;
3021 static int  myMin       = 0;
3022 static int  myMax       = 0;
3023 static int  myCur       = 0;
3024 static bool priocntl_enable = false;
3025 
3026 static const int criticalPrio = FXCriticalPriority;
3027 static int java_MaxPriority_to_os_priority = 0; // Saved mapping
3028 
3029 
3030 // lwp_priocntl_init
3031 //
3032 // Try to determine the priority scale for our process.
3033 //
3034 // Return errno or 0 if OK.
3035 //
3036 static int lwp_priocntl_init() {
3037   int rslt;
3038   pcinfo_t ClassInfo;
3039   pcparms_t ParmInfo;
3040   int i;
3041 
3042   if (!UseThreadPriorities) return 0;
3043 
3044   // If ThreadPriorityPolicy is 1, switch tables
3045   if (ThreadPriorityPolicy == 1) {
3046     for (i = 0; i < CriticalPriority+1; i++)
3047       os::java_to_os_priority[i] = prio_policy1[i];
3048   }
3049   if (UseCriticalJavaThreadPriority) {
3050     // MaxPriority always maps to the FX scheduling class and criticalPrio.
3051     // See set_native_priority() and set_lwp_class_and_priority().
3052     // Save original MaxPriority mapping in case attempt to
3053     // use critical priority fails.
3054     java_MaxPriority_to_os_priority = os::java_to_os_priority[MaxPriority];
3055     // Set negative to distinguish from other priorities
3056     os::java_to_os_priority[MaxPriority] = -criticalPrio;
3057   }
3058 
3059   // Get IDs for a set of well-known scheduling classes.
3060   // TODO-FIXME: GETCLINFO returns the current # of classes in the
3061   // the system.  We should have a loop that iterates over the
3062   // classID values, which are known to be "small" integers.
3063 
3064   strcpy(ClassInfo.pc_clname, "TS");
3065   ClassInfo.pc_cid = -1;
3066   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3067   if (rslt < 0) return errno;
3068   assert(ClassInfo.pc_cid != -1, "cid for TS class is -1");
3069   tsLimits.schedPolicy = ClassInfo.pc_cid;
3070   tsLimits.maxPrio = ((tsinfo_t*)ClassInfo.pc_clinfo)->ts_maxupri;
3071   tsLimits.minPrio = -tsLimits.maxPrio;
3072 
3073   strcpy(ClassInfo.pc_clname, "IA");
3074   ClassInfo.pc_cid = -1;
3075   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3076   if (rslt < 0) return errno;
3077   assert(ClassInfo.pc_cid != -1, "cid for IA class is -1");
3078   iaLimits.schedPolicy = ClassInfo.pc_cid;
3079   iaLimits.maxPrio = ((iainfo_t*)ClassInfo.pc_clinfo)->ia_maxupri;
3080   iaLimits.minPrio = -iaLimits.maxPrio;
3081 
3082   strcpy(ClassInfo.pc_clname, "RT");
3083   ClassInfo.pc_cid = -1;
3084   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3085   if (rslt < 0) return errno;
3086   assert(ClassInfo.pc_cid != -1, "cid for RT class is -1");
3087   rtLimits.schedPolicy = ClassInfo.pc_cid;
3088   rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
3089   rtLimits.minPrio = 0;
3090 
3091   strcpy(ClassInfo.pc_clname, "FX");
3092   ClassInfo.pc_cid = -1;
3093   rslt = priocntl(P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
3094   if (rslt < 0) return errno;
3095   assert(ClassInfo.pc_cid != -1, "cid for FX class is -1");
3096   fxLimits.schedPolicy = ClassInfo.pc_cid;
3097   fxLimits.maxPrio = ((fxinfo_t*)ClassInfo.pc_clinfo)->fx_maxupri;
3098   fxLimits.minPrio = 0;
3099 
3100   // Query our "current" scheduling class.
3101   // This will normally be IA, TS or, rarely, FX or RT.
3102   memset(&ParmInfo, 0, sizeof(ParmInfo));
3103   ParmInfo.pc_cid = PC_CLNULL;
3104   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3105   if (rslt < 0) return errno;
3106   myClass = ParmInfo.pc_cid;
3107 
3108   // We now know our scheduling classId, get specific information
3109   // about the class.
3110   ClassInfo.pc_cid = myClass;
3111   ClassInfo.pc_clname[0] = 0;
3112   rslt = priocntl((idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo);
3113   if (rslt < 0) return errno;
3114 
3115   if (ThreadPriorityVerbose) {
3116     tty->print_cr("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
3117   }
3118 
3119   memset(&ParmInfo, 0, sizeof(pcparms_t));
3120   ParmInfo.pc_cid = PC_CLNULL;
3121   rslt = priocntl(P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
3122   if (rslt < 0) return errno;
3123 
3124   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3125     myMin = rtLimits.minPrio;
3126     myMax = rtLimits.maxPrio;
3127   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3128     iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3129     myMin = iaLimits.minPrio;
3130     myMax = iaLimits.maxPrio;
3131     myMax = MIN2(myMax, (int)iaInfo->ia_uprilim);       // clamp - restrict
3132   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3133     tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3134     myMin = tsLimits.minPrio;
3135     myMax = tsLimits.maxPrio;
3136     myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
3137   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3138     fxparms_t *fxInfo = (fxparms_t*)ParmInfo.pc_clparms;
3139     myMin = fxLimits.minPrio;
3140     myMax = fxLimits.maxPrio;
3141     myMax = MIN2(myMax, (int)fxInfo->fx_uprilim);       // clamp - restrict
3142   } else {
3143     // No clue - punt
3144     if (ThreadPriorityVerbose) {
3145       tty->print_cr("Unknown scheduling class: %s ... \n",
3146                     ClassInfo.pc_clname);
3147     }
3148     return EINVAL;      // no clue, punt
3149   }
3150 
3151   if (ThreadPriorityVerbose) {
3152     tty->print_cr("Thread priority Range: [%d..%d]\n", myMin, myMax);
3153   }
3154 
3155   priocntl_enable = true;  // Enable changing priorities
3156   return 0;
3157 }
3158 
3159 #define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
3160 #define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
3161 #define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
3162 #define FXPRI(x)        ((fxparms_t *)((x).pc_clparms))
3163 
3164 
3165 // scale_to_lwp_priority
3166 //
3167 // Convert from the libthread "thr_setprio" scale to our current
3168 // lwp scheduling class scale.
3169 //
3170 static int scale_to_lwp_priority(int rMin, int rMax, int x) {
3171   int v;
3172 
3173   if (x == 127) return rMax;            // avoid round-down
3174   v = (((x*(rMax-rMin)))/128)+rMin;
3175   return v;
3176 }
3177 
3178 
3179 // set_lwp_class_and_priority
3180 int set_lwp_class_and_priority(int ThreadID, int lwpid,
3181                                int newPrio, int new_class, bool scale) {
3182   int rslt;
3183   int Actual, Expected, prv;
3184   pcparms_t ParmInfo;                   // for GET-SET
3185 #ifdef ASSERT
3186   pcparms_t ReadBack;                   // for readback
3187 #endif
3188 
3189   // Set priority via PC_GETPARMS, update, PC_SETPARMS
3190   // Query current values.
3191   // TODO: accelerate this by eliminating the PC_GETPARMS call.
3192   // Cache "pcparms_t" in global ParmCache.
3193   // TODO: elide set-to-same-value
3194 
3195   // If something went wrong on init, don't change priorities.
3196   if (!priocntl_enable) {
3197     if (ThreadPriorityVerbose) {
3198       tty->print_cr("Trying to set priority but init failed, ignoring");
3199     }
3200     return EINVAL;
3201   }
3202 
3203   // If lwp hasn't started yet, just return
3204   // the _start routine will call us again.
3205   if (lwpid <= 0) {
3206     if (ThreadPriorityVerbose) {
3207       tty->print_cr("deferring the set_lwp_class_and_priority of thread "
3208                     INTPTR_FORMAT " to %d, lwpid not set",
3209                     ThreadID, newPrio);
3210     }
3211     return 0;
3212   }
3213 
3214   if (ThreadPriorityVerbose) {
3215     tty->print_cr ("set_lwp_class_and_priority("
3216                    INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
3217                    ThreadID, lwpid, newPrio);
3218   }
3219 
3220   memset(&ParmInfo, 0, sizeof(pcparms_t));
3221   ParmInfo.pc_cid = PC_CLNULL;
3222   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
3223   if (rslt < 0) return errno;
3224 
3225   int cur_class = ParmInfo.pc_cid;
3226   ParmInfo.pc_cid = (id_t)new_class;
3227 
3228   if (new_class == rtLimits.schedPolicy) {
3229     rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
3230     rtInfo->rt_pri     = scale ? scale_to_lwp_priority(rtLimits.minPrio,
3231                                                        rtLimits.maxPrio, newPrio)
3232                                : newPrio;
3233     rtInfo->rt_tqsecs  = RT_NOCHANGE;
3234     rtInfo->rt_tqnsecs = RT_NOCHANGE;
3235     if (ThreadPriorityVerbose) {
3236       tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
3237     }
3238   } else if (new_class == iaLimits.schedPolicy) {
3239     iaparms_t* iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
3240     int maxClamped     = MIN2(iaLimits.maxPrio,
3241                               cur_class == new_class
3242                               ? (int)iaInfo->ia_uprilim : iaLimits.maxPrio);
3243     iaInfo->ia_upri    = scale ? scale_to_lwp_priority(iaLimits.minPrio,
3244                                                        maxClamped, newPrio)
3245                                : newPrio;
3246     iaInfo->ia_uprilim = cur_class == new_class
3247                            ? IA_NOCHANGE : (pri_t)iaLimits.maxPrio;
3248     iaInfo->ia_mode    = IA_NOCHANGE;
3249     if (ThreadPriorityVerbose) {
3250       tty->print_cr("IA: [%d...%d] %d->%d\n",
3251                     iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
3252     }
3253   } else if (new_class == tsLimits.schedPolicy) {
3254     tsparms_t* tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
3255     int maxClamped     = MIN2(tsLimits.maxPrio,
3256                               cur_class == new_class
3257                               ? (int)tsInfo->ts_uprilim : tsLimits.maxPrio);
3258     tsInfo->ts_upri    = scale ? scale_to_lwp_priority(tsLimits.minPrio,
3259                                                        maxClamped, newPrio)
3260                                : newPrio;
3261     tsInfo->ts_uprilim = cur_class == new_class
3262                            ? TS_NOCHANGE : (pri_t)tsLimits.maxPrio;
3263     if (ThreadPriorityVerbose) {
3264       tty->print_cr("TS: [%d...%d] %d->%d\n",
3265                     tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
3266     }
3267   } else if (new_class == fxLimits.schedPolicy) {
3268     fxparms_t* fxInfo  = (fxparms_t*)ParmInfo.pc_clparms;
3269     int maxClamped     = MIN2(fxLimits.maxPrio,
3270                               cur_class == new_class
3271                               ? (int)fxInfo->fx_uprilim : fxLimits.maxPrio);
3272     fxInfo->fx_upri    = scale ? scale_to_lwp_priority(fxLimits.minPrio,
3273                                                        maxClamped, newPrio)
3274                                : newPrio;
3275     fxInfo->fx_uprilim = cur_class == new_class
3276                            ? FX_NOCHANGE : (pri_t)fxLimits.maxPrio;
3277     fxInfo->fx_tqsecs  = FX_NOCHANGE;
3278     fxInfo->fx_tqnsecs = FX_NOCHANGE;
3279     if (ThreadPriorityVerbose) {
3280       tty->print_cr("FX: [%d...%d] %d->%d\n",
3281                     fxLimits.minPrio, maxClamped, newPrio, fxInfo->fx_upri);
3282     }
3283   } else {
3284     if (ThreadPriorityVerbose) {
3285       tty->print_cr("Unknown new scheduling class %d\n", new_class);
3286     }
3287     return EINVAL;    // no clue, punt
3288   }
3289 
3290   rslt = priocntl(P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
3291   if (ThreadPriorityVerbose && rslt) {
3292     tty->print_cr ("PC_SETPARMS ->%d %d\n", rslt, errno);
3293   }
3294   if (rslt < 0) return errno;
3295 
3296 #ifdef ASSERT
3297   // Sanity check: read back what we just attempted to set.
3298   // In theory it could have changed in the interim ...
3299   //
3300   // The priocntl system call is tricky.
3301   // Sometimes it'll validate the priority value argument and
3302   // return EINVAL if unhappy.  At other times it fails silently.
3303   // Readbacks are prudent.
3304 
3305   if (!ReadBackValidate) return 0;
3306 
3307   memset(&ReadBack, 0, sizeof(pcparms_t));
3308   ReadBack.pc_cid = PC_CLNULL;
3309   rslt = priocntl(P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ReadBack);
3310   assert(rslt >= 0, "priocntl failed");
3311   Actual = Expected = 0xBAD;
3312   assert(ParmInfo.pc_cid == ReadBack.pc_cid, "cid's don't match");
3313   if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
3314     Actual   = RTPRI(ReadBack)->rt_pri;
3315     Expected = RTPRI(ParmInfo)->rt_pri;
3316   } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
3317     Actual   = IAPRI(ReadBack)->ia_upri;
3318     Expected = IAPRI(ParmInfo)->ia_upri;
3319   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
3320     Actual   = TSPRI(ReadBack)->ts_upri;
3321     Expected = TSPRI(ParmInfo)->ts_upri;
3322   } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
3323     Actual   = FXPRI(ReadBack)->fx_upri;
3324     Expected = FXPRI(ParmInfo)->fx_upri;
3325   } else {
3326     if (ThreadPriorityVerbose) {
3327       tty->print_cr("set_lwp_class_and_priority: unexpected class in readback: %d\n",
3328                     ParmInfo.pc_cid);
3329     }
3330   }
3331 
3332   if (Actual != Expected) {
3333     if (ThreadPriorityVerbose) {
3334       tty->print_cr ("set_lwp_class_and_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
3335                      lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
3336     }
3337   }
3338 #endif
3339 
3340   return 0;
3341 }
3342 
3343 // Solaris only gives access to 128 real priorities at a time,
3344 // so we expand Java's ten to fill this range.  This would be better
3345 // if we dynamically adjusted relative priorities.
3346 //
3347 // The ThreadPriorityPolicy option allows us to select 2 different
3348 // priority scales.
3349 //
3350 // ThreadPriorityPolicy=0
3351 // Since the Solaris' default priority is MaximumPriority, we do not
3352 // set a priority lower than Max unless a priority lower than
3353 // NormPriority is requested.
3354 //
3355 // ThreadPriorityPolicy=1
3356 // This mode causes the priority table to get filled with
3357 // linear values.  NormPriority get's mapped to 50% of the
3358 // Maximum priority an so on.  This will cause VM threads
3359 // to get unfair treatment against other Solaris processes
3360 // which do not explicitly alter their thread priorities.
3361 
3362 int os::java_to_os_priority[CriticalPriority + 1] = {
3363   -99999,         // 0 Entry should never be used
3364 
3365   0,              // 1 MinPriority
3366   32,             // 2
3367   64,             // 3
3368 
3369   96,             // 4
3370   127,            // 5 NormPriority
3371   127,            // 6
3372 
3373   127,            // 7
3374   127,            // 8
3375   127,            // 9 NearMaxPriority
3376 
3377   127,            // 10 MaxPriority
3378 
3379   -criticalPrio   // 11 CriticalPriority
3380 };
3381 
3382 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3383   OSThread* osthread = thread->osthread();
3384 
3385   // Save requested priority in case the thread hasn't been started
3386   osthread->set_native_priority(newpri);
3387 
3388   // Check for critical priority request
3389   bool fxcritical = false;
3390   if (newpri == -criticalPrio) {
3391     fxcritical = true;
3392     newpri = criticalPrio;
3393   }
3394 
3395   assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
3396   if (!UseThreadPriorities) return OS_OK;
3397 
3398   int status = 0;
3399 
3400   if (!fxcritical) {
3401     // Use thr_setprio only if we have a priority that thr_setprio understands
3402     status = thr_setprio(thread->osthread()->thread_id(), newpri);
3403   }
3404 
3405   int lwp_status =
3406           set_lwp_class_and_priority(osthread->thread_id(),
3407                                      osthread->lwp_id(),
3408                                      newpri,
3409                                      fxcritical ? fxLimits.schedPolicy : myClass,
3410                                      !fxcritical);
3411   if (lwp_status != 0 && fxcritical) {
3412     // Try again, this time without changing the scheduling class
3413     newpri = java_MaxPriority_to_os_priority;
3414     lwp_status = set_lwp_class_and_priority(osthread->thread_id(),
3415                                             osthread->lwp_id(),
3416                                             newpri, myClass, false);
3417   }
3418   status |= lwp_status;
3419   return (status == 0) ? OS_OK : OS_ERR;
3420 }
3421 
3422 
3423 OSReturn os::get_native_priority(const Thread* const thread,
3424                                  int *priority_ptr) {
3425   int p;
3426   if (!UseThreadPriorities) {
3427     *priority_ptr = NormalPriority;
3428     return OS_OK;
3429   }
3430   int status = thr_getprio(thread->osthread()->thread_id(), &p);
3431   if (status != 0) {
3432     return OS_ERR;
3433   }
3434   *priority_ptr = p;
3435   return OS_OK;
3436 }
3437 
3438 
3439 // Hint to the underlying OS that a task switch would not be good.
3440 // Void return because it's a hint and can fail.
3441 void os::hint_no_preempt() {
3442   schedctl_start(schedctl_init());
3443 }
3444 
3445 ////////////////////////////////////////////////////////////////////////////////
3446 // suspend/resume support
3447 
3448 //  The low-level signal-based suspend/resume support is a remnant from the
3449 //  old VM-suspension that used to be for java-suspension, safepoints etc,
3450 //  within hotspot. Currently used by JFR's OSThreadSampler
3451 //
3452 //  The remaining code is greatly simplified from the more general suspension
3453 //  code that used to be used.
3454 //
3455 //  The protocol is quite simple:
3456 //  - suspend:
3457 //      - sends a signal to the target thread
3458 //      - polls the suspend state of the osthread using a yield loop
3459 //      - target thread signal handler (SR_handler) sets suspend state
3460 //        and blocks in sigsuspend until continued
3461 //  - resume:
3462 //      - sets target osthread state to continue
3463 //      - sends signal to end the sigsuspend loop in the SR_handler
3464 //
3465 //  Note that the SR_lock plays no role in this suspend/resume protocol,
3466 //  but is checked for NULL in SR_handler as a thread termination indicator.
3467 //  The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
3468 //
3469 //  Note that resume_clear_context() and suspend_save_context() are needed
3470 //  by SR_handler(), so that fetch_frame_from_ucontext() works,
3471 //  which in part is used by:
3472 //    - Forte Analyzer: AsyncGetCallTrace()
3473 //    - StackBanging: get_frame_at_stack_banging_point()
3474 //    - JFR: get_topframe()-->....-->get_valid_uc_in_signal_handler()
3475 
3476 static void resume_clear_context(OSThread *osthread) {
3477   osthread->set_ucontext(NULL);
3478 }
3479 
3480 static void suspend_save_context(OSThread *osthread, ucontext_t* context) {
3481   osthread->set_ucontext(context);
3482 }
3483 
3484 static PosixSemaphore sr_semaphore;
3485 
3486 void os::Solaris::SR_handler(Thread* thread, ucontext_t* context) {
3487   // Save and restore errno to avoid confusing native code with EINTR
3488   // after sigsuspend.
3489   int old_errno = errno;
3490 
3491   OSThread* osthread = thread->osthread();
3492   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3493 
3494   os::SuspendResume::State current = osthread->sr.state();
3495   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3496     suspend_save_context(osthread, context);
3497 
3498     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3499     os::SuspendResume::State state = osthread->sr.suspended();
3500     if (state == os::SuspendResume::SR_SUSPENDED) {
3501       sigset_t suspend_set;  // signals for sigsuspend()
3502 
3503       // get current set of blocked signals and unblock resume signal
3504       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3505       sigdelset(&suspend_set, ASYNC_SIGNAL);
3506 
3507       sr_semaphore.signal();
3508       // wait here until we are resumed
3509       while (1) {
3510         sigsuspend(&suspend_set);
3511 
3512         os::SuspendResume::State result = osthread->sr.running();
3513         if (result == os::SuspendResume::SR_RUNNING) {
3514           sr_semaphore.signal();
3515           break;
3516         }
3517       }
3518 
3519     } else if (state == os::SuspendResume::SR_RUNNING) {
3520       // request was cancelled, continue
3521     } else {
3522       ShouldNotReachHere();
3523     }
3524 
3525     resume_clear_context(osthread);
3526   } else if (current == os::SuspendResume::SR_RUNNING) {
3527     // request was cancelled, continue
3528   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3529     // ignore
3530   } else {
3531     // ignore
3532   }
3533 
3534   errno = old_errno;
3535 }
3536 
3537 void os::print_statistics() {
3538 }
3539 
3540 bool os::message_box(const char* title, const char* message) {
3541   int i;
3542   fdStream err(defaultStream::error_fd());
3543   for (i = 0; i < 78; i++) err.print_raw("=");
3544   err.cr();
3545   err.print_raw_cr(title);
3546   for (i = 0; i < 78; i++) err.print_raw("-");
3547   err.cr();
3548   err.print_raw_cr(message);
3549   for (i = 0; i < 78; i++) err.print_raw("=");
3550   err.cr();
3551 
3552   char buf[16];
3553   // Prevent process from exiting upon "read error" without consuming all CPU
3554   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
3555 
3556   return buf[0] == 'y' || buf[0] == 'Y';
3557 }
3558 
3559 static int sr_notify(OSThread* osthread) {
3560   int status = thr_kill(osthread->thread_id(), ASYNC_SIGNAL);
3561   assert_status(status == 0, status, "thr_kill");
3562   return status;
3563 }
3564 
3565 // "Randomly" selected value for how long we want to spin
3566 // before bailing out on suspending a thread, also how often
3567 // we send a signal to a thread we want to resume
3568 static const int RANDOMLY_LARGE_INTEGER = 1000000;
3569 static const int RANDOMLY_LARGE_INTEGER2 = 100;
3570 
3571 static bool do_suspend(OSThread* osthread) {
3572   assert(osthread->sr.is_running(), "thread should be running");
3573   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
3574 
3575   // mark as suspended and send signal
3576   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
3577     // failed to switch, state wasn't running?
3578     ShouldNotReachHere();
3579     return false;
3580   }
3581 
3582   if (sr_notify(osthread) != 0) {
3583     ShouldNotReachHere();
3584   }
3585 
3586   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
3587   while (true) {
3588     if (sr_semaphore.timedwait(0, 2000 * NANOSECS_PER_MILLISEC)) {
3589       break;
3590     } else {
3591       // timeout
3592       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
3593       if (cancelled == os::SuspendResume::SR_RUNNING) {
3594         return false;
3595       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
3596         // make sure that we consume the signal on the semaphore as well
3597         sr_semaphore.wait();
3598         break;
3599       } else {
3600         ShouldNotReachHere();
3601         return false;
3602       }
3603     }
3604   }
3605 
3606   guarantee(osthread->sr.is_suspended(), "Must be suspended");
3607   return true;
3608 }
3609 
3610 static void do_resume(OSThread* osthread) {
3611   assert(osthread->sr.is_suspended(), "thread should be suspended");
3612   assert(!sr_semaphore.trywait(), "invalid semaphore state");
3613 
3614   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
3615     // failed to switch to WAKEUP_REQUEST
3616     ShouldNotReachHere();
3617     return;
3618   }
3619 
3620   while (true) {
3621     if (sr_notify(osthread) == 0) {
3622       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
3623         if (osthread->sr.is_running()) {
3624           return;
3625         }
3626       }
3627     } else {
3628       ShouldNotReachHere();
3629     }
3630   }
3631 
3632   guarantee(osthread->sr.is_running(), "Must be running!");
3633 }
3634 
3635 void os::SuspendedThreadTask::internal_do_task() {
3636   if (do_suspend(_thread->osthread())) {
3637     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
3638     do_task(context);
3639     do_resume(_thread->osthread());
3640   }
3641 }
3642 
3643 // This does not do anything on Solaris. This is basically a hook for being
3644 // able to use structured exception handling (thread-local exception filters) on, e.g., Win32.
3645 void os::os_exception_wrapper(java_call_t f, JavaValue* value,
3646                               const methodHandle& method, JavaCallArguments* args,
3647                               Thread* thread) {
3648   f(value, method, args, thread);
3649 }
3650 
3651 // This routine may be used by user applications as a "hook" to catch signals.
3652 // The user-defined signal handler must pass unrecognized signals to this
3653 // routine, and if it returns true (non-zero), then the signal handler must
3654 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
3655 // routine will never retun false (zero), but instead will execute a VM panic
3656 // routine kill the process.
3657 //
3658 // If this routine returns false, it is OK to call it again.  This allows
3659 // the user-defined signal handler to perform checks either before or after
3660 // the VM performs its own checks.  Naturally, the user code would be making
3661 // a serious error if it tried to handle an exception (such as a null check
3662 // or breakpoint) that the VM was generating for its own correct operation.
3663 //
3664 // This routine may recognize any of the following kinds of signals:
3665 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, BREAK_SIGNAL, SIGPIPE, SIGXFSZ,
3666 // ASYNC_SIGNAL.
3667 // It should be consulted by handlers for any of those signals.
3668 //
3669 // The caller of this routine must pass in the three arguments supplied
3670 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3671 // field of the structure passed to sigaction().  This routine assumes that
3672 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3673 //
3674 // Note that the VM will print warnings if it detects conflicting signal
3675 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3676 //
3677 extern "C" JNIEXPORT int JVM_handle_solaris_signal(int signo,
3678                                                    siginfo_t* siginfo,
3679                                                    void* ucontext,
3680                                                    int abort_if_unrecognized);
3681 
3682 
3683 void signalHandler(int sig, siginfo_t* info, void* ucVoid) {
3684   int orig_errno = errno;  // Preserve errno value over signal handler.
3685   JVM_handle_solaris_signal(sig, info, ucVoid, true);
3686   errno = orig_errno;
3687 }
3688 
3689 // This boolean allows users to forward their own non-matching signals
3690 // to JVM_handle_solaris_signal, harmlessly.
3691 bool os::Solaris::signal_handlers_are_installed = false;
3692 
3693 // For signal-chaining
3694 bool os::Solaris::libjsig_is_loaded = false;
3695 typedef struct sigaction *(*get_signal_t)(int);
3696 get_signal_t os::Solaris::get_signal_action = NULL;
3697 
3698 struct sigaction* os::Solaris::get_chained_signal_action(int sig) {
3699   struct sigaction *actp = NULL;
3700 
3701   if ((libjsig_is_loaded)  && (sig <= Maxsignum)) {
3702     // Retrieve the old signal handler from libjsig
3703     actp = (*get_signal_action)(sig);
3704   }
3705   if (actp == NULL) {
3706     // Retrieve the preinstalled signal handler from jvm
3707     actp = get_preinstalled_handler(sig);
3708   }
3709 
3710   return actp;
3711 }
3712 
3713 static bool call_chained_handler(struct sigaction *actp, int sig,
3714                                  siginfo_t *siginfo, void *context) {
3715   // Call the old signal handler
3716   if (actp->sa_handler == SIG_DFL) {
3717     // It's more reasonable to let jvm treat it as an unexpected exception
3718     // instead of taking the default action.
3719     return false;
3720   } else if (actp->sa_handler != SIG_IGN) {
3721     if ((actp->sa_flags & SA_NODEFER) == 0) {
3722       // automaticlly block the signal
3723       sigaddset(&(actp->sa_mask), sig);
3724     }
3725 
3726     sa_handler_t hand;
3727     sa_sigaction_t sa;
3728     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3729     // retrieve the chained handler
3730     if (siginfo_flag_set) {
3731       sa = actp->sa_sigaction;
3732     } else {
3733       hand = actp->sa_handler;
3734     }
3735 
3736     if ((actp->sa_flags & SA_RESETHAND) != 0) {
3737       actp->sa_handler = SIG_DFL;
3738     }
3739 
3740     // try to honor the signal mask
3741     sigset_t oset;
3742     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3743 
3744     // call into the chained handler
3745     if (siginfo_flag_set) {
3746       (*sa)(sig, siginfo, context);
3747     } else {
3748       (*hand)(sig);
3749     }
3750 
3751     // restore the signal mask
3752     pthread_sigmask(SIG_SETMASK, &oset, 0);
3753   }
3754   // Tell jvm's signal handler the signal is taken care of.
3755   return true;
3756 }
3757 
3758 bool os::Solaris::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3759   bool chained = false;
3760   // signal-chaining
3761   if (UseSignalChaining) {
3762     struct sigaction *actp = get_chained_signal_action(sig);
3763     if (actp != NULL) {
3764       chained = call_chained_handler(actp, sig, siginfo, context);
3765     }
3766   }
3767   return chained;
3768 }
3769 
3770 struct sigaction* os::Solaris::get_preinstalled_handler(int sig) {
3771   assert((chainedsigactions != (struct sigaction *)NULL) &&
3772          (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3773   if (preinstalled_sigs[sig] != 0) {
3774     return &chainedsigactions[sig];
3775   }
3776   return NULL;
3777 }
3778 
3779 void os::Solaris::save_preinstalled_handler(int sig,
3780                                             struct sigaction& oldAct) {
3781   assert(sig > 0 && sig <= Maxsignum, "vm signal out of expected range");
3782   assert((chainedsigactions != (struct sigaction *)NULL) &&
3783          (preinstalled_sigs != (int *)NULL), "signals not yet initialized");
3784   chainedsigactions[sig] = oldAct;
3785   preinstalled_sigs[sig] = 1;
3786 }
3787 
3788 void os::Solaris::set_signal_handler(int sig, bool set_installed,
3789                                      bool oktochain) {
3790   // Check for overwrite.
3791   struct sigaction oldAct;
3792   sigaction(sig, (struct sigaction*)NULL, &oldAct);
3793   void* oldhand =
3794       oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
3795                           : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
3796   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3797       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3798       oldhand != CAST_FROM_FN_PTR(void*, signalHandler)) {
3799     if (AllowUserSignalHandlers || !set_installed) {
3800       // Do not overwrite; user takes responsibility to forward to us.
3801       return;
3802     } else if (UseSignalChaining) {
3803       if (oktochain) {
3804         // save the old handler in jvm
3805         save_preinstalled_handler(sig, oldAct);
3806       } else {
3807         vm_exit_during_initialization("Signal chaining not allowed for VM interrupt signal.");
3808       }
3809       // libjsig also interposes the sigaction() call below and saves the
3810       // old sigaction on it own.
3811     } else {
3812       fatal("Encountered unexpected pre-existing sigaction handler "
3813             "%#lx for signal %d.", (long)oldhand, sig);
3814     }
3815   }
3816 
3817   struct sigaction sigAct;
3818   sigfillset(&(sigAct.sa_mask));
3819   sigAct.sa_handler = SIG_DFL;
3820 
3821   sigAct.sa_sigaction = signalHandler;
3822   // Handle SIGSEGV on alternate signal stack if
3823   // not using stack banging
3824   if (!UseStackBanging && sig == SIGSEGV) {
3825     sigAct.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
3826   } else {
3827     sigAct.sa_flags = SA_SIGINFO | SA_RESTART;
3828   }
3829   os::Solaris::set_our_sigflags(sig, sigAct.sa_flags);
3830 
3831   sigaction(sig, &sigAct, &oldAct);
3832 
3833   void* oldhand2 = oldAct.sa_sigaction ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3834                                        : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3835   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3836 }
3837 
3838 
3839 #define DO_SIGNAL_CHECK(sig)                      \
3840   do {                                            \
3841     if (!sigismember(&check_signal_done, sig)) {  \
3842       os::Solaris::check_signal_handler(sig);     \
3843     }                                             \
3844   } while (0)
3845 
3846 // This method is a periodic task to check for misbehaving JNI applications
3847 // under CheckJNI, we can add any periodic checks here
3848 
3849 void os::run_periodic_checks() {
3850   // A big source of grief is hijacking virt. addr 0x0 on Solaris,
3851   // thereby preventing a NULL checks.
3852   if (!check_addr0_done) check_addr0_done = check_addr0(tty);
3853 
3854   if (check_signals == false) return;
3855 
3856   // SEGV and BUS if overridden could potentially prevent
3857   // generation of hs*.log in the event of a crash, debugging
3858   // such a case can be very challenging, so we absolutely
3859   // check for the following for a good measure:
3860   DO_SIGNAL_CHECK(SIGSEGV);
3861   DO_SIGNAL_CHECK(SIGILL);
3862   DO_SIGNAL_CHECK(SIGFPE);
3863   DO_SIGNAL_CHECK(SIGBUS);
3864   DO_SIGNAL_CHECK(SIGPIPE);
3865   DO_SIGNAL_CHECK(SIGXFSZ);
3866   DO_SIGNAL_CHECK(ASYNC_SIGNAL);
3867 
3868   // ReduceSignalUsage allows the user to override these handlers
3869   // see comments at the very top and jvm_solaris.h
3870   if (!ReduceSignalUsage) {
3871     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3872     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3873     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3874     DO_SIGNAL_CHECK(BREAK_SIGNAL);
3875   }
3876 }
3877 
3878 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3879 
3880 static os_sigaction_t os_sigaction = NULL;
3881 
3882 void os::Solaris::check_signal_handler(int sig) {
3883   char buf[O_BUFLEN];
3884   address jvmHandler = NULL;
3885 
3886   struct sigaction act;
3887   if (os_sigaction == NULL) {
3888     // only trust the default sigaction, in case it has been interposed
3889     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3890     if (os_sigaction == NULL) return;
3891   }
3892 
3893   os_sigaction(sig, (struct sigaction*)NULL, &act);
3894 
3895   address thisHandler = (act.sa_flags & SA_SIGINFO)
3896     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3897     : CAST_FROM_FN_PTR(address, act.sa_handler);
3898 
3899 
3900   switch (sig) {
3901   case SIGSEGV:
3902   case SIGBUS:
3903   case SIGFPE:
3904   case SIGPIPE:
3905   case SIGXFSZ:
3906   case SIGILL:
3907   case ASYNC_SIGNAL:
3908     jvmHandler = CAST_FROM_FN_PTR(address, signalHandler);
3909     break;
3910 
3911   case SHUTDOWN1_SIGNAL:
3912   case SHUTDOWN2_SIGNAL:
3913   case SHUTDOWN3_SIGNAL:
3914   case BREAK_SIGNAL:
3915     jvmHandler = (address)user_handler();
3916     break;
3917 
3918   default:
3919       return;
3920   }
3921 
3922   if (thisHandler != jvmHandler) {
3923     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
3924     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
3925     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
3926     // No need to check this sig any longer
3927     sigaddset(&check_signal_done, sig);
3928     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
3929     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
3930       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
3931                     exception_name(sig, buf, O_BUFLEN));
3932     }
3933   } else if(os::Solaris::get_our_sigflags(sig) != 0 && act.sa_flags != os::Solaris::get_our_sigflags(sig)) {
3934     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
3935     tty->print("expected:");
3936     os::Posix::print_sa_flags(tty, os::Solaris::get_our_sigflags(sig));
3937     tty->cr();
3938     tty->print("  found:");
3939     os::Posix::print_sa_flags(tty, act.sa_flags);
3940     tty->cr();
3941     // No need to check this sig any longer
3942     sigaddset(&check_signal_done, sig);
3943   }
3944 
3945   // Print all the signal handler state
3946   if (sigismember(&check_signal_done, sig)) {
3947     print_signal_handlers(tty, buf, O_BUFLEN);
3948   }
3949 
3950 }
3951 
3952 void os::Solaris::install_signal_handlers() {
3953   signal_handlers_are_installed = true;
3954 
3955   // signal-chaining
3956   typedef void (*signal_setting_t)();
3957   signal_setting_t begin_signal_setting = NULL;
3958   signal_setting_t end_signal_setting = NULL;
3959   begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3960                                         dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3961   if (begin_signal_setting != NULL) {
3962     end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3963                                         dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3964     get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3965                                        dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3966     get_libjsig_version = CAST_TO_FN_PTR(version_getting_t,
3967                                          dlsym(RTLD_DEFAULT, "JVM_get_libjsig_version"));
3968     libjsig_is_loaded = true;
3969     if (os::Solaris::get_libjsig_version != NULL) {
3970       int libjsigversion =  (*os::Solaris::get_libjsig_version)();
3971       assert(libjsigversion == JSIG_VERSION_1_4_1, "libjsig version mismatch");
3972     }
3973     assert(UseSignalChaining, "should enable signal-chaining");
3974   }
3975   if (libjsig_is_loaded) {
3976     // Tell libjsig jvm is setting signal handlers
3977     (*begin_signal_setting)();
3978   }
3979 
3980   set_signal_handler(SIGSEGV, true, true);
3981   set_signal_handler(SIGPIPE, true, true);
3982   set_signal_handler(SIGXFSZ, true, true);
3983   set_signal_handler(SIGBUS, true, true);
3984   set_signal_handler(SIGILL, true, true);
3985   set_signal_handler(SIGFPE, true, true);
3986   set_signal_handler(ASYNC_SIGNAL, true, true);
3987 
3988   if (libjsig_is_loaded) {
3989     // Tell libjsig jvm finishes setting signal handlers
3990     (*end_signal_setting)();
3991   }
3992 
3993   // We don't activate signal checker if libjsig is in place, we trust ourselves
3994   // and if UserSignalHandler is installed all bets are off.
3995   // Log that signal checking is off only if -verbose:jni is specified.
3996   if (CheckJNICalls) {
3997     if (libjsig_is_loaded) {
3998       if (PrintJNIResolving) {
3999         tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4000       }
4001       check_signals = false;
4002     }
4003     if (AllowUserSignalHandlers) {
4004       if (PrintJNIResolving) {
4005         tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4006       }
4007       check_signals = false;
4008     }
4009   }
4010 }
4011 
4012 
4013 void report_error(const char* file_name, int line_no, const char* title,
4014                   const char* format, ...);
4015 
4016 // (Static) wrappers for the liblgrp API
4017 os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home;
4018 os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init;
4019 os::Solaris::lgrp_fini_func_t os::Solaris::_lgrp_fini;
4020 os::Solaris::lgrp_root_func_t os::Solaris::_lgrp_root;
4021 os::Solaris::lgrp_children_func_t os::Solaris::_lgrp_children;
4022 os::Solaris::lgrp_resources_func_t os::Solaris::_lgrp_resources;
4023 os::Solaris::lgrp_nlgrps_func_t os::Solaris::_lgrp_nlgrps;
4024 os::Solaris::lgrp_cookie_stale_func_t os::Solaris::_lgrp_cookie_stale;
4025 os::Solaris::lgrp_cookie_t os::Solaris::_lgrp_cookie = 0;
4026 
4027 static address resolve_symbol_lazy(const char* name) {
4028   address addr = (address) dlsym(RTLD_DEFAULT, name);
4029   if (addr == NULL) {
4030     // RTLD_DEFAULT was not defined on some early versions of 2.5.1
4031     addr = (address) dlsym(RTLD_NEXT, name);
4032   }
4033   return addr;
4034 }
4035 
4036 static address resolve_symbol(const char* name) {
4037   address addr = resolve_symbol_lazy(name);
4038   if (addr == NULL) {
4039     fatal(dlerror());
4040   }
4041   return addr;
4042 }
4043 
4044 void os::Solaris::libthread_init() {
4045   address func = (address)dlsym(RTLD_DEFAULT, "_thr_suspend_allmutators");
4046 
4047   lwp_priocntl_init();
4048 
4049   // RTLD_DEFAULT was not defined on some early versions of 5.5.1
4050   if (func == NULL) {
4051     func = (address) dlsym(RTLD_NEXT, "_thr_suspend_allmutators");
4052     // Guarantee that this VM is running on an new enough OS (5.6 or
4053     // later) that it will have a new enough libthread.so.
4054     guarantee(func != NULL, "libthread.so is too old.");
4055   }
4056 
4057   int size;
4058   void (*handler_info_func)(address *, int *);
4059   handler_info_func = CAST_TO_FN_PTR(void (*)(address *, int *), resolve_symbol("thr_sighndlrinfo"));
4060   handler_info_func(&handler_start, &size);
4061   handler_end = handler_start + size;
4062 }
4063 
4064 
4065 int_fnP_mutex_tP os::Solaris::_mutex_lock;
4066 int_fnP_mutex_tP os::Solaris::_mutex_trylock;
4067 int_fnP_mutex_tP os::Solaris::_mutex_unlock;
4068 int_fnP_mutex_tP_i_vP os::Solaris::_mutex_init;
4069 int_fnP_mutex_tP os::Solaris::_mutex_destroy;
4070 int os::Solaris::_mutex_scope = USYNC_THREAD;
4071 
4072 int_fnP_cond_tP_mutex_tP_timestruc_tP os::Solaris::_cond_timedwait;
4073 int_fnP_cond_tP_mutex_tP os::Solaris::_cond_wait;
4074 int_fnP_cond_tP os::Solaris::_cond_signal;
4075 int_fnP_cond_tP os::Solaris::_cond_broadcast;
4076 int_fnP_cond_tP_i_vP os::Solaris::_cond_init;
4077 int_fnP_cond_tP os::Solaris::_cond_destroy;
4078 int os::Solaris::_cond_scope = USYNC_THREAD;
4079 bool os::Solaris::_synchronization_initialized;
4080 
4081 void os::Solaris::synchronization_init() {
4082   if (UseLWPSynchronization) {
4083     os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_lock")));
4084     os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_trylock")));
4085     os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("_lwp_mutex_unlock")));
4086     os::Solaris::set_mutex_init(lwp_mutex_init);
4087     os::Solaris::set_mutex_destroy(lwp_mutex_destroy);
4088     os::Solaris::set_mutex_scope(USYNC_THREAD);
4089 
4090     os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("_lwp_cond_timedwait")));
4091     os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("_lwp_cond_wait")));
4092     os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_signal")));
4093     os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("_lwp_cond_broadcast")));
4094     os::Solaris::set_cond_init(lwp_cond_init);
4095     os::Solaris::set_cond_destroy(lwp_cond_destroy);
4096     os::Solaris::set_cond_scope(USYNC_THREAD);
4097   } else {
4098     os::Solaris::set_mutex_scope(USYNC_THREAD);
4099     os::Solaris::set_cond_scope(USYNC_THREAD);
4100 
4101     if (UsePthreads) {
4102       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_lock")));
4103       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_trylock")));
4104       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_unlock")));
4105       os::Solaris::set_mutex_init(pthread_mutex_default_init);
4106       os::Solaris::set_mutex_destroy(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("pthread_mutex_destroy")));
4107 
4108       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("pthread_cond_timedwait")));
4109       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("pthread_cond_wait")));
4110       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_signal")));
4111       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_broadcast")));
4112       os::Solaris::set_cond_init(pthread_cond_default_init);
4113       os::Solaris::set_cond_destroy(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("pthread_cond_destroy")));
4114     } else {
4115       os::Solaris::set_mutex_lock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_lock")));
4116       os::Solaris::set_mutex_trylock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_trylock")));
4117       os::Solaris::set_mutex_unlock(CAST_TO_FN_PTR(int_fnP_mutex_tP, resolve_symbol("mutex_unlock")));
4118       os::Solaris::set_mutex_init(::mutex_init);
4119       os::Solaris::set_mutex_destroy(::mutex_destroy);
4120 
4121       os::Solaris::set_cond_timedwait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP_timestruc_tP, resolve_symbol("cond_timedwait")));
4122       os::Solaris::set_cond_wait(CAST_TO_FN_PTR(int_fnP_cond_tP_mutex_tP, resolve_symbol("cond_wait")));
4123       os::Solaris::set_cond_signal(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_signal")));
4124       os::Solaris::set_cond_broadcast(CAST_TO_FN_PTR(int_fnP_cond_tP, resolve_symbol("cond_broadcast")));
4125       os::Solaris::set_cond_init(::cond_init);
4126       os::Solaris::set_cond_destroy(::cond_destroy);
4127     }
4128   }
4129   _synchronization_initialized = true;
4130 }
4131 
4132 bool os::Solaris::liblgrp_init() {
4133   void *handle = dlopen("liblgrp.so.1", RTLD_LAZY);
4134   if (handle != NULL) {
4135     os::Solaris::set_lgrp_home(CAST_TO_FN_PTR(lgrp_home_func_t, dlsym(handle, "lgrp_home")));
4136     os::Solaris::set_lgrp_init(CAST_TO_FN_PTR(lgrp_init_func_t, dlsym(handle, "lgrp_init")));
4137     os::Solaris::set_lgrp_fini(CAST_TO_FN_PTR(lgrp_fini_func_t, dlsym(handle, "lgrp_fini")));
4138     os::Solaris::set_lgrp_root(CAST_TO_FN_PTR(lgrp_root_func_t, dlsym(handle, "lgrp_root")));
4139     os::Solaris::set_lgrp_children(CAST_TO_FN_PTR(lgrp_children_func_t, dlsym(handle, "lgrp_children")));
4140     os::Solaris::set_lgrp_resources(CAST_TO_FN_PTR(lgrp_resources_func_t, dlsym(handle, "lgrp_resources")));
4141     os::Solaris::set_lgrp_nlgrps(CAST_TO_FN_PTR(lgrp_nlgrps_func_t, dlsym(handle, "lgrp_nlgrps")));
4142     os::Solaris::set_lgrp_cookie_stale(CAST_TO_FN_PTR(lgrp_cookie_stale_func_t,
4143                                                       dlsym(handle, "lgrp_cookie_stale")));
4144 
4145     lgrp_cookie_t c = lgrp_init(LGRP_VIEW_CALLER);
4146     set_lgrp_cookie(c);
4147     return true;
4148   }
4149   return false;
4150 }
4151 
4152 // int pset_getloadavg(psetid_t pset, double loadavg[], int nelem);
4153 typedef long (*pset_getloadavg_type)(psetid_t pset, double loadavg[], int nelem);
4154 static pset_getloadavg_type pset_getloadavg_ptr = NULL;
4155 
4156 void init_pset_getloadavg_ptr(void) {
4157   pset_getloadavg_ptr =
4158     (pset_getloadavg_type)dlsym(RTLD_DEFAULT, "pset_getloadavg");
4159   if (pset_getloadavg_ptr == NULL) {
4160     log_warning(os)("pset_getloadavg function not found");
4161   }
4162 }
4163 
4164 int os::Solaris::_dev_zero_fd = -1;
4165 
4166 // this is called _before_ the global arguments have been parsed
4167 void os::init(void) {
4168   _initial_pid = getpid();
4169 
4170   max_hrtime = first_hrtime = gethrtime();
4171 
4172   init_random(1234567);
4173 
4174   page_size = sysconf(_SC_PAGESIZE);
4175   if (page_size == -1) {
4176     fatal("os_solaris.cpp: os::init: sysconf failed (%s)", os::strerror(errno));
4177   }
4178   init_page_sizes((size_t) page_size);
4179 
4180   Solaris::initialize_system_info();
4181 
4182   int fd = ::open("/dev/zero", O_RDWR);
4183   if (fd < 0) {
4184     fatal("os::init: cannot open /dev/zero (%s)", os::strerror(errno));
4185   } else {
4186     Solaris::set_dev_zero_fd(fd);
4187 
4188     // Close on exec, child won't inherit.
4189     fcntl(fd, F_SETFD, FD_CLOEXEC);
4190   }
4191 
4192   clock_tics_per_sec = CLK_TCK;
4193 
4194   // check if dladdr1() exists; dladdr1 can provide more information than
4195   // dladdr for os::dll_address_to_function_name. It comes with SunOS 5.9
4196   // and is available on linker patches for 5.7 and 5.8.
4197   // libdl.so must have been loaded, this call is just an entry lookup
4198   void * hdl = dlopen("libdl.so", RTLD_NOW);
4199   if (hdl) {
4200     dladdr1_func = CAST_TO_FN_PTR(dladdr1_func_type, dlsym(hdl, "dladdr1"));
4201   }
4202 
4203   main_thread = thr_self();
4204 
4205   // dynamic lookup of functions that may not be available in our lowest
4206   // supported Solaris release
4207   void * handle = dlopen("libc.so.1", RTLD_LAZY);
4208   if (handle != NULL) {
4209     Solaris::_pthread_setname_np =  // from 11.3
4210         (Solaris::pthread_setname_np_func_t)dlsym(handle, "pthread_setname_np");
4211   }
4212 }
4213 
4214 // To install functions for atexit system call
4215 extern "C" {
4216   static void perfMemory_exit_helper() {
4217     perfMemory_exit();
4218   }
4219 }
4220 
4221 // this is called _after_ the global arguments have been parsed
4222 jint os::init_2(void) {
4223   // try to enable extended file IO ASAP, see 6431278
4224   os::Solaris::try_enable_extended_io();
4225 
4226   // Check and sets minimum stack sizes against command line options
4227   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4228     return JNI_ERR;
4229   }
4230 
4231   Solaris::libthread_init();
4232 
4233   if (UseNUMA) {
4234     if (!Solaris::liblgrp_init()) {
4235       UseNUMA = false;
4236     } else {
4237       size_t lgrp_limit = os::numa_get_groups_num();
4238       int *lgrp_ids = NEW_C_HEAP_ARRAY(int, lgrp_limit, mtInternal);
4239       size_t lgrp_num = os::numa_get_leaf_groups(lgrp_ids, lgrp_limit);
4240       FREE_C_HEAP_ARRAY(int, lgrp_ids);
4241       if (lgrp_num < 2) {
4242         // There's only one locality group, disable NUMA.
4243         UseNUMA = false;
4244       }
4245     }
4246     if (!UseNUMA && ForceNUMA) {
4247       UseNUMA = true;
4248     }
4249   }
4250 
4251   Solaris::signal_sets_init();
4252   Solaris::init_signal_mem();
4253   Solaris::install_signal_handlers();
4254 
4255   // initialize synchronization primitives to use either thread or
4256   // lwp synchronization (controlled by UseLWPSynchronization)
4257   Solaris::synchronization_init();
4258 
4259   if (MaxFDLimit) {
4260     // set the number of file descriptors to max. print out error
4261     // if getrlimit/setrlimit fails but continue regardless.
4262     struct rlimit nbr_files;
4263     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4264     if (status != 0) {
4265       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4266     } else {
4267       nbr_files.rlim_cur = nbr_files.rlim_max;
4268       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4269       if (status != 0) {
4270         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4271       }
4272     }
4273   }
4274 
4275   // Calculate theoretical max. size of Threads to guard gainst
4276   // artifical out-of-memory situations, where all available address-
4277   // space has been reserved by thread stacks. Default stack size is 1Mb.
4278   size_t pre_thread_stack_size = (JavaThread::stack_size_at_create()) ?
4279     JavaThread::stack_size_at_create() : (1*K*K);
4280   assert(pre_thread_stack_size != 0, "Must have a stack");
4281   // Solaris has a maximum of 4Gb of user programs. Calculate the thread limit when
4282   // we should start doing Virtual Memory banging. Currently when the threads will
4283   // have used all but 200Mb of space.
4284   size_t max_address_space = ((unsigned int)4 * K * K * K) - (200 * K * K);
4285   Solaris::_os_thread_limit = max_address_space / pre_thread_stack_size;
4286 
4287   // at-exit methods are called in the reverse order of their registration.
4288   // In Solaris 7 and earlier, atexit functions are called on return from
4289   // main or as a result of a call to exit(3C). There can be only 32 of
4290   // these functions registered and atexit() does not set errno. In Solaris
4291   // 8 and later, there is no limit to the number of functions registered
4292   // and atexit() sets errno. In addition, in Solaris 8 and later, atexit
4293   // functions are called upon dlclose(3DL) in addition to return from main
4294   // and exit(3C).
4295 
4296   if (PerfAllowAtExitRegistration) {
4297     // only register atexit functions if PerfAllowAtExitRegistration is set.
4298     // atexit functions can be delayed until process exit time, which
4299     // can be problematic for embedded VM situations. Embedded VMs should
4300     // call DestroyJavaVM() to assure that VM resources are released.
4301 
4302     // note: perfMemory_exit_helper atexit function may be removed in
4303     // the future if the appropriate cleanup code can be added to the
4304     // VM_Exit VMOperation's doit method.
4305     if (atexit(perfMemory_exit_helper) != 0) {
4306       warning("os::init2 atexit(perfMemory_exit_helper) failed");
4307     }
4308   }
4309 
4310   // Init pset_loadavg function pointer
4311   init_pset_getloadavg_ptr();
4312 
4313   return JNI_OK;
4314 }
4315 
4316 // Mark the polling page as unreadable
4317 void os::make_polling_page_unreadable(void) {
4318   if (mprotect((char *)_polling_page, page_size, PROT_NONE) != 0) {
4319     fatal("Could not disable polling page");
4320   }
4321 }
4322 
4323 // Mark the polling page as readable
4324 void os::make_polling_page_readable(void) {
4325   if (mprotect((char *)_polling_page, page_size, PROT_READ) != 0) {
4326     fatal("Could not enable polling page");
4327   }
4328 }
4329 
4330 // Is a (classpath) directory empty?
4331 bool os::dir_is_empty(const char* path) {
4332   DIR *dir = NULL;
4333   struct dirent *ptr;
4334 
4335   dir = opendir(path);
4336   if (dir == NULL) return true;
4337 
4338   // Scan the directory
4339   bool result = true;
4340   char buf[sizeof(struct dirent) + MAX_PATH];
4341   struct dirent *dbuf = (struct dirent *) buf;
4342   while (result && (ptr = readdir(dir, dbuf)) != NULL) {
4343     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4344       result = false;
4345     }
4346   }
4347   closedir(dir);
4348   return result;
4349 }
4350 
4351 // This code originates from JDK's sysOpen and open64_w
4352 // from src/solaris/hpi/src/system_md.c
4353 
4354 int os::open(const char *path, int oflag, int mode) {
4355   if (strlen(path) > MAX_PATH - 1) {
4356     errno = ENAMETOOLONG;
4357     return -1;
4358   }
4359   int fd;
4360 
4361   fd = ::open64(path, oflag, mode);
4362   if (fd == -1) return -1;
4363 
4364   // If the open succeeded, the file might still be a directory
4365   {
4366     struct stat64 buf64;
4367     int ret = ::fstat64(fd, &buf64);
4368     int st_mode = buf64.st_mode;
4369 
4370     if (ret != -1) {
4371       if ((st_mode & S_IFMT) == S_IFDIR) {
4372         errno = EISDIR;
4373         ::close(fd);
4374         return -1;
4375       }
4376     } else {
4377       ::close(fd);
4378       return -1;
4379     }
4380   }
4381 
4382   // 32-bit Solaris systems suffer from:
4383   //
4384   // - an historical default soft limit of 256 per-process file
4385   //   descriptors that is too low for many Java programs.
4386   //
4387   // - a design flaw where file descriptors created using stdio
4388   //   fopen must be less than 256, _even_ when the first limit above
4389   //   has been raised.  This can cause calls to fopen (but not calls to
4390   //   open, for example) to fail mysteriously, perhaps in 3rd party
4391   //   native code (although the JDK itself uses fopen).  One can hardly
4392   //   criticize them for using this most standard of all functions.
4393   //
4394   // We attempt to make everything work anyways by:
4395   //
4396   // - raising the soft limit on per-process file descriptors beyond
4397   //   256
4398   //
4399   // - As of Solaris 10u4, we can request that Solaris raise the 256
4400   //   stdio fopen limit by calling function enable_extended_FILE_stdio.
4401   //   This is done in init_2 and recorded in enabled_extended_FILE_stdio
4402   //
4403   // - If we are stuck on an old (pre 10u4) Solaris system, we can
4404   //   workaround the bug by remapping non-stdio file descriptors below
4405   //   256 to ones beyond 256, which is done below.
4406   //
4407   // See:
4408   // 1085341: 32-bit stdio routines should support file descriptors >255
4409   // 6533291: Work around 32-bit Solaris stdio limit of 256 open files
4410   // 6431278: Netbeans crash on 32 bit Solaris: need to call
4411   //          enable_extended_FILE_stdio() in VM initialisation
4412   // Giri Mandalika's blog
4413   // http://technopark02.blogspot.com/2005_05_01_archive.html
4414   //
4415 #ifndef  _LP64
4416   if ((!enabled_extended_FILE_stdio) && fd < 256) {
4417     int newfd = ::fcntl(fd, F_DUPFD, 256);
4418     if (newfd != -1) {
4419       ::close(fd);
4420       fd = newfd;
4421     }
4422   }
4423 #endif // 32-bit Solaris
4424 
4425   // All file descriptors that are opened in the JVM and not
4426   // specifically destined for a subprocess should have the
4427   // close-on-exec flag set.  If we don't set it, then careless 3rd
4428   // party native code might fork and exec without closing all
4429   // appropriate file descriptors (e.g. as we do in closeDescriptors in
4430   // UNIXProcess.c), and this in turn might:
4431   //
4432   // - cause end-of-file to fail to be detected on some file
4433   //   descriptors, resulting in mysterious hangs, or
4434   //
4435   // - might cause an fopen in the subprocess to fail on a system
4436   //   suffering from bug 1085341.
4437   //
4438   // (Yes, the default setting of the close-on-exec flag is a Unix
4439   // design flaw)
4440   //
4441   // See:
4442   // 1085341: 32-bit stdio routines should support file descriptors >255
4443   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4444   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4445   //
4446 #ifdef FD_CLOEXEC
4447   {
4448     int flags = ::fcntl(fd, F_GETFD);
4449     if (flags != -1) {
4450       ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4451     }
4452   }
4453 #endif
4454 
4455   return fd;
4456 }
4457 
4458 // create binary file, rewriting existing file if required
4459 int os::create_binary_file(const char* path, bool rewrite_existing) {
4460   int oflags = O_WRONLY | O_CREAT;
4461   if (!rewrite_existing) {
4462     oflags |= O_EXCL;
4463   }
4464   return ::open64(path, oflags, S_IREAD | S_IWRITE);
4465 }
4466 
4467 // return current position of file pointer
4468 jlong os::current_file_offset(int fd) {
4469   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4470 }
4471 
4472 // move file pointer to the specified offset
4473 jlong os::seek_to_file_offset(int fd, jlong offset) {
4474   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4475 }
4476 
4477 jlong os::lseek(int fd, jlong offset, int whence) {
4478   return (jlong) ::lseek64(fd, offset, whence);
4479 }
4480 
4481 char * os::native_path(char *path) {
4482   return path;
4483 }
4484 
4485 int os::ftruncate(int fd, jlong length) {
4486   return ::ftruncate64(fd, length);
4487 }
4488 
4489 int os::fsync(int fd)  {
4490   RESTARTABLE_RETURN_INT(::fsync(fd));
4491 }
4492 
4493 int os::available(int fd, jlong *bytes) {
4494   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
4495          "Assumed _thread_in_native");
4496   jlong cur, end;
4497   int mode;
4498   struct stat64 buf64;
4499 
4500   if (::fstat64(fd, &buf64) >= 0) {
4501     mode = buf64.st_mode;
4502     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4503       int n,ioctl_return;
4504 
4505       RESTARTABLE(::ioctl(fd, FIONREAD, &n), ioctl_return);
4506       if (ioctl_return>= 0) {
4507         *bytes = n;
4508         return 1;
4509       }
4510     }
4511   }
4512   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4513     return 0;
4514   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4515     return 0;
4516   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4517     return 0;
4518   }
4519   *bytes = end - cur;
4520   return 1;
4521 }
4522 
4523 // Map a block of memory.
4524 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4525                         char *addr, size_t bytes, bool read_only,
4526                         bool allow_exec) {
4527   int prot;
4528   int flags;
4529 
4530   if (read_only) {
4531     prot = PROT_READ;
4532     flags = MAP_SHARED;
4533   } else {
4534     prot = PROT_READ | PROT_WRITE;
4535     flags = MAP_PRIVATE;
4536   }
4537 
4538   if (allow_exec) {
4539     prot |= PROT_EXEC;
4540   }
4541 
4542   if (addr != NULL) {
4543     flags |= MAP_FIXED;
4544   }
4545 
4546   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4547                                      fd, file_offset);
4548   if (mapped_address == MAP_FAILED) {
4549     return NULL;
4550   }
4551   return mapped_address;
4552 }
4553 
4554 
4555 // Remap a block of memory.
4556 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4557                           char *addr, size_t bytes, bool read_only,
4558                           bool allow_exec) {
4559   // same as map_memory() on this OS
4560   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4561                         allow_exec);
4562 }
4563 
4564 
4565 // Unmap a block of memory.
4566 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4567   return munmap(addr, bytes) == 0;
4568 }
4569 
4570 void os::pause() {
4571   char filename[MAX_PATH];
4572   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4573     jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4574   } else {
4575     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4576   }
4577 
4578   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4579   if (fd != -1) {
4580     struct stat buf;
4581     ::close(fd);
4582     while (::stat(filename, &buf) == 0) {
4583       (void)::poll(NULL, 0, 100);
4584     }
4585   } else {
4586     jio_fprintf(stderr,
4587                 "Could not open pause file '%s', continuing immediately.\n", filename);
4588   }
4589 }
4590 
4591 #ifndef PRODUCT
4592 #ifdef INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
4593 // Turn this on if you need to trace synch operations.
4594 // Set RECORD_SYNCH_LIMIT to a large-enough value,
4595 // and call record_synch_enable and record_synch_disable
4596 // around the computation of interest.
4597 
4598 void record_synch(char* name, bool returning);  // defined below
4599 
4600 class RecordSynch {
4601   char* _name;
4602  public:
4603   RecordSynch(char* name) :_name(name) { record_synch(_name, false); }
4604   ~RecordSynch()                       { record_synch(_name, true); }
4605 };
4606 
4607 #define CHECK_SYNCH_OP(ret, name, params, args, inner)          \
4608 extern "C" ret name params {                                    \
4609   typedef ret name##_t params;                                  \
4610   static name##_t* implem = NULL;                               \
4611   static int callcount = 0;                                     \
4612   if (implem == NULL) {                                         \
4613     implem = (name##_t*) dlsym(RTLD_NEXT, #name);               \
4614     if (implem == NULL)  fatal(dlerror());                      \
4615   }                                                             \
4616   ++callcount;                                                  \
4617   RecordSynch _rs(#name);                                       \
4618   inner;                                                        \
4619   return implem args;                                           \
4620 }
4621 // in dbx, examine callcounts this way:
4622 // for n in $(eval whereis callcount | awk '{print $2}'); do print $n; done
4623 
4624 #define CHECK_POINTER_OK(p) \
4625   (!Universe::is_fully_initialized() || !Universe::is_reserved_heap((oop)(p)))
4626 #define CHECK_MU \
4627   if (!CHECK_POINTER_OK(mu)) fatal("Mutex must be in C heap only.");
4628 #define CHECK_CV \
4629   if (!CHECK_POINTER_OK(cv)) fatal("Condvar must be in C heap only.");
4630 #define CHECK_P(p) \
4631   if (!CHECK_POINTER_OK(p))  fatal(false,  "Pointer must be in C heap only.");
4632 
4633 #define CHECK_MUTEX(mutex_op) \
4634   CHECK_SYNCH_OP(int, mutex_op, (mutex_t *mu), (mu), CHECK_MU);
4635 
4636 CHECK_MUTEX(   mutex_lock)
4637 CHECK_MUTEX(  _mutex_lock)
4638 CHECK_MUTEX( mutex_unlock)
4639 CHECK_MUTEX(_mutex_unlock)
4640 CHECK_MUTEX( mutex_trylock)
4641 CHECK_MUTEX(_mutex_trylock)
4642 
4643 #define CHECK_COND(cond_op) \
4644   CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu), (cv, mu), CHECK_MU; CHECK_CV);
4645 
4646 CHECK_COND( cond_wait);
4647 CHECK_COND(_cond_wait);
4648 CHECK_COND(_cond_wait_cancel);
4649 
4650 #define CHECK_COND2(cond_op) \
4651   CHECK_SYNCH_OP(int, cond_op, (cond_t *cv, mutex_t *mu, timestruc_t* ts), (cv, mu, ts), CHECK_MU; CHECK_CV);
4652 
4653 CHECK_COND2( cond_timedwait);
4654 CHECK_COND2(_cond_timedwait);
4655 CHECK_COND2(_cond_timedwait_cancel);
4656 
4657 // do the _lwp_* versions too
4658 #define mutex_t lwp_mutex_t
4659 #define cond_t  lwp_cond_t
4660 CHECK_MUTEX(  _lwp_mutex_lock)
4661 CHECK_MUTEX(  _lwp_mutex_unlock)
4662 CHECK_MUTEX(  _lwp_mutex_trylock)
4663 CHECK_MUTEX( __lwp_mutex_lock)
4664 CHECK_MUTEX( __lwp_mutex_unlock)
4665 CHECK_MUTEX( __lwp_mutex_trylock)
4666 CHECK_MUTEX(___lwp_mutex_lock)
4667 CHECK_MUTEX(___lwp_mutex_unlock)
4668 
4669 CHECK_COND(  _lwp_cond_wait);
4670 CHECK_COND( __lwp_cond_wait);
4671 CHECK_COND(___lwp_cond_wait);
4672 
4673 CHECK_COND2(  _lwp_cond_timedwait);
4674 CHECK_COND2( __lwp_cond_timedwait);
4675 #undef mutex_t
4676 #undef cond_t
4677 
4678 CHECK_SYNCH_OP(int, _lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4679 CHECK_SYNCH_OP(int,__lwp_suspend2,       (int lwp, int *n), (lwp, n), 0);
4680 CHECK_SYNCH_OP(int, _lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4681 CHECK_SYNCH_OP(int,__lwp_kill,           (int lwp, int n),  (lwp, n), 0);
4682 CHECK_SYNCH_OP(int, _lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4683 CHECK_SYNCH_OP(int,__lwp_sema_wait,      (lwp_sema_t* p),   (p),  CHECK_P(p));
4684 CHECK_SYNCH_OP(int, _lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4685 CHECK_SYNCH_OP(int,__lwp_cond_broadcast, (lwp_cond_t* cv),  (cv), CHECK_CV);
4686 
4687 
4688 // recording machinery:
4689 
4690 enum { RECORD_SYNCH_LIMIT = 200 };
4691 char* record_synch_name[RECORD_SYNCH_LIMIT];
4692 void* record_synch_arg0ptr[RECORD_SYNCH_LIMIT];
4693 bool record_synch_returning[RECORD_SYNCH_LIMIT];
4694 thread_t record_synch_thread[RECORD_SYNCH_LIMIT];
4695 int record_synch_count = 0;
4696 bool record_synch_enabled = false;
4697 
4698 // in dbx, examine recorded data this way:
4699 // for n in name arg0ptr returning thread; do print record_synch_$n[0..record_synch_count-1]; done
4700 
4701 void record_synch(char* name, bool returning) {
4702   if (record_synch_enabled) {
4703     if (record_synch_count < RECORD_SYNCH_LIMIT) {
4704       record_synch_name[record_synch_count] = name;
4705       record_synch_returning[record_synch_count] = returning;
4706       record_synch_thread[record_synch_count] = thr_self();
4707       record_synch_arg0ptr[record_synch_count] = &name;
4708       record_synch_count++;
4709     }
4710     // put more checking code here:
4711     // ...
4712   }
4713 }
4714 
4715 void record_synch_enable() {
4716   // start collecting trace data, if not already doing so
4717   if (!record_synch_enabled)  record_synch_count = 0;
4718   record_synch_enabled = true;
4719 }
4720 
4721 void record_synch_disable() {
4722   // stop collecting trace data
4723   record_synch_enabled = false;
4724 }
4725 
4726 #endif // INTERPOSE_ON_SYSTEM_SYNCH_FUNCTIONS
4727 #endif // PRODUCT
4728 
4729 const intptr_t thr_time_off  = (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
4730 const intptr_t thr_time_size = (intptr_t)(&((prusage_t *)(NULL))->pr_ttime) -
4731                                (intptr_t)(&((prusage_t *)(NULL))->pr_utime);
4732 
4733 
4734 // JVMTI & JVM monitoring and management support
4735 // The thread_cpu_time() and current_thread_cpu_time() are only
4736 // supported if is_thread_cpu_time_supported() returns true.
4737 // They are not supported on Solaris T1.
4738 
4739 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4740 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4741 // of a thread.
4742 //
4743 // current_thread_cpu_time() and thread_cpu_time(Thread *)
4744 // returns the fast estimate available on the platform.
4745 
4746 // hrtime_t gethrvtime() return value includes
4747 // user time but does not include system time
4748 jlong os::current_thread_cpu_time() {
4749   return (jlong) gethrvtime();
4750 }
4751 
4752 jlong os::thread_cpu_time(Thread *thread) {
4753   // return user level CPU time only to be consistent with
4754   // what current_thread_cpu_time returns.
4755   // thread_cpu_time_info() must be changed if this changes
4756   return os::thread_cpu_time(thread, false /* user time only */);
4757 }
4758 
4759 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4760   if (user_sys_cpu_time) {
4761     return os::thread_cpu_time(Thread::current(), user_sys_cpu_time);
4762   } else {
4763     return os::current_thread_cpu_time();
4764   }
4765 }
4766 
4767 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4768   char proc_name[64];
4769   int count;
4770   prusage_t prusage;
4771   jlong lwp_time;
4772   int fd;
4773 
4774   sprintf(proc_name, "/proc/%d/lwp/%d/lwpusage",
4775           getpid(),
4776           thread->osthread()->lwp_id());
4777   fd = ::open(proc_name, O_RDONLY);
4778   if (fd == -1) return -1;
4779 
4780   do {
4781     count = ::pread(fd,
4782                     (void *)&prusage.pr_utime,
4783                     thr_time_size,
4784                     thr_time_off);
4785   } while (count < 0 && errno == EINTR);
4786   ::close(fd);
4787   if (count < 0) return -1;
4788 
4789   if (user_sys_cpu_time) {
4790     // user + system CPU time
4791     lwp_time = (((jlong)prusage.pr_stime.tv_sec +
4792                  (jlong)prusage.pr_utime.tv_sec) * (jlong)1000000000) +
4793                  (jlong)prusage.pr_stime.tv_nsec +
4794                  (jlong)prusage.pr_utime.tv_nsec;
4795   } else {
4796     // user level CPU time only
4797     lwp_time = ((jlong)prusage.pr_utime.tv_sec * (jlong)1000000000) +
4798                 (jlong)prusage.pr_utime.tv_nsec;
4799   }
4800 
4801   return (lwp_time);
4802 }
4803 
4804 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4805   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
4806   info_ptr->may_skip_backward = false;    // elapsed time not wall time
4807   info_ptr->may_skip_forward = false;     // elapsed time not wall time
4808   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
4809 }
4810 
4811 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4812   info_ptr->max_value = ALL_64_BITS;      // will not wrap in less than 64 bits
4813   info_ptr->may_skip_backward = false;    // elapsed time not wall time
4814   info_ptr->may_skip_forward = false;     // elapsed time not wall time
4815   info_ptr->kind = JVMTI_TIMER_USER_CPU;  // only user time is returned
4816 }
4817 
4818 bool os::is_thread_cpu_time_supported() {
4819   return true;
4820 }
4821 
4822 // System loadavg support.  Returns -1 if load average cannot be obtained.
4823 // Return the load average for our processor set if the primitive exists
4824 // (Solaris 9 and later).  Otherwise just return system wide loadavg.
4825 int os::loadavg(double loadavg[], int nelem) {
4826   if (pset_getloadavg_ptr != NULL) {
4827     return (*pset_getloadavg_ptr)(PS_MYID, loadavg, nelem);
4828   } else {
4829     return ::getloadavg(loadavg, nelem);
4830   }
4831 }
4832 
4833 //---------------------------------------------------------------------------------
4834 
4835 bool os::find(address addr, outputStream* st) {
4836   Dl_info dlinfo;
4837   memset(&dlinfo, 0, sizeof(dlinfo));
4838   if (dladdr(addr, &dlinfo) != 0) {
4839     st->print(PTR_FORMAT ": ", addr);
4840     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4841       st->print("%s+%#lx", dlinfo.dli_sname, addr-(intptr_t)dlinfo.dli_saddr);
4842     } else if (dlinfo.dli_fbase != NULL) {
4843       st->print("<offset %#lx>", addr-(intptr_t)dlinfo.dli_fbase);
4844     } else {
4845       st->print("<absolute address>");
4846     }
4847     if (dlinfo.dli_fname != NULL) {
4848       st->print(" in %s", dlinfo.dli_fname);
4849     }
4850     if (dlinfo.dli_fbase != NULL) {
4851       st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4852     }
4853     st->cr();
4854 
4855     if (Verbose) {
4856       // decode some bytes around the PC
4857       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4858       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4859       address       lowest = (address) dlinfo.dli_sname;
4860       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
4861       if (begin < lowest)  begin = lowest;
4862       Dl_info dlinfo2;
4863       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4864           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4865         end = (address) dlinfo2.dli_saddr;
4866       }
4867       Disassembler::decode(begin, end, st);
4868     }
4869     return true;
4870   }
4871   return false;
4872 }
4873 
4874 // Following function has been added to support HotSparc's libjvm.so running
4875 // under Solaris production JDK 1.2.2 / 1.3.0.  These came from
4876 // src/solaris/hpi/native_threads in the EVM codebase.
4877 //
4878 // NOTE: This is no longer needed in the 1.3.1 and 1.4 production release
4879 // libraries and should thus be removed. We will leave it behind for a while
4880 // until we no longer want to able to run on top of 1.3.0 Solaris production
4881 // JDK. See 4341971.
4882 
4883 #define STACK_SLACK 0x800
4884 
4885 extern "C" {
4886   intptr_t sysThreadAvailableStackWithSlack() {
4887     stack_t st;
4888     intptr_t retval, stack_top;
4889     retval = thr_stksegment(&st);
4890     assert(retval == 0, "incorrect return value from thr_stksegment");
4891     assert((address)&st < (address)st.ss_sp, "Invalid stack base returned");
4892     assert((address)&st > (address)st.ss_sp-st.ss_size, "Invalid stack size returned");
4893     stack_top=(intptr_t)st.ss_sp-st.ss_size;
4894     return ((intptr_t)&stack_top - stack_top - STACK_SLACK);
4895   }
4896 }
4897 
4898 // ObjectMonitor park-unpark infrastructure ...
4899 //
4900 // We implement Solaris and Linux PlatformEvents with the
4901 // obvious condvar-mutex-flag triple.
4902 // Another alternative that works quite well is pipes:
4903 // Each PlatformEvent consists of a pipe-pair.
4904 // The thread associated with the PlatformEvent
4905 // calls park(), which reads from the input end of the pipe.
4906 // Unpark() writes into the other end of the pipe.
4907 // The write-side of the pipe must be set NDELAY.
4908 // Unfortunately pipes consume a large # of handles.
4909 // Native solaris lwp_park() and lwp_unpark() work nicely, too.
4910 // Using pipes for the 1st few threads might be workable, however.
4911 //
4912 // park() is permitted to return spuriously.
4913 // Callers of park() should wrap the call to park() in
4914 // an appropriate loop.  A litmus test for the correct
4915 // usage of park is the following: if park() were modified
4916 // to immediately return 0 your code should still work,
4917 // albeit degenerating to a spin loop.
4918 //
4919 // In a sense, park()-unpark() just provides more polite spinning
4920 // and polling with the key difference over naive spinning being
4921 // that a parked thread needs to be explicitly unparked() in order
4922 // to wake up and to poll the underlying condition.
4923 //
4924 // Assumption:
4925 //    Only one parker can exist on an event, which is why we allocate
4926 //    them per-thread. Multiple unparkers can coexist.
4927 //
4928 // _Event transitions in park()
4929 //   -1 => -1 : illegal
4930 //    1 =>  0 : pass - return immediately
4931 //    0 => -1 : block; then set _Event to 0 before returning
4932 //
4933 // _Event transitions in unpark()
4934 //    0 => 1 : just return
4935 //    1 => 1 : just return
4936 //   -1 => either 0 or 1; must signal target thread
4937 //         That is, we can safely transition _Event from -1 to either
4938 //         0 or 1.
4939 //
4940 // _Event serves as a restricted-range semaphore.
4941 //   -1 : thread is blocked, i.e. there is a waiter
4942 //    0 : neutral: thread is running or ready,
4943 //        could have been signaled after a wait started
4944 //    1 : signaled - thread is running or ready
4945 //
4946 // Another possible encoding of _Event would be with
4947 // explicit "PARKED" == 01b and "SIGNALED" == 10b bits.
4948 //
4949 // TODO-FIXME: add DTRACE probes for:
4950 // 1.   Tx parks
4951 // 2.   Ty unparks Tx
4952 // 3.   Tx resumes from park
4953 
4954 
4955 // value determined through experimentation
4956 #define ROUNDINGFIX 11
4957 
4958 // utility to compute the abstime argument to timedwait.
4959 // TODO-FIXME: switch from compute_abstime() to unpackTime().
4960 
4961 static timestruc_t* compute_abstime(timestruc_t* abstime, jlong millis) {
4962   // millis is the relative timeout time
4963   // abstime will be the absolute timeout time
4964   if (millis < 0)  millis = 0;
4965   struct timeval now;
4966   int status = gettimeofday(&now, NULL);
4967   assert(status == 0, "gettimeofday");
4968   jlong seconds = millis / 1000;
4969   jlong max_wait_period;
4970 
4971   if (UseLWPSynchronization) {
4972     // forward port of fix for 4275818 (not sleeping long enough)
4973     // There was a bug in Solaris 6, 7 and pre-patch 5 of 8 where
4974     // _lwp_cond_timedwait() used a round_down algorithm rather
4975     // than a round_up. For millis less than our roundfactor
4976     // it rounded down to 0 which doesn't meet the spec.
4977     // For millis > roundfactor we may return a bit sooner, but
4978     // since we can not accurately identify the patch level and
4979     // this has already been fixed in Solaris 9 and 8 we will
4980     // leave it alone rather than always rounding down.
4981 
4982     if (millis > 0 && millis < ROUNDINGFIX) millis = ROUNDINGFIX;
4983     // It appears that when we go directly through Solaris _lwp_cond_timedwait()
4984     // the acceptable max time threshold is smaller than for libthread on 2.5.1 and 2.6
4985     max_wait_period = 21000000;
4986   } else {
4987     max_wait_period = 50000000;
4988   }
4989   millis %= 1000;
4990   if (seconds > max_wait_period) {      // see man cond_timedwait(3T)
4991     seconds = max_wait_period;
4992   }
4993   abstime->tv_sec = now.tv_sec  + seconds;
4994   long       usec = now.tv_usec + millis * 1000;
4995   if (usec >= 1000000) {
4996     abstime->tv_sec += 1;
4997     usec -= 1000000;
4998   }
4999   abstime->tv_nsec = usec * 1000;
5000   return abstime;
5001 }
5002 
5003 void os::PlatformEvent::park() {           // AKA: down()
5004   // Transitions for _Event:
5005   //   -1 => -1 : illegal
5006   //    1 =>  0 : pass - return immediately
5007   //    0 => -1 : block; then set _Event to 0 before returning
5008 
5009   // Invariant: Only the thread associated with the Event/PlatformEvent
5010   // may call park().
5011   assert(_nParked == 0, "invariant");
5012 
5013   int v;
5014   for (;;) {
5015     v = _Event;
5016     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5017   }
5018   guarantee(v >= 0, "invariant");
5019   if (v == 0) {
5020     // Do this the hard way by blocking ...
5021     // See http://monaco.sfbay/detail.jsf?cr=5094058.
5022     int status = os::Solaris::mutex_lock(_mutex);
5023     assert_status(status == 0, status, "mutex_lock");
5024     guarantee(_nParked == 0, "invariant");
5025     ++_nParked;
5026     while (_Event < 0) {
5027       // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5028       // Treat this the same as if the wait was interrupted
5029       // With usr/lib/lwp going to kernel, always handle ETIME
5030       status = os::Solaris::cond_wait(_cond, _mutex);
5031       if (status == ETIME) status = EINTR;
5032       assert_status(status == 0 || status == EINTR, status, "cond_wait");
5033     }
5034     --_nParked;
5035     _Event = 0;
5036     status = os::Solaris::mutex_unlock(_mutex);
5037     assert_status(status == 0, status, "mutex_unlock");
5038     // Paranoia to ensure our locked and lock-free paths interact
5039     // correctly with each other.
5040     OrderAccess::fence();
5041   }
5042 }
5043 
5044 int os::PlatformEvent::park(jlong millis) {
5045   // Transitions for _Event:
5046   //   -1 => -1 : illegal
5047   //    1 =>  0 : pass - return immediately
5048   //    0 => -1 : block; then set _Event to 0 before returning
5049 
5050   guarantee(_nParked == 0, "invariant");
5051   int v;
5052   for (;;) {
5053     v = _Event;
5054     if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5055   }
5056   guarantee(v >= 0, "invariant");
5057   if (v != 0) return OS_OK;
5058 
5059   int ret = OS_TIMEOUT;
5060   timestruc_t abst;
5061   compute_abstime(&abst, millis);
5062 
5063   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5064   int status = os::Solaris::mutex_lock(_mutex);
5065   assert_status(status == 0, status, "mutex_lock");
5066   guarantee(_nParked == 0, "invariant");
5067   ++_nParked;
5068   while (_Event < 0) {
5069     int status = os::Solaris::cond_timedwait(_cond, _mutex, &abst);
5070     assert_status(status == 0 || status == EINTR ||
5071                   status == ETIME || status == ETIMEDOUT,
5072                   status, "cond_timedwait");
5073     if (!FilterSpuriousWakeups) break;                // previous semantics
5074     if (status == ETIME || status == ETIMEDOUT) break;
5075     // We consume and ignore EINTR and spurious wakeups.
5076   }
5077   --_nParked;
5078   if (_Event >= 0) ret = OS_OK;
5079   _Event = 0;
5080   status = os::Solaris::mutex_unlock(_mutex);
5081   assert_status(status == 0, status, "mutex_unlock");
5082   // Paranoia to ensure our locked and lock-free paths interact
5083   // correctly with each other.
5084   OrderAccess::fence();
5085   return ret;
5086 }
5087 
5088 void os::PlatformEvent::unpark() {
5089   // Transitions for _Event:
5090   //    0 => 1 : just return
5091   //    1 => 1 : just return
5092   //   -1 => either 0 or 1; must signal target thread
5093   //         That is, we can safely transition _Event from -1 to either
5094   //         0 or 1.
5095   // See also: "Semaphores in Plan 9" by Mullender & Cox
5096   //
5097   // Note: Forcing a transition from "-1" to "1" on an unpark() means
5098   // that it will take two back-to-back park() calls for the owning
5099   // thread to block. This has the benefit of forcing a spurious return
5100   // from the first park() call after an unpark() call which will help
5101   // shake out uses of park() and unpark() without condition variables.
5102 
5103   if (Atomic::xchg(1, &_Event) >= 0) return;
5104 
5105   // If the thread associated with the event was parked, wake it.
5106   // Wait for the thread assoc with the PlatformEvent to vacate.
5107   int status = os::Solaris::mutex_lock(_mutex);
5108   assert_status(status == 0, status, "mutex_lock");
5109   int AnyWaiters = _nParked;
5110   status = os::Solaris::mutex_unlock(_mutex);
5111   assert_status(status == 0, status, "mutex_unlock");
5112   guarantee(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5113   if (AnyWaiters != 0) {
5114     // Note that we signal() *after* dropping the lock for "immortal" Events.
5115     // This is safe and avoids a common class of  futile wakeups.  In rare
5116     // circumstances this can cause a thread to return prematurely from
5117     // cond_{timed}wait() but the spurious wakeup is benign and the victim
5118     // will simply re-test the condition and re-park itself.
5119     // This provides particular benefit if the underlying platform does not
5120     // provide wait morphing.
5121     status = os::Solaris::cond_signal(_cond);
5122     assert_status(status == 0, status, "cond_signal");
5123   }
5124 }
5125 
5126 // JSR166
5127 // -------------------------------------------------------
5128 
5129 // The solaris and linux implementations of park/unpark are fairly
5130 // conservative for now, but can be improved. They currently use a
5131 // mutex/condvar pair, plus _counter.
5132 // Park decrements _counter if > 0, else does a condvar wait.  Unpark
5133 // sets count to 1 and signals condvar.  Only one thread ever waits
5134 // on the condvar. Contention seen when trying to park implies that someone
5135 // is unparking you, so don't wait. And spurious returns are fine, so there
5136 // is no need to track notifications.
5137 
5138 #define MAX_SECS 100000000
5139 
5140 // This code is common to linux and solaris and will be moved to a
5141 // common place in dolphin.
5142 //
5143 // The passed in time value is either a relative time in nanoseconds
5144 // or an absolute time in milliseconds. Either way it has to be unpacked
5145 // into suitable seconds and nanoseconds components and stored in the
5146 // given timespec structure.
5147 // Given time is a 64-bit value and the time_t used in the timespec is only
5148 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5149 // overflow if times way in the future are given. Further on Solaris versions
5150 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5151 // number of seconds, in abstime, is less than current_time  + 100,000,000.
5152 // As it will be 28 years before "now + 100000000" will overflow we can
5153 // ignore overflow and just impose a hard-limit on seconds using the value
5154 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5155 // years from "now".
5156 //
5157 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5158   assert(time > 0, "convertTime");
5159 
5160   struct timeval now;
5161   int status = gettimeofday(&now, NULL);
5162   assert(status == 0, "gettimeofday");
5163 
5164   time_t max_secs = now.tv_sec + MAX_SECS;
5165 
5166   if (isAbsolute) {
5167     jlong secs = time / 1000;
5168     if (secs > max_secs) {
5169       absTime->tv_sec = max_secs;
5170     } else {
5171       absTime->tv_sec = secs;
5172     }
5173     absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5174   } else {
5175     jlong secs = time / NANOSECS_PER_SEC;
5176     if (secs >= MAX_SECS) {
5177       absTime->tv_sec = max_secs;
5178       absTime->tv_nsec = 0;
5179     } else {
5180       absTime->tv_sec = now.tv_sec + secs;
5181       absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5182       if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5183         absTime->tv_nsec -= NANOSECS_PER_SEC;
5184         ++absTime->tv_sec; // note: this must be <= max_secs
5185       }
5186     }
5187   }
5188   assert(absTime->tv_sec >= 0, "tv_sec < 0");
5189   assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5190   assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5191   assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5192 }
5193 
5194 void Parker::park(bool isAbsolute, jlong time) {
5195   // Ideally we'd do something useful while spinning, such
5196   // as calling unpackTime().
5197 
5198   // Optional fast-path check:
5199   // Return immediately if a permit is available.
5200   // We depend on Atomic::xchg() having full barrier semantics
5201   // since we are doing a lock-free update to _counter.
5202   if (Atomic::xchg(0, &_counter) > 0) return;
5203 
5204   // Optional fast-exit: Check interrupt before trying to wait
5205   Thread* thread = Thread::current();
5206   assert(thread->is_Java_thread(), "Must be JavaThread");
5207   JavaThread *jt = (JavaThread *)thread;
5208   if (Thread::is_interrupted(thread, false)) {
5209     return;
5210   }
5211 
5212   // First, demultiplex/decode time arguments
5213   timespec absTime;
5214   if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5215     return;
5216   }
5217   if (time > 0) {
5218     // Warning: this code might be exposed to the old Solaris time
5219     // round-down bugs.  Grep "roundingFix" for details.
5220     unpackTime(&absTime, isAbsolute, time);
5221   }
5222 
5223   // Enter safepoint region
5224   // Beware of deadlocks such as 6317397.
5225   // The per-thread Parker:: _mutex is a classic leaf-lock.
5226   // In particular a thread must never block on the Threads_lock while
5227   // holding the Parker:: mutex.  If safepoints are pending both the
5228   // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5229   ThreadBlockInVM tbivm(jt);
5230 
5231   // Don't wait if cannot get lock since interference arises from
5232   // unblocking.  Also. check interrupt before trying wait
5233   if (Thread::is_interrupted(thread, false) ||
5234       os::Solaris::mutex_trylock(_mutex) != 0) {
5235     return;
5236   }
5237 
5238   int status;
5239 
5240   if (_counter > 0)  { // no wait needed
5241     _counter = 0;
5242     status = os::Solaris::mutex_unlock(_mutex);
5243     assert(status == 0, "invariant");
5244     // Paranoia to ensure our locked and lock-free paths interact
5245     // correctly with each other and Java-level accesses.
5246     OrderAccess::fence();
5247     return;
5248   }
5249 
5250   OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5251   jt->set_suspend_equivalent();
5252   // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5253 
5254   // Do this the hard way by blocking ...
5255   // See http://monaco.sfbay/detail.jsf?cr=5094058.
5256   if (time == 0) {
5257     status = os::Solaris::cond_wait(_cond, _mutex);
5258   } else {
5259     status = os::Solaris::cond_timedwait (_cond, _mutex, &absTime);
5260   }
5261   // Note that an untimed cond_wait() can sometimes return ETIME on older
5262   // versions of the Solaris.
5263   assert_status(status == 0 || status == EINTR ||
5264                 status == ETIME || status == ETIMEDOUT,
5265                 status, "cond_timedwait");
5266 
5267   _counter = 0;
5268   status = os::Solaris::mutex_unlock(_mutex);
5269   assert_status(status == 0, status, "mutex_unlock");
5270   // Paranoia to ensure our locked and lock-free paths interact
5271   // correctly with each other and Java-level accesses.
5272   OrderAccess::fence();
5273 
5274   // If externally suspended while waiting, re-suspend
5275   if (jt->handle_special_suspend_equivalent_condition()) {
5276     jt->java_suspend_self();
5277   }
5278 }
5279 
5280 void Parker::unpark() {
5281   int status = os::Solaris::mutex_lock(_mutex);
5282   assert(status == 0, "invariant");
5283   const int s = _counter;
5284   _counter = 1;
5285   status = os::Solaris::mutex_unlock(_mutex);
5286   assert(status == 0, "invariant");
5287 
5288   if (s < 1) {
5289     status = os::Solaris::cond_signal(_cond);
5290     assert(status == 0, "invariant");
5291   }
5292 }
5293 
5294 extern char** environ;
5295 
5296 // Run the specified command in a separate process. Return its exit value,
5297 // or -1 on failure (e.g. can't fork a new process).
5298 // Unlike system(), this function can be called from signal handler. It
5299 // doesn't block SIGINT et al.
5300 int os::fork_and_exec(char* cmd) {
5301   char * argv[4];
5302   argv[0] = (char *)"sh";
5303   argv[1] = (char *)"-c";
5304   argv[2] = cmd;
5305   argv[3] = NULL;
5306 
5307   // fork is async-safe, fork1 is not so can't use in signal handler
5308   pid_t pid;
5309   Thread* t = Thread::current_or_null_safe();
5310   if (t != NULL && t->is_inside_signal_handler()) {
5311     pid = fork();
5312   } else {
5313     pid = fork1();
5314   }
5315 
5316   if (pid < 0) {
5317     // fork failed
5318     warning("fork failed: %s", os::strerror(errno));
5319     return -1;
5320 
5321   } else if (pid == 0) {
5322     // child process
5323 
5324     // try to be consistent with system(), which uses "/usr/bin/sh" on Solaris
5325     execve("/usr/bin/sh", argv, environ);
5326 
5327     // execve failed
5328     _exit(-1);
5329 
5330   } else  {
5331     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5332     // care about the actual exit code, for now.
5333 
5334     int status;
5335 
5336     // Wait for the child process to exit.  This returns immediately if
5337     // the child has already exited. */
5338     while (waitpid(pid, &status, 0) < 0) {
5339       switch (errno) {
5340       case ECHILD: return 0;
5341       case EINTR: break;
5342       default: return -1;
5343       }
5344     }
5345 
5346     if (WIFEXITED(status)) {
5347       // The child exited normally; get its exit code.
5348       return WEXITSTATUS(status);
5349     } else if (WIFSIGNALED(status)) {
5350       // The child exited because of a signal
5351       // The best value to return is 0x80 + signal number,
5352       // because that is what all Unix shells do, and because
5353       // it allows callers to distinguish between process exit and
5354       // process death by signal.
5355       return 0x80 + WTERMSIG(status);
5356     } else {
5357       // Unknown exit code; pass it through
5358       return status;
5359     }
5360   }
5361 }
5362 
5363 // is_headless_jre()
5364 //
5365 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5366 // in order to report if we are running in a headless jre
5367 //
5368 // Since JDK8 xawt/libmawt.so was moved into the same directory
5369 // as libawt.so, and renamed libawt_xawt.so
5370 //
5371 bool os::is_headless_jre() {
5372   struct stat statbuf;
5373   char buf[MAXPATHLEN];
5374   char libmawtpath[MAXPATHLEN];
5375   const char *xawtstr  = "/xawt/libmawt.so";
5376   const char *new_xawtstr = "/libawt_xawt.so";
5377   char *p;
5378 
5379   // Get path to libjvm.so
5380   os::jvm_path(buf, sizeof(buf));
5381 
5382   // Get rid of libjvm.so
5383   p = strrchr(buf, '/');
5384   if (p == NULL) {
5385     return false;
5386   } else {
5387     *p = '\0';
5388   }
5389 
5390   // Get rid of client or server
5391   p = strrchr(buf, '/');
5392   if (p == NULL) {
5393     return false;
5394   } else {
5395     *p = '\0';
5396   }
5397 
5398   // check xawt/libmawt.so
5399   strcpy(libmawtpath, buf);
5400   strcat(libmawtpath, xawtstr);
5401   if (::stat(libmawtpath, &statbuf) == 0) return false;
5402 
5403   // check libawt_xawt.so
5404   strcpy(libmawtpath, buf);
5405   strcat(libmawtpath, new_xawtstr);
5406   if (::stat(libmawtpath, &statbuf) == 0) return false;
5407 
5408   return true;
5409 }
5410 
5411 size_t os::write(int fd, const void *buf, unsigned int nBytes) {
5412   size_t res;
5413   RESTARTABLE((size_t) ::write(fd, buf, (size_t) nBytes), res);
5414   return res;
5415 }
5416 
5417 int os::close(int fd) {
5418   return ::close(fd);
5419 }
5420 
5421 int os::socket_close(int fd) {
5422   return ::close(fd);
5423 }
5424 
5425 int os::recv(int fd, char* buf, size_t nBytes, uint flags) {
5426   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5427          "Assumed _thread_in_native");
5428   RESTARTABLE_RETURN_INT((int)::recv(fd, buf, nBytes, flags));
5429 }
5430 
5431 int os::send(int fd, char* buf, size_t nBytes, uint flags) {
5432   assert(((JavaThread*)Thread::current())->thread_state() == _thread_in_native,
5433          "Assumed _thread_in_native");
5434   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5435 }
5436 
5437 int os::raw_send(int fd, char* buf, size_t nBytes, uint flags) {
5438   RESTARTABLE_RETURN_INT((int)::send(fd, buf, nBytes, flags));
5439 }
5440 
5441 // As both poll and select can be interrupted by signals, we have to be
5442 // prepared to restart the system call after updating the timeout, unless
5443 // a poll() is done with timeout == -1, in which case we repeat with this
5444 // "wait forever" value.
5445 
5446 int os::connect(int fd, struct sockaddr *him, socklen_t len) {
5447   int _result;
5448   _result = ::connect(fd, him, len);
5449 
5450   // On Solaris, when a connect() call is interrupted, the connection
5451   // can be established asynchronously (see 6343810). Subsequent calls
5452   // to connect() must check the errno value which has the semantic
5453   // described below (copied from the connect() man page). Handling
5454   // of asynchronously established connections is required for both
5455   // blocking and non-blocking sockets.
5456   //     EINTR            The  connection  attempt  was   interrupted
5457   //                      before  any data arrived by the delivery of
5458   //                      a signal. The connection, however, will  be
5459   //                      established asynchronously.
5460   //
5461   //     EINPROGRESS      The socket is non-blocking, and the connec-
5462   //                      tion  cannot  be completed immediately.
5463   //
5464   //     EALREADY         The socket is non-blocking,  and a previous
5465   //                      connection  attempt  has  not yet been com-
5466   //                      pleted.
5467   //
5468   //     EISCONN          The socket is already connected.
5469   if (_result == OS_ERR && errno == EINTR) {
5470     // restarting a connect() changes its errno semantics
5471     RESTARTABLE(::connect(fd, him, len), _result);
5472     // undo these changes
5473     if (_result == OS_ERR) {
5474       if (errno == EALREADY) {
5475         errno = EINPROGRESS; // fall through
5476       } else if (errno == EISCONN) {
5477         errno = 0;
5478         return OS_OK;
5479       }
5480     }
5481   }
5482   return _result;
5483 }
5484 
5485 // Get the default path to the core file
5486 // Returns the length of the string
5487 int os::get_core_path(char* buffer, size_t bufferSize) {
5488   const char* p = get_current_directory(buffer, bufferSize);
5489 
5490   if (p == NULL) {
5491     assert(p != NULL, "failed to get current directory");
5492     return 0;
5493   }
5494 
5495   jio_snprintf(buffer, bufferSize, "%s/core or core.%d",
5496                                               p, current_process_id());
5497 
5498   return strlen(buffer);
5499 }
5500 
5501 #ifndef PRODUCT
5502 void TestReserveMemorySpecial_test() {
5503   // No tests available for this platform
5504 }
5505 #endif
5506 
5507 bool os::start_debugging(char *buf, int buflen) {
5508   int len = (int)strlen(buf);
5509   char *p = &buf[len];
5510 
5511   jio_snprintf(p, buflen-len,
5512                "\n\n"
5513                "Do you want to debug the problem?\n\n"
5514                "To debug, run 'dbx - %d'; then switch to thread " INTX_FORMAT "\n"
5515                "Enter 'yes' to launch dbx automatically (PATH must include dbx)\n"
5516                "Otherwise, press RETURN to abort...",
5517                os::current_process_id(), os::current_thread_id());
5518 
5519   bool yes = os::message_box("Unexpected Error", buf);
5520 
5521   if (yes) {
5522     // yes, user asked VM to launch debugger
5523     jio_snprintf(buf, sizeof(buf), "dbx - %d", os::current_process_id());
5524 
5525     os::fork_and_exec(buf);
5526     yes = false;
5527   }
5528   return yes;
5529 }